Section 3 Classical Studies

### **Chapter 5**

## Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations into English

*Leonor Scliar-Cabral*

### **Abstract**

The aim of this chapter is a discussion about the criteria used on the translation of Brazilian Portuguese poetry into American English. I thus exemplify it with the translations of Quevedo's sonnets, namely the sonnet "*Desde a torre*," "From the tower." As one goes through Quevedo's sonnets, one can notice the recurrence of the semantic fields "fire" and "prison." The first appears until fatigue in the opposites game within the Petrarquean pairs tradition like fire/snow~water. In order to enjoy the multiple readings that Quevedo offers, it is necessary to delve into the disappointments of which he was a victim and the seventeenth-century Spain collapse, a tumultuous scene of the Baroque: Spain was ravaged by the Thirty Years' War and defeated by Holland and France. I based the criteria upon translation theories, more specifically upon poetic translation. I describe in detail the genre sonnet, particularly its metrics and rhymes, and the difficulties of translating them. I end the chapter with a microanalysis of the poem "*Desde a torre*" translation into American English.

**Keywords:** poetry translation, Brazilian Portuguese, American English, Quevedo's sonnet, Baroque

#### **1. Introduction**

In order to enjoy the multiple readings that Quevedo (Francisco Gómez de Quevedo y Villegas, 1580–1645) offers, it is necessary to delve into the disappointments of which he was a victim and the seventeenth-century Spain collapse, a tumultuous scene of the Baroque: Spain was ravaged by the Thirty Years**'** War and defeated by Holland and France. It is against this backdrop that the Spanish literary Baroque thrives, whose main characteristics are cultism and conceptism, with Quevedo being an adept of the latter, unlike Gôngora.

The sonnet's author considered the most beautiful in the Spanish language by Alonso [1], "*Cerrar podrá mis ojos la postrera/ sombra*," "My eyes will close the ultimate, and last/shadow," was exiled three times due to palace intrigues, and it is none other than one of these that inspired the title of the sonnet "From the tower."

Mentioning the topos versed by him, from the satirical to the amorous, it is undoubtedly when dealing with death and life brevity that conceptism is handled most vigorously: neostoicism bases his ideas, in the wake of Seneca**'**s thought, revisited by Christianity. Conceptism must be understood as a style that can be either epigrammatic or vigorous, whose concise words or phrases shine with the force and speed of lightning. There is a lot of wordplay and lively metaphors. But Quevedo's characteristic touch, as Alonso points out [1] in several steps, is his affectivity: the "eruptive vitality," the "forging fury," and the "condensation."

Quevedo is, perhaps, among the poets whom I translated [2], the most Hispanic without, however, ceasing to be the most universal and, at the same time, historical: imprisoned emotions that finally explode, the satirization of human villainies, the panel of a Spain in decadence are portrayed with the formal domain of someone who masters Petrarque**'**s technique, Mannerism and Baroque, although he does not allow himself to be subjugated by them.

The imprisoned passions idea has its source more in Quevedo**'**s existential vision than perhaps in experiences with real women, but he had a very concrete prison experience three times. In the last arrest, at the end of his life, from 1639 to 1643, he remained under house arrest, under suspicion of political satire against King Filipe IV (poor whoever fell under Quevedo**'**s sights!).

#### **2. Quevedo's topos**

With a vast work, whose chronology has not yet been established with precision, the theme that stands out in the sonnets I have translated [2] is the explosion of affections.

As one goes through Quevedo's sonnets, one can notice the recurrence of the semantic fields "fire" and "prison." The first appears until fatigue in the opposites game within the Petrarquean pairs tradition like fire/snow ~ water, which in the collection I translated [2] is illustrated in the sonnets: "*Cerrar podrá mis ojos la postrera*," "My eyes will close the ultimate, and last"; "*En los claustros del alma la herid*a," "In the cloisters of soul the suffered wound."

But the obsession with the inner fire also appears in his lexical and metaphorical preferences by terms such as "marrows" (in "My eyes will close the ultimate, and last, "verse 11); (in "In the cloisters of soul the suffered wound," verse 4).

A Bosch of literature, Quevedo hyperbolically emphasizes the human grotesque, illustrated below with the sonnet that satirizes female old age.1 It's exactly on the satirical side where Quevedo presents one of his great originalities, drinking in us popular records, he populates his verses with Spanish words and expressions that were in people's mouths, such as *antaño* (yesteryear), *morenos* (dark skinned), and *présteme un ochavo* (lend me an ochavo), contradicting the prevailing trend for cultisms.

On the other hand, a true panel of Hispanic decadence, where his preference for the oppressed is clear, parades before us, this time reminiscent of Goya. And not least, he was the victim of inquisitorial processes, for mocking hell itself. His appeals for death to welcome him reach the maximum lyricism in the sonnet "Already great and formidable sounds" (**Figure 1**):

Many times, almost translating the authors within this cultural tradition (which does not diminish its literary value, as the great contributions of Plautus and Terence are examples), Quevedo presents originalities, as I have already pointed out: his pagan-Christian

<sup>1</sup> A topic dear to Greek and Latin epigramists, cf. the Palatine Anthology, book XI, [3]**.**

*Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

**Figure 1.**

*"Se um feliz descanso" into English.*

syncretism, whose inspiration can be traced back to the stoicism of Seneca's or to Marcus Aurelius' theme of life fleetingness, never obscures the strongest side of his literary production, which is to express the human passions and the formal resources core: they are not an end in themselves, but create tensions, which mirror the most intimate energies, starting with syntactic parallels and forming a type of correlation.

#### **3. Is poetic translation possible?**

The debate on poetry translatability is inserted, firstly, in violating the original text of translation in general. According to Paes [4] "From its earliest beginnings, translation can thus be seen as an effort to make the positive impulse of language towards openness and communication prevail over the negativity of its other impulse towards closure and communication exclusion" [...] "the translator is a builder of linguistic bridges that link the closed-in-itself idiomatic islands to one another."

Paes [4] is, therefore, among those, who admit translatability, citing Walter Benjamin [5], when referring to the restoration in human memory of the "traces of the Adamic language erased by Babel curse": the existence of a Universal human cognition, despite linguistic diversity, would support translatability, since in the translation process the translator starts with the text interpretation in the source language and concludes his work, providing the receiver with the same interpretation in the target text.

According to Gadamer [6] "The meaning must be maintained, but because it has to be understood in a new linguistic world, it will come into force in a new way. That is why all translation is interpretation," an idea shared by Theodor [7]. Further on, Gadamer [6] emphasizes the role of recreation in the translation process.

In the opposite direction, there are several translation theorists, whose epistemological basis is in line with the linguistic relativism of Sapir [8] and Whorf [9], that linguistic systems, conditioned spatially and temporally, shape the way of perceiving reality and, therefore, of thinking and this would make it impossible to translate different worldviews.

In a period (nineteenth century beginning), when translation still focuses heavily on literal, word-for-word translation, Wilhelm von Humboldt [10] states in the preface to his translation of Aeschylus' *Agamemnon*: "Analysis and experience confirm that, which has been observed more than once: apart from expressions, which designate only physical objects, no word in one language is perfectly the same as one in another."

Coseriu [11] criticizes the word-for-word approach to literal translation, asserting that "only texts are translated," taking into account, in addition, the various extralinguistic contexts involved: empirical, historical, cultural, etc.

Paes [4], however, while accepting translatability, as I commented above, with regard to poetry, states: "In no other sector of translation activity is the struggle against the isolationist impulse of language more fierce than in poetry translation" [...] "This insofar as, to mean the most with the least, it makes use of all language

expressive resources, largely peculiar, exclusive to it" [...] "To the despair of the poetry translator, the phonic extract is always the most idiosyncratic" [...] "the poetry translator is sometimes led to more radical solutions than those commonly adopted in the routine of translating. This can creatively influence the very use of the language into which he translates, enriching it with resources of 'strangeness'." However, Humboldt [10] warns: "Insofar as it makes the strangeness feel rather than the strangeness, translation has achieved its highest ends, however, the moment the strangeness itself appears, perhaps even obscuring the strange, the translator reveals not to be up to his original."

Humboldt [10] also recognizes the impossibility of translating poetry, when referring to *Agamemnon*. The uniqueness of each message is linked to the concept of energy: "each one must carry within himself his work and that of all the others, this emergence should resemble the emergence of an ideal figure in the artist's imagination. Nor can it be extracted from something real, it arises by a pure spirit energy, and rather out of nothing, but from that moment on it starts to live and be real and lasting."

A very strong line of poetic translation defends the idea that only a poet is competent to do it. The arguments rest on the eminently creative character of poetic translation with all its implications for the permissible audacity of violating the original text according to Paz [12] and Islam [13].

Eco [14] points out that the full meaning of the poetic text also rests on the phonic suggestions, as well as on the rhyme and other sound aspects, such as paronomasia, alluding to Jakobson's analysis of the example "I like Ike." It is necessary to emphasize that Jakobson [15] was not referring strictly to poetry, but rather to language poetic function, which he defined as having the message itself as its focus (also for Silvestri [16]. Thus, the poetic function may be the predominant one in the girl's choice of the adjective "horrible," in the expression "The horrible Harry," instead of "dreadful," "terrible," "frightful," or "disgusting," because it is paronomasia. Eco [14] concludes that the notion of propositional content only applies to utterances that unambiguously represent states of the world and never to those, in which the focus is the message itself, as occurs when the poetic function predominates. The expressive substance, therefore, for Eco [14] is "fundamental with regard to phono-stylistic topics and discursive rhythm in general." But it is precisely on the expressive substance that many deny the possibility of translation, as for instance, Silvestri [16].

The difficulties in translating poetry and artistic prose had already been pointed out by Schleiermacher [17], for whom "the language musical element that manifests itself in rhythm and intonation is also of special and superior importance" [...] "Therefore, what strikes the sensitive reader of the original work in this respect as characteristic, intentional and effective in tone and mood terms, and as decisive for the speech rhythmic or musical accompaniment must also be transmitted by the translator."

In an intermediate position, there are theorists such as Dollerup [18], who claim that it is not possible to find equivalences in poetic translation, but rather compensatory strategies, such as the use of traits considered poetic by a given culture.

The concept of seeking equivalences in the target language has been one of the most debated by translation theorists. Schleiermacher [17] in his classic essay on the different methods of translation, when commenting on literary and scientific translation difficulties, points out the difference between "equivalent expression," which is impossible and "closer" one, which is possible. On the other hand, Coseriu [11], based on his tripartite model of Meaning 1 (the meaning of the source language), Meaning 2 (the meaning in the target language), and Designation, that is, the referent, only

#### *Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

admits equivalences in the designation. For him, "in the translation process itself – it is about finding meanings in the target language that can designate the same thing." In fact, Coseriu [11] argues that linguistic contents are in an irrational relationship, which he calls incommensurability: in his view, it is the main problem of translation theory, that is, "the problem of identical designation," with different linguistic means" [11]. However, the identical designation collides with the differences between cultures, spatially and temporally distanced, as ironically emphasizes Nietzsche [19]: "There are translations with honest intentions that are almost falsifications and involuntary vulgarizations of the original, only because their cheerful and courageous time could not be translated – time which, providentially omitted, overcomes all that is dangerous in things and words."

When dealing with the issue of incommensurability, Eco [14] asserts that, although it is a fact, it cannot be equated with incomparability. The possibility of comparing broadens the translatability horizons, as long as the translator is not restricted to linguistic limits, making use of intertextuality, narrativity, and psychological aspects.

Eco [14] approaches the issue from another perspective, by proposing the existence of a deep meaning in the texts, to which the translator has access through interpretation, superficializing it, when translating it into the target language, even at the cost of lexical and referential violations.

When stating that translations "are not about linguistic types, but about linguistic occurrences" and, when recalling the Chomskyan dichotomy of deep and superficial structures, several implications follow: on the deep meaning, superficialized in the utterances, several factors intervene, which Coseriu [11] calls contexts.

A first factor that Eco [14] mentions is the total work context, exemplified by *al di là dela siepe* translation of his book *Il pendolo di Foucault* (Foucault's Pendulum). It is a literary quotation from the sonnet *L*'*infinito* by Giacomo Leopardi, like hundreds of others that fill the book, to highlight the style of the three characters. The English translator Weaver did the best translation, with an "explicit reference to Keats": "we glimpsed endless vistas." Another factor pointed out by Eco is the cultural differences that interfere in the translation of even banal situations. The French expression "*Cherchez la femme,*" cannot be translated as "Look for the woman," because its meaning is "where there is trouble, the cause is in some woman" [14].

#### **4. The need for poetic translation**

Despite the immense difficulties that poetic translation faces at, it is necessary. Several theorists point out that it allows those, who do not know the source language, to have access to new art forms and new worldviews, in addition to enriching the target language with new forms of expression. Humboldt [10] cites the example of German meter enrichment, after the Greek classics' translation by Klopstock. I cite the contribution of Plautus and Terence, when they translated the Greek New Comedy and, thus, established the Latin meter.

The enrichment of the target language and culture is not only provided by poetic translation: the impact on the German language and culture with the Bible translation into the vernacular by Luther was pointed out by Humboldt [10]; Eco [14] mentions the radical change in the style of the French philosophical genre, after Heidegger's translation; the same happened with the Italian narratives, after the American authors' translation, before the Second World War. As a result, according to Eco, the

axis of the debate shifts from the relationship between source and target language to the "effect that the translated text has on the target culture."

#### **5. Difficulties in poetic translation**

Translating poetry is a challenge, as it implies finding the stylistic resources in the target language, that is, finding, among the parameters that define a given genre, those that will cause on the reader a similar effect to those felt by the readers of the source text.

Borges [20] confessed that if he could translate music from English or German into Spanish, he would be a great poet. Paraphrasing, it can be said that only those who deal with musical effects can translate poetry and this is the biggest challenge, that is, reconciling the meanings intended by the author with aesthetic solutions, finding their aesthetic availability in the target language, or, as stated by Dámaso Alonso [1], "in poetry there is always a motivated link between signifier and signified."

Given the difficulties in finding equivalences in the target language, particularly with regard to poetic translation, theorists have suggested compensatory strategies to obtain similar effects during reception. Gadamer [6] explains the sometime painful path, in which the translator seeks a middle ground to reconcile the interpretation he makes of the source language text in order to make it available in the target language, through back and forth.

Such a middle ground is also suggested by Goethe [21] when he explains that there are two maxims in translation, the first one being to bring the culture from which the text to be translated comes to the culture of the target language, with an incorporation, while the other consists of an inverse trajectory, that is, subjection "to the conditions, their way of speaking, their particularities." When in doubt, Goethe suggests the first option.

