Set up a digital-twin diagnostic model with deep learning

This paper addresses the resource-intensive nature of traditional intelligent fault diagnosis by proposing a computationally efficient yet highly effective digital twin diagnostic model. We transform time-series fault data into 2D image-like representations, known as Gramian Angular Fields (GAF), and then integrate octave convolution into the ResNet50 architecture to extract robust features from machine data. By leveraging the lower complexity of octave convolution, our approach significantly enhances diagnostic efficiency. Experimental results demonstrate that our method achieves over 95% accuracy while reducing computational costs by 42%. And this algorithm can be used for lightweight and efficient fault diagnosis.

I. Introduction

In recent years, with the development of Industry 4.0[1], the demand for equipment fault diagnosis has become increasingly urgent. Efficient and high-quality fault diagnosis of equipment is the guarantee for ensuring its normal operation [2]. The digital-twin diagnostic model uses sensor data from device operation to establish a diagnostic model that can reflect the actual operating status of the device [3].