Beyond Prediction: Integrating Physics-of-Failure Analysis with Predictive Maintenance for Failure Prevention A Critical Review of Predictive Maintenance Approaches and a Case for Physics-of-Failure Prevention

Abstract

Predictive maintenance (PdM) has emerged as the dominant paradigm in industrial maintenance research, driven by advances in IoT, machine learning, and data analytics. However, this critical narrative review argues that the field’s near-exclusive focus on predicting when failures will occur has overshadowed a more fundamental question: why do failures happen? We critically examine PdM research (2010–2025) and identify a persistent gap: most predictive models detect degradation without addressing the underlying physical mechanisms that cause failure. Consequently, even successful predictions do not prevent recurrence. We propose that a physics-of-failure (PoF) approach—rooted in root cause failure analysis (RCFA) and reliability-centered maintenance (RCM)—offers a complementary and more durable solution when integrated with data-driven methods. A case study of 7,300 recurring spring leaf failures in a heavy truck fleet illustrates this gap: a single overlooked failure mechanism—shock absorber wear causing increased dynamic stress on springs—perpetuated a cycle that no predictive model could break, yet was resolved through root cause elimination. We propose an integrated framework combining data-driven monitoring with mechanism-based root cause analysis and design feedback, arguing that the ultimate goal of maintenance should be to eliminate the causes of failure, not merely predict them.