Fault Prediction Model for Motor and Generative Adversarial Networks for Acceleration Signal Generation

The manufacturing process must continuously be improved. One of the most efficient strategies is maintenance scheduling by predictive maintenance for early fault detection and assisting with real-time decisions. The major concern of developing a predictive maintenance system is the lack of abnormal data and the cost of a high-specification sensor device for collecting data. This paper introduces the unsupervised learning model called Generative Adversarial Networks (GANs) for generating abnormal data in the form of acceleration signals to provide a dataset for developing an early fault prediction model and assisting a real-time decision on a low-frequency sensor device. The prediction model dataset is labeled on IS010816 to classify the label of data by Velocity Vibration (mm/s). The machine learning classifier model implements a hyperparameters optimization framework called OPTUNA to provide the best model performance. The proposed system aims to assist in real-time decision and maintenance schedules for the injection molding machine and offer the prediction model based on low-frequency sensor data from a drive motor.