Rotating machines play a crucial role in industrial applications, requiring efficient maintenance to ensure their safe and continuous operation. Predictive maintenance is preferred in automated factories and process plants, and vibration analysis is widely used for fault detection. Artificial intelligence, with machine learning and deep learning algorithms, can automate the diagnostic process, reducing maintenance cycles and improving accuracy. These algorithms can learn complex patterns and relationships in the collected data from rotating machines and, by training a model with this data, artificial intelligence can identify characteristic patterns of different types of faults. The main objective of this work is to contribute to the study of the application of artificial intelligence methods in diagnosing faults in rotating machines through vibration signal analysis. The MAFAULDA database, which contains vibration measurements collected in a fault simulation test rig with different machine health conditions, and three classifier algorithms were selected to obtain the model with the best performance in identifying these conditions. A data preprocessing step was performed to use the vibration data as input for the classifier algorithms, including feature extraction, feature standardization, and data balancing. Then, the three classifier algorithms were implemented and evaluated. Two different classification configurations were tested, the first considering 42 classes with sublevels of each defect, and the second more general, with 10 classes. The second configuration achieved the best results: the Extremely Randomized Trees classifier showed the best overall performance with 96% accuracy, followed by Random Forest and Artificial Neural Network with 95% and 93% accuracy, respectively. These techniques demonstrate significant results, highlighting their potential to improve the efficiency and reliability of predictive maintenance.
|
