Title: Physically-Constrained Mamba-SDE for Remaining Useful Life Prediction under Irregular Observations

Original: arXiv:2606.01894v1 Announce Type: new Abstract: Accurate Remaining Useful Life prediction is critical for industrial predictive maintenance. However, real-world deployment is challenging due to the irregular nature of sensor observations, characterized by asynchronous sampling, burst missingness, and temporal jitter. Compounding this issue, purely data-driven models often generate physically implausible degradation trajectories that violate the irreversible nature of damage accumulation. To address this, we propose PC-MambaSDE, a unified continuous-time framework for robust RUL prediction under irregular observations. Specifically, we design a Mask-Aware Continuous Mamba Encoder that explicitly leverages observation masks to extract context-rich control signals. Furthermore, we introduce a Physics-Guided Latent SDE with parametrically rectified hybrid drift, superimposing a global physical bias to enforce monotonic degradation even amid severe observation gaps. Additionally, we formulate RUL prediction as a boundary value problem via a Terminal Degradation Penalty, which decouples a Health Index dimension and applies a penalty loss to guide trajectories toward the failure state. Theoretically, we prove that our variational objective is mathematically equivalent to minimizing the KL divergence via Girsanov's theorem, and we guarantee the global asymptotic stability of the learned dynamics through Lyapunov analysis. To enable rigorous evaluation, we develop a Hybrid Irregularity Generation Scheme that simulates realistic industrial imperfections. Extensive experiments on public benchmarks demonstrate that PC-MambaSDE significantly outperforms state-of-the-art methods, particularly under extreme observation scarcity, validating the efficacy of embedding physical priors into continuous-time latent dynamics.

Rewritten: Title: Physically-Constrained Mamba-SDE for Remaining Useful Life Prediction under Irregular Observations

Original: arXiv:2606.01894v1 Announce Type: new Abstract: Accurate Remaining Useful Life prediction is critical for industrial predictive maintenance. However, real-world deployment is challenging due to the irregular nature of sensor observations, characterized by asynchronous sampling, burst missingness, and temporal jitter. Compounding this issue, purely data-driven models often generate physically implausible degradation trajectories that violate the irreversible nature of damage accumulation. To address this, we propose PC-MambaSDE, a unified continuous-time framework for robust RUL prediction under irregular observations. Specifically, we design a Mask-Aware Continuous Mamba Encoder that explicitly leverages observation masks to extract context-rich control signals. Furthermore, we introduce a Physics-Guided Latent SDE with parametrically rectified hybrid drift, superimposing a global physical bias to enforce monotonic degradation even amid severe observation gaps. Additionally, we formulate RUL prediction as a boundary value problem via a Terminal Degradation Penalty, which decouples a Health Index dimension and applies a penalty loss to guide trajectories toward the failure state. Theoretically, we prove that our variational objective is mathematically equivalent to minimizing the KL divergence via Girsanov's theorem, and we guarantee the global asymptotic stability of the learned dynamics through Lyapunov analysis. To enable rigorous evaluation, we develop a Hybrid Irregularity Generation Scheme that simulates realistic industrial imperfections. Extensive experiments on public benchmarks demonstrate that PC-MambaSDE significantly outperforms state-of-the-art methods, particularly under extreme observation scarcity, validating the efficacy of embedding physical priors into continuous-time latent dynamics.

Rewritten: Accurate estimation of Remaining Useful Life (RUL) is a cornerstone of effective industrial predictive maintenance. Nevertheless, implementing these systems in practice is hindered by the irregularity inherent in sensor data, which includes issues such as asynchronous sampling, sudden bursts of missing data, and temporal jitter. This challenge is exacerbated by the tendency of purely data-driven models to produce degradation paths that defy physical reality, specifically by ignoring the irreversible accumulation of damage. To overcome these limitations, we introduce PC-MambaSDE, a comprehensive continuous-time framework designed for resilient RUL prediction despite irregular data inputs. Central to our approach is a Mask-Aware Continuous Mamba Encoder, which utilizes observation masks to derive control signals enriched with contextual information. We also incorporate a Physics-Guided Latent Stochastic Differential Equation (SDE) featuring a hybrid drift term that is parametrically rectified. By applying a global physical bias, this component ensures monotonic degradation trajectories, even when facing significant data gaps. Moreover, we treat RUL prediction as a boundary value problem using a Terminal Degradation Penalty. This mechanism isolates a Health Index dimension and employs a penalty loss to steer the predicted trajectories toward the failure threshold. From a theoretical standpoint, we demonstrate through Girsanov’s theorem that our variational objective corresponds to minimizing the Kullback-Leibler (KL) divergence. Furthermore, Lyapunov analysis confirms the global asymptotic stability of the dynamics learned by the model. For thorough assessment, we created a Hybrid Irregularity Generation Scheme to mimic the complex imperfections found in industrial settings. Our extensive testing on public benchmarks reveals that PC-MambaSDE substantially surpasses current state-of-the-art techniques, especially in scenarios with severe data scarcity. These results confirm the value of integrating physical priors into continuous-time latent dynamic models.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC