An engine-fault-diagnosis system based on sound intensity analysis and wavelet packet pre-processing neural network. (September 2020)

• Record Type: Journal Article
• Authors: Wang, Y.S.; Liu, N.N.; Guo, H.; Wang, X.L.
• Is Part Of: Engineering applications of artificial intelligence. Volume 94(2020)
• Publication Date: 2020-09
• Subjects: Engine fault diagnosis -- Sound intensity analysis -- Noise-based fault recognition -- Wavelet package analysis (WPA) -- Artificial neural network (ANN); Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2020.103765
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Based on the techniques of sound intensity analysis, incomplete wavelet packet analysis (WPA) and artificial neural network (ANN), a WPA pre-processing method for noise-based engine fault diagnosis (EFD), so-called WPA–ANN model, is presented in this paper. The noises of an EFI gasoline engine under normal and fault states are measured and their contours of sound intensity level (SIL) are calculated by interpolation approach to initially investigate the possibility of a SIL-based EFD. Furthermore, an incomplete WPA model, which consists of a five-level discrete wavelet transform (DWT) and a four-level WPA, is developed and applied to the measured noise signals for extracting fault features of the engine, as is a multi-layered ANN model for engine failure classification by using the extracted features of the noises. To verify the proposed approach, the WPA–ANN model is extended to recognize other noise-related faults of the engine. The results suggest that the noise-based WPA–ANN models are effective for engine fault diagnosis. Due to its time–frequency characteristics and pattern recognition capacity, the WPA–ANN can be used to process both the stationary and nonstationary signals. In view of the applications, the proposed WPA–ANN model can be directly used in vehicle EFDs, and may be extended to other sound-related fields for failure diagnosis in engineering. Highlights: A novel noise-based method for engine-fault diagnosis (EFD) is proposed. An incomplete WPA model is established for sound intensity feature extraction. A three-layer ANN is specially designed for fault feature recognition of engine. An empirical formula for EFD decision is defined by setting threshold values. Predicted EFD results suggest a very good accuracy of the WPA–ANN model.