Hardware-in-the-Loop Evaluation of On-Chip Deep Learning and Smart Contract Execution for Securing Embedded AMI Devices Against FDI Attacks

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Advanced Metering Infrastructure (AMI) devices face growing threats from false data injection (FDI) attacks that can manipulate energy consumption data and compromise grid operations. While machine learning and blockchain approaches have shown promise for detecting such attacks, their practical deployment on resource-constrained embedded smart meters remains unvalidated. This study presents a hardware-in-the-loop (HIL) evaluation framework that integrates an on-chip deep learning classifier with smart contract-based integrity verification for embedded AMI devices. The methodology employs a two-stage architecture: a quantized neural network deployed on an ARM Cortex-M7 microcontroller for real-time FDI detection, and an Ethereum-compatible smart contract executing on an FPGA-based hardware security module for tamper-proof logging. The HIL testbed reproduces realistic AMI communication constraints using MQTT protocol and evaluates detection performance under single and multi-feature FDI attack scenarios. Experimental results demonstrate that the hybrid architecture achieves 89.4% detection accuracy with 12.7 ms inference latency and 94.2% smart contract execution success rate, while maintaining energy consumption below 45 mJ per inference. The framework outperforms centralized detection baselines by 8.3 percentage points in recall for stealthy attacks. These findings establish that on-chip deep learning combined with smart contract verification is feasible for embedded AMI security, providing a replicable HIL evaluation methodology for resource-constrained cyber-physical systems.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.