Fault Detection of Axle Box Bearing Based on Matter-Element and Negative Selection Algorithm

In view of the difficulty in obtaining the fault data of high-speed train axle box bearings, a fault detection method combined matter-element model and negative selection algorithm (NSA) was proposed, which did not need prior knowledge. Firstly, detectors of NSA were constructed by means of multi-dimensional matter-element model, and the comprehensive correlation degree (CCD) between detectors and training samples was used as the matching rule. In order to achieve greater coverage of the non-self space by the detector, a control parameter was introduced within the constraint range of CCD. Then, the fitness function was constructed according to the matching rule and the control parameter, and the particle swarm optimization (PSO) algorithm was utilized to generate candidate detectors. The influence of control parameter on detector generation and convergence speed of particle swarm optimization algorithm was analyzed. In order to reduce the redundancy of the candidate detector set, the merging rules of characteristic parameters intervals of candidate detectors were proposed based on the correlation function, and the number of mature detectors was reduced to 18. Various fault signals of axle box bearings were generated by signal simulation method, 100 sets of test samples were established, and 18 mature detectors were used for fault detection. The results show that the mature detectors have good detection performance for different types of bearing fault, and the detector activation rate of normal sample is 1.11%, while its minimum value of fault samples is 96.67%.