Embedded TinyML for Predictive Maintenance: Vibration Analysis on ESP32 with Real-Time Fault Detection in Industrial Equipment
Year: 2025
Type: article
Abstract: The rapid evolution of embedded intelligence within industrial environments has catalyzed the development of lightweight, real-time predictive maintenance systems. Conventional fault diagnosis approaches often depend on centralized, resource-intensive infrastructures that are ill-suited for distributed and energy-constrained settings. Addressing these limitations, this paper introduces a TinyML-based framework for real-time vibration analysis and fault detection deployed on the ESP32 microcontroller, a cost-effective, ultra-low-power embedded platform. Vibration data—acquired using a triaxial ADXL345 accelerometer—serve as key indicators of mechanical integrity, enabling the early identification of anomalies such as misalignment, imbalance, and bearing defects. The proposed system features an optimized 1D convolutional neural network (CNN) designed to operate within the memory and processing limitations of the ESP32. The architecture incorporates adaptive sampling, in-situ feature extraction, and edge-based classification, allowing for autonomous decision-making without cloud dependency. A custom dataset encompassing four machine states—normal, misaligned, imbalanced, and bearing-worn—is created using controlled experimental setups to simulate real-world operational conditions. Two deep learning models are implemented and compared for performance in terms of accuracy, memory usage, and inference time on-device. Results demonstrate that the proposed TinyML approach achieves over 92% fault detection accuracy while maintaining a compact computational footprint. This framework offers a scalable, low-latency solution for predictive maintenance in Industry 4.0 applications, reducing unplanned downtime and enhancing machine reliability. The integration of vibration-based analysis with embedded machine learning advances the field toward decentralized, real-time condition monitoring in smart industrial systems.
Authors: Shubbham Gupta, Shiv Naresh Shivhare
Institutions: University College Dublin, Bennett University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4410417356