A B-spline and OS-ELM Fusion Approach for Prognostics with Singularity Problems

Liyue Yan, Houjun Wang, Hao Wang, Zhen Liu

Abstract

In practice, the degradation process of electronic products is usually accompanied with singularities caused by intermittent and transient interference, improper conduct on singularity processing would inevitably and seriously affect the accuracy of products’ life-time prediction. Taking advantage of rapid development of AI technology recent year, a new surrogate approach based on spline function and online sequential extreme learning machine (OS-ELM) is developed in this paper, to address these issue. This fusion approach takes the cubic non - polynomial spline function as the prediction cell of the output from OS-ELM, the second derivative of the spline model can be adopted and calculated to form a series of observation frames, meanwhile, an improved particle swarm optimization (PSO) is used to optimize the parameters of ELM hidden network to help with forecasting the observation sequence and rebuild the spline function. In the verification stage, two numerical simulation examples and a practical application involving typical time series data with singularities demonstrate the effectiveness of this proposed fusion method, respectively.

Keywords

Prognostic; Fusion; Spline; OS-ELM; Singularity; PSO.

Download

IMEKO-TC10-2020-022.pdf

DOI

-

IMEKO TC

TC10 - Measurement for Diagnostics, Optimization and Control

Back to the proceedings