测控技术2025，Vol.44Issue(4)：25-34,41,11.DOI:10.19708/j.ckjs.2025.02.211
基于特征重要性的质量相关故障诊断与量化评估
Quality-Related Fault Diagnosis and Quantitative Evaluation Method Based on Feature Importance
华宇辉 1胡文军 1崔胜男 1王士同2
作者信息
- 1. 湖州师范学院信息工程学院,浙江湖州 313000||浙江省现代农业资源智慧管理与应用研究重点实验室,浙江湖州 313000
- 2. 江南大学人工智能与计算机学院,江苏无锡 214000
- 折叠
摘要
Abstract
Quality related fault diagnosis is one of the current research hotspot in the field of process control.However,the complex relationship between process variables and quality variables poses a serious challenge to improving fault detection performance and visualization.Therefore,a new quality-related fault diagnosis method and fault quantitative evaluation criterion are proposed.Firstly,based on the boosting tree model,the correlation between process variables and quality variables is analyzed,and the feature importance score is defined accord-ing to the three characteristics of the boosting tree model.Then,the feature importance score is applied to the trade-off parameter of the support vector data description to construct a feature importance regularized support vector data description(FIR-SVDD).Finally,fault quantitative evaluation is performance using the projected sphere distance metric in kernel space.Experimental results on Tennessee-Eastman Chemical data and HYDAC hydraulic system data show that the proposed method has better performance.关键词
支持向量数据描述/过程变量/质量变量/提升树模型/距离度量Key words
support vector data description/process variables/quality variables/boosting tree model/distance metric分类
计算机与自动化引用本文复制引用
华宇辉,胡文军,崔胜男,王士同..基于特征重要性的质量相关故障诊断与量化评估[J].测控技术,2025,44(4):25-34,41,11.
