International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 04 | Apr 2024
p-ISSN: 2395-0072
www.irjet.net
Advancements in Fault Diagnosis by Integrating Graph-Based Representations and Advanced Machine Learning Techniques Mrs. Suchetha1,2, Dr. Asha Saraswathi B2 1Department of Mathematics, Sahyadri College of Engineering and Management, Adyar,
574007, India
2Department of Mathematics, Srinivas University, College of Engineering and Technology, Mukka, 574146, India
---------------------------------------------------------------------***--------------------------------------------------------------------emerging as a promising model. Predictive maintenance Abstract - Fault diagnosis in complex engineering systems is
seeks to anticipate equipment malfunctions in advance, allowing for timely interventions and preventive actions to minimize potential hazards. Predictive maintenance systems utilize data analytics, sensor technology, and machine learning algorithms to analyse previous performance data. This analysis helps detect patterns and anomalies that indicate potential breakdowns in the future. This proactive strategy not only reduces the amount of time that systems are not functioning, and the expenses associated with maintenance, but also improves the effectiveness of operations and the dependability of assets [3]. Figure 1 illustrates the distinct approaches of reactive maintenance, preventive maintenance, and predictive maintenance within the context of industrial operations.
a significant activity that affects operational efficiency, safety, and maintenance costs. This paper provides a comprehensive review of innovative methodology and techniques for a fault diagnosis, with a focus on the application of graph-based representations and sophisticated machine learning algorithms. The assessment emphasizes the difficulties associated with standard data-driven methodologies for properly leveraging the correlation and geometric structure present in vast amounts of unlabeled industrial data. To address these problems, the review investigates novel technologies such as hypergraphs for representing equipment structure, deep hypergraph autoencoder embedding (DHAEE) for defect detection, and multiresolution hypergraph neural networks for discovering higher-order correlations in data. Furthermore, the study investigates the combination of modelbased and data-driven approaches, as demonstrated by the series configuration approach, which combines Bayesian Networks with adaptive gas path analysis. While these approaches present intriguing opportunities for enhancing fault detection accuracy and efficacy, issues like as algorithm complexity, data availability, and result interpretability remain relevant. The survey results highlight the need of using integrated and creative approaches to problem diagnosis, which have the potential to improve operational reliability, minimize downtime, and optimize maintenance procedures in complex engineering systems. Key Words: Graph Convolution Neural network, Knowledge graph, fault diagnosis, hypergraphs, industry 4.0
1.INTRODUCTION In the landscape of industrial operations, the concept of maintenance has undergone a profound evolution. Historically, maintenance methods have predominantly followed a reactive strategy, responding to equipment failures or malfunctions with repairs or replacements. The reactive paradigm, which has been widespread for many years, has notable disadvantages, including unforeseen periods of inactivity, decreased production, and increased expenses for upkeep [1,2]. Furthermore, it frequently neglects to tackle fundamental problems that cause recurring failures, leading to subpar asset performance and a shortened lifespan. Acknowledging these constraints, the industrial sector has shifted towards more proactive maintenance practices, with predictive maintenance
© 2024, IRJET
|
Impact Factor value: 8.226
Fig- 1: Maintenance Strategies However, the reliance on pre-established rules or statistical models, which may overlook intricate interconnections and dynamic interactions within industrial systems, often limits the effectiveness of conventional predictive maintenance methods. Furthermore, as industrial infrastructures grow more interconnected and reliant on large amounts of data, traditional approaches have challenges dealing with the size and intricacy of modern operational settings. In order to tackle these difficulties and capitalize on fresh possibilities for predictive maintenance, there has been a rising fascination with incorporating cognitive technologies like artificial intelligence (AI), machine learning (ML), and
|
ISO 9001:2008 Certified Journal
|
Page 1186