Autonomous Sustainable Sensing Nodes Based on Joint Design of Edge AI and Computational Waveform in Industrial IoT

Abstract

Industrial IoT sensing nodes face a fundamental tension among limited energy budgets, constrained computational resources, and growing demands for intelligent real-time sensing, which severely restricts the feasibility of large-scale autonomous deployment. To address this, a joint design framework is proposed that incorporates edge AI inference configuration and computational sensing waveform parameters into a unified energy consumption model. By explicitly establishing the coupling relationships among channel signal-to-noise ratio, lightweight neural network inference overhead, and energy harvesting constraints, a mixed-integer nonlinear programming objective function is constructed and solved via an alternating optimization algorithm that decomposes the original problem, enabling real-time scheduling complexity to meet the processing capability constraints of embedded nodes. Experimental results demonstrate that the proposed scheme reduces average per-cycle node energy consumption by 31.4% compared to a separated-design baseline, maintains sensing accuracy above 92.3% under dynamic industrial channel conditions, and achieves continuous power-on survival throughout a 72-hour validation period. Although the system maintained electrical viability, it experienced brief transitions into Minimum Survival Mode to prioritize energy replenishment, during which high-frequency sensing was temporarily suspended to prevent complete depletion.