DAMAGE PATTERNS RECOGNITION STUDY BASED ON ADAPTIVE PROBABILISTIC NEURAL NETWORK

An adaptive probabilistic neural netwok (APNN) patterns recognition method is proposed based on the traditional probabilistic neural network (PNN). The ranges of σ parameters are estimated preliminarily by the Gap-Based method and then the genetic algorithm (GA) is adopted to optimize them. In the case study, damage patterns recognition of a two-span bridge benchmark problem proposed by Bridge Health Monitoring Committee is carried out by APNN which selects energy feature vectors of structure dynamic response signal in the sweep sine shaker test and the traffic excitation simulation test as the input variables and uses wavelet packet anslysis for data processing. The results indicate that APNN not only has high identification accuracy and strong noise resistance but also can improve net's learning efficiency by automatic feature selection and dimensionality reduction.