Embedded AI-based induction motor diagnosis and fault classification

Abstract

In any industry, regardless of how reliable the equipment is, it is prone to failures and degradation due to many factors ranging from environmental factors to simply reaching the end of the life cycle. This made fault diagnosis a necessary and reliable tool, to maintain equipment and extend their lifetime. This report focuses on presenting the development of an AI-based system for fault diagnosis of an induction motor, using classi?cation techniques and deploying it on an embedded system, with the aim of extending equipment lifetime and maintaining its e?ciency. The project utilizes data collected from accelerometers and acoustic sensors to train multiple AI models mainly: Support-vector-machines, k-nearest-neighbor, Decision tree, Random forest, Feed forward neural network, and Long-short-term memory. As a result, the feed forward neural network model is found to perform the best in terms of accuracy, model size, and testing time among the evaluated models. Moreover, the system we deployed on an ESP32-S3 SoC, performed well and proved to be reliable for industrial application when tested with new data. Consequently, the ?ndings highlight the reliability and precision of AI models in fault diagnosis tasks, and showed how to bene?t from deploying such systems on embedded platforms. Overall, the presented report emphasizes the importance of fault diagnosis in industrial settings and showcases the practicality and e?ectiveness of AI on embedded systems in this domain.