A Lightweight Parallel Convolutional Model for Abnormal Detection and Classification of Universal Robots Under Varied Load Conditions

With the advancement of modern industrial automation and smart manufacturing, the demand for robots to perform precise operations has increased dramatically. Robots, with their highly repetitive movements and operations in diverse and complex environments, are prone to faults, posing challenges to production efficiency and equipment reliability. In order to avoid the cost of incorporating additional sensors, this study directly uses the feedback data generated by the intrinsic control system of universal robots for condition monitoring. An innovative lightweight parallel convolutional model is developed to facilitate the extraction and learning of multi-layered features, which leverages position and force data as inputs. The design of the dual-stream residual structure allows the model to capture feature information with lower parameter complexity, enhancing data processing efficiency. The multi-scale feature enhancement module improves the adaptability and robustness of the model under different working conditions, providing technical support for rapid diagnostics in practice. Experimental datasets demonstrate the model's capability in abnormal detection and classification under various load conditions.