1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2940099, IEEE
Transactions on Industrial Informatics
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS 1
Performance Supervised Fault Detection Schemes
for Industrial Feedback Control Systems and their
Data-Driven Implementation
Linlin Li, Member, IEEE , and Steven X. Ding
Abstract—This paper addresses performance supervised fault
detection issues for industrial feedback control systems based
on performance degradation prediction. To be specific, three
performance indicators are first introduced based on Bellman
equation to predict system performance degradations for in-
dustrial processes with the aid of machine learning techniques.
Based on them, three performance supervised fault detection
schemes are proposed by embedding the performance indicators
as supervising information. In this context, the data-driven imple-
mentation of performance supervised fault detection schemes are
investigated for linear systems with unmeasurable state variables.
A case study on rolling mill process, a typical benchmark in the
steel manufacturing processes, is given at the end of this paper to
illustrate the applications of the proposed fault detection schemes.
Index Terms—Data-driven, performance supervised fault de-
tection, performance prediction, Bellman equation
I. INTRODUCTION
W
ITH enhanced requirements for reliability as well as
safety in industrial processes, fault detection (FD) has
received considerable attention in both research and industrial
application domains [1]–[5]. Due to their close relation with
control system design, observer-based FD methods have been
widely applied for detecting faults in automatic control systems
[3], [4], [6], [7]. A great number of methods have been
published on designing observers and post-filters towards
enhancing fault detection performance. Most of them can be
classified either as data-driven or model-based methods [1], [2],
[8]. The essential idea behind model-based FD methods consists
in the integration of a process model into the fault detection
system in form of an observer, which demands for considerable
modeling efforts. In recent years, enormous research effort has
been dedicated to the development on data-driven FD methods.
In general, the data-driven FD approaches are mainly based on
the analysis of correlation relations between the (measurable)
process variables [9]–[14].
It is known that modern industrial processes have continu-
ously increasing demands on system performance, efficiency
and reliability. Driven by it, intensive attention has been
This work has been supported by the National Natural Science Foundation
of China under grant 61603033. (Corresponding author: Linlin Li.)
L. Li is with the Key Laboratory of Knowledge Automation for Industrial
Processes of Ministry of Education, School of Automation and Electrical
Engineering, University of Science and Technology Beijing, Beijing 100083,
P. R. China. Email: linlin.li@ustb.edu.cn.
S. X. Ding is with the Institute for Automatic Control and Complex Systems
(AKS), University of Duisburg-Essen, Germany. Email: steven.ding@uni-
due.de.
paid to performance monitoring approaches over the past
decade [15]–[20]. Among the involved studies, the data-
driven methods have been recently applied to the computation
and monitoring of system key performance indicators (KPIs)
[15], [16], [21]–[23]. Moreover, a performance-based fault
detection scheme has been investigated by establishing the
relation between stability performance degradation and residual
signal [18]. To our best knowledge, most of the existing
performance monitoring studies can be classified as component
fault oriented detection, while limited attention has been
paid on system performance degradation oriented detection.
System performance degradation could be caused not only
by system component faults, but also by e.g. mismatching of
controller parameters (in large scale systems), even though
there exists no fault in the system components. On the other
hand, reviewing the recent research shows that few effort
has been made on i) predicting the operation performance
of feedback control systems and ii) handling the performance
monitoring in the presence of the model uncertainties and
unknown deterministic inputs. Moreover, the timely detection
of the potential faults are of significant application interests,
which allows the process to take actions to avoid further
degradation of the system performance. These observations
motivate us to investigate detection schemes by predicting
system performance degradation with respect to the running
the control and optimization policy using (actually) process
data, aiming at achieving an early/real-time fault detection.
The main objective of this paper is to develop performance
supervised fault detection (PSFD) schemes for industrial
feedback control systems. To predict the system performance
degradations, three performance indicators are introduced first
based on Bellman equation for feedback control systems.
Driven by information delivered by the performance indicators,
three performance supervised fault detection schemes are
proposed with the aid of machine learning techniques. The data-
driven design approaches of the PSFD schemes are addressed
for linear systems with unmeasurable state variables. The
novelties and contributions of the proposed PSFD schemes
lie in the following aspects:
•
detecting the degradations in the system performance
which may be caused not only by the components faults,
but also by the mismatching of controller parameters in
large-scale systems;
•
promising real-time fault detection by predicting the
system performance degradations using the online data;