XJTU-SY+IMS数据集

针对工业环境中广泛在多工况下多滚动轴承实时状态监测的需求和部署环境受限的挑战，提出一种基于卷积神经网络(Convolutional Neural Network, CNN)的面向多传感器滚动轴承运行状态监控方法。该方法将两个不同工况下的一维时间序列数据集以均方根(Root Mean Square, RMS)指标标注，并通过对一维时间序列多传感器数据重构为二维空间张量的形式输入卷积神经网络训练。最后利用层融合和16比特量化优化，将网络部署到FPGA上，用以解决CNN的计算开销。实验结果表明，在结合了两种不同工况的数据集下，网络测试推理准确度依然高达99.24%，比多层感知机实现高10.48%，比多层感知机结合支持向量机的实现高2.91%，该算法对于新加入的数据集也有较强的鲁棒性，经过重训练，新加入的数据集准确率可以达到99.17%的准确率。基于FPGA部署优化的网络的峰值能效为76.217GPOS/W，为CPU实现的33.09倍，GPU实现的5.39倍。其中，16比特精度部署的网络测试精度相较32比特精度实现仅降低0.001%。

Aiming at the demand for real-time condition monitoring of multiple rolling bearings under various working conditions in industrial environments and the challenge of limited deployment environments, this work proposes a multi-sensor rolling bearing operational condition monitoring method based on Convolutional Neural Network (CNN). This method labels one-dimensional time series datasets collected under two distinct working conditions using the Root Mean Square (RMS) metric, and feeds the reconstructed two-dimensional spatial tensors converted from multi-sensor one-dimensional time series data into the CNN for training. Finally, layer fusion and 16-bit quantization optimization are applied to deploy the trained network onto FPGAs, mitigating the computational overhead of CNNs. Experimental results demonstrate that even when trained on the dataset combining two different working conditions, the network still achieves a test inference accuracy as high as 99.24%, which is 10.48% higher than that of a Multi-Layer Perceptron (MLP), and 2.91% higher than that of an MLP combined with a Support Vector Machine (SVM). The proposed algorithm also demonstrates strong robustness against newly added datasets; after retraining, the accuracy of the model on the newly added datasets can reach 99.17%. The FPGA-deployed optimized network has a peak energy efficiency of 76.217 GPOS/W, which is 33.09 times that of the CPU implementation and 5.39 times that of the GPU implementation. Specifically, the test accuracy of the network deployed with 16-bit precision only decreases by 0.001% compared to the 32-bit precision implementation.

搜集汇总

背景与挑战

背景概述

XJTU-SY+IMS数据集是一个针对工业环境中多工况下滚动轴承实时状态监测的数据集，采用卷积神经网络(CNN)方法进行状态监控，并通过FPGA部署优化实现高效计算。数据集包含34.44MB的301个文件，主要格式为tif和docx，实验结果显示其具有高准确率和强鲁棒性。