Latent Anomalies: Advanced Detection in Industrial Machinery via Convolutional Autoencoder

The paper presents a framework that utilizes Convolutional Autoencoders (CAEs) for anomaly detection in industrial settings. The framework focuses on ensuring product quality and identifying defects in industrial manufacturing, with an emphasis on distinguishing between normal and anomalous occurrences. The proposed model is trained using 25 epochs with a batch size of 1024 and employs Mean Square Error (MSE) and Structural Similarity Index Measure (SSIM) to optimize pixel-wise accuracy and perceptual quality. The Adam optimizer is used to minimize the reconstruction loss, and the model’s performance is evaluated on the MVTec Anomaly Detection dataset, achieving an accuracy of 77.1 on the validation set. The study highlights the potential of CAEs in enhancing industrial quality control and emphasizes future work focusing on optimizing model architecture, real-time anomaly detection, and exploring transfer learning techniques for improved versatility and robustness. The experimental results, visualizations, and implications underscore the significance of the proposed framework in addressing the challenges of anomaly detection in industrial manufacturing.