Automated Industrial Load Station Inspection

The manufacturing sector heavily relies on fast and accurate material inspection to improve product quality and productivity. Computer vision approaches based on deep learning provide superior performance for automatic material inspection over traditional methods. However, collecting a large amount of data to train a deep learning model is a challenging and expensive task, especially in an industrial environment. To address this challenge, we propose a synthetic augmentation approach, in which source data is prepared with a hybrid technique, combining synthetic data and real data to train a deep learning model, with performance evaluated on the target data. The applied approach consists of three key steps and performance evaluation indicates the ability to learn the domainvariant features and enhance the model's generalisation ability. For inspection purposes, two well-known models, YOLOv5 and YOLOv8, were employed utilizing the optimizers SGD and AdamW, respectively, which provided significant test accuracy. Building on the results, the applied approach may be recommended for similar inspection use cases in smart industry applications.