Schleiermacher [17] conceives the ideal that translation would be, either enabling the reader to meet the author or conversely, for the author to meet the reader. In the second hypothesis, the author would be able to write the work in the target language, which would be the perfect translation. It is surprising that in 1813 Schleiermacher already mentioned "We could outline rules for each of the two methods, taking into account the different genres of discourse" [17].

To get an idea of the difficulties in translating a Petrarquean sonnet, such as the sonnet "From the tower," I will begin by examining the questions of meter and rhyme. To solve them, strategies are used that involve inversions, additions, omissions, and substitutions of items, preferably purely grammatical ones, although additions and losses of items and/or expressions that refer to external meaning are also used; major displacements generally imply morphosyntactic changes.

In the sphere of meaning, resources are used, such as the use of items belonging to the same semantic field and figures as metaphors. Sometimes, some more audacious resources are used, interpreted as the space of freedom left to the poetic creation of the translator.

Such difficulties increase when the differences between the source language and the target language are compared, as it occurs in the translation from Portuguese into English. The following are systematized and further elaborated on:

a.American English (AE) has almost 13 oral vowels, including lax and tense ones, unlike Brazilian Portuguese (BP), which has seven oral vowels and five nasal

*Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

ones. The contrast between syllables in words, in BP, is marked by the stressed and unstressed syllables, while in AE, it is between lax and tense syllables, which influences the rhythm in both languages;


#### **6. More serious rhyming problems**

The most serious rhyming issues in translation from BP into AE stem from differences between the two phonological systems, particularly with regard to the vowel system, since, as seen, BP has five nasal vowels and AE does not have any although vowel letters in the sonnets are the same, as, a, e, i, o, u, since both written languages adopt the same script, the Latin one. There are also differences regarding nasal consonants that, in English, can appear in syllabic locking, unlike Portuguese, which compensates for this gap with nasalized diphthongs, which are absent in English.

Failure to observe these differences caused several stumbling blocks in the translation by Horta, Vianna and Rivera [22], as shown in the following example, in Cristóbal de Castillejo's sonnet "*Sonho*" ("Dreaming") the translators rhyme "cousa" (/´kowza/), with "saborosa" (/sabo´rɔza/), and "formosa" (/foR´mɔza/).

This constitutes a trap for translating the rhymes, because, although the letters are the same, the grapheme values depend on the phonological system of each language.

Meter problems, particularly in heroic verses, that is, decasyllables with ictus on the sixth foot, stem from many factors, firstly, morphosyntactic issues, such as: the AE does not admit the subject postponed to the verb in declarative affirmative sentences, nor the postposition of the adjective to the noun, as well as it does not admit the ellitic or null subject; secondly, prosody questions, since the contrast between syllables in words, in BP, is marked by the stressed and unstressed syllables, while in AE, it is between lax and tense syllables. Therefore, as I stated above, several strategies are necessary that I will specify below:

a.inversions, additions, omissions, and substitutions

Example of inversion, addition, and substitution in the translation of the sonnet "From the tower," I found in the fourth verse of the first stanza (**Figure 2**):

b.another strategy I will comment on deals with semantics and figures, particularly metaphors. It must be assumed that approximate meanings can be found in the source language and in the target one but never the senses, since their respective

#### **Figure 2.**

*"e os mortos eu escuto" into English.*


**Figure 3.** *"Alma que todo um deus" into English.*

> readers do not share the same sociocultural experiences, with spatiotemporal coordinates being implicit. Such a difference already allows the literary translator a wide margin of creativity, without too much discrepancy from the original meanings. Therefore, he will use resources such as replacing words with items belonging to the same semantic field, paraphrases, and tropos. Sometimes the translator eliminates items not essential to the central idea.

c.The last strategy deals with syntactic parallelism, exemplified in the bellow sonnet "The eyes will close the last." There are three subjects of split clauses, asyndetically coordinated, each followed by an adjective clause (in the first tercet), creating a tension that is solved in the last stanza, with the implicit resumption of the three vocatives, first, in an asyndetically coordinated clause, then in a coordinated adversative clause, and finally, in a clause in which only the predicate is coordinated by the adversative (**Figure 3**):

#### **7. The sonnet**

The sonnet is a poetic form that dates from the thirteenth century, having originated in Sicily, at the court of Frederick II, and coexists with Provence poetry, but the poet Guittone D'Arezzo was the one who established its form. The first great classics of the genre are Petrarch and Dante. The first bequeathed its name to one of the most widespread sonnet forms, the Petrarquean sonnet, adopted by Camões. Shakespeare composed his sonnets in the English form, with three quartets and a couplet. In Portugal, Sá de Miranda introduced the sonnet and several other poetic genres, constituting the so-called *dolce stil nuovo*, after having made a trip to Italy in the first quarter of the fifteenth century.

The structure adopted in Quevedo's sonnets is Petrarchean. Despite being criticized, the sonnet continues to be widely practiced. The great Brazilian poet Francisco

#### *Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

Carvalho [23] asks: "After all, if the sonnet is really out of fashion, outdated in form and content, why do so many people continue to write it with such conviction?"

I will begin by explaining the units of which the verse is constituted in the sonnet. They are the feet (probably a metonymy derived from the rhythm marking), and they have their origin in the Greek lyric, later adopted by the Romans. Plautus and Terence were the Latin meter creators, when they translated the New Greek Comedy. We can therefore conclude that the translator's role goes far beyond translation. In both Greek and Latin, the phonological accent is based on duration, as it is the case in English, and therefore, so it is the meter. Thus, what counts is how the long and short syllables are articulated among themselves, forming the units we call feet, depending on how many they are and the position they occupy. It is a binary system: the limit of possible combinations is determined by our processing capacity. In languages whose phonological accent is based on stress, as is the case of Portuguese and Spanish, an equivalence is made while translating poetry.

See which feet usually occur, with their respective names and formalizations, in the decasyllable (ten-syllable verse), used in the sonnet "From the tower":

Iambic (I), whose formalization is - /, meaning a short or unstressed syllable, followed by a long or stressed syllable, as in the beginning of the first verse of the second stanza of "From the tower": "If not always implied, they are revealed."

Trocheu (T), whose formalization is/ -, meaning a long or stressed syllable, followed by a short or unstressed syllable, as in the last three words of the second verse: "*com poucos, porém doutos livros juntos*,", in English, "with few, although be learned readings jointly." In Brazilian Portuguese, the trochaic form predominates in nouns, adjectives, and verbs.

Pirríqueo (P), whose formalization is - -, meaning a short or unstressed syllable, followed by another short or unstressed syllable, as in the beginning of the third verse of the second stanza: "*e em silenciosos, músicos conjuntos*", in English, "in a counterpoint music silently."

Espondeu (E), whose formalization is//, meaning long or stressed syllable, followed by long or stressed syllable. The single example in "From the tower" is "Dom Ioseph."

Anapesto (A), whose formalization is - - /, meaning short or unstressed syllable, followed by another short or unstressed syllable, and another long or stressed syllable, like the first feet of the first line of the sonnet: "Retirado na paz destes desertos," in English, "In the peace of these deserts, my heart broken."

Dactylic (D), whose formalization is/- -, meaning a long or stressed syllable, followed by two short or unstressed syllables, as in the first feet of "*destes desertos*" in English, "peace of these deserts," from the first verse of the first stanza: "*Retirado na paz destes desertos,*" in English, "In the peace of these deserts, my heart broken."

#### **8. Micro-analysis of "From the tower" translation**

See **Figures 4** and **5**.

**Figure 4.** *"Da Torre" (BP version) into English.*

**Figure 5.** *Desde la torre (Quevedo's original).* *Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

#### **9. Stylistic analysis and the search for aesthetic effects in translation**

In the BP version first word, "*Retirado,"* we are faced at its polysemy, as it can have the meaning of someone who seeks peace in a retreat, as well as that of the banishment imposed on Quevedo: I lean toward the second interpretation, and I used the metaphor "my heart broken," image of suffering. In the second verse, the author alludes to the confiscation of his books, when arrested in 1639. The adjective "*doutos,*" in the second verse, translated into "learned" is repeated in the last verse of the first tercet: the same positions are maintained in the English translation.

In the first stanza, there were only difficulties in the fourth verse, as "mortos" does not rhyme with "deserts": I made syntactic transpositions, dragging "*os mortos*" to the beginning of the fourth verse (the same with "the deads" in the English translation) and drawing "*despertos*,'" which was at the end of the second stanza, to crown the end of the first one (the same with "vigilant, open'" in the English translation).

The effect of these changes was the topicalization of "the deads." It is worth emphasizing that this verse is one of the most beautiful, created by Quevedo. Not only does the author use synesthesia and an oxymoron, but, as it is so often the case in literature, he anticipates neuroscience scientific findings, by proving that, after the written word recognition by the reading neurons, its acoustic images are internally heard. A similar anticipation of the unconscious was recorded by Virgil [24], in the *Aeneid*, when he uses the expression "*alta in mente."*

In the second stanza, I made some lexical changes in the last word of the third verse, due to the rhymes. I believe they did not cause major semantic violations. The first verse is bimembre, and this resource, very Petrarquean, was kept. The third verse, however, was problematic, as "counterpoints" does not end in "ly." To keep the meter, I reversed the order and translated "*e em silenciosos, músicos conjuntos"* into *"'*in a counterpoint music silently" and the oxymoron was kept. The word "counterpoint," in addition to be in Quevedo's original version, connoting antithesis, was one of the Baroque musical forms, cultivated by J. S. Bach. It was possible to keep in the translation the anastrophe in the fourth verse, another Baroque characteristic.

It was not possible to keep always the same lexicon to obtain the rhymes, because the word endings and its suffixes are phonologically distinct, in the two languages. Sometimes, I used a syntactic resource, transforming an adjective into an adverb, like "jointly" for "*juntos*" (first stanza, second verse) or "silently" for "*silenciosos*" (second stanza, third verse); other times I added words, reinforcing the original semantic field as the word "coyly" from "in conversation with deads*,* I live coyly," for "*vivo em conversação com os defuntos,*" (first stanza, third verse), or as the word "vigilant" from "I hear the deads, the eyes vigilant, open" for "*e os mortos eu escuto, olhos despertos*." The dearest topos to Spanish poetry, the brevity of life, opens the last tercet, closed in the last two lines by a correlation.

#### **10. Computational semantics contributions**

The methodology of the present research resulted primarily from manually constructed sources that could have benefited from computational supervised learning, since the manual specification and the automatic acquisition of knowledge are solidly interrelated; however, the automatic induction of semantic information is guided and dependent on manually specified information.

One of the issues most worked on by Computational Semantics is the automatic disambiguation of the meaning of words: "Senseval was the first open communitybased evaluation exercise for Word Sense Disambiguation programs" [25], essential to literary translation, in particular, poetry. From the polysemic nature of words [26], ambiguity multiplies in the situational context of statements, in which the enunciator, although using the same *significant* [27], denotes a new meaning to the referent, thanks to time, which is never repeated, with all the repercussions on his/her prior knowledge.

Computational Semantics seeks to solve this paradox, making use of databases and computational statistics, through which it was possible to make available the contextual patterns where the most frequent meanings of a significant data occur [28].

The following questions remain for literary translation:


#### **11. Conclusions**

To understand the strategies used for translating Quevedo's sonnet "From the tower," I framed them in a theoretical discussion: translation is a creative process. It allows interpreting, according to Gadamer [6] and Theodor [7], as the source text, capturing the linguistic meanings emerged from the author's intratextual and intertextual relationships with extra-linguistic information, coming from his experiences, the moment and the historical-cultural space, according to Coseriu [11]. For this reason, I began the chapter with a historical context and with an evocation of the remarkable episodes in Quevedo's life that served as a background for the topos in his work and, in particular, for the sonnet that is the main focus of the paper. Only in this way we will be able to capture the allusion to prison and isolation, in the first verse of "From the tower" and the solace he finds with the books company.

However, when it comes to poetry, in addition to the general translation characterization, this genre requires that aesthetic aspects be prioritized, as mentioned by Jakobson [15], Paes [4], Eco [14], Silvestri [16], and Schleiermacher [17], such as meter, rhythm, rhymes, alliteration, paronomasia, and figures that relate to the signifier. Some authors even deny the possibility of translating poetry like Humboldt [10]. For this reason, a great deal of space was devoted to the examination of the sonnet form and rhymes.

In this chapter, I presented the strategies I used in the translation of the sonnet "From the tower" to achieve aesthetic effects similar to those of the Brazilian Portuguese version, given the phonological and morphological differences between it and American English, so, it was quite difficult to benefit from Computational Semantics supervised learning,

*Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

#### **Acknowledgements**

The author published another article in Portuguese, based on the same theoretical references, dealing with the difficulties of poetic translation of Quevedo's sonnets, however, from the original Spanish into Brazilian Portuguese, entitled: "OS MORTOS EU ESCUTO, OLHOS DESPERTOS: HOMENAGEM A MOACYR SCLIAR" (THE DEAD I LISTEN WITH MY OPEN EYES: HOMAGE TO MOACYR SCLIAR), in the journal Lingüística Vol. 32-1, June 2016: 79-93 ISSN 2079-312X online ISSN 1132-0214 printed DOI: 10.5935/2079-312X.20160005.

