Optimization of Post-Production Quality Control of Steel Products Using Autoencoder Algorithm for AI-Based Anomaly Detection

Post-production quality control of steel products plays a crucial role in reducing waste and increasing customer satisfaction. In this study, an Autoencoder-based approach with an Encoder Network is examined to enhance the automated quality control of steel. The Encoder Network compresses input data into a lower-dimensional latent space, extracting the essential features from steel surface images and allowing for efficient anomaly detection by filtering out irrelevant data. This encoded data is then used to reconstruct the original input through the Decoder Network, with reconstruction errors signaling the presence of defects. This algorithm, by learning patterns from natural data, has the ability to identify anomalies and surface defects in steel products. The performance of the model was evaluated using the NEU Surface Defects Database, which contains high-resolution images of surface defects in steel. This dataset is recognized as one of the reliable sources for research in quality control and defect detection in steel products, covering a variety of defects that allow researchers to evaluate the performance of different algorithms. The results showed that the proposed model can detect anomalies with 96% accuracy. Additionally, the analysis of reconstruction error for normal and anomalous data demonstrated that this approach offers higher precision in defect detection and lower production costs compared to traditional methods. One of the key achievements of this study is the high speed of defect detection and the model's ability to accurately predict potential issues in the early stages of production, which reduces inspection time and increases production line efficiency. The implementation of this model in production lines can lead to improvements in quality control processes and reduce waste. Furthermore, this study shows that the Autoencoder model, by analyzing historical data, can prevent the recurrence of similar defects, significantly aiding in process optimization and resource conservation.