AI-Assisted Predictive Maintenance in Smart Factories Using Vibration Signal Analysis

Abstract

Due to Industry 4.0, smart factories require smart, data-based maintenance in order to be efficient and maintain a minimal amount of downtimes. It presented a robust predictive maintenance system, which is based on the knowledge of vibration signals to identify issues with the rotating equipment at a very early fault stage. Vibration signals are sensitive to machine changes, and hence they can be used to monitor the health of the machine without causing further damage to the machine. A suggested architecture consists of three large blocks: preprocessing the signals, time-frequency features extraction with STFT and CWT and faults classification with CNN-LSTM. It makes room to accommodate visual characteristics in addition to recording the sequencing of the vibration occurrences with time. Experiments were conducted by integrating CWRU bearing dataset with data in smart factory testbeds. Compared to the performance of CNN-LSTM, fault classification achieved 94.6 percent accuracy that surpassed the classification performance of typical machine learning or deep learning models. Moreover, the system performed effectively in various cases and the percentage of incorrect reports was less than 5%. Based on these findings, CBM reliability is enhanced because the structure identifies small problems before they evolve into significant defects. In all fairness, this study introduces a highly adaptable and successful model of predictive maintenance that could be effective in the case of smart manufacturing systems.