Generalized composite multiscale improved diversity entropy and its application to diesel injector fault feature extraction and diagnosis

The diesel injector is the core component of marine diesel engines, and its health monitoring is one of the keys to ensuring the efficient and reliable operation of diesel engines and even ships. However, extracting effective feature information from nonlinear dynamic signals is the difficulty in realizing intelligent fault diagnosis of diesel fuel injectors. The existing diversity entropy (DE) is an advanced algorithm for extracting fault features. Still, it has the problem of failing to obtain sufficiently effective fault features, thus affecting diagnostic accuracy due to its inherent defects. To solve the above problem, a new generalized composite multiscale improved diversity entropy (GCMIDE) is first proposed, and the comparative advantages of the proposed entropy in terms of consistency, robustness, and computational efficiency are verified with common noise-simulated signals. GCMIDE is further utilized as an excellent means of feature extraction, and a new injector fault diagnosis method, GCMIDE-Relieff-SSA-BP, is proposed to meet the needs of efficiency and practicality in fault diagnosis. Finally, the method is adopted for the diagnostic experiments of diesel fuel injectors with different fault types and different fault degrees for full validation. The experimental results all indicate that, in comparison with six commonly used entropy-based fault diagnosis methods, the proposed method has the best feature extraction performance and fault diagnosis accuracy under a different number of features or different training sample ratios. This provides an effective tool for fault diagnosis of injectors and even intelligent operation and maintenance of diesel engines.