**3.2. Algorithm of QFD-based knowledge base for product configuration**

• Decision rules—which contain expressions such as IF *x1* is *F1* **and/or** *x2* is *F2* **and/or** … *xn* is *Fn*, THEN *y* is *P* where *x1, x2, ….. xn, y* denotes objects or attributes and *F1, F2, ……. Fn*, *P* denote values. Decision rules describe both information elements (expressions) and relations between them, and therefore, a set of such rules (*r*) defines a knowledge base:

• Decision trees—which are graph representations of the decision process. The inspection of the condition in the decision path starts from the beginning node called the root and ends

• Frames are used when information units are characterized by many important features. The structure of a simple frame contains three different lines: a heading with the frame name, a pointer to another frame with appropriate relation, and slots defining attribute

• Semantic networks capture knowledge as a graph, in which nodes represent pieces of information (objects, concepts, or situations in the problem domain), and the arcs represent

• Artificial neural networks (ANNs) are inspired by neurons in the brain and have become a popular knowledge representation useful for learning [35]. Among many kinds of ANNs, feed-forward ones are widely used by researchers who apply them as a tool for data classification or as a predicator. The idea of ANN usage is to create a learning set, which includes data characterized by input and output features. During training, ANNs create a model which is able to transform input features into output features of a data set. If the predicted or classified

data depends on many variables (features), ANNs are a convenient tool for analyses.

• Case-based reasoning (CBR), in which the problem-solving method is focused on finding the solution in the base of examples (cases). The case which has been found will be adapted to the new usage. This method is applied when knowledge is presented as a description of cases.

Knowledge can take many forms, and it is necessary to identify the kind of knowledge repre-

In the presented customized product configuration QFD-KB, the following methods of

• Procedural knowledge used for identifying the product features recognized by the cus-

• Artificial neural network (ANN), used for assessing the missing manufacturing process

The data and knowledge generated and used during manufacturing may be related to products, machines, processes, materials, inventories, maintenance, planning and control, assem-

sentation method which is the most suitable for solving a particular problem.

tomer and identifying the product features recognized by the producer.

• Case-based reasoning (CBR), applied for identifying product alternatives.

• Declarative knowledge applied to define the evaluation rules.

*KB* = {*r*1*, r*2*,…*}.

names and values.

in the leaves which give the decision.

56 Product Lifecycle Management - Terminology and Applications

relations or associations between them.

knowledge representation were used [32]:

bly, logistics, performances, etc. [33].

parameters.

Enterprises develop data bases to store different types of data, e.g., data orders, codes of products, technical documentation related to products and the manufacturing processes, and product and process failure data.

Taking into consideration categories mentioned above, product configuration needs information related to customer requirements, product use circumstances, needed product characteristic analyzed from the functional point of view, product portfolio, parts characteristics, and manufacturing process characteristics.

The problem of determining product configuration can be structured according to the decision method presented in **Figure 3**. The presented approach developing web-based selection system was described by Gibson et al. [36].

Product configuration is divided into three levels including product-level configuration, component-level configuration, and manufacturing parameter-level configuration [37]. These three levels can be developed with the use of QFD series of matrices.

In the algorithm of QFD-KB for product configuration presented in **Figure 4** [32], the methods of knowledge representation such as rules from an expert, case-based reasoning and neural networks were applied.

Product offer preparation requires information regarding product portfolio offered by the enterprise and an evaluation of differences between customer requirements and the offered

**Figure 3.** A decision model for product configuration.


**Figure 4.** Algorithm of QFD-KB for product configuration.

products. Customer service department staff should know how the product characteristics needed by the customer are different from the product characteristics offered by the enterprise and what kind of changes it is possible to implement in the product.

Product offer preparation needs a product requirement analysis, which includes analyzing product functions, reliability, safety, environment, packaging, transportation, storage, etc.

The decision problem solved with the use of QFD-KB for product configuration is how to choose and evaluate the right product from the product portfolio and adopt it to particular customer needs. The knowledge needed to solve this problem could origin from, e.g., experienced staff, databases, and documentation.

Possible data sources used in product configuration are presented in **Figure 5**.