#### **Author details**

Leonor Scliar-Cabral Federal University of Santa Catarina (UFSC), Florianópolis, SC, Brazil

\*Address all correspondence to: ppgl@contato.ufsc.br; eonorsc20@gmail.com

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

### **References**

[1] Alonso D. Poesia espanhola, ensaio de métodos e limites estilísticos. 3rd ed. Brasília: National Institute Library (INL)/ Culture Education Ministry (MEC); 1960. p. 471. (pp. 395, 428/9, 376, 24)

[2] Scliar-Cabral L. Poesia espanhola do século de ouro. 1st ed. Letras Contemporâneas: Florianópolis; 1998. p. 104

[3] Grotius H, Dübner F, Cougny E. Epigrammatum Anthologia Pallatina. 1st ed. Paris: Editore Ambrosio Firmin; 1871. p. 150

[4] Paes JP. Gaveta de tradutor. 1st ed. São Paulo: Letras Contemporâneas; 1996. p. 160

[5] Arendt H. Walter Benjamin and the Task of the Translator. 1st ed. Schocken: Rio de Janeiro; 1968. p. 278. (pp. 79-80)

[6] Gadamer HG. Hermenêutica da obra de arte. 1st ed. São Paulo: Martins Fontes; 2010. p. 506. (pp. 241, 243, 237)

[7] Theodor E. Tradução – Ofício e Arte. 3rd ed. São Paulo: Cultrix; 1986. p. 150

[8] Sapir E. The emergence of the concept of personality in a study of cultures. In: Mandelbaum DG, editor. Selected Writings of Edward Sapir in Language Culture Personality. Berkeley, Los Angeles: University of California; 2020. pp. 194-208. DOI: 10.1525/9780520311893-010

[9] Whorf BL. In: Carroll JB, Levinson SC, Lee P, editors. Language, thought, and Reality: Selected Writings of Benjamin Lee Whorf. 2nd ed. Cambridge, MA: MIT Press; 2012. p. 457

[10] Humboldt W. Introdução ao Agamênon. In: Heidermann W, editor. Antologia Bilíngue. Clássicos da Teoria da Tradução. Alemão-Português. 2nd ed. Florianópolis: UFSC/Núcleo de Pesquisas em Literatura e Tradução; 2010. pp. 104-117

[11] Coseriu E. O falso e o verdadeiro na Teoria da Tradução. In: Heidermann W, editor. Antologia Bilíngue. Clássicos da Teoria da Tradução. Alemão-Português. 2nd ed. Florianópolis: UFSC/Núcleo de Pesquisas em Literatura e Tradução; 2010. pp. 250-289

[12] Paz O. Translation: Literature and letters. In: Schulte R, Biguenet J, editors. Theories of Translation: An Anthology of Essays from Dryden to Derrida. Chicago: The University of Chicago Press; 1992. pp. 152-162

[13] Manzoorul IS. Translatability and untranslatability. The case of Tagore's poems. Perspectives Studies in Translatology. 1995;**3**(1):55-65. DOI: 10.1080/0907676X.1995.9961248

[14] Eco U. Experiences in Translation. 1st ed. Toronto, Buffalo: University of Toronto Press; 2001. p. 135. (pp. 12, 13, 88, 14, 16, 21, 105)

[15] Jakobson R. Concluding statement: Linguistics and poetics. In: Sebeok T, editor. Style in Language. 2nd ed. Cambridge, MA: The MIT Press; 1964. pp. 350-377

[16] Silvestri L. Nugae – Teoría de la traducción. 1st ed. Buenos Aires: Simurg; 2003. p. 80. (pp. 16, 12)

[17] Schleiermacher E. Sobre os diferentes métodos de tradução. In: Heidermann W, editor. Antologia Bilíngue. Clássicos da Teoria da Tradução. Alemão-Português. 2nd ed. Florianópolis: UFSC/Núcleo

*Perspective Chapter: Difficulties for Translating Quevedo's Sonnets from Portuguese Translations... DOI: http://dx.doi.org/10.5772/intechopen.109699*

de Pesquisas em Literatura e Tradução; 2010. pp. 38-101

[18] Dollerup C. Poetry: Theory, practice and feed-back to theory. Tradterm. 1997;**4**(2):129-147. DOI: 10.11606/ issn.2317-9511.tradterm.1997.49856

[19] Nietzsche. Sobre o problema da tradução. In: Heidermann W, editor. Antologia Bilíngue. Clássicos da Teoria da Tradução. Alemão-Português. 2nd ed. Florianópolis: UFSC/Núcleo de Pesquisas em Literatura e Tradução; 2010. pp. 195-199

[20] Borges JL. Obras completas (v. 2). 1st ed. São Paulo: Globo; 1999. p. 565. (p. 258)

[21] Goethe JW. Três trechos sobre tradução. In: Heidermann W, editor. Antologia Bilíngue. Clássicos da Teoria da Tradução. Alemão-Português. 2nd ed. Florianópolis: UFSC/Núcleo de Pesquisas em Literatura e Tradução; 2010. pp. 29-35

[22] Horta AB, Vianna FM, Rivera JJ. Poetas do século de ouro espanhol. Poetas del siglo de oro español. 1st ed. Brasília: Embajada de España; 2000. p. 343. (p. 57)

[23] Carvalho F. O sonho é nossa chama. 1st ed. João Pessoa: Expressão Gráfica Editora; 2010. p. 98

[24] Virgil L' Énéide. Nouvelle Édition Revue et Augmentée avec Introduction, Notes, Appendices et Index par Maurice Rat. 2nd ed. Paris: Garnier; 1955. p. 439. (p. 4, line 26)

[25] Kilgarriff A, Palmer M. Introduction, Special Issue on SENSEVAL: Evaluating Word Sense Disambiguation Programs. Computers and the Humanities. 2000;**34**:1-2

[26] Ravin Y, Leacock C. Polysemy: Theoretical and Computational Approaches. Oxford: OUP; 2000. p. 242 [27] De Saussure F. Cours de Linguistique Générale. Édition critique preparée par Mauro TD. Paris: Payot; 1972. p. 510

[28] Schütze R. Disambiguation and connectionism. In: Ravin Y, Leacock C, editors. Polysemy: Theoretical and Computational Approaches. Oxford: OUP; 2000. pp. 205-219

Section 4

## Semantic Analysis in Computing

## Toward Lightweight Cryptography: A Survey

*Mohammed Abujoodeh, Liana Tamimi and Radwan Tahboub*

#### **Abstract**

The main problem in Internet of Things (IoT) security is how to find lightweight cryptosystems that are suitable for devices with limited capabilities. In this paper, a comprehensive literature survey that discusses the most prominent encryption algorithms used in device security in general and IoT devices in specific has been conducted. Many studies related to this field have been discussed to identify the most technical requirements of lightweight encryption systems to be compatible with variances in IoT devices. Also, we explored the results of security and performance of the AES algorithm in an attempt to study the algorithm performance for keeping an acceptable security level which makes it more adaptable to IoT devices as a lightweight encryption system.

**Keywords:** cyber, information security, IOT security, networks, cryptography, AES, lightweight cryptography

#### **1. Introduction**

An information system is a set of interconnected components that collect, process, store, and transfer information. These components include the physical and software components and the communication networks [1].

Networks enable communications between many devices by connecting them and enabling the most reliable possible connection. Moreover, networks are subject to many attacks due to users and their different directions. Here, the challenge lies in maintaining the security of these networks with their resources and data while maintaining high performance [1–3].

In its simplest sense, Internet of Things (IoT) is a system of various intelligent devices known in our daily lives. These things link and communicate between them and ensure data transfer between them independently via the network without human interaction, a self-control system [4–6]. Smart Cities played an essential role in highlighting IoT. Smart Cities express the concept that depends on the city's technology, as these cities are linked to each other electronically. Information is collected continuously from sensors, monitoring, and computers covering the whole city [5, 6]. "Thing" term can be a sensor network, as safe houses, or in general, any device that can take an IP address and can interact through a network [5].

Security plays an essential role in judging IoT applications strengths. Users wish to have secure IoT software hat is secure in all respects. IoT application's security includes a secure transfer of data, protection from eavesdropping, and unauthorized

access. The system security has become one of the essential critical requirements of the system's core functions [2, 3]. Furthermore, the security aims to achieve what is known as the Confidentiality, Integrity, and Availability (CIA) triad. Finally, one of the most critical security goals is to control access through the Authentication, Authorization, and Accounting (AAA) framework [3, 7].

IoT causes a massive increase in the volume of data. Securing such enormous amount of data requires special efforts. Several technologies may serve this purpose. But the devices used in the combination of smart cities and the IoT vary among themselves in capabilities. Moreover, most of these devices have limited specifications and restrictions [5, 6]. Hence the need to find new technologies that work on these limited capabilities and achieve an acceptable degree of security. Furthermore, since the capabilities are limited, these technologies should be lightweight and rely on simple operations without consuming energy, storage, and processing capacity.

The rest of the chapter organized as follows. We provide clarifications for some concepts related to this field considering cryptography and Lightweight Cryptography (LWC) areas in Section 2. Section 3 provides some researches related to LWC, and Advanced Standard Algorithm (AES). In Section 4, recommendations and findings are discussed. Finally, we conclude the paper and present a vision for future work in Section 5.

#### **2. Background**

This section introduces the concepts of IoT, Cryptography, and LWC in Subsections 2.1, 2.2, and 2.3 respectively.

#### **2.1 Internet of thing**

IoT today is a hot topic in research. The importance of IoT comes because of keeping pace with the variables of life that call us to exploit everything new in technology, such as computers, cars, TV, refrigerators, and washing machines [5, 6]. **Figure 1** shows IoT Reference Architecture. The figure shows that the IoT system consists of data collector's devices as a sensor used to get the data and data analyzer device like a mobile phone used for data processing to make a decision. These two subsystems communicate and transfer data via a network [5, 6, 8].

**Figure 1.** *IoT reference architecture [8].*

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

IoT has dramatically helped to increase the efficiency of work and operations. It relies on a system of self-interaction that means reducing the waiting time for response. As a result, performance gains, and therefore the number of completed processes increases, giving users access to the best possible user services, enhancing the work's actual value [5, 8]. In general, IoT provides a wide range of benefits at the enterprise and individual levels.

**Figure 2** presents the concept of IoT. There are many valuable and significant applications for IoT, such as Safe Houses, Health Care, and Farming systems.

Despite the significant benefits of IoT, the IoT suffers from a lack of standardization and is vulnerable to cyber-attacks, data theft, data fraud, botnet attacks, and physical compromises. The reason for this is that the IoT differs from traditional networks. There are two types of IoT devices: those rich in resources, like computers, and those with limited resources, like sensors. The real challenges are in the second type, which has low memory and computing power, short battery life, and Low bandwidth to connect [6, 8]. So, we should be careful about security and privacy [8]. Hence, the challenge is how to design an IoT system efficiently and securely.

#### **2.2 Cryptography**

Cryptography is a way to protect data and communications by ensuring that those not authorized to access sent data cannot read and process it [3, 9]. The goals of the encryption process revolve around guaranteeing each of the following [3, 9, 10]:


**Figure 2.** *Concept of IoT [4].*


**Figure 3** summarizes CIA triad and AAA framework.

#### *2.2.1 Cryptography algorithms*

In cryptography science, encryption transforms original messages (Plain Text) to non-readable data (Cipher Text) using an encryption algorithm. This Cipher Text cannot give anyone any information about the Plain Text except those with the encryption key [9, 12–17]. Therefore, we can perform a simple encryption example by replacing every character in the plain text with its next character in aliphatic order.

```
P = "Thesis".
Alg.: substitution Pi = Pi + 1.
C = "uiftjt".
```
There are two main types of encryptions: Asymmetric cipher, and Symmetric cipher, as shown in **Figure 4** [9, 12–17].

#### **2.3 Asymmetric cipher**

Asymmetric cipher is conjointly referred to as public-key cryptography. Associate cryptography technique uses a mix of public key and private key. The sender has the receiver's public key, whereas the private key is not known. The receiver ought to produce his try of the general public and private key, publish his public key while not

**Figure 3.** *CIA triad and AAA framework [10, 11].*

#### **Figure 4.** *Encryption models [9].*

considering its security. The private key should be a procedure not possible to seek out through the general public key.

Uneven cryptography is employed in authentication and digital signatures. A signed message with the sender's private key proves the sender's identity, and anyone who has that sender's public key can verify it. Thus, the receiver may ensure that the message has not been changed or replaced by the other one that confirms the sender's identity [9, 15, 16].

Rivest*–*Shamir*–*Adleman (RSA) algorithm one of the most popular and widely used asymmetric encryption algorithms. It was developed in 1977 by Ron Rivest, Adi Shamir, Leonard Adleman and took its name from them. Besides Encryption, Digital signatures and key exchange are possible using RSA [17, 18].

RSA gained its strength by relying on parsing large integers in the formation of keys. First, two prime numbers are manipulated to create the user's public and private keys. Then, the message is encrypted using the recipient's public key and decrypted exclusively with the recipient's private key. **Figure 5** shows the RSA Process [17].

Although RSA is the most popular and secure asymmetric encryption algorithm in terms of key difficulty, it takes a long time to encrypt and decrypt. Besides, a security flaw appears that encrypting the same message again produces the same encrypted message [18].

ElGamal is an asymmetric cipher based on Diffie–Hellman key exchange. This algorithm gains its strength through the difficulty of finding discrete logarithms. For example, even though we know *G<sup>x</sup>* and *G<sup>y</sup>* , it is challenging to find *Gxy*. This algorithm consists of key generation, encryption, and decryption processes. **Figure 6** shows each of them [19–21].

**Figure 5.** *RSA process [18].*

Elliptic Curve Cryptography (ECC) it uses the mathematics on elliptic curves. ECC is widely used due to its high security and small size. The difficulty in cracking the elliptic curves that underpin key strength has made ECC more secured and considered as the next generation of RSA [21, 22].

The main difference between ECC and RSA is the strength of the key. A 160-bit key in ECC is equivalent in power to a 1024-bit key in RSA. Considering that there is no linear relationship, doubling the size of the RSA key does not mean that we need to double the size of the RSA key. ECC is characterized by the speed of obtaining the keys and less memory to store them. On the other hand, a challenge for ECC is that it cannot be implemented as efficiently as RSA [22]. **Figure 7** presents the ECC.

Digital Signature Algorithm (DSA) is an algorithm that uses discrete logarithms and standard bases to introduce and validate the notion of a digital signature. Compared to RSA, DSA provides faster key generation.

As a result, it is slower in the encryption process, but it offers better results in the decryption process. DSA is mainly used to verify the sender's identity of a message since it bears his signature, which cannot be duplicated [23]. **Figure 8** presents the DSA mechanism.

**Figure 7.** *ECC basics [22].*

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

#### **2.4 Asymmetric cipher summary**

This section discussed various Asymmetric Cipher algorithms such as RSA, ElGamal, DSA, and ECC. **Table 1** highlight the most comparison points between them. This type of algorithm offers high strength in terms of security, it requires a large amount of processing, which means low performance and draining resources. Therefore, based on the preceding, these algorithms are not compatible with the discrepancy in the capabilities of IoT devices and therefore cannot be used in building security systems in term of encryption. Hence, we find that symmetric encryption is more suitable for such systems. However, this does not detract from its value, as it cannot be dispensed with in verification, key exchange, and signature operations.


**Table 1.** *Asymmetric ciphers comparison.*

#### **2.5 Symmetric cipher**

Each sender and receiver share the same secret key in this kind of Encryption. Hence, it uses within the encryption and decryption processes. However, symmetric Encryption has better speed but a lower security level than asymmetric [9, 12, 16, 24]. **Figure 9** shows the general structure of this encryption model. Symmetric ciphers can be used as a block cipher or stream cipher. We will discuss both types in detail in this section.

#### *2.5.1 Stream cipher*

In this type of encryption, the data are encrypted bit by bit. Because every encrypted bit is independent of other bits, diffusion and confusion properties are not achieved [9].

