A systematic review of knowledge distillation in industrial predictive maintenance: Applications, methods and challenges

This is the authors' abstract. We don't add key points for this paper.

Deploying deep learning models for predictive maintenance (PdM) is often constrained by high computational costs, limiting real-time industrial deployment. Knowledge distillation (KD) offers a lightweight alternative by transferring knowledge from large teachers to compact students. Despite growing research on KD in PdM, no systematic review has consolidated existing progress. This paper fills that gap by analyzing 48 KD-based PdM studies, identifying six key paradigms and analyzing their efficiency–accuracy trade-offs. This review highlights unresolved challenges and outlines future directions toward adaptive, cross-domain, and resource-efficient KD frameworks for intelligent industrial maintenance.