A Trimaran Structure Damage Identification Method Based on Machine Learning

A trimaran structure damage identification method based on machine learning is proposed. In the damage identification method, the wavelet transform and an improved adaptive threshold function are combined to realise wavelet filtering (WF), and the structural curvature difference based on fractal box dimension is used as a critical index to train a back-propagation (BP) neural network for identifying the damage degree and location. The improved particle swarm optimisation (IPSO) is adopted to develop an IPSO-WF-BP strategy for synergistically optimising the BP neural network and the wavelet filter. In addition, to make the local structure response simulation more realistic, the monitoring load data from a trimaran model test are extracted and combined with the finite element method to simulate multiple damage occurrences under different hull deformations. Comparative analysis has finally demonstrated the reliability of the IPSO-WF-BP model in different wave environments and noise signals. This damage identification method will provide technical support for auxiliary decision-making and damage diagnosis of intelligent trimarans.