This encryption type mainly uses as simple as possible operators in this type of cipher. In most cases, it uses the XOR operation between the plaintext bits and the corresponding key bits. As a result, stream cipher throughput (speed of Encryption) is much *higher* than the block cipher but is considered less secure than Block Cipher [9, 12–16, 24].

Rivest Cipher 4 (RC4) is a stream cipher algorithm proposed by Ron Rivest in 1987. It later became a widely used algorithm from being a personal algorithm due to its speed and simplicity. RC4 has been frequently used to encrypt network traffic [25]. This algorithm uses byte-oriented operations with a variable key size. Simply put, RC4 relies on an XOR operation between each piece of plaintext with a small portion of the key to produce the ciphertext. And the decoding process is only a reflection of this process. However, with the development of computers, it became possible to break this algorithm easily. However, RC4 can be considered secure if the initial bytes of the key are ignored [25–27].

Salsa20 is a synchronous stream cipher suggested by Bernstein. The number 20 indicates the number of rounds, but this can be reduced to 12 or 8 as needed. *Salsa20* relies on simple operations such as rotation, addition, and XOR, making it a highspeed algorithm, which makes it secure against timing attacks [27, 28].

Sosemanuk is a synchronous stream cipher with variable key length. It has good properties of confusion and diffusion for a low cost. Furthermore, the Mux operation

**Figure 9.** *Simple symmetric model [9].*

is secure against algebraic and fast correlation attacks. Finally, Sosemanuk has good performance due to the internal static data [29].

**Table 2** provides a brief comparison of these algorithms, following our discussion and our review of their definitions and specifications.

From this comparison, we note that the RC4 algorithm is optimal for use, as it is more robust and available in more than one version to suit the system in which it will be used. However, in light of the fact that stream ciphers offer high speed and low security and the requirement for keys to be the same size as plaintext, none of these algorithms are suitable for use as a foundation for building an IoT system.

#### *2.5.2 Block cipher*

In Block Cipher, the plaintext is divided into blocks based on encryption algorithm structure [12]. This type of Encryption has an execution time slower than the stream cipher. So, the encryption throughput of stream cipher is much higher than the block cipher [9, 23]. In contrast, a block cipher provides better security than the stream cipher against some well-known attacks. Moreover, the essential properties of the secure ciphertext, which are the confusion and the diffusion properties, are included inside block ciphering algorithms. Based on these facts, we can nominate one block cipher algorithm to build our algorithm for the IoT after reviewing it and choosing the most appropriate based on its specification and results.

Data Encryption Standard (DES) is a symmetric encryption algorithm that uses a seemingly 64-bit key, of which 56 bits are used as the practical key over 16 rounds of the 48-bit subkeys, to encrypt data of a fixed length of 64 bits. The apparent key's remaining 8 bits are utilized to verify for parity. In decryption, the same process is employed in reverse [30, 31]. **Figure 10** shows an example of DES encryption.

Even though this algorithm has been widely adopted due to its speed and ease of use, it suffers from a serious security weakness in reality. The use of DES with a short key makes it very fragile, especially using a brute force attack, which is easy to use in this case. In addition, there are many attacks, such as Davie's attack and offensive Linear and differential cryptanalysis, which are theoretical attacks [30, 31].

An improved version of the encryption algorithm has been created to solve the security issues with DES. This method is as simple as applying the DES algorithm precisely three times. We now have three keys, each of which is 56 bits long. As a result, the implementation technique differed in the keys utilized. There were several versions because the relationship of the three keys affects the extent of the algorithm's power in the previously described. Triple DES (3DES), which used three distinct keys with a total of 156 actual bits, was thought to be very powerful [28]. However, 3DES will not be used by the end of 2023 as we move to more secure generations for encryption [32].


**Table 2.** *Stream cipher comparison.*

**Figure 10.** *DES algorithm [30].*

Blowfish is a symmetric cipher technique that uses a 64-bit block and a variablelength encryption key as needed. In terms of speed, Blowfish is a good algorithm, but the amount of security it provides varies depending on the length of the key employed. As a result, even though no genuine threats have been detected, it has gotten less attention than other algorithms [32, 33].

AES is one of the most famous and prominent symmetric encryption algorithms that has been introduced to be a quantum leap in this field. AES has outstanding performance and an excellent security level compared to its peers.

AES deals with data blocks with a fixed size of 128 bits in length, in addition to providing flexibility in choosing the size of the key according to the required degree of security. From here, it appears that AES has three versions according to the size of the key, namely AES-128, AES-192, and AES-256 with 10, 12, and 14 rounds, respectively. Each process uses several operations to encrypt a data block [34–36]. **Figure 11** represents the flow of the AES algorithm.

The working mechanism of AES is based on the use of the design principle known as the permutation and substitution network, and this mechanism is represented by using the following arithmetic operations:


The need for key expansion comes from the fact that each AES round needs a key of a specific length based on the criteria mentioned earlier. Therefore, when using AES-128, we need 11 keys depending on the number of rounds. Key derivation is done using the AES Key Schedule algorithm, which expands the key using a key schedule [34, 35].

AES distinguished itself from its peers in improving its performance for systems dealing with large amounts of data by integrating these steps and running them on a *Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

#### **Figure 11.** *AES algorithm [31].*

byte-oriented approach. This approach only converts its arithmetic operations into a series of look-up tables [35].

### **3. Modes of operation**

In Block Cipher, a fixed size block is handled at a time. Usually, the data size is much larger than the block size. Hence, the data is divided into a set of blocks. Each block is encrypted as one unit, the relationship, and dependency between encrypted blocks relaying on the encryption mode. Several modes have been developed to accommodate the variety of applications that will use Encryption. The process of selecting the required mode depends on many factors such as error propagation, the level of security, pre-processing, parallelization, and the speed of Encryption and decryption [12, 36]. These modes are as follows:


In general terms, without going into details of each mode. **Table 3** compares these modes.

After discussing the previous block cipher algorithms such as DES, 3DES, Blowfish, and AES, after reviewing the definition and specifications of each, **Table 4** provides a brief comparison of these algorithms.

From this comparison, we found that the stream has better performance and complexity, but it is not guaranteeing the diffusion, can be reversed easily, and it is providing less security. Because of that, we conclude that the block cipher is better solution since it provides more security in the case of text-based and image-based encryption.

#### **3.1 Symmetric cipher summary**

In this section, we summarize the symmetric cipher algorithms. **Table 5** compare stream and block cipher algorithms.

#### **3.2 Cryptography summary**

After discussing the cryptography algorithms and classifying them into Asymmetric and Symmetric, we reviewed their definition and specifications of each type. **Table 6** provides a brief comparison of these algorithms.


**Table 3.** *Encryption modes comparison.*

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*


#### **Table 4.**

*Block cipher comparison.*


**Table 5.** *Stream vs. block cipher.*

#### **3.3 Lightweight cryptography**

NIST defined LWC as a cryptosystem whose features have been optimized to meet the requirements of devices of varying specifications, especially resource-constrained devices [37]. From this definition, we conclude that all cryptography terms can be LWC if it is possible to legalize its need for resources to ensure the desired effect. Thus, asymmetric Encryption is an exception due to its complexity and demand for high resources. On the other hand, symmetric Encryption can be used in these systems if it is properly exploited.

Depending on the critical challenges mentioned before, we found that the LWC algorithm should use little memory and power and provide good performance while maintaining the required level of security [38]. Therefore, the factors of LWC requirements can be explained as follows [39]:



#### **Table 6.**

*Asymmetric cipher comparison.*


Many LWC algorithms provide different performance and security strengths. And after studying many related studies, we find that there have been some trends in relying on stream cipher due to its high efficiency in terms of performance. Still, most of the algorithms were based on block cipher since it offers better security but with a significant performance improvement [38]. We highlight some of these LWC algorithms in the following sections depending on its base as a stream or block.

#### **3.4 Stream LWC**

This section presents some LWC algorithms based on stream cipher methodologies. A4 is a very efficient lightweight stream cipher that uses LFSR and FCSR. The key feature of A4 is the ease of implementation and high security. In addition, A4 has proven itself in resistance to brute-force and algebraic attacks [39].

New Lightweight Stream Cipher (NLSC) is a chaos-based algorithm that uses an 80-bit secret key, two Nonlinear Feedback Shift Registers (NFCR), and three multiplexers. NFCR has good security, making it resistant to statistical attacks and providing good performance [38, 39].

#### **3.5 Block LWC**

This section presents some LWC algorithms based on block cipher methodologies.

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

PRESENT is an LCW algorithm that relies on Substitution-Permutation Network (SPN). It was suitable for limited hardware as it uses an 80-bit key. However, it was noted that it takes 32 rounds to encrypt 64 bits. Another version uses a 128-bit key, but it requires more computations [38].

GIFT is an enhanced PRESENT version; it uses a lighter S-Box with minimal rounds and a faster key scheduling algorithm. These properties enable it to provide more throughput. It is also available in more than one version depending on the required throughput. These versions are; GIFT-64 and GIFT-128. With a 64-bit block size that requires 28 rounds and a 128-bit block size that requires 40 rounds, respectively [38].

KATAN is an algorithm that outperforms PRESENT by saving 48% of the power. KATAN uses an 80-bit key and handles different text sizes 32, 48, and 64 bits. However, its downside is that it requires 254 rounds to complete this process [38].

The National Security Agency developed Simon as an improved algorithm that uses rounds cycles but uses a lot of arithmetic operations. It offers many different key sizes as 64-bit, 72-bit, 96-bit, 128-bit, 144-bit, 192-bit, and 256-bit that handle 32-bit, 48-bit, 64-bit, 96-bit, and 128-bit block size through 32, 36, 42, 44, 52, 54, 68, 69, and 72 rounds. While SPECK is the same as SIMON, it supports exact block sizes and keys, but 22, 23, 26-29, and 32-34 rounds [38].

RECTANGLE is a very LWC algorithm, which is different from PRESENT. It Relies on lighter SPN with 25 rounds. This reduced algorithm significantly speeded up the execution based on Bit-slice, as it relies on parallel swapping and replacement [40].

SIT is an algorithm that combines Feistel and SP network and takes five rounds to handle 64 bits of text with 64 bits of text as a key. It mainly consists of two parts: the first is for key expansion, and the second is for the encryption section. Key expansion


**Table 7.** *LWC summary.* relies on simple operations such as concatenation, shifting, addition, and XOR. As a result, this algorithm achieves high throughput and low power consumption [38].

#### **3.6 LWC summary**

After discussing various LWC algorithms such as LSC, A4, NLSC, PRESENT, GIFT, KATAN, SIMON, and SPECK, RECTANGLE, and SIT. **Table 7** highlight the most comparison points between them.

#### **4. Literature review**

This section discusses the most recent related research. After studying these researches, we categorized them into two groups. The first group, including [40–50], reviews LWC and defines its essential requirements. The second group discusses AES versions that are proposed to be compatible with LWC requirements [51–60].

#### **4.1 Lightweight cryptography related works**

This section summarizes some researches that introduce the concept of LWC in terms of terminology, requirements, and how to implement them in line with the available capabilities.

Manifavas et al. [40] discussed lightweight encryption algorithms, focusing on streaming encryption, which provides high performance with simple operations, making it suitable for the capabilities of IoT devices, especially when the text length is unknown or continuous. The results showed the superiority of symmetric encryption in performance. Still, most of the streaming algorithms were not secure, as after analyzing 31 algorithms, it was found that only 6 were secure.

Buchanan et al. [41] emphasized the IoT's security and privacy challenges. Also, the researchers review the trends of designing lightweight algorithms after explaining alternatives to traditional cryptography methods that fit the composition of the IoT. Finally, after reviewing the challenges in terms of physical and software implementation, the study recommended that when developing LWC solutions, the following should be noted:


Based on the previously mentioned recommendations, the following are the methods presented by this study that can be included when designing a lightweight security system for the IoT [41]:

• *Hashing*: is a mathematical algorithm that assigns data of arbitrary size (often called "message") to a fixed-size bit matrix (the "message summary"). It is a one-way

function that is practically useless to reverse or reverse the account. Ideally, the only way to find a message that produces a particular hash is to forcibly search for potential inputs to see if they have a match or use a rainbow table of identical hash.


Sehrawat et al. [42] presented a detailed comparison between several algorithms compatible with the IoT and after conducting cryptanalysis attacks. This study also showed that block ciphers had attracted the attention of many researchers as a basis for developing LWC algorithms. Finally, this study also recommended the requirements for the future of LWC algorithms.

Dutta et al. [43], reviewed the encryption solutions that can be used in the IoT by comparing some LWC that can fit with the nature of IoT devices. Researchers believe that symmetric encryption is the closest to suit the heart of the IoT. They also found that the modified AES algorithm provides a suitable security solution to the restrictions imposed by the capabilities of IoT devices after studying many algorithms like *DES, 3DES, Blowfish*, etc.

After choosing AES as a standard and reliable algorithm and achieving the desired goal, the researchers analyzed the performance of a set of versions of the algorithm implemented in previous studies by sorting them into two parts as follow [43]:


The researchers also presented a study of attacks on AES that should be monitored and found solutions such as Differential Fault Analysis Attacks and wireless interceptive side-channel attack techniques. These attacks can be resisted through the use of dummy keys and XOR operations [43].

Rajesh et al. [44] presented the Novel Tiny Symmetric encryption Algorithm (NTSA), which provides better confusion for each round which leads to better security level. The comparison centered with the TEA algorithm is considered one of the most attractive algorithms because of its ease of implementation and less memory usage. Its main problem is to use the same key for all rounds, which reduces the level of security and its poor performance. The results show that NTSA outperforms many other security algorithms and achieves better performance, making it more suitable for IoT and embedded devices.

Gunathilake et al. [45] discussed the future applications of LWC, how to implement it, and the challenges it faces. The study also touched on the existing LWC algorithms previously mentioned in our research and confirmed the effectiveness of the modified AES algorithm in this field.

Usman et al. [46] reviews the light encryption algorithms that fit the nature of the IoT after identifying the obstacles to using traditional algorithms, such as the low power capacity of the devices. Researchers believe that the security of big data flowing through the IoT is the main problem, as this weakness may overwhelm the advantages of IoT applications. Therefore, considering the capabilities of these devices represented in the low capacities, it was necessary to think of new methods that require simpler arithmetic operations and less memory while providing an acceptable degree of security. In addition to what has been mentioned, these methods must consider the diversity of devices, their different capabilities, and the protocols used to have the ability to integrate and adapt to this diversity. And now we still have the issue of privacy, as the IoT, with the vast amounts of data circulating, must provide the user with the possibility of appropriate control over his data [46]. The researchers considered that symmetric encryption is best suited for the IoT because asymmetric encryption requires higher capabilities. And the following are some of the symmetric encryption algorithms that have been reviewed [46].

