Distributed Predictive Maintenance Architecture for Edge Sensors Networks: An Optimal Regression Based Machine Learning Model

Abstract

In predictive maintenance (PdM), implementing machine learning models (ML) on edge sensor hardware is particularly challenging. This is due to power constraints which significantly reduce computational performance in conventional embedded processors such as central processing units (CPUs) and microcontroller units (MCUs). However, Field Programmable Gate Arrays (FPGAs) have been identified as an ideal processing unit to overcome this, providing hardware acceleration of models on the edge. With low-precision data, FPGAs have been shown to outperform conventional processing units both in terms of giga-operations-per-second (GOPS) and power consumption. This research seeks to establish an effective methodology for implementing high-level ML regression models on FPGAs within edge sensors.