A Diesel Engine Misfire Diagnosis Method Based on Envelop Curve and 1D Residual Convolutional Neural Network

To achieve efficient diagnosis of diesel engine misfire faults, an intelligent diagnostic method based on vibration signals and a one-dimensional residual convolutional neural network was proposed. Misfire faults under different operating conditions were injected into the engine test bench, and vibration signals from the cylinder head surface were collected to construct a dataset. To reduce system complexity and cost, a signal similarity analysis method was employed to optimize the vibration sensor layout, achieving a reduction in the number of sensors. Misfire states were classified and identified by the residual convolutional neural network model which was trained using the vibration signal envelope. Experimental results showed that the proposed method achieved an average accuracy of 96.4% under varying operating conditions while reducing sensor usage, showing significant improvement compared to traditional convolutional neural networks（CNN） at 85.2% and deep neural networks（DNN） at 79.7%, which validated the effectiveness of the proposed strategy in engineering applications and provided a cost-effective and high-precision diagnostic approach for diesel engine misfire faults.