- This event has passed.
Thesis Defence: Data-Driven Fault Classification and Operator Action Recommendation in Industrial Control Systems
May 9 at 9:00 am - 12:00 pm
Negar Yassaie, supervised by Dr. Ahmad Al-Dabbagh, will defend their thesis titled “Data-Driven Fault Classification and Operator Action Recommendation in Industrial Control Systems” in partial fulfillment of the requirements for the degree of Master of Applied Science in Electrical Engineering.
An abstract for Negar’s thesis is included below.
Defence are open to all members of the campus community as well as the general public. Email email@example.com to obtain the zoom link for this defence.
The significant increase in industrial data generation and the development of machine learning and data mining tools and techniques, has led to a surge of interest in data-driven approaches for addressing industrial challenges. Specifically, this thesis aims to propose novel data-driven approaches for fault classification in industrial systems using the data from a reduced number of process variables data, and designing a recommender system for human operators to help them take corrective action when such anomalies happen in the system.
First, to reduce computational expenses and make the procedure capable of classifying faults even when data of some process variables (i.e., measurements) are/become unavailable, a novel three-stage computational procedure is proposed which uses collected data of process variables (i.e., measurements), and achieves the following: (i) time series data clustering using k-means clustering and soft dynamic time warping, (ii) variable selection using a modified LASSO-based or Elastic Net-based regression methods, (iii) data-to-image conversion and fault classification using convolutional neural networks. The proposed procedure can be applicable in industrial systems such as chemical production plants and power generation systems.
Second, a decision-support tool for human operators in industrial automation and control systems is proposed to help them take proper actions when anomalies such as faults occur, which we believe is the first in the literature. The proposed recommender system uses the historical data of the system, is based on a three-stage computational procedure, and achieves the following: (i) clustering historical process data segments containing timestamped events using a k-medoids-based algorithm, where instead of distance metrics, structural similarity of the segments is used and obtained using a Smith-Waterman-based algorithm, (ii) predicting the missing action ratings using a modified collaborative filtering technique applicable for time sequences, (iii) presenting proper and improper actions to the human operators. The recommender system can effectively be used in industrial systems where a human operator is responsible for making decisions and taking corrective actions.