Perturbation Analysis of Statistical Feature-based Evidence

In complex equipment, the evidence reasoning method based on statistical feature extraction can realize health assessment under high dimensional correlation data. However, in the actual operation of the equipment, affected by environmental factors and operation modes, the monitoring data will be disturbed, which will lead to changes in the extracted features and affect the evaluation results. In order to analyze the influence of perturbation on the evaluation results, a perturbation analysis method for statistical feature-based evidence is proposed. By introducing perturbations into statistical evidence reasoning models, we analyze the effects of perturbations from variables to features to evidence construction and output results. Then, the relative utility variation factor is defined to judge the degree of perturbation influence on the evaluation results and the acceptable range of the model. Finally, the effectiveness of the proposed method is verified by the perturbation analysis experiment of the inertial measurement unit.