Joint Self-learning and Fuzzy Clustering Algorithm for Early Warning Detection of Railway Running Gear Defects

Abstract

The paper proposes a new feature pattern recognition method for early warning of defects of the railway vehicle running gear. Based on a large amount of historical data, a joint self-learning and fuzzy clustering algorithm was developed. The joint algorithm combines the advantages of the fuzzy clustering algorithm and of the self-learning algorithm; the fuzzy clustering algorithm has been widely applied in fault diagnosis of conventional mechanical systems, but is difficult to be applied for the fault diagnosis of railway vehicle running gears in the specific track-vehicle environment, due to the track irregularities. When combined with the self-learning algorithm, the new joint algorithm converts original featured values into clustering series as new judgement criteria by clustering samples in the same section, and then obtains the dynamic early warning threshold to realize the vibration monitoring and early warning of the railway vehicle running gear. A mechanical vibration test rig was built to verify the new joint algorithm. A monitoring and early warning software platform based on the joint algorithm was also developed to monitor and early warn the abnormal vibrations of the railway vehicle in real time. The experimental results show that the new method can efficiently identify the abnormal vibrations in the case of mechanical failure.