Evidence Fusion Method for Fault Diagnosis Based on Optimal Coordination and Iteration Correction

In fault diagnosis, multi-source information fusion (MSIF) is usually more reliable than single-source information, and Dempster–Shafer (D-S) evidence theory provides a universal and popular decision-level fusion framework for MSIF. However, existing evidence fusion methods still have two limitations: (1) the overweighting effect of high information volume on unreliable evidence is ignored, and (2) the fusion accuracy cannot be further improved, as only one-time evidence correction is considered. To overcome these limitations, an evidence fusion method based on optimal coordination and iterative correction is proposed for fault diagnosis. Firstly, the credibility and information volume of each piece of evidence are quantified by the Jousselme distance and Deng entropy, respectively. Then, using game theory combination weighting (GTCW), credibility and information volume are optimally coordinated to correct all pieces of evidence, which are then initially fused with Dempster’s rule. Ultimately, taking the initial fusion result as the reference, the credibility is iteratively recalculated to correct and fuse all pieces of evidence until the fusion result converges. The optimal coordination suppresses the overweighting effect caused by high information volume, and the iterative correction breaks the limitation of one-time fusion. Experimental results demonstrate that the proposed method outperforms existing methods and can significantly improve the fusion results in fault diagnosis.