Comparative Analysis of MFCC and Mel-Spectrogram Features in Pump Fault Detection Using Autoencoder

Pump maintenance plays a pivotal role in industrial operations, where timely detection of faults is key to avoiding costly downtimes. This research explores the influence of two audio feature extraction techniques, Mel-Frequency Cepstral Coefficients (MFCC) and Mel-spectrograms, on the effectiveness of autoencoders in detecting pump faults. Using the Malfunctioning Industrial Machine Investigation and Inspection (MIMII) dataset, the study trained an auto encoder on normal pump sounds and evaluated it against a balanced test set of normal and anomalous sounds. The results present the superiority of Mel-spectrograms over MFCCs in various performance metrics. These findings emphasize the critical role of feature selection in autoencoder-based pump fault detection, marking a significant stride towards optimizing predictive maintenance strategies.