The object of this study is the vibration signals received from engines with existing defects. The problem that was solved within the framework of this work arises from the need to construct an accurate and reliable system of prognostic diagnostics, capable of automatically recognizing malfunctions in electric motors, despite the influence of external noises, complex operating conditions, and the similarity of characteristics of signals of various types of defects. The essence of the results is the devised methodology, which includes several stages of vibration signal processing and the use of a convolutional neural network (CNN) for the identification and classification of engine states. At the first stage, the signal is processed in the time domain, in which its main characteristics are analyzed. The signal is then transformed into the frequency domain using a Fast Fourier Transform (FFT) to extract its spectral components. To obtain a more informative representation of the signal, the shorttime Fourier transform (STFT) is applied, which makes it possible to obtain a time-frequency characteristic in the form of a spectrogram. The resulting spectrograms represent a vibration signal in a form suitable for processing by a convolutional neural network, which performs their further analysis. The use of CNN as the main analysis tool allowed us to achieve high results in the classification of engine states. According to the results of experiments, the model showed 100 % accuracy in detecting various types of engine malfunctions, including the most difficult to diagnose conditions. This high level of accuracy is due to the neural network’s ability to efficiently process spectrograms and detect hidden patterns in the data. In addition, the application of STFT ensured the preservation of critical time-frequency information that is not available for use with only conventional FFT. The main advantage of the proposed approach is its versatility and adaptability to different types of engines and malfunctions. The methodology can be used under industrial conditions for automated monitoring of equipment condition. This makes it possible to accidentally detect malfunctions, prevent emergencies, reduce maintenance costs, and increase the overall reliability of the equipment. The proposed approach is particularly useful in applications in which high diagnostic accuracy and fast response to engine state changes are required Keywords: predictive maintenance, machine learning, vibration analysis, frequency analysis, neural networks
Loading....