Data-based fault detection in chemical processes: Managing records with operator intervention and uncertain labels

作者：

Highlights：

• A fault detection system under uncertain labels and dynamic attributes is proposed.

• A systematic framework for improving operator-process interaction is developed.

• An iterative scheme for refining a database and retraining classifiers is introduced.

• Application on case studies and industrial data shows potential performance gains.

摘要

•A fault detection system under uncertain labels and dynamic attributes is proposed.•A systematic framework for improving operator-process interaction is developed.•An iterative scheme for refining a database and retraining classifiers is introduced.•Application on case studies and industrial data shows potential performance gains.

论文关键词：Fault detection,Mislabeling,Label noise,Underlying states,Operational intelligence,Interactive learning

论文评审过程：Received 27 January 2016, Revised 17 June 2016, Available online 23 June 2016, Version of Record 30 June 2016.

论文官网地址：https://doi.org/10.1016/j.eswa.2016.06.040