Machine learning methods for fault detection and diagnosis of digitalized processing system

Arunthavanathan, Rajeevan (2022) Machine learning methods for fault detection and diagnosis of digitalized processing system. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (3MB)


This thesis presents novel development and applications of machine learning techniques for process fault detection, diagnosis, and prognosis from safety and predictive maintenance perspectives. The main contributions of this thesis include the development of (i) new algorithms to diagnose the unlabelled faults; (ii) a self-learning tool for fault detection and diagnosis of untrained faults; (iii) a forecast model for fault conditions; (iv) a framework for root cause analysis in an automated environment; and (v) a methodology for estimating the remaining useful life. In the context of Industry 4.0, process plants’ operations have become increasingly autonomous and run in an intelligent mode. An intelligent process operation takes advantage of online data, uses advanced modelling approaches and utilizes automation to achieve a flexible, smart, and reconfigurable operation. In such an autonomous environment, process fault detection, diagnosis, and prognosis play critical roles in ensuring its safety and integrity. In this study intelligent fault detection and diagnosis methods are developed based on state-of-the-art machine learning techniques. Further, this study is extended to calculate the remaining useful life online, using the fault to failure transmission time. The research study results in five signification contributions. First, a comprehensive review of the existing fault detection and diagnosis approaches was conducted to identify the knowledge gaps and to develop fault detection and diagnosis approaches that are best suited for Industry 4.0. Second, a cognitive fault detection and diagnosis technique using unlabelled process data and an anomaly detection technique using machine learning were developed. Third, a self-learning neural network and permutation algorithm were developed for prediction of the root cause of a detected fault. Fourth, a methodology was developed to early predict faults, based on monitoring the fault symptoms using a deep learning algorithm. Fifth, a model was developed to estimate the remaining useful life using the system’s failure threshold and a degradation model. In this research work, all the proposed models were developed using self-learning methodologies. Therefore, the work constitutes an essential step towards developing an autonomous fault detection, diagnosis, and remaining useful life estimation tool. The proposed frameworks are validated using experimental data and simulated process system data. The findings from this study highlight that by integrating unsupervised and supervised learning, without prior knowledge of the fault condition, the proposed machine learning model was able to detect and diagnose the fault conditions. Unsupervised learning was used to detect the unknown fault conditions, and a neural network permutation algorithm was used to identify the root cause for the detected unknown faults. This work also used supervised learning to classify the known fault conditions. Furthermore, by investigating the failure condition of the identified root cause variable or feature, remaining useful life was estimated by developing a regression model. Likewise, this thesis finds the solution for early fault detection in real-time by integrating the deep learning tools with unsupervised learning.

Item Type: Thesis (Doctoral (PhD))
Item ID: 15449
Additional Information: Includes bibliographical references.
Keywords: fault detection and diagnosis, machine learning algorithms, failure prognosis, remaining useful life, self learning models
Department(s): Engineering and Applied Science, Faculty of
Date: May 2022
Date Type: Submission
Digital Object Identifier (DOI):
Library of Congress Subject Heading: Machine learning; Computer algorithms; Fault location (Engineering); Computer simulation.

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics