Research on fusion algorithm for service life prediction based on kernel functions

Real-time service life prediction offers an effective means of assessing the remaining service life of operational control devices in rail transit systems. Currently, the primary methods employed for this purpose are based on physics-of-failure analysis and degradation data, respectively. Early in the lifecycle of these control devices, the lack of real-time degradation data makes the physics-of-failure-based methods more accurate. However, as these devices approach the end of their lifecycle, a large amount of degradation data accumulates, allowing the degradation-data-based methods to yield more precise predictions. This paper proposes a kernel function-based fusion algorithm for service life prediction that combines predictions from both methods to enhance accuracy in predictions throughout the entire lifecycle. Accelerated degradation experiments were conducted using an optical coupler on a core board of a rail transit control device to compare the results from the physics-of-failure-based method, the degradation-data-based method, and the proposed fusion algorithm. The results demonstrate that the fusion algorithm achieves higher prediction accuracy throughout the entire lifecycle compared with the other two methods.