Abutair et al. [47] believe that despite their importance, smart cities still face the challenge of balancing the quality of service and maintaining the privacy and security of information. This study summarized the difficulty of achieving this balance as follows:


The researchers studied many lightweight algorithms used in the IoT. Based on this study, an infrastructure has been proposed that provides a specific degree of privacy and security for the IoT. This study concluded that some modern algorithms such as *CLEFIA* and *TRIVIUM* achieved terrible results compared to the old algorithms, especially *TRIVIUM*, which gave disastrous results [48]. The study explains the structure of smart cities. Without going into details here, the aspect that concerns us is the necessity of providing IoT devices with algorithms that meet the guarantee of authentication, integration, and confidentiality to protect the network from threats. Such as *Corrupted Data, Replay Attacks, IP Spoofing, Identity Usurpation, DoS/DDoS Attacks, and, Data Leakage* [47]*.* This study presented a new design that depends on the capabilities of the device that will be added. Based on these capabilities, the

appropriate lightweight algorithm is selected for it. The mechanism of this design can be summarized as follows:


After testing many algorithms by changing some factors, the researchers found that the algorithm closest to adapting to the majority of IoT devices is the *AES* algorithm, with the need to reduce its resources [47].

Ramadan et al. [48] introduced a LWC algorithm called LBC-IoT that handles 32 bit blocks with a key of up to 80 bits. This algorithm is based mainly on the Feistel structure, along with simple operations such as XOR that do not consume power and 4-bit S-boxes. The results indicate the strength of this algorithm against attacks in addition to its acceptable performance, and it is considered a promising algorithm for implementation on small and very restricted devices.

Periasamy et al. [49] proposed a lightweight block cipher mechanism that works on 8-bit processing, as their study indicates that this algorithm is superior to its counterparts. According to the researchers, this algorithm derives its strength from the strength of the encryption in the compensation boxes. In terms of performance, the design of the compensation boxes played marginally using the Multi sequence Linear Feedback Shift Register and reliance on simple operations such as XOR, shifting, and registers to reduce space required and optimization in power consumption and speed.

Thabit et al. [50], researchers introduced a New LWC Algorithm (NLCA) to secure cloud computing applications. This algorithm uses a 16-byte key based on Feistel and substitution permutation. This algorithm succeeded in achieving confusion and diffusion by introducing some logical operations into the algorithm's formula, such as Shifting, Swapping, and XOR. One of the advantages of this algorithm is the flexibility, such as AES, where the number of rounds and the length of the key are variable according to the application's needs. The results also indicate that this algorithm provides a good level of security and performance, which makes it suitable for these applications.

In this section, we discuss many LWC related researches. **Table 8**, focus on the key points that have been discussed in IoT cryptography related works and summarize them.

#### **4.2 AES related works**

In this section, we summarize some researches that present some AES-based system, discuss these systems and highlight the differences in these AES versions to reach the best possible ways to improve the performance and strength of this algorithm more.

Javed et al. [51], presented a new design for the AES algorithm to make it suitable for mobile devices and speed it up despite the limitations of the hardware specifications. After reviewing the mechanism of the standard AES algorithm, the researchers discuss the improvement that was made to AES implementation and the motives that were relied upon in this optimization as follows:


#### **Table 8.**

*LWC related works summary.*


*generation* approach is costly in clock rounds and need 16 bytes of additional memory to store the last round keys for the decryption [51].

The results of this study showed that the performance of the proposed method gives better results, as it provides 3 times better encryption speed and is about 20 times better in round keys calculations. This design outperformed its predecessor by 20 times while reading data from the hard disk and encrypting it if the data was greater or equal to 1 MB [53].

Abhijith et al. [52], presented an improved model for implementing the AES algorithm by slicing and integrating the internal processes of the algorithm. This new version used Block-Ram and 10 levels of pipelines to improve efficiency and productivity. The results indicate that this enhanced version significantly enhances performance and the possibility of integrating it with other systems.

Bui et al. [53] worked on finding an improved version of AES in several ways. First, reduce the combinational logic and number of records by organizing the data path. Second, the clock gateway strategy, key expansion, and minimization of data activities contributed to reducing the algorithm's energy use. Here are the modifications that have been implemented to achieve the above improvements:


These modifications were additions that can be used without modifying the algorithm. As for the fundamental alterations in the algorithm, they were represented as follows [53]:


of improvement is to reduce consumption, to suit it for mobile applications, the structure is directed to minimize space.

The results show that the proposed version offers the same PRESENT algorithm in energy use. Also, the proposed system is resistant to the attack of power correlation analysis with less than 20,000 traces, which seeks to expose the data path. Also, the data path in case of parallelism provides it with more robustness. Finally, this design uses different key sizes, which contributes to providing various levels of security as needed [55].

Mamoun et al. [54] provided a comprehensive explanation of the AES algorithm. The study presented a new model for the AES algorithm to enhance its security level by adding an XOR operation to an extra byte of s-box and using an additional random key. The results indicate that this modification contributed to improving the level of AES security variably due to the randomness of the added key. The results also showed that this modification improved confusion and increased time security.

Umer et al. [55] tested AES using different techniques depending on the resources of the target devices, the results were characterized by varying in nature according to the techniques used. Among these techniques were used; Parallelization and storage of s-box and key expansion, as it has been noted that the introduction of such technologies helps in optimizing the exploitation of resources to provide better results.

Daoud et al. [56], the researchers present an optimization of the AES algorithm using Vivado High-Level Synthesis (HLS), and their results show significant progress in increasing the throughput of the proposed algorithm, which was implemented on the FPGA only using flip flops and look-up tables. Since optimizing commands in Hardware Description Languages (HDL) is not easy and time-consuming, HLS improves the algorithm with less effort. HLS is an automated process that deals with high-level programming languages such as C that is used to ease the struggles that HDL requires in the development process, debugging, and provide flexibility in meeting system requirements. HLS tool synthesized compiled core AES functions in an RTL block, and sub-functions were divided into sub-blocks at higher system levels. Below is a review of the improvements that this study made to the AES algorithm [58]:


The main objective of this study was to achieve the maximum throughput in encryption. The process that most positively affected the results is integrating key expansion with encryption. By comparing the effects of frequency, productivity, and area utilization, it appears to us that the proposed design in this study has outperformed the previous strategies [56].

Proceeding from the fact that the AES algorithm is considered the best secure algorithm currently available and can be adapted to IoT devices. Rokan et al. [57] provided an integrated security system for the IoT called *Modified Lightweight AES* *Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

(*MLAES*) that includes two integrated systems; The first one is a *Secure Encryption* based on a lightweight version AES integrated with Chaos Maps. The second is a *Secure Authentication* using a chaotic hash function based on SHA3-256-bits. The following is a review of the three main phases of this system:


The study results indicate that despite the modification to AES, the level of security remained strong, in addition to the significant improvement in its performance and the specifications required for its operation. Perhaps the most prominent result was that this system passed the NIST tests, which means that the system is resistant to linear differential attacks and brute force attacks [57].

Farooq et al. [58], given the discrepancy between the capabilities of IoT devices, explored five implementations of the AES algorithm. These applications use modifications and improvements to the AES algorithm. The applications indicate the disparity in the results, as each of these applications fits a specific category of IoT devices. Therefore, the study recommended moving away from comprehensiveness and not limiting encryption to one algorithm for all devices, but instead relying on the device's capabilities to choose the optimal AES version for use.

Nagalakshmi et al. [59], given the discrepancy between the capabilities of IoT devices, presented some strategies for implementing AES with a set of other systems to suit these devices of varying powers, and the study also touched on the use of LFSR. The results indicate a security improvement, the ability to check signatures, and random checks without significantly affecting performance.

Salim et al. [60] presented the development of an AES algorithm called multi-key AES. The name came concerning the fact that this proposal uses the AES algorithm but uses several keys as the secret key is used to configure a variable number of keys using ECC. The study specialized in implementing this algorithm in the IoT, provided that it is used on devices capable of running this algorithm. The results indicated that this


**Table 9.**

*AES related works summary.*

modification did not affect the algorithm's performance, but it contributed to improving its security.

In this section, we discuss many AES-based related researches. **Table 9**, present the summary of some researches that worked on modifying AES to adapt it with IoT.

#### **5. Evaluation**

This section presents the ways of evaluating algorithms and a brief discussion of this study.

#### **5.1 Evaluation**

The evaluation process should address performance evaluation and security evaluation to ensure the power of the algorithm. To evaluate performance, we will initially need to calculate the following:


As for security, we will initially need to account for:



**Table 10.** *AES evaluation results.*

#### **5.2 Summary**

Based on all that was mentioned previously, studies have confirmed that stream cipher provides better performance than block cipher. Still, a block cipher is superior to a stream cipher in terms of security especially when we looking to better confidentiality. Some previous studies also indicated that lightweight stream cipher did not succeed much on the security front. From here, we can be sure that the basis in our research should be based on a block cipher with its security strength while trying to improve it in the level of performance [64].

We believe that using a recognized and standard algorithm to improve it would be better at the current stage. Most previous studies confirmed that the choice fell on AES due to its superiority. In appendix A, we review the summary of the results of the AES algorithm test in terms of performance and security to be a starting point for improvement [64]. These results are shown in **Table 10**.

These results showed that AES provide an acceptable degree of security according to this evaluation criteria, such as Key security, Histogram, NIST, Confusion, and Diffusion. But to prove that, we will use more security tests in future work such as Mapping, Correlation, Unified averaged changed intensity, and Number of Changing pixel Rate. On other hand, the result of performance testing can be improved by changing or replacing some core functions on AES.

#### **6. Conclusion**

In this chapter, a detailed study of computer security has been conducted. After clarifying different kinds of cryptography, LWC has been addressed, considering its basics and requirements. Some of the presented algorithms highlight the essential needs for LWC algorithms and the importance of making them compatible with the resources of IoT devices. This study also discussed the latest studies related to each of Lightweight Cryptography, Lightweight AES-based algorithms, and the most prominent evaluation criteria used to judge the suitability of an algorithm. Finally, this study presented the results of testing the AES algorithm according to the specified criteria. We believe that these results constitute a starting point for future work as promising results in the field of LWC algorithms and their suitability to the resources of IoT devices.

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

### **Author details**

Mohammed Abujoodeh\*, Liana Tamimi and Radwan Tahboub College of IT and Computer Engineering, Palestine Polytechnic University, Hebron, Palestine

\*Address all correspondence to: 131089@ppu.edu.ps

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

### **References**

[1] Alfred Y. Network Security. Malaysia: Asia Pacific University; 2019. pp. 5-11. DOI: 10.13140/RG.2.2.19900.59526

[2] Manuj A. Network Security with pfSense: Architect, Deploy, and Operate Enterprise-Grade Firewalls. 1st ed. Birmingham: PACKT Publishing; 2018

[3] William S. Cryptography and Network Security: Principles and Practice. 8th ed. London: Pearson; 2017

[4] Mohammed Z, Ahmed E. Internet of things applications, challenges and related future technologies. Journal of World Scientific News. 2017;**67**(2):126-148

[5] Available from: https://internetofth ingsagenda.techtarget.com/Ultimate-IoT-implementation-guide-for-busine sses [Accessed: February 15, 2022]

[6] Mista S, Roy C, Mukherjee A. Introduction to Industrial Internet of Things and Industry 4.0. 1st ed. Florida: CRC Press; 2021

[7] Qabajeh L. A more secure and scalable routing protocol for mobile ad hoc networks. Security and Communication Networks. 2013;**6**:286-308

[8] Makhdoom I, Abolhasan M, Ni W. Blockchain for IoT: The challenges and a way forward. International Council for Evangelical Theological Education. 2018: 594-605

[9] Salhab O, Jweihan N, AbuJoodeh M, Abutaha M, Farajallah M. Survey paper: Pseudo-random number generators and security tests. Journal of Theoretical and Applied Information Technology. 2018;**96**: 1951-1970

[10] Abutaha M, Farajallah M, Tahboub RM. Survey paper:

Cryptography is the science of information security. International Journal of Computer Science and Security. 2011;**5**:475

[11] Available from: https://geek-unive rsity.com/ccna-security/aaa-explained [Accessed: February 15, 2022]

[12] Elminaam DA, Abdual-Kader HM, Hadhoud MM. Evaluating the performance of symmetric encryption algorithms. IJ Network Security. 2010; **10**:216-222

[13] Guanrong C, Ybin M, Charles C. A symmetric image encryption scheme based on 3D chaotic cat maps. Chaos, Solitons & Fractals. 2004;**21**:749-761

[14] Zhu S. Algorithm design of secure data message transmission based on Openssl and Vpn. Journal of Theoretical & Applied Information Technology. 2013;**48**:562-569

[15] Bellare M, Rogaway P. Optimal asymmetric encryption. In: Workshop on the Theory and Application of Cryptographic Techniques. Berlin, Heidelberg: Springer; 1994. pp. 92-111

[16] Simmons G. Symmetric and asymmetric Encryption. ACM Computing Surveys (CSUR). 1979;**1**:305-330

[17] Shamir A, Adleman L, Rivest R. A method for obtaining digital signatures and public-key cryptosystems. Communications of the ACM. 1978;**21**: 120-126

[18] Alsadeh A, Karakra A. A-RSA: Augmented RSA. 40th conf of SAI Computing. 2016:1016-1023

[19] Gamal T. A public-key cryptosystem and a signature scheme based on discrete *Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

logarithms. IEEE Transactions on Information Theory. 1985;**31**:469-472

[20] Available from: https://www.que s10.com/p/33937/el-gamal-cryptograph y-algorithm-1/ [Accessed: February 15, 2022]

[21] Maurer U, Wolf S. The Diffie– Hellman protocol. Designs Codes and Cryptography. 2000;**19**:147-171

[22] Mihailescu M, Nita S. "Elliptic-curve cryptography", elliptic-curve cryptography. In: Pro Cryptography and Cryptanalysis. Berkeley, CA: Apress; 2021. DOI: 10.1007/978-1-4842-6367-9\_1

[23] Al-Absi M, Abdullaev A, Absi A, Sain M, Lee H. Cryptography survey of DSS and DSA. In: Advances in Materials and Manufacturing Engineering, Lecture Notes in Mechanical Engineering. Singapore: Springer; 2020. pp. 661-669

[24] Agrawal M, Mishra P. A comparative survey on symmetric key encryption techniques. International Journal on Computer Science and Engineering. 2012;**4**:877-882

