Learning-based Fault Detection and Mitigation Model in Industrial Control Systems Based on Measurement Data

Authors

Sohrab MokhtariFollow

Document Type

Dissertation

Degree

Doctor of Philosophy (PhD)

Major/Program

Electrical and Computer Engineering

First Advisor's Name

Kang K Yen

First Advisor's Committee Title

Committee Chair

Second Advisor's Name

Arif I. Sarwat

Second Advisor's Committee Title

Committee Member

Third Advisor's Name

Jean H. Andrian

Third Advisor's Committee Title

Committee Member

Fourth Advisor's Name

Deng Pan

Fourth Advisor's Committee Title

Committee Member

Keywords

Industrial Control Systems, False Data Injection Attack, Learning-based fault detection

Date of Defense

11-1-2022

Abstract

Industrial control systems (ICSs) play a significant role in supervising, controlling, and automating critical infrastructures, such as power plants, water treatment, and civil transportation. In the past years, ICSs have employed open technologies to communicate data over other ICS or non-ICS networks. Although wireless communication has the privilege of having access to the system from far distances, it opens new points of intrusion for adversaries. Fault detection problems in ICSs are commonly known as a network traffic monitoring scheme for detecting abnormal activities. However, a network-based intrusion detection system (NIDS) can be deceived by attackers that imitate the system's normal activity. This research is devoted to develop model-based and learning-based fault detection and mitigation designs, by focusing on the data-driven methods. The main contribution of this study is two-folded. First, it proposes a novel machine learning-based approach for fault detection and mitigation in ICSs based on measurement data in the supervisory control and data acquisition (SCADA) system. The proposed fault detection, isolation, and identification (FDI) approach is called measurement intrusion detection system (MIDS), which enables the system to detect, locate, and identify the type of any abnormal activity in the system even if the attacker tries to conceal it in the system's control layer. Second, a learning-based control design would compensate for the detected fault to converge the system's deviation to zero. The proposed controller consists of a set of deep learning algorithms that learns the normal behavior of the ICS by monitoring the input and output data. This design allows ICSs to be fault-tolerant even if the mathematical model is unavailable. Also, it can be a complementary protection system along with the conventional NIDS to improve the security and reliability of ICSs remarkably. The proposed mechanism is implemented and tested on two test-beds. First, employing experimental data, the MIDS is implemented on a dataset exploited from a test-bed consisting of a boiler, turbine, water-treatment, and a hardware-in-the-loop (HIL) simulator. Second, the performance of the MIDS and the learning-based controller is evaluated on a two-area interconnected power system with a load frequency control section. The results show a very successful and reliable performance for the proposed mechanism to protect ICSs against faults and attacks.

Identifier

FIDC010862

ORCID

0000-0001-9363-1706

Recommended Citation

Mokhtari, Sohrab, "Learning-based Fault Detection and Mitigation Model in Industrial Control Systems Based on Measurement Data" (2022). FIU Electronic Theses and Dissertations. 5216.
https://digitalcommons.fiu.edu/etd/5216