Condition monitoring of internal combustion engines with vibration signals and fault detection by using machine learning techniques

Internal combustion engines are frequently used in transportation, power plants, and in many other applications for industrial purposes. For this reason, it is very important that the maintenance is done systematically and that the faults are detected correctly. In this study, two different methods were used for the detection of the healthy internal combustion engine (H) and faulty internal combustion engines (single-cylinder misfire-F1, two-cylinder misfire-F2). In the first method, classical signal features were extracted from engine vibration measurements and used in the training of artificial neural networks (ANNs) and support vector machine (SVM). In the second method, convolutional neural networks (CNNs), a deep learning method in which features are extracted automatically, are used. Spectrograms of engine vibration signals were used to train pre-trained CNNs with different structures. Spectrograms were obtained by applying short-time Fourier transform (STFT) to vibration signals. The results of GoogleNet and ResNet-50 models trained with spectrograms were compared with the results obtained from models based on ANNs and SVM.