Akademik Veri Yönetim Sistemi
IEEE Access, 2026 (SCI-Expanded, Scopus)
Today, by analysing real-time operating data from mechanical systems, the failure condition of the mechanical components of the system can be determined. However, most of the available diagnostic techniques are time-consuming and usually performed by specialists. In this work, we argue that classification of spectrum images obtained from acoustic signals using artificial intelligence (AI) algorithms is a powerful approach for fault diagnosis of electric machines that can be used even in noisy environments. For this purpose, we tried to find the most suitable AI approach. In the study, datasets consisting of audio signals containing background noise representing normal and abnormal operating conditions of three different electrical machines were used. The spectrogram images obtained from the noisy audio signals using Short-Time Fourier Transform (STFT) are classified as normal or abnormal states of the machines using various Convolutional Neural Networks (CNN). Different Machine Learning (ML) approaches were also used to compare the results. While both CNN-based and ML-based methods achieve favorable performance, the results indicate that CNNs offer higher accuracy in distinguishing healthy and faulty machine conditions from noisy acoustic data.