CNN-LSTM-AE Based Predictive Maintenance Using STFT for Rotating Machinery

In modern manufacturing after the Fourth Industrial Revolution, fast and accurate production is essential, and it is important to maintain and improve process reliability and productivity. Beyond the traditional conservative maintenance method, it is important to monitor the condition of facilities and take preventive measures. This study aims to analyze vibration data of manufacturing equipment and predict rotor defects using deep learning technology. An anomaly detection model that distinguishes between normal and abnormal conditions of the equipment is trained, and proactive problem prediction through internal and external data (sensor data, operation logs, etc.) is emphasized. Vibration data is converted into two-dimensional image data through Short Time Fourier Transform (STFT), and feature extraction is performed through Convolutional Neural Network (CNN) algorithm. Then, an Autoencoder (AE) based time series anomaly detection technique is used to make efficient and accurate predictions. In order to strengthen the competitiveness of the manufacturing industry, a deep learning-based rotor defect prediction model was developed, which is expected to improve the stability and productivity of the manufacturing process.