Online Dynamic Semi-supervised Broad Learning System for Real-time Fault Diagnosis with Variable Working Conditions
作者单位：MCC5 Group Shanghai Co.LTD
会议名称：《第35届中国过程控制会议》
会议日期：2024年
学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 08[工学] 0835[工学-软件工程] 0802[工学-机械工程] 0811[工学-控制科学与工程] 080201[工学-机械制造及其自动化] 0812[工学-计算机科学与技术（可授工学、理学学位）]
关 键 词：Multi-mode fault diagnosis broad learning system transitional conditions online semi-supervised learning
摘 要：The multi-mode fault diagnosis method under time-varying working conditions presents a significant challenge in process control. Production processes often exhibit multiple steady conditions due to variations in external environmental conditions, changes in production schemes, and fluctuations in raw material properties. The influence of operating conditions on monitoring data features may outweigh that of faults, thus significantly impacting fault diagnosis effectiveness. Effectively handling transitional conditions in fault diagnosis necessitates updating models with online monitoring data due to practical limitations in obtaining sufficient historical data. In this paper, we propose an online dynamic semi-supervised broad learning system for real-time fault diagnosis under variable working conditions. A dynamic representation center-based anchor set construction principle is designed to adapt to the evolving distribution of monitoring data over time. We also derive a model updating rule to regulate the evolutionary direction of online updates for the diagnostic model with dynamic manifold regularization term. Realistic gearbox fault diagnosis experiments under various working conditions demonstrate its effectiveness in mitigating adverse effects of transitional conditions on diagnostic performance.