238 / 2025-06-15 16:21:44

Industrial Robot Joint Fault Diagnosis with CNN and Refined D-S Evidence Theory

Industrial robots, Convolutional Neural Network, bulldozer distance, D-S evidence theory, fault diagnosis

The industrial robot joint’s structure is complex, making it difficult to accurately diagnose with a single information source. Uncertainty in fault symptom derived from multi-physical source data, which easily causes problems such as conflicting diagnosis results and misdiagnosis. In this paper, a method for diagnosing faults in industrial robot joints based on the data fusion of CNN and improved D-S evidence theory is proposed. To begin with, joint vibration and current data are gathered through a self-constructed robotic arm test bench. Then, CNN are employed to train and test normal joint operation data as well as six types of faulty data, obtaining recognition accuracy, and generating basic probability assignment functions through the total probability formula. Subsequently, Introducing the Earth Mover's Distance (EMD), a refined data fusion approach with D-S evidence theory is proposed, demonstrating effectiveness in handling completely conflicting evidence. Finally, industrial robot joint fault diagnosis is carried out using the D-S fusion rule. Experiments show that compared with other data fusion methods, the proposed method can effectively improve the accuracy of industrial robot joint fault diagnosis. It can avoid the subjectivity of obtaining basic probability assignment functions, and has higher generalization and stability.