This paper presents a deployable and interpretable framework for the health monitoring of high -power electro-mechanical actuators (EMAs) used in non-road mobile machinery (NRMM), turning standard sensor data into compact predictors, a health index, and finally a fused ensemble that tracks degradation for real-time maintenance.Using experimental run-to-life data, four correlation-based coefficients identified in a previous study are computed to capture cross-sensor dynamics.An early-life baseline is established, and a bounded, monotone health index is derived using Mahalanobis distancebased deviation, persistence gating, and outlier-robust accumulation.Several machine learning regression models are trained and compared; three complementary models-Gaussian process regression, a bagged-tree ensemble, and a compact multilayer perceptron are selected for further process.The models are then deployed in Simulink where raw sensor signals are transformed into correlation-based coefficients, which feed three trained regressors.The outputs of these individual regressors are fused via a performance-weighted scheme and smoothed with a rolling-window average to produce a bounded health index.In a separate 200-cycle dataset withheld from all training, the Simulink implementation tracks degradation accurately and suppresses transients.After an initial overshoot due to a shorter adaptive window, the HI remains close to 0.4 on a scale of 0 to 1, which is within the expected range for the withheld data from the HI map of the training data.The fused HI remains interpretable, monotonic in expectation, robust to brief operational disturbances, and consistent with training behaviour, and providing a practical basis for real-time maintenance decisions for high-power EMAs used in NRMM.