[25] Stallings W. The RC4 stream encryption algorithm. In: Cryptography and Network Security, Prentice Hall, 2005

[26] Mister S, Tavares S. Cryptanalysis of RC4-like ciphers. Selected Areas in Cryptography. 1998;**1556**:131-143

[27] Robshaw M, Billet O. New Stream Cipher Designs: The eSTREAM Finalists. New York: Springer; 2008

[28] Bernstein D. The Salsa20 family of stream ciphers. In: Robshaw M, Billet O, editors. New Stream Cipher Designs. Lecture Notes in Computer Science. Vol. 4986. Berlin, Heidelberg: Springer; 2008

[29] Berbain C, Billet O, Canteaut A, Courtois N, Gilbert H, Henri L, et al. SOSEMANUK: A Fast Software-Oriented Stream Cipher. New York: Springer; 2008. pp. 98-118

[30] Coppersmith D, Holloway C, Matyas S, Zunic N. The data encryption standard. Information Security Technical Report. 1997;**2**:22-24

[31] Daemen J, Joan, Rijmen V. The data encryption standard. In: The Design of Rijndael. Springer; 2002. pp. 81-87

[32] Available from: https://www.c ryptomathic.com/news-events/blog/3de s-is-officially-being-retired. [Accessed: February 15, 2022]

[33] Bhat P, Deepthi. Comparison of MD5 and blowfish algorithm. International Journal of Innovative Research in Science Engineering and Technology. 2016;**5**:506-511

[34] Kalaiselvi RC, Vennila M. An analysis of AES, RSA, and blowfish - A review. The International Journal of Analytical and Experimental Modal Analysis. 2020;**XII**:568-588

[35] Blumenthal U, Fabio M, Keith M. The Advanced Encryption Standard (AES) Cipher Algorithm in the SNMP User-Based Security Model. Vol. No. RFC 3826. USA: Bell Labs; 2004

[36] Dworkin M. Recommendation for Block Cipher Modes of Operation: Methods and Techniques. SP: NIST; 2001. pp. 800-38A

[37] Cruz-Cunha M, Portela I. Handbook of Research on Digital Crime, Cyberspace Security, and Information Assurance. 1st ed. Pennsylvania: IGI Global; 2014

[38] Thakor V, Razzaque MA, Khandaker M. Lightweight cryptography algorithms for resource-constrained IoT devices: A review, comparison, and research opportunities. IEEE Access. 2021;**9**:28177-28193

[39] Muhammad R, Quazi M, Rafiqul I. Current lightweight cryptography protocols in Smart City IoT networks: A survey. ArXiv. 2010;**00852**: 2020

[40] Manifavas H, Hatzivasilis G, Fysarakis K, Papaefstathiou J. A survey of lightweight stream ciphers for embedded systems. Security and Communication Networks. 2015;**9**: 1226-1246

[41] Buchanan W, Li S, Asif R. Lightweight cryptography methods. Journal of Cyber Security Technology. 2018;**1**:187-201

[42] Sehrawat D, Gill N. Lightweight block ciphers for IoT based applications: A review. International Journal of Applied Engineering Research. 2018;**13**: 2258-2270

[43] Dutta I, Ghosh B, Bayoumi N. Lightweight Cryptography for Internet of Insecure Things: A Survey. 2019 IEEE 9th Annual Computing and Communication Workshop and Conference; 2019. pp. 0475-0481

[44] Rajesh S, Paul V, Menon V, Khosravi MM. A secure and efficient lightweight symmetric encryption scheme for transfer of text files between embedded IoT devices. Symmetry. 2019;**11**(2):293

[45] Gunathilake N, Buchanan W, Asif R, Rameez. Next Generation Lightweight Cryptography for Smart IoT Devices: Implementation, Challenges and Applications. IEEE 5th World Forum on Internet of Things (WF-IoT); 2019. pp. 707-710

[46] Usman M. Lightweight encryption for the low powered IOT devices. arXiv. 2020;**2012**:00193

[47] Abu-tair M, Djahel S, Perry P, Scotney B, Zia U, Carracedo J, et al. Towards secure and privacy-preserving IoT enabled smart home: Architecture and experimental study. Sensor. 2020; **20**:6131

[48] Ramadan R, Aboshosha B, Yadav K, Alseadoon I, Kashout M, Elhoseny M. LBC-IoT: Lightweight block cipher for IoT constraint devices. CMC-computers Materials Continua. 2021;**67**:3563-3579

[49] Prakasam P, Madheswaran M, Sujith KP, Shohel S. An enhanced energy efficient lightweight cryptography method for various IoT devices. ICT Express. 2021;**7**:487-492

[50] Thabit F, Alhomdy S, Al-ahdal A, Abdulrazzaq, Jagtap P. A new lightweight cryptographic algorithm for enhancing data security In cloud computing. Global Transitions. 2021;**2**: 91-99

[51] Javed A. Fast Implementation of AES on Mobile Devices. Proc. 8th Int. Netw. Conf.; 2010. pp. 133-142

[52] Abhijith P, Goswami M, Tadi S, Pandey K. Optimized architecture for AES. Cryptology ePrint Archive: Report. 2014;**1**:540

[53] Bui D, Puschini D, Bacles-Min S, Beigné E, Tran X-T. AES Datapath optimization strategies for low-power low-energy multisecurity-level internetof-thing applications. IEEE Transactions on Very Large-Scale Integration (VLSI) Systems. 2017;**25**:3281-3290

[54] Mamun A, Rahman S, Shaon T, Hossain A. Security analysis of AES and enhancing its security by modifying

*Toward Lightweight Cryptography: A Survey DOI: http://dx.doi.org/10.5772/intechopen.109334*

S-box with an additional byte. International Journal of Computer Networks & Communications. 2017;**9**: 69-88

[55] Farooq U, Aslam F. Comparative analysis of different AES implementation techniques for efficient resource usage and better performance of an FPGA. Journal of King Saud University - Computer and Information Sciences. 2017;**29**:295-302

[56] Daoud L, Hussein F, Rafla N. Optimization of Advanced Encryption Standard (AES) Using Vivado High Level Synthesis (HLS). Proceedings of 34th International Conference on Computers and Their Applications. 2019: 36-44

[57] Naif R, Abdul-Majeed GH, Farhan AK. Secure IOT system based on chaos-modified lightweight AES. 2019 International Conference on Advanced Science and Engineering (ICOASE). 2019:1-6

[58] Farooq U, Mushtaq M, Bhatti M. Efficient AES implementation for better resource usage and performance of IoTs. In: CYBER 2020 - 5th International Conference on Cyber-Technologies and Cyber-Systems. France: Nice; 2020

[59] Nagalakshmi E, Mohan V, Kumar D. AES datapath optimization strategies for low-power low-energy multi securitylevel internet-of-thing applications. International Journal of Advanced Research in Science, Engineering and Technology. 2020;**2**:347-355

[60] Salim K, Alalak S, Jawad M. Improved image security in internet of thing (IoT) using multiple key AES. Baghdad Science Journal. 2021;**18**: 417-429

[61] Jagtap S, Thabit F, Alhomdy S. Security analysis and performance evaluation of a new lightweight cryptographic algorithm for cloud computing environment. Global Transitions Proceedings. 2021;**2**:100-110

[62] Coskun B, Memon N. Confusion/ diffusion capabilities of some robust hash functions. 40th Conference Information Sciences and Systems. 2006; **CISS'6**:1188-1193

[63] Farajallah M, Abutaha M, Abujoodeh M, Salhab O, Jweihan N. Pseudo-random number generator based on look-up table and chaotic maps. Journal of Theoretical and Applied Information Technology. 2020;**98**:3130

[64] Abujoodeh M, Tamimi L, Tahboub R. Exploring and Adapting AES Algorithm for Optimal Use as a Lightweight IoT Crypto Algorithm, [master thesis]. Palestine: Palestine Polytechnic University; 2022. Available from: https://scholar.ppu.edu/handle/ 123456789/8635

#### **Chapter 7**

## Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection and Predictive Maintenance

*Cristian Bosch and Ricardo Simon-Carbajo*

#### **Abstract**

Early power loss detection in wind turbines is a key for the wind energy industry to avoid elevated maintenance costs and reduce the uncertainty regarding generated power estimations. Location, especially of those wind farms isolated offshore, causes the strategy of scheduled-only maintenance inefficient and very costly, additionally presenting a typically long downtime after a breakdown. These problems point to the creation of predictive solutions to anticipate the maintenance procedure, preparing the necessary parts and avoiding the possibility of destructive failures. Predicting failures in structures of such complexity requires modeling their multiple components individually in addition to the whole system. For this purpose, physics-based and data-driven models are used, which have proven themselves in this context. Machine learning has proven to be a valuable resource for solving a variety of problems in this industry. Thus, we will propose data-driven Deep Learning methods to compute the Power output of wind turbines with respect to all the mechanical and electrical features by using two types of Deep Neural Networks: a simpler combination of linear layers and a Long-Short Term Memory Neural Network. Then, with the use of a onedimensional Convolutional Neural Network we will predict the time to failure of the system.

**Keywords:** wind-turbine, predictive maintenance, computational semantics, deep learning, LSTM, time series, regression, classification, CNN

#### **1. Introduction**

As per WindEurope [1], wind energy represents the second biggest provider of energy in the European Union (EU), accounting for a 18.8% of the capacity, behind gas. Ireland, for instance, represents the 3.5% of the EU's combined capacity, and wind energy covers a 28% of the country's energy demand. In this specific circumstance, it is interesting to point that the maintenance cost of a wind turbine can go from a 16% to a 30% [2, 3] of the Levelized Cost Of Electricity (LCOE). Wind power technologies, on the rise despite being mature, would more greatly consolidate by

adopting solid solutions to the maintenance problem. Subsequently, the predictive approach for wind turbines and farms has become a priority, introducing techniques aiming for downtime reduction and wind turbine lifespan extension.

While physics-based modeling systems exists, our purpose is to approach the problem through the application of Deep Learning (DL) algorithms on the data collected by the Condition Monitoring System (CMS) and the SCADA control system from several wind turbines. For this work, we have obtained data from an onshore Siemens SWT-2.3-101 wind turbine. The target is, based on historical data, to predict the anomalous behavior of the system and a fault with enough anticipation.

This chapter will explain the phases of data preparation, acknowledging there is a time series behavior and the application of Deep Learning (DL) solutions. Two Deep Neural Network architectures designed with the purpose of data-driven prediction of the SCADA measured "Power" output have been trained in a semi-supervised paradigm, assuming the training samples belonged to what could be considered the normal state of the wind turbine. We will address whether there is a significant improvement if the regression on the power output is done with a basic Artificial Neural Network (ANN) mainly built on linear layers or using a memory cell, as it is the case in Long-Short Term Memory ANNs (LSTMs). For the latter, our choice will be applying directly of Bidirectional LSTMs (BiLSTMs) as they are expected to offer a performance superior to that of the base architecture [4]. Then, the anomalies in the prediction of validation and test datasets will be found using several methods that assume that, ideally, the prediction deviations from real data will follow a normal distribution (which will be the driver of the hyperparameter optimization). These methods include the standard deviation, a Monte Carlo dropout and a fewshot dropout using less predictions and fitting a t-Student distribution on them, all three at 95% CL. The anomalies will be statistically analyzed afterwards aiming for testing hypothesis about their relationship to the Time-to-Fault, which is highly complex to generalize in the case of these datasets with errors of different origin and short periods between them (the maximum constant performance period being of 11 days). The statistical significance of the anomaly appearance will be a motivator for building a one-dimensional Convolutional Neural Network (1-dim CNN) to predict downtime with an appropriate time-to-fault. In this classification task, we will consider our "Prefault" label for a processed sample as another hyperparameter and draw critical conclusions of its possible values. For a previous reference on our work in this dataset, we studied data augmentation through optimal transport dataset aggregation in [5].

The book chapter has the following structure: The subsequent subsection displays the state of the art with regards to wind turbine fault forecasting and general anomaly detection using several Artificial Intelligence methods. After that, we explain the methodology of our approach to solving this problem, analyzing the data with regards to discriminating them semantically as a preparation for the computation of the power output and the detection of anomalous behavior. Then, we will explain the methodology for our data-driven regression of the power output with the use of two different deep neural networks and the possible the extraction of the anomalies in the predicted data. After showing statistical interest in these anomalies, we will proceed to show our strategy to classify data samples with a 1D-CNN and present how an interesting choice of metrics and hyperparameter space can help solve this problem. We finalize the book chapter by sharing the conclusions of our research.

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

#### **1.1 State of the art**

Wind power generators are composed of different rotating components that undergo an extensive overall performance during its lifetime. Condition Monitoring Systems (CMSs) are common within the modern industry and are a set of sensors that screen the state of the turbine's one-of-a-kind components in real time. The topic has extensively been reviewed in [6], where the advantages of Fault Detection Systems are outlined, ranging from cost reduction to the improvement of the Capacity Factor, since the ability to forecast and anomaly can optimize the moment when the maintenance is applied, avoiding any stops at high energy output periods. A CMS can collect data from a wide range of sensors focused on vibration, component temperatures, oil levels and electricity voltages and currents. Fault forecasting can combine this strategy with monitoring processes affecting the wind turbine, such as: crack detection, strain, thermographic analysis, electrical conditions, signal and performance monitoring and acoustic analysis.

There are methods inside the literature which put their attention on modeling the activity of the wind turbine parts by means of their physical behavior [7] and enhance this model with CMSs data from the wind turbine to create an approach favored by this hybridization. However, the challenge of our research is to consider only the data aggregated from the CMS to model the behavior of the wind turbines and make the model useful to predict periods of downtime and determine if it must be in a general way or with specific information about the upcoming fault.

Concerning the employment of these data using ML algorithms, the problem of defect detection can be resolved with two different strategies: i) Modeling the normal performance and detecting anomalies as they arise and ii) Evaluating data from time spanning before faults to anticipate defective behavior. We will face the problem by one or a combination of both these approaches.

