Adaptive water-mist infrared signature suppression for naval vessels via MPCM-constrained LSTM

Abstract

Infrared stealth is critical to the survivability and combat effectiveness of modern naval vessels, and fine water-mist spraying, as a mature and effective infrared-stealth technique, reduces ship detectability in the $8\sim 12\mu m$ infrared band through cooling and scattering. However, conventional water-mist systems lack responsiveness to dynamic environments, resulting in unstable stealth performance. To address this issue, an adaptive water-mist infrared-stealth optimisation approach integrating a multi-physics coupling model (MPCM) and a long short-term memory (LSTM) neural network is proposed in this study. First, environmental, navigational and device-level data are collected and fused to construct a unified input state; then, an MPCM is established to simulate the coupled physical processes of ship infrared radiation, temperature distribution and water-mist diffusion, thereby producing physics-constrained high-fidelity labels for training the control model; subsequently, an LSTM model is trained on historical and real-time feature windows to predict the optimal spraying parameters for the next time step; finally, background-difference-ratio-based thresholding is combined with virtual spray optimisation (VSO) to realise a dual closed-loop feedback mechanism. Experimental results indicate that, compared with non-adaptive baseline schemes, the proposed method reduces the peak infrared radiance by $18.3\%$, decreases the number of extreme hot spots by $57.1\%$, compresses the target–background temperature difference to $7.63°C$, and lowers the total water consumption over 13 h to $8,075m^3$. Moreover, the control system operates stably at 1 Hz with an end-to-end latency below 0.451 s, demonstrating that the method simultaneously achieves stronger suppression, reduced water consumption and real-time compliance, thereby providing a feasible route for the engineering deployment of shipborne infrared stealth.