Design of a low-power intelligent system for optimizing railway tunnel settlement data stream processing using TinyMLt

Abstract

The advancement of TinyML technologies has opened new avenues for designing low-power intelligent systems, significantly enhancing the processing capabilities of embedded devices in critical infrastructure. To overcome these limitations, we propose a Dynamic Geomechanical Prediction Framework (DGPF) integrated with a novel Adaptive Learning-Informed Simulation (ALIS) strategy. The DGPF combines physics-informed simulations with data-driven corrections, leveraging finite element modeling and machine learning techniques to capture intricate soil-tunnel interactions. The ALIS strategy enhances model adaptability and efficiency by incorporating real-time monitoring data, adaptive parameter calibration, and multi-fidelity modeling. This study introduces a low-power intelligent framework for railway tunnel settlement prediction, leveraging TinyML to enhance scalability and energy efficiency. By integrating physics-informed simulations with data-driven corrections through a Dynamic Geomechanical Prediction Framework (DGPF), the proposed approach addresses the nonlinear complexities of settlement processes. The framework achieves a 35% reduction in prediction errors and a 50% improvement in computational efficiency compared to traditional methods. Experimental results demonstrate its practical applicability for real-time settlement monitoring in safety-critical geotechnical engineering scenarios.