Edge-AI Power Auditor: Decentralized Harmonic Analysis and NILM-based Diagnostics

Abstract

Conventional energy meters measure only total consumption and give no information about harmonic distortion or load faults. This paper presents an edge intelligence power auditor on an STM32F411 ARM Cortex-M4 that performs real-time harmonic analysis and three-state fault classification with no cloud dependency. Voltage and current are acquired using a ZMPT101B sensor and SCT-013 clamp with LM358 conditioning, sampled at 5000 Hz through the 12-bit ADC. A 1024-point CMSIS-DSP FFT extracts harmonics up to the seventh order and computes THD, RMS voltage, RMS current, and power factor. A rule-based classifier using THD, power factor, and H3/H1 ratio identifies load conditions as Healthy, Degraded, or Faulty, with the 10% THD boundary aligned with IEEE 519. Hardware testing confirmed correct classification across all three states, with THD of 1.97% healthy and 39.84% faulty. Results are shown on an SSD1306 OLED, and a relay disconnects the load on fault detection. All processing runs on-chip with no external data transmission, making the system suitable for rural, off-grid, and small-industry use.