Diesel Engine Fault Diagnosis Based on Convolutional Autoencoder Using Vibration Signals

The diesel engine is the power source and core equipment of large mechanical systems such as ships. Thus, the engine must be maintained in good working conditions for the smooth operation of the mechanical system. Vibration signals of diesel engines are caused by the actions of piston slapping and valves, from which fault information can be obtained. As the vibration characteristics are more obvious during acceleration or deceleration, the signals can be used for speedily and accurately diagnosing the fault state of the diesel engine. In this study, vibration signal diagnosis methods for the diesel engine were developed. The methods were based on the convolutional autoencoder. The auto-encoder was trained using the vibration signals from normal working states, and the reconstruction error was used for fault diagnosis. Subsequently, the performances of three autoencoders and stacked autoencoders for fault detection and classification were analyzed and compared. The results showed that the stacked autoencoder was the most effective in fault diagnosis and classification. The proposed method can be applied to fault detection and classification for diesel engines using vibration signals.