Application of t-SNE and PSO-BPNN for Identification of Penetration State in Ship GMAW

Due to the redundancy of high-dimensional feature vectors extracted from arc sound signal during ship GMAW, it is difficult to effectively identify the penetration state. The method combination of t-distributed stochastic neighbor embedding (t-SNE) algorithm and particle swarm optimization-back propagation neural network (PSO-BPNN) model are proposed in this paper. Firstly, the arc sound experimental system of ship GMAW is established, and then frequency domain feature parameters of different frequency bands and MFCC are extracted. t-SNE, PCA and LPP are used to reduce the dimensionality of the features, respectively. Using PSO algorithm to optimize the weights and thresholds of BPNN, the BPNN converges faster during the training process and improves network performance. Finally, PSO-BPNN and BPNN are used to train and identify three types of dimensionality reduction data. The simulation results show that compared with the other five method combinations, the t-SNE and PSO-BPNN proposed in this paper can more effectively identify partial penetration, full penetration and excessive penetration states, thus providing strong technical support for online monitoring of GMAW penetration state in ship.