**5. Soft computing**

Soft computing includes a series of strategies, that aim to exploit tolerance for imprecision, uncertainty, and partial truth to establish robustness and flexibility along with low solution cost. Major soft computing techniques and topics are summarized in **Figure 11**. In bridge monitoring, different soft computing techniques have been utilized for damage detection and system identification. The followings present the most commonly used soft computing applications for bridge monitoring.

Applications of data mining in SHM have recently been reported [44, 45], though due to the novelty of data mining, the application of data mining in SHM is still controversial and is not as much as expected. Therefore, it seems necessary studies are required to advance the data mining application in SHM. One of the most widespread systematic data mining tools is Cross Industry Standard Process for Data Mining (CRISP-DM), which was introduced by a consortium of several companies such as National Cash Register (NCR) System Engineering Copenhagen from the USA and Denmark, Integral Solutions Ltd. (ISL)/SPSS from the USA, Daimler Chrysler AG from Germany and an insurance corporation in the Netherlands, called OHRA [46, 47]. A generalized form of CRISP-DM based on the SHM system has been developed by the authors of [48]. The proposed data mining-based damage identification approach consisted of six new defined stages: target identification, data exploration, database construction, pattern identification, pattern evaluation, and knowledge extraction. At the first stage, specimen description and experimental setup have been presented using a slab-on-girder bridge structure (see **Figure 12**). In the second stage, vibration data were collected from experimental modal analysis of healthy (baseline) and damaged

*DOI: http://dx.doi.org/10.5772/intechopen.104905 Introduction to Monitoring of Bridge Infrastructure Using Soft Computing Techniques*

> **Figure 11.** *Soft computing techniques.*

structures. Analysis of collected data was done in the third stage to generate datasets using the first four flexural modes and all corresponding mode shape values of doublepoint damage cases as inputs for next stage, which was pattern identification. In this stage, Artificial Neural Network (ANN)-based Imperial Competitive Algorithm (ICA) was employed to train datasets and build a model for damage identification of the structure. Then, in the fifth stage, model performance was assessed using evaluation methods and ANN approach. Finally, the last stage extracted valuable knowledge and damage identification.

A multi-layer perceptron ANN-based damage detection of truss bridge joints was proposed by Mehrjoo et al. [49]. The proposed network was conducted using a single hidden layer and 729 training patterns obtained from the first five modes of the structure. It should be noted that the stiffness reduction of members was considered as damage in this study. **Figure 13** illustrates the fatigue damage in truss bridge joints. Around 40% of these types of structures usually experience fatigue damage in their joints during their service life. Louisville bridge was used as a test specimen for this research. The standard back-propagation (BP) algorithm with 75,000 epochs and the root mean square (RMS) was employed to train the networks and evaluation of patterns, respectively. The average error of the predicted damage percentage value for all joints was 1.28%. The authors summarized that their presented methodology was effective in system identification of truss bridges.

**Figure 12.** *Experimental test of the lab-scale bridge [37].*

**Figure 13.** *Fatigue damage in truss bridge joints in Louisville bridge truss [49].*

## *DOI: http://dx.doi.org/10.5772/intechopen.104905 Introduction to Monitoring of Bridge Infrastructure Using Soft Computing Techniques*

A bridge monitoring scheme using operational modal analysis was developed by [50] through a hybrid Fuzzy Krill Herd approach. The proposed fuzzy logic-based SHM diagram is displayed in **Figure 14a**. Two types of bridges, i.e., Banafjäl bridge in Sweden and the Tirehrood bridge in Iran, were considered as test specimens for this paper. The damage scenarios were presented by the output of the fuzzy logic-based SHM approach, as shown in **Figure 14b**. The outcomes revealed the proficiency of the proposed approach in achieving precise knowledge in the existence of noisy data.

Padil et al. [51] proposed a PCA-based non-probabilistic technique. The proposed method was verified using a big Frequency Response Function (FRF) matrix comprising 1200 FRFs with 512 frequency points obtained from intact and damaged simply-supported steel truss bridge model (see **Figure 15a**). To aid the aim, a number of damage scenarios were considered by cutting the structural members, i.e., M1 and

**Figure 14.** *(a) The proposed fuzzy logic-based SHM diagram, and (b) damage scenarios by [50].*

M2 in the main girder bar, and W1 in the web bar, as shown in **Figure 15b**. In this study, it was shown that the results of hybrid PCA were better than traditional PCA.

The application of deep learning in SHM was implemented in [52] by different PCA-based methods, i.e., the deep principal component analysis (DPCA), nonlinear principal component analysis (NLPCA), and kernel principal component analysis (KPCA). The damage-sensitive features of Z24 and Tamar Bridges were used to evaluate the applicability of the aforementioned algorithms (see **Figure 16**). The DPCA showed the best performance.

**Figure 16.**

*(a) Z24 bridge, and (d) Tamar suspension bridge [52].*

**Figure 17.** *SVM results for composite bridge structure using different kernel functions [36].*
