Marine diesel engine piston ring fault diagnosis based on LSTM and improved beluga whale optimization

Abstract

The operational state of piston rings in marine diesel engines significantly influences the overall performance of the machinery. However, traditional data-driven diagnosis methods have problems with relying on manual feature extraction and failing to adequately leverage the temporal characteristics inherent in fault vibration signals. Therefor a fault diagnosis method based on long short-term memory neural network (LSTM) optimized by the improved beluga whale optimization algorithm (IBWO) is proposed in this paper. The LSTM process vibration signals, leveraging their gating mechanism for temporal feature extraction before classification via softmax. Setting optimal combinations of hidden layers and learning rates is difficult due to complexity and lengthy training times, making parameter optimization a significant challenge. The beluga whale optimization (BWO) algorithm for parameter optimization is employed to address this. Additionally, to reduce the risk of convergence to local optima, the balance factor is improved by replacing the linear function with a nonlinear function in the original algorithm. Finally, IBWO-LSTM is compared with BWO-LSTM, FOA-LSTM, PSO-LSTM and LSTM. Experimental validation shows that IBWO-LSTM outperforms BWO-LSTM, FOA-LSTM, PSO-LSTM, and standard LSTM, with an average accuracy higher than 90 %. Therefore, the IBWO-LSTM demonstrates better fault identification accuracy, providing a more precise solution for marine diesel engine piston ring fault diagnosis.