Embedded AI for Autonomous Wireless Sensor Networks in Aerospatial Vehicle SHM

The Structural Health Monitoring (SHM) of aerospace vehicles faces extreme challenges due to harsh operating conditions, stringent latency requirements, high reliability constraints, and limited energy resources. While conventional wired SHM architectures remain reliable, they significantly increase the total weight and complexity of the vehicle’s electrical system. This weight penalty is particularly pronounced in commercial and military aircraft, where the extensive wiring infrastructure has become a critical design constraint. According to the Wireless Avionics Intra-Communications (WAIC) initiative, modern aircraft can have approximately 100,000 wires totaling 470 km in length and weighing around 5,700 kg, with about 30% of these wires being potential candidates for wireless substitution to reduce weight and simplify installation. Consequently, there is a growing trend toward adopting wireless sensor networks (WSNs) to maintain reliability while reducing wiring mass. While WSNs offer promising weight reduction benefits, current implementations lack the adaptive intelligence required to satisfy the complex, multi-dimensional QoS constraints of aerospace systems, particularly when facing simultaneous demands for minimal latency, maximum reliability, and optimal energy consumption. These multi-objective constraints necessitate more intelligent network management, that can autonomously adapt to changing mission requirements and environmental conditions. This work presents an embedded Artificial Intelligence (AI) framework for autonomous WSNs dedicated to aerospace vehicle SHM. The proposed system leverages Time Slotted Channel Hopping (TSCH) and RPL routing protocols, enhanced by an adaptive AI-based controller (CEREBRO), that dynamically optimizes communication schedules based on real-time network conditions and environmental variations. The proposed AI framework combines convolutional layers for capturing local spatial structures with specialized temporal processing blocks designed to model sequential dependencies and temporal dynamics in the data. The combined model outputs adaptive routing and scheduling vectors, which CEREBRO uses to adjust slot allocations and routing paths based on network operations and conditions, ensuring that critical measurements are delivered reliably and efficiently while extending network lifetime in complex aerospace environments. The proposed approach has been experimentally validated on a real testbed using the FIT IoT- LAB platform, showing significant improvements in reliability, latency, and energy consumption compared to static TSCH configurations. The resulting architecture provides a scalable, lightweight, and energy-efficient foundation for embedded SHM in aerospace structures. Overall, this work highlights the potential of embedded intelligence to transform SHM systems into fully autonomous and adaptive diagnostic networks, capable of operating reliably and efficiently in the extreme environments of aerospace systems. To accommodate different deployment scenarios, CEREBRO is available in both a CPU implementation for resource-constrained nodes and a GPU- accelerated version on NVIDIA Jetson platforms for high-performance edge computing. The framework advances the paradigm of edge-based diagnostics by bringing intelligence directly to the acquisition system, enabling real-time analytics and adaptive decision-making without relying on centralized cloud processing.