Application of Bayesian-designed artificial neural networks in phase II structural health monitoring benchmark studies

Abstract

This paper presents the results of a study into the use of pattern recognition as a method for detecting damage in structures. Pattern recognition is achieved by the use of artificial neural networks (ANNs), however, these require careful design because the number of hidden layers and the number of neurons in each hidden layer are critical to the ANN's performance. In the current study, a Bayesian model class selection method was employed to select an optimal ANN model class that avoids ad hoc assumptions and subjective decisions in the ANN design. The objective of the research was to provide an extended study of the proposed method using the IASC-ASCE Structural Health Monitoring Phase II Simulated Benchmark Structure. Damage-induced modal parameter changes were used as a pattern feature in damage detection. Analysis showed that the proposed method is able to successfully identify damages in the benchmark structure.