A Self-Learning Framework for Fault Detection in Industrial Components

The fault detection of industrial components is crucial for ensuring the safety and reliability of industrial systems. This paper presents a change detection-based approach for fault detection in industrial components, employing a ResNet34based neural network. Data augmentation techniques are used to simulate these fault scenarios, and the image features are extracted for feature differentiation and fusion. A weighted crossentropy loss function is applied to address the class imbalance issue. In addition, a self-learning method is proposed, which effectively addresses the challenge of insufficient negative samples in industrial datasets. Experimental results based on real and simulated fault conditions show that the model successfully detects various types of faults. This approach offers a promising solution for the automated detection and maintenance of faults in industrial components.