Identification of Defects in the Production of Powered Window Regulators Using Deep Learning on Vibration Data

Powered window regulators are a crucial component of automobile. Its smooth operation and timely response is important for the better functionality of automobile. Defects like noisy operation and slow or no response imply manufacturing faults. Conventional quality inspection uses auditory assessment method, which is labor insensitive and error prone. This study suggests an artificial intelligence driven solution using deep learning for automated fault detection using a modified pre-trained VGG16 convolutional neural network to classify faulty and non-faulty powered window regulators using spectrogram derived from vibration data. An accelerometer is employed to collect vibration data and labelled as ‘OK’ for non-faulty and ‘NOK’ for faulty powered window regulators. The time series vibration data is transformed into spectrograms using short-time Fourier transform. The modified pre-trained VGG16 model achieved 100% accuracy, precision, recall and F1 score both on validation and testing datasets.The results show the model is able to accurately classify faulty and non-faulty powered window regulators. It eliminates human error and increases productivity by reducing inspection time by 60%. This approach aligns with Industry 4.0 standards, providing a scalable solution for quality assurance in manufacturing. In future work, the model will be deployed on an edge device, and a user interface will also be developed for front-end interaction and visualizing the results, remote supervision and alerts for recurring issues.