Enhanced engine misfire diagnosis through integration of vibration and acoustic emission signals using artificial neural networks

Abstract An accurate Engine Misfire Detection diagnosis ensures the engine runs well and reduces emissions. An ANN has successfully combined vibration and acoustic emission (AE) signals to improve the detection of misfires in engines. The vibration and AE signals obtained during normal use and with individual cylinder misfires allowed 20 time- and frequency-domain statistical features such as (range, skewness, kurtosis, etc.) to be extracted. Data was acquired systematically, and all parameters used to gather information were standardized for each engine condition. After all the features were put into a complete matrix, the selection algorithms chose the most essential ones for classifying the data. Furthermore, optimizing different hyperparameters multiple times using Bayesian Optimization by MATLAB Classification Learner helped the ANN deliver more accurate results with noise and vibration data. Using two data types at the network input (AE and vibration) helped the ANN model achieve a better accuracy (94.11%) than when only one signal was used for recognition. Using several sensors has increased accuracy since AE data may record information about combustion events that vibration data might miss. Multi-signal methods in engine diagnostics were proven effective, providing a pathway for predictive maintenance of IC vehicles.