Purpose <p>Accurate performance degradation assessment (PDA) of harmonic reducer is essential for sustaining the stability and secure operation of industrial robotic joints. A common strategy for PDA involves developing a health indicator (HI) that exhibits a monotonic trend and can effectively differentiate between distinct degradation phases. However, existing HI construction approaches relying on single-source signals often struggle to capture critical degradation information.</p> Method <p>To address the limitation, an innovative adaptive multi-source information fusion approach based on Newton–Raphson Optimizer (NRBO) is proposed for building a HI that can effectively represent the health state of harmonic reducer. Firstly, the dynamic monotonicity strength index (DMSI) is proposed to address the limitations inherent in the monotonicity strength index (MSI). Subsequently, DMSI, correlation, prognosability, and detectability of acoustic emission (AE) and micro-vibration (MV) signal features are calculated, and the optimal features are selected based on the composite evaluation index (CEI). Finally, NRBO is employed to fuse the optimal AE and MV features, yielding the final HI.</p> Results and Conclusions <p>Experimental results validate that the constructed fusion HI outperforms other HIs in comprehensive performance, demonstrating superior predictive accuracy in terms of root mean square error (RMSE), mean absolute percentage error (MAPE), and cumulative relative accuracy (CRA). This provides robust support for preventive maintenance decision-making for the harmonic reducer.</p>