CLASSIFICATION OF BEARING FAILURES ON AN EXPERIMENTAL BENCH: COMPARISON BETWEEN CLASSICAL MODELS AND QUANTUM MACHINE LEARNING

Vibration analysis is one of the main techniques used in condition monitoring of rotating machinery. In this context, this work evaluates the performance of classical and quantum-classical hybrid models in the classification of bearing failures from signals obtained on an experimental bench. Two operating conditions were considered: aligned and balanced (AB) and misaligned and unbalanced (DD). The signals were segmented, and statistical characteristics were extracted to compose the input set of the models. A model based on Multi-Layer Perceptron (MLP) and approaches with Quantum Machine Learning were evaluated, using angle encoding and amplitude encoding. In AB condition, the models showed an accuracy of up to 0.91. In DD, a significant reduction was observed, with accuracy of around 0.71 for MLP and angle encoding, and 0.62 for amplitude encoding. The results indicate a strong influence of operating conditions on the performance of the models, with no significant gains from quantum approaches in this scenario.