An integrated intelligent diagnosis framework for harmonic drives under fast-varying operating conditions

Abstract

Harmonic drives are indispensable for precision motion in industrial robots and space manipulators, yet their reliability is critically challenged under fast-varying operating conditions. The complex rigid-flexible coupling dynamics and resulting non-stationary vibration signals render traditional fault diagnosis methods ineffective. This paper proposes an integrated intelligent framework that synergizes physics-based modeling with data-driven analytics to address this challenge. The framework is built upon three pillars: 1) A high-fidelity dynamic model and finite element analysis to elucidate fault mechanisms in the flexspline and bearings. 2) A semi-supervised anomaly detection method combining Multiscale Diversity Entropy (MDE) with a One-Class Support Vector Machine (OCSVM) for identifying unknown faults using only healthy data. 3) An advanced fault identification scheme employing Variational Embedding Multiscale Divergence Entropy (veMDE) fused with an Extreme Learning Machine (ELM) for accurate classification and fault location. Extensive validation on dedicated test platforms demonstrates the framework’s superior performance. Our proposed methods achieve high accuracy in both anomaly detection and fault identification tasks, significantly outperforming other benchmark techniques. This work provides a robust and practical solution for the intelligent health management of harmonic drives in demanding dynamic applications.