Multi-mode fusion BP neural network model with vibration and acoustic emission signals for process pipeline crack location

Vibration fatigue occurs easily in offshore platform pipelines that have been in service for several years. Timely damage detection and the detection of fatigue cracks are of great significance to ensure safe production. To accurately locate pipe cracks, a multi-mode fusion crack location model is proposed in this paper. The model combines two location methods based on vibration signals and acoustic emission (AE) signals through a backpropagation (BP) neural network. The natural frequencies of a large number of crack models at different locations were obtained by combining vibration modal test and finite element simulation results. A large number of acoustic emission signal time differences of cracks at different positions were obtained by simulating the acoustic emission signals at the initiation and propagation stages of cracks with the pencil-lead break method. The multi-mode fusion model was trained by experimental and simulation data, and the positioning effect of the model was verified. The results showed that training the neural network based on multi-mode data could effectively avoid the over-fitting of the network and improve the training effect. Compared with the method based on single information or only a weighted average method, this model could effectively improve the accuracy of crack location and reduce the fluctuations of the location errors, and the location results were more reliable. •Establishing backpropagation (BP) neural network based on vibration to determine the location of cracks.•Establishing backpropagation (BP) neural network based on acoustic emission to determine the location of cracks.•Proposing a multi-mode fusion crack location model to accurately locate pipe cracks.•Comparing with single-information network, the mean square error of the multi-mode network is significantly lower.