Regarding anomaly detection, there is an advantage in modeling the normal output periods of a wind turbine, which is the use of most of the data collected by the CMS, as the datasets present a big imbalance with normal regime being the most populated class. These are known as semi-supervised models, since faults and time ranges of data close to faults are removed purposely before training the algorithms. A representative solution of this strategy are autoencoders. Autoencoders are Deep Neural Networks (NNs) with a symmetrical architecture with encoder layers that arrive to a bottleneck that stores the encoded representation of the data. These data will be decoded afterwards, with an output that preserves the important features. Autoencoders have become a reference in early fault detection and have been proven capable of discriminating the parts originating the failure [8]. There are other possibilities for early anomaly detection present in the literature: We can find NN architectures trained only with normal regime data that predict the power output expected at the next time iteration which, once compared against the actual generated output of the turbine, can determine if its behavior is unexpected. The parts responsible for such faulty behavior can be traced through a Principal Component Analysis (PCA) analysis [9]. We will partially follow this technique, modeling the power output through our dataset and then deviating from that work in the way anomalies are studied. Classification methods that constrain normal behavior periods to be modeled only if far prior or far posterior from a fault have been proven to be a competent way of discriminating the SCADA delivered features worth of consideration [10].

Moving to putting the focus on the historical faults of a device, the range of techniques is diverse too. The literature contains classifiers relying on supervised training, which uses datasets with every sample labeled. One example that differs from the analysis of SCADA data turns to visually inspect the turbines with drones and then trains Convolutional Neural Networks (CNNs) to detect usual vortex generator damage indicators such as erosion or missing gear teeth [11]. These datasets, while based on image collection instead of SCADA, feature high data imbalance too, which requires compensation from software, providing complex architectures for the CNNs proposed for the task [12]. With respect to using turbine sensor data in fully supervised training scenarios, multiclass classification with Support Vector Machines (SVM) has been undertaken with simulated turbine data, showing success in discriminating faults according to their nature [13]. SVMs gained early popularity in predictive maintenance field, though the main models currently employed are decision trees and gradient boosting. This is shown in the benchmarking of Random Forest and XGBoost classifiers present in [14]. Signal analysis has been experimented on too [15], where interference in the currents of the Double-Fed Inductor Generator (DFIG), originated by the vibrations of the faulty gearbox are studied through autoencoders and NN classifiers for anomaly detection.

Focusing more on the work related to anomaly detection [9], we can also find several applications of ANNs for different time series problems. We will follow the strategy of dividing the dataset in four pieces [16], emphasizing that training must be done with what is considered normal regime data [17], which we will apply to both our ANN based on linear layers and an BiLSTM. These methods rely on the deviations from predictions to the real output to fall in a normal distribution, so as to have a reliable way of computing confidence intervals for the anomalies to be spotted [18, 19]. The confidence intervals associated to the prediction will be computed making use of the Monte Carlo dropout, activating the dropout layers in the evaluation time for the computation of a big number of predictions, a method use by Uber [20]. In order to prove the viability of less resource-consuming, a small number of dropout activated predictions will be computed and then fitted to a t-Student distribution [21]. We will try to predict anomalous behavior using the global standard deviation of predictions in the normal validation dataset too, for a more complete analysis.

Wind turbine data feature engineering can be a complicated task. The idea of using CNNs, famous for extracting the interesting features of images, in the field of fault prediction, has been explored in the literature too, as a feature information extraction tool to be combined with LSTMs [22] and as an independent predictor for software bugs [23]. Another model that is a combination of adaptive feature engineering and a CNN for fault forecasting can be found in [24]. Techniques aiming for automated feature engineering and modeling are a greatly explored topic in the field of data science, as they ease building models when domain expertise is not available. Among these, a couple of very relevant toolkits are AutoML [25, 26] and the H20.ai package [27, 28].

#### **2. Methodology**

#### **2.1 Data**

Our data originates from a Siemens SWT-2.3-101 turbine. Samples were collected every ten minutes and the dataset spans for a period of nearly four years. From the features included in the dataset, we choose those that refer to weather conditions (wind speed, temperature, etc.), have mechanical origin (gear bearing temperatures,

#### *Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

blade angles or pressure, etc.) or electrical measurements (different voltages and currents) to train our models.

The dataset is cleaned and labeled according to the status flag associated to each sample and assumptions with regards to posterior status flags. The computation of the regression requires a semi-supervised approach: After we make an initial split in training, validation and test sets, being the test a fourth part of the original data and the validation part a fourth portion of the remainder, we purge the training data from what we cannot confirm as well-behaved samples, and then we make the same in the validation dataset, thus creating one with only well-behaved data and a full validation dataset. All these portions are chronologically ordered, since we are dealing with a time series. This could have implications when training the BiLSTM, as there would be cuts due to the preprocessing purges. We will later assess if this has caused relevant effects on the models trained.

Since we are using a real industrial dataset that has not been curated for research purposes, there is no labeling beyond the indication of a fault happening (status flag). The definition of the normal or good behavior we seek to isolate is not clear (normal is not used as belonging to a Gaussian distribution here). As a way to deal with it, we define normal data based on the number of days before a fault is registered. This number of days will be considered a hyperparameter and thus the labeling process is included as part of the optimization to avoid putting a human bias greater than using full days as a time reference, which is a flexible enough decision. Trimming the data this way ensures that the power regression is computed with samples that are not semantically different, which will help achieving a well-behaved prediction distribution. A scaler is fit in what is considered normal data and the transformation is applied on the whole dataset, as the normal data is confirmed not to contain any outliers on features that would provoke a loss of information after scaling with the use of extreme points. The posterior classification task will have a similar labeling strategy, where a logistic regression will be performed for classifying each sample as either "Normal" (or 0) or "Prefault" (or 1). These prefault periods will be determined by a hyperparameter as well, which will then be determined by the best model during optimization.

As we are aiming for a full data-driven regression, after the selection of the features, there will be no dimensionality reduction as we want to include every detail in the prediction of the power. This contrasts with theoretical approaches that would try to reproduce the power curve, which is considered a relationship mostly exclusive between "Wind Speed" and "Power".

Moving to statistical features of the data, as it being normally distributed, it is a fact that the wind turbine works within a regime that makes the power curve a relation but not a function (**Figure 1**), since each value of the domain can correspond to several values of the codomain. This has, as we will expand later, significance when the intent is to have a regression where its errors are normally distributed. Our intent is that the deviations of the real data from the prediction (on the well-behaved regime used for training and validation of the regression) are normally distributed which, due to the power output being multivalued, is not trivial (see **Figure 2**).

Considering that the dataset only once shows a period of 11 days going without any downtime and that we will assume that, for the convenience of scheduling the maintenance, at least 24 hours of anticipation are needed, we will set the days of well-behaved regime hyperparameter between 2 and 6 days before a fault, so the optimization will decide the best possible outcome without slicing dramatically the amount of training samples.

The metrics of the regression will be measured in the well-behaved split of the validation dataset, and it will be the full validation dataset the one used to compute statistics referring to the appearance of anomalies in the prediction of the power

**Figure 1.** *Power curve, relating the wind speed and the power generated by the turbine.*

**Figure 2.** *Wind speed (left) and power (right) histograms.*

output. An anomaly will be defined by real data escaping the prediction intervals computed by the regression and posterior techniques with statistical significance.

Once the anomalies in the regression of the power have been computed, this anomalous data will be included in the 1D-CNN as features too, simply as the deviations between registered and predicted power for each sample. As we aim to predict a failure with an anticipation that is suitable for performing early maintenance, this is a very complicated task, which we will try to compensate with feature engineering by adding rolling averages of the different original features to the data, in an agnostic manner. The time windows will be considered hyperparameters and these newly engineered features will be created anew for each hyperparameter optimization step, whereas the original features will be a constant during training of the CNN.

#### **2.2 Deep learning models**

Our first goal is to find deviations from a data-driven regression of the power prediction and the real power output feature registered by the SCADA monitoring system. The next step will be to study these deviations statistically and determine

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

whether their appearance is significant as the time before a fault gets shorter. Finally, we will build a classification model of the data samples for pinpointing an impending downtime of the wind turbine with sufficient anticipation. For succeeding in all these tasks, we will make use of three different neural network architectures. The power regression will be performed through two different approaches, a deep neural network (NN) that is made by a succession of Linear and Dropout layers (with ReLu activation layers within) and by using a BiLSTM with dropout, to check if the inclusion of memory cell can improve the regression even if we are not interested in using previous power predictions (autoregression). Power will always be our dependent variable. In a sense, we are extending the "Power curve" concept of computing power from wind speed but with the corrections of the other features included. As we mentioned before, creating the training dataset for this regression implied cutting out many samples, which could go against the philosophy of using recurring neural networks. However, a good performance in the first simple NN attempt would make these cuts irrelevant.

The architecture of the simpler regression model is built using PyTorch [29] ModuleList class, which allows us to build the NN with generality, determining the hyperparameters set to define it by using Bayesian optimization with Weights & Biases [30]. The BiLSTM will be defined using the LSTM class from PyTorch, with dropout and bidirectional parameters set as true. These NNs will have the following hyperparameters:

Both architectures:


#### ANN:

• Number of fully connected layers.

#### BiLSTM:


Other hyperparameters related to their training are:


As previously shown, Power is not Gaussian distributed. This may affect our predictions as their deviations from the real data may not be normally distributed either. However, we need them to be if we want to spot anomalies in the prediction. Thus, we will establish a custom metric that ensures this requisite is met, with the following definition:

$$metric = \left(r^2\right)^{100 \times Validation\text{ }lau} \tag{1}$$

where *r* 2 is the coefficient of determination in a Q-Q plot representing the deviations on the prediction with respect to the real "Power" feature in the well-behaved validation dataset against the 45° line. The data included in this Q-Q plot is extracted by modifying the StatsModels library [31] so as to retrieve the slope and ordinate at the origin of the "s" line when building the plot. By requesting the maximization of this custom metric, since 0 < *r* 2 < 1, we are ensuring that the prediction errors in the normal dataset follow a Gaussian distribution and, at the same time, NNs that present a high validation loss are penalized, with the loss being the Mean Square Error.

After the optimization of these NNs, a prediction interval based on the standard deviation of these prediction errors will be computed for both the full validation and test sets. Then, a t-Student distribution prediction interval based on 10 runs with the Dropout layers in training mode (at evaluation time) and a full Monte Carlo prediction interval with 100 runs of Dropout in training mode will be added. These last two intervals have the advantage of changing sample by sample, instead of being completely general as in the case of the standard deviation. We will define all three prediction intervals at a 95% confidence level, as we want to prove there is statistical significance in when these anomalies appear with respect to a fault.

ANOVA tests will be performed on both the full validation and test datasets, with the hypothesis of anomalies having different distributions according to the Remaining Useful Life (RUL) or time-to-fault with respect to the total anomalies recorded for a time series slice between two faults and the total samples contained in said slice.

Regarding the classification task, our choice for performing logistic regression on the data as normal or prefault has been a Tensorflow [32] based CNN architecture. Since we are aware of the challenge inherent to fault forecasting, we decided to directly implement an architecture that gets higher features but, otherwise, it is quite simple: we will have four one-dimensional convolutional layers with a max pooling layer after each two of them. We will train minimizing validation loss (binary crossentropy), despite it not being the focus of interest in our classification though.

We will evaluate a list of metrics. The most usual in classification tasks: precision, recall and f1-score, will of course be evaluated, according to:

$$\begin{aligned} Precision &= \frac{TP}{TP + FP} & Recall &= \frac{TP}{TP + FN} \\ F1 - score &= 2 \frac{Precision \times Recall}{Precision + Recall} \end{aligned} \tag{2}$$

where TP, FP and FN are "True Positives", "False Positives" and "False negatives" correspondingly.

However, our hyperparameter optimization goal will be to maximize the Matthews correlation coefficient (MCC), since it is more complete by taking into consideration "True Negative" (TN) predictions, as shown in Eq. 3. This is a very appropriate metric four our problem, as we have no previous knowledge of the correct labeling, and we are facing a dataset that can become greatly imbalanced towards the well-behaved or normal label of the turbine.

$$\text{MCC} = \frac{\text{TP} \times \text{TN} - \text{FP} \times \text{FN}}{\sqrt{(\text{TP} + \text{FP})(\text{TP} + \text{FN})(\text{TN} + \text{FP})(\text{TN} + \text{FN})}} \tag{3}$$

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

#### **3. Results**

#### **3.1 Regression with a deep NN**

After 100 runs of a Sweep (term used in Weights & Biases for a search in the hyperparameter space) with Bayesian optimization, the value obtained for the metric shown in Eq. 1 is *metric* = 0.98385. This metric has the double goal of favoring models that are statistically appropriate for anomaly extraction and fit correctly the data. Thus, a metric this close to 1 proves that a NN without a memory cell can fit the multivariate Power curve excellently. In **Figure 3**, we present the Q-Q plot showing how the deviations of our regression from the real "power" values belong to a Gaussian distribution (computed with the well-behaved data of the validation dataset).

Since the model has shown a good fit between Gaussianity and prediction deviations, the next step is to study the whole validation dataset, which includes data close in time to faults, as we want to prove that these data present anomalies. Our statistical study will study the deviations from data and prediction in the following ways: using a Monte Carlo dropout with 100 iterations and establishing a threshold of 1.96 times the standard deviation (95%) of the predictions for each sample; then computing a smaller Monte Carlo dropout sample of only 10 iterations and obtaining the t-Student distribution at a 95% confidence level and using 1.96 times the standard deviation of the global errorin-prediction distribution. These three strategies are presented as they represent and increasing speed of computation. A Monte Carlo dropout consists of setting the dropout layers of our NN architecture in train mode and perform a number of predictions for every sample, which can be quite slow. Therefore, it is interesting to find a lighter process to find an anomaly, such as computing the confidence interval defined by the mentioned t-Student distribution (Eq. 4) with a very limited Monte Carlo sampling.

$$CI\_{t(0.95)}^{\mathbb{F}} = \overline{\mathfrak{X}}\_{prediction} \pm t\_{\omega\_{2,n-1}} \times s \sqrt{1 + \frac{\mathbf{1}}{n}} \tag{4}$$

where *s* is the standard deviation of the *n* samples used (10 in our case) and *t* is the *p*th percentile of the t-Student distribution with *n*-1 degrees of freedom.

#### **Figure 3.**

*Q-Q plot of the deviations between prediction and real data in the well-behaved wind turbine part of the validation dataset.*


#### **Table 1.**

*ANOVA results the hypothesis of anomalies appearing when time is closer to the fault.*

Any prediction exceeding the high or low limits established by these methods will be considered an anomaly according to those particular statistics. Once these anomalies have been pinpointed, our interest moves towards determining if these anomalies arise significantly as the time-to-fault reduces. We present the results of these ANOVA tests in **Table 1** with their statistical significance.

