Reliable heavy machinery requires accurate health assessments of its hydraulic systems. Existing data-driven models often fail to track long-term degradation trends while concurrently ignoring the physical laws governing wear. This oversight produces predictions that contradict the natural irreversible progression of equipment faults. This study introduces the Physics-Informed Monotonic Conformer to address this specific problem. The proposed model combines convolutional inductive biases with global self-attention to merge multi-scale spatiotemporal features. We also implement a monotonicity loss function to enforce physical degradation constraints. This step grounds the purely data-driven network in actual physical realities. Testing on an electrohydrostatic actuator dataset shows the new method surpasses current baseline models. The regularization mechanism also significantly improves physical consistency, yielding a high Spearman rank correlation. The resulting health indicators offer the numerical precision and physical reliability necessary for safety-critical aerospace deployment.
He et al. (Wed,) studied this question.