Quantized Probabilistic AI for Gear Fault Diagnosis in Motor Drives

TL;DR

This paper introduces a quantization-aware training method for Bayesian Neural Networks, enabling efficient gear fault diagnosis in motor drives with minimal accuracy loss and significant computational gains.

Contribution

It demonstrates a novel quantization approach that reduces model precision to INT8 without sacrificing accuracy or uncertainty estimates in power electronics applications.

Findings

Quantization reduces computational load by 30-45%.

Accuracy and uncertainty estimates are preserved after quantization.

Enables deployment of lightweight AI models on low-cost edge processors.

Abstract

Deploying large artificial intelligence (AI) models in power electronics often demands high computational resources. Driven by the quantization paradigm, this digest proposes a quantization-aware training (QAT) principle to substantially minimize the number of bits required and simultaneously maximize the accuracy of computations in pre-trained AI models. Considering a pre-trained probabilistic Bayesian Neural Network (BNN) for gear fault diagnosis in motor drives as an example, we quantize its weights and activation functions from floating-point FP32 to low-precision INT8 values, which enhances the computational efficiency by a significant margin of 30-45% (for different model versions) without any compromise in the accuracy and uncertainty estimates. This substantiates a sustainable mechanism of deploying most quantized light-weight AI models into low-cost edge processors for power electronic applications.