Teknik & Rekayasa editorial
Open AccessOA2026
An Explainable AI and Machine Learning Framework for Integrated Predictive Maintenance, Structural Health Monitoring, Energy Optimization, and Reliability Assessment Across Mechanical, Civil, and Electrical Engineering Systems
This review examines explainable machine learning applications across mechanical predictive maintenance, civil structural health monitoring, electrical energy optimization, and reliability assessment. It proposes a conceptual framework linking data collection, prediction, explanation, reliability assessment, and engineering action, while highlighting challenges such as poor sensor data, limited failure history, and unstable explanations.
A. Nalbandh; Ashutosh Sudhirrao Werulkar; J. Pandya; M. Vyas; Ravi G. Mehta; Bhupendra M. Parmar; Ankita Kumariยท Computer and Decision Making: An International Journalยท 2026ยท DOI 10.1922/comdem.v3i6s.18910
The core problem
Artificial intelligence (AI) and machine learning (ML) are increasingly used to monitor engineering assets, predict faults, and support operational decisions. However, when engineers cannot understand the evidence behind predictions or assess uncertainty, the usefulness of these predictions is limited. This review addresses the need for explainable machine learning (XML) in four key application areas: mechanical predictive maintenance, civil structural health monitoring, electrical energy optimization, and reliability assessment. The authors argue that while these domains share common methodological needs, they also differ in data sources, prediction targets, explanation requirements, and validation approaches. The paper aims to synthesize these perspectives and propose a conceptual framework that integrates data collection, prediction, explanation, reliability assessment, and engineering action. The framework is intended to make alerts interpretable and actionable for construction, structural assessment, maintenance programs, and energy management, considering the potential impact of failures.
Innovation
The review adopts a conceptual and comparative approach. It examines the similarities and differences across the four application domains in terms of information sources, prediction objectives, explanation needs, and validation requirements. Based on this analysis, the authors propose a conceptual framework that connects data acquisition, predictive modeling, explanation generation, reliability assessment, and decision-making. The framework emphasizes the importance of physically meaningful uncertainty expressions to enhance relevance. The paper does not present empirical testing; instead, it outlines typical challenges such as poor sensor information, inadequate failure history, varying operating conditions, and unstable explanations. The methodology is thus grounded in literature synthesis and conceptual development rather than experimental validation.
Introduction
Artificial intelligence (AI) and machine learning (ML) are increasingly used to monitor engineering assets, predict faults, and support operational decisions. However, when engineers cannot understand the evidence behind predictions or assess uncertainty, the usefulness of these predictions is limited. This review addresses the need for explainable machine learning (XML) in four key application areas: mechanical predictive maintenance, civil structural health monitoring, electrical energy optimization, and reliability assessment. The authors argue that while these domains share common methodological needs, they also differ in data sources, prediction targets, explanation requirements, and validation approaches. The paper aims to synthesize these perspectives and propose a conceptual framework that integrates data collection, prediction, explanation, reliability assessment, and engineering action. The framework is intended to make alerts interpretable and actionable for construction, structural assessment, maintenance programs, and energy management, considering the potential impact of failures.
Why it matters
The discussion centers on the challenges of implementing explainable ML in engineering systems. Poor sensor data quality, insufficient failure histories, and varying operating conditions can undermine model reliability. Explanations themselves may be unstable or plausible but misleading. The authors argue that physically meaningful uncertainty expressions can improve the relevance of explanations. They also stress the need for validation across assets and sites to determine whether explanations actually influence engineering decisions. The framework's integration of reliability assessment ensures that explanations are linked to failure consequences. The paper concludes that while explainable ML holds promise for integrated predictive maintenance, structural health monitoring, energy optimization, and reliability assessment, further empirical evaluation is needed. Until then, engineers should exercise caution and maintain monitoring in high-stakes situations.
Who should read this
Pembaca Research Digest dan praktisi rumpun ini.
Opening member contentโฆ