As we can see, the deep NN is successful enough both for performing a regression of the power output accurately and to find statistically significant anomalies. Let us compare now these results with those obtained by training the BiLSTM for regression. This comparison is relevant as both NNs differ greatly in the training and prediction times, being the BiLSTM much slower and with greater need of resources. First of all, we present in **Figure 4** the Q-Q plot of the best model obtained after the Bayesian optimization of the hyperparameters. This model has a metric (defined by Eq. 1) value of *metric* = 0.9732, which is still a good result for our purposes though worse than the previously obtained, with the drawback of going through a much slower training. As seen in the figure, the Q-Q plot does not fit that well the line of gaussianity in the deviation between prediction and real value.

We reproduce the one-way ANOVE tests the same way as with the previous architecture, presenting the results in **Table 2**.

This time, results are not favorable to the null hypothesis. The metric and the Q-Q plot were not as good as with the previous architecture, which may make the anomalies less reliable than those predicted by the simpler ANN, since the deviations in the validation split that only contains good behavior of the turbine are not a good fit into a Gaussian distribution, which is required to obtain reliable anomalies. It is

#### **Figure 4.**

*Q-Q plot of the deviations between prediction and real data in the well-behaved wind turbine part of the validation dataset for the BiLSTM.*

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*


#### **Table 2.**

*ANOVA results the hypothesis of anomalies appearing when time is closer to the fault for the BiLSTM architecture.*

also relevant to remind the computational semantics of data preprocessing at this stage, as the well-behaved data isolation for the training split causes cuts that can affect the memory cells of the architecture. Along this research, LSTMs have proven difficult to manage in terms of reproducibility as determinism is difficult to achieve and we included dropout to have the chance of doing the Monte Carlo sampling of predictions.

Nevertheless, these results are motivating if contradictory, which suggests the need for a way that arbitrates if it is possible to predict a fault and arranging maintenance with enough anticipation. This is where the 1-dim CNN enters. To recap the previous discussion, this is a challenging dataset where feature engineering is not an easy task and "Power" is dominated by the "Wind Speed". The use of convolution takes care of part of the feature engineering and the rolling averages with time windows defined by hyperparameters (a different one for each of the original features) ensure a non-biased feature extraction. This agnosticism is also represented in the CNN architecture, where kernel sizes and filters are defined as hyperparameters too. This search of the hyperparameter space has a dimensionality too high for Bayesian optimization to work, so we will perform an extensive random search.

Since one of our purposes is to prove the convenience of the MCC metric in classification tasks, our figures will present MCC and f1-score, showing that the latter can be in a range that is considered good despite the former being too far from 1. The Matthews correlation coefficient can range from −1 < MCC < 1 and close to zero is a bad fit, though it is considered a good metric when MCC > 0.5 (negative numbers mean that there is anticorrelation). We show the values of MCC and f1-score in **Figure 5** according to the time-to-fault, which we will plot as the number of samples labeled as 1 (we are performing logistic regression) from the fault backwards in time, which is the anticipation we seek for maintenance.

There are interesting results in this plot. There is a curve where most of the models fall and show a steady increase in both metrics. This increase is explained as more data samples are labeled as 1 or "Prefault", which biases the model making it seemingly more accurate. From the different outliers to this curve, one is very interesting, as it greatly exceeds the curve at an interesting time-to-fault. The metrics for this particular hyperparameters are:

MCC = 0.6418

F1-Score = 0.8701

Time-to-fault = 197 samples (~1 day 9 h).

It must be said that at this time-to-fault the labeling of the whole dataset is very balanced, being the "prefault" (1) samples around a 48% of the data. As it has been a very specific result without nearby hyperparameter space realizations with similar metrics, we then fix the time-to-fault and scan the remainder of hyperparameter space to determine if there are more models converging into the metrics found. The

#### *Computational Semantics*

**Figure 5.** *MCC (left) and F1-score (color) with respect to time-to-fault (as samples labeled 1).*

**Figure 6.** *MCC with respect to the F1-score for a fixed time-to-fault of 197 samples (colors represent validation loss).*

#### *Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

results, shown in **Figure 6**, prove that after 400 random hyperparameter runs, a good f1-score is commonly achieved but it is indeed complicated to reach a good MCC metric (worse metrics than shown are cut from the figure).

Thus, it is proven that, despite the difficulty of this endeavor, it is possible to train a 1-dimensional CNN to reliable predict wind turbine faults causing downtime and schedule maintenance with at least more than a day of anticipation. In **Figure 5** we can see other time-to-fault values that are promising too, though it is understandably more complicated to predict an error the further back we move on time. For this purpose, it is highly recommendable to use the Matthews correlation coefficient instead of relying solely in the f1-score, as it is a more complete metric including True Negative samples in its computation.

#### **4. Conclusions**

Wind turbine datasets entail a high complexity for succeeding in the task of predictive maintenance. Through this chapter, we have proven that an artificial neural network can be built so as to train a regressor for the power output of the wind turbine according to the other features, where the main influence is the wind speed traditionally, as in theoretical power curves. The smart definition of metrics is the best ally to obtain a model that fits the target variable with high accuracy and can be used for computing anomalies, which requires deviations in the prediction of "Power" to fall in a Gaussian distribution for the wind turbine regime considered normal or without any fault in the time vicinity. Besides, we have shown that training a LSTM increases the difficulty of achieving these goals, which requires more computing time and resources to achieve a subpar result compared to that of the simpler NN architecture.

In addition to the regression of the power, we have developed a one-dimensional CNN architecture capable of, after an extensive hyperparameter optimization, classify any new registered data sample as a normal state or indicative of an impending fault that will cause downtime, with at least 1 day and 9 hours of anticipation. This was our main purpose, and for its achievement it has been necessary to solve both feature engineering through convolution and, as the original data is not labeled (only faults are annotated once they happen), finding the correct annotation of samples through the optimization with a powerful metric that is robust against class imbalance, such as the Matthews correlation coefficient.

To sum up, the problem of predictive maintenance without the aid of domain expertise or annotated training data can be solved with a patient hyperparameter optimization and the evaluation of strategic metrics powerful enough to train our neural networks correctly for the task.

#### **Acknowledgements**

The authors thank Enterprise Ireland and the European Union's Horizon 2020 research and innovation programme for funding under the Marie Skłodowska-Curie grant agreement No. 713654.

### **Conflict of interest**

The authors declare no conflict of interest.

### **Author details**

Cristian Bosch\* and Ricardo Simon-Carbajo CeADAR, University College Dublin, Dublin, Ireland

\*Address all correspondence to: cristian.boschserrano@ucd.ie

© 2023 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

#### **References**

[1] WindEurope Business Intelligence; Wind energy in Europe in 2018. Trends and Statistics. windeurope.org. Feb 2019

[2] Taylor M, Ralon P, Al-Zoghoul S, Jochum M, Gielen D. Renewable Power Generation Costs in 2021. IRENA; 2022

[3] Feng Y, Tavner PJ, Long H. Early experiences with UK round 1 offshore wind farms. Proceedings of the Institution of Civil Engineers - Energy. 2010;**163**:167-181

[4] Siami-Namini S, Tavakoli N, Namin AS. The performance of LSTM and BiLSTM in forecasting time series. IEEE International Conference on Big Data (Big Data). 2019;**2019**:3285-3292

[5] Bosch C, Simon-Carbajo R. Machine learning for wind turbine fault prediction through the combination of datasets from same type turbines, floating offshore energy devices: GREENER. Materials Research Proceedings. 2022;**20**:45-57

[6] Hameed Z, Hong YS, Cho YM, Ahn SH, Song CK. Condition monitoring and fault detection of wind turbines and related algorithms: A review. Renewable and Sustainable Energy Reviews. 2009;**13**:1-39

[7] Breteler D, Kaidis C, Tinga T, Loendersloot R. Physics based methodology for wind turbine failure detection, diagnostics & prognostics. EWEA. 2015;**2015**:1-9

[8] Zhao H, Liu H, Hu W, Yan X. Anomaly detection and fault analysis of wind turbine components based on deep learning network. Renewable Energy. 2018;**127**:825-834

[9] Mazidi P, Bertling-Tjernberg L, Sanz-Bobi MA. Performance analysis and anomaly detection in wind turbines based on neural networks and principal component analysis. In: 12th Workshop on Industrial Systems and Energy Technologies (JOSITE2017). Madrid: comillas.edu; 2017

[10] Felgueira T, Rodrigues S, Perone CS, Castro R. The impact of feature causality on Normal behaviour models for SCADAbased wind turbine fault detection. ICML 2019 Workshop - Climate Change, How Can AI Help; arXiv:1906.12329; 2019

[11] Shihavuddin ASM, Chen X, Fedorov V, Christensen AN, Riis NAB, Branner K, et al. Wind turbine surface damage detection by deep learning aided drone inspection analysis. Energies. 2019;**12**:676

[12] Anantrasirichai N, Bull D. DefectNET: Multi-class fault detection on highly-imbalanced datasets. IEEE International Conference on Image Processing (ICIP). 2019;**2019**:2481-2485

[13] Mokhtari A, Belkheiri M. Fault diagnosis of a wind turbine benchmark via statistical and support vector machine. International Journal of Engineering Research in Africa. 2018;**37**:29-42

[14] Zhang DH, Qian LY, Mao BJ, Huang C, Huang B, Si YL. A datadriven Design for Fault Detection of wind turbines using random forests and XGboost. IEEE Access. 2018;**6**:21020-21031

[15] Cheng FZ, Wang J, Qu LY, Qiao W. Rotor current-based fault diagnosis for DFIG wind turbine drivetrain gearboxes using frequency analysis and a deep classifier. IEEE Industry Applications Society Annual Meeting. 2017;**54**(2017):1062-1071

[16] Chauhan S, Vig L. Anomaly detection in ECG time signals via deep long short-term memory networks. IEEE International Conference on Data Science and Advanced Analytics (DSAA). 2015;**2015**:1-7

[17] Maya S, Ueno K, Nishikawa T. dLSTM: A new approach for anomaly detection using deep learning with delayed prediction. International Journal of Data Science and Analytics. 2019;**8**:137-164

[18] Bakhtawar Shah M. Anomaly Detection in Electricity Demand Time Series Data. Sweden: KTH Royal Institute of Technology; 2019

[19] Malhotra P, Vig L, Shroff G, Agarwal P. Long short term memory networks for anomaly detection in time series. In: 23rd European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. Bruges: ESANN; 2015. pp. 89-94

[20] Zhu L, Laptev N. Deep and confident prediction for time series at Uber. In: 2017 IEEE International Conference on Data Mining Workshops (ICDMW). New Orleans, LA, USA: IEEE; 2017. pp.103-110

[21] Hill DJ, Minsker BS. Anomaly detection in streaming environmental sensor data: A data-driven modeling approach. Environmental Modelling and Software. 2010;**25**:1014-1022

[22] Zheng L, Xue W, Chen F, Guo P, Chen J, Chen B, et al. A fault prediction of equipment based on CNN-LSTM network. IEEE International Conference on Energy Internet (ICEI). 2019;**2019**:537-541

[23] Al Qasem O, Akour M. Software fault prediction using deep learning algorithms. International Journal of Open Source Software and Processes. 2019;**10**:1-19

[24] He J, Wang J, Dai L, Zhang J, Bao J. An adaptive interval forecast CNN model for fault detection method. In: 15th IEEE International Conference on Automation Science and Engineering, (CASE). IEEE; 2019. pp. 602-607

[25] Guyon I, Bennett K, Cawley G, Escalante HJ, Escalera S, Tin Kam H, et al. Design of the 2015 ChaLearn AutoML challenge. In: 2015 International Joint Conference on Neural Networks (IJCNN). Killarney, Ireland: IJCNN; 2015. pp. 1-8. DOI: 10.1109/IJCNN.2015.7280767

[26] Guyon I, Sun-Hosoya L, Boullé M, Escalante HJ, Escalera S, Liu Z, et al. Analysis of the AutoML challenge series 2015-2018. In: Frank Hutter LK, Vanschoren J, editors. Automated Machine Learning. Cham: Springer; 2019. pp. 177-219

[27] Stetsenko P. Machine Learning with Python and H2O; docs.H2O.ai; 2020. Available from: https://docs.h2o.ai/h2o/ latest-stable/h2o-docs/faq/general.html#iam-writing-an-academic-research-paperand-i-would-like-to-cite-h2o-in-mybibliography-how-should-i-do-that

[28] Stetsenko P. Machine Learning with Python and H2O. 2020. Available from: https://docs.h2o.ai/h2o/ latest-stable/h2o-docs/faq/general. html#i-am-writing-an-academicresearch-paper-and-i-would-like-tocite-h2o-in-my-bibliography-howshould-i-do-that)

[29] Paszke A, Gross S, Massa F, Lerer A, Bradbury J, Chanan G, et al. PyTorch: An imperative style, high-performance deep learning library. In: Wallach H, Larochelle H, Beygelzimer A, Garnett R, editors. Advances in Neural Information Processing Systems 32. Vancouver, Canada: Curran Associates Inc; 2019. pp. 8024-8035

*Perspective Chapter: Computation of Wind Turbine Power Generation, Anomaly Detection… DOI: http://dx.doi.org/10.5772/intechopen.109698*

[30] Biewald L. "Experiment Tracking with Weights and Biases," Weights & Biases. 2022. [Online]. Available from: wandb.com

[31] Seabold S, Perktold J. Statsmodels: Econometric and Statistical Modeling with Python. In: Proceedings of the 9th Python in Science Conference; SciPy; 2010

[32] Abadi M, Agarwal A, Barham P, Brevdo E, Chen Z, Citro C, et al. TensorFlow: Large-scale machine learning on heterogeneous distributed systems. 2016. ArXiv, abs/1603.04467

### *Edited by George Dekoulis and Jainath Yadav*

This book analyzes the application of computer science and artificial intelligence (AI) techniques in the semantics' analysis for linguistics, classical studies, and philosophy. Similar techniques can be implemented to incorporate the fields of education, psychology, humanities, law, maritime, data science and business intelligence. The book is suitable for the broader audience interested in the emerging scientific field of formal and Natural Language Processing (NLP). The significance of incorporating all aspects of logic design right at the beginning of the creation of a new NLP system is emphasized and analyzed throughout the book. NLP and AI systems offer an unprecedented set of virtues to society. However, the principles of ethical logic design and operation of primitive to deep learning NLP products must be considered in the future, even via the preparation of legislation if needed. As law applications are already taking advantage of the techniques mentioned, the manufacturers should apply the laws and the possible knowledge development of the NLP products could even be monitored after sales. This will minimize the drawbacks of implementing such intelligent technological solutions. NLP systems are a digital representation of ourselves and may even interact with each other in the future. Learning from them is also a way to improve ourselves.

Published in London, UK © 2023 IntechOpen © metamorworks / iStock

Computational Semantics

Computational Semantics

*Edited by George Dekoulis* 

*and Jainath Yadav*