Machine learning-based wear fault diagnosis for marine diesel engine by fusing multiple data-driven models

Abstract

Wear fault is one of the dominant causes for marine diesel engine damage which significantly influences ship safety. By taking full advantage of the data generated in engine operation, machine learning-based wear fault diagnostic model can help engineers to determine fault modes correctly and take quick action to avoid severe accidents. To identify wear faults more accurately, a multi-model fusion system based on evidential reasoning (ER) rule is proposed in this paper. The outputs of three data-driven models including an artificial neural network (ANN) model, a belief rule-based inference (BRB) model, and an ER rule model are used as pieces of evidence to be fused in decision level. In this paper, the fusion system defines reliability and importance weight of every single model respectively. A novel method is presented to determine the reliability of evidence by considering the accuracy and stability of every single model. The importance weight is optimized by genetic algorithm to improve the performance of the fusion system. The proposed machine learning-based diagnostic system is validated by a set of real samples acquired from marine diesel engines in operation. The test results show that the system is more accurate and robust, and the fault tolerant ability is improved remarkably compared with every single data-driven diagnostic model.