Utilizing data fusion and machine learning in fault detection & diagnosis

Abdulaban, Abdula (2023) Utilizing data fusion and machine learning in fault detection & diagnosis. Masters 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 (9MB)


Increasing complexity and high number of variables in the process industry have increased the interest in autonomous process health monitoring among process experts and production line operators. Modern information technology makes data ubiquitous, which makes data-driven approaches popular in ensuring product quality and process safety. However, data-driven approaches require efficient tools for processing data to extract relevant, valuable information about a process. The proposed research in this thesis is mainly focused on two aspects: multiple data fusion to segment and cluster faulty data and root-cause analysis to label faults clusters. Most existing solutions rely on a single type of data source, need significant knowledge about the process, and require a data set with labels for normal and abnormal events. The first step proposes a new segmentation and clustering technique for extracting distinct operating scenarios from numerous data sources. Knowledge is extracted from each data source separately and fused in a multiple-stage data fusion technique. The final output is clusters that have segments of similar patterns. In the second step, a rootcause method is proposed for labeling the clusters. The two-step PCMCI approach uses graph theory and structural causation to identify the relevant variables in the first step and then evaluates the relationships between the variables in the second step to finding the root cause. Therefore, faulty clusters will be recognized by their root cause. Finally, a cloud platform was developed that includes the above-mentioned tools, in addition to a number of other necessary tools for managing, analyzing, and designing alarms. The proposed approaches were tested using a benchmark simulation model of the Tennessee Eastman process and a large set of real data from a refinery.

Item Type: Thesis (Masters)
URI: http://research.library.mun.ca/id/eprint/15812
Item ID: 15812
Additional Information: Includes bibliographical references (pages 99-112)
Keywords: fault detection, clustering, segmentation, data fusion
Department(s): Engineering and Applied Science, Faculty of
Date: February 2023
Date Type: Submission
Library of Congress Subject Heading: Machine learning; Fault location (Engineering); Computer simulation; Multisensor data fusion; Graph theory; Cluster analysis

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics