多源数据融合与改进注意力机制的轴承智能故障诊断

Intelligent Fault Diagnosis of Bearing Based on Multi-Source Data Fusion and Improved Attention Mechanism

关键词:  多源数据融合  注意力机制  卷积神经网络  特征提取  故障诊断  滚动轴承

Key words:  Multi-source data fusion  Attention mechanism  Convolutional neural network  Feature extraction  Fault diagnosis  Rolling bearing

摘要:

单一传感器信号不能全面表达机械设备的运行特征且容易受到自身品质、性能的影响，为此本文提出了一种多源数据融合与改进注意力机制相结合的滚动轴承智能诊断方法。采集不同位置的传感器振动信号作为模型的输入向量，每一个传感器信号作为一个通道，将多通道信号同时送入模型特征输入层；引入改进注意力机制建立各通道和空间动态权重参数，随着模型训练，不断增强故障特征、弱化无用特征；运用深度卷积神经网络模型的卷积、池化等操作将多传感器信号进一步融合并提取故障特征，输出诊断结果。在进行滚动轴承故障诊断实验时，该方法诊断准确率达到100%，高于准确率最佳值为97.42%的单传感器。与其他方法相比，本文方法可以自适应融合多传感器数据以满足诊断任务的要求，具有良好的自适应性和鲁棒性，为滚动轴承的故障诊断提供了一种可行的方法。

Abstract:

A single sensor signal cannot fully express the operating characteristics of mechanical equipment and is easily affected by its quality and performance. Therefore， an intelligent diagnosis method combining multi-source data fusion and an improved attention mechanism is proposed. Firstly， the sensor vibration signals at different positions are collected as the input vector of the model， each sensor signal is taken as a channel， and the multi-channel signals are input to the feature input layer of the model at the same time. Secondly， the improved attention mechanism is introduced to establish the dynamic weight parameters of each channel and space. With the model training， fault features are continuously enhanced， and useless features are constantly weakened. Finally， through the convolution and pooling of the deep convolution neural network model， the multi-sensor signals are further fused， the fault features are extracted， and the diagnosis results are output. In the case of rolling bearing fault diagnosis， the improved method had a diagnostic accuracy of 100%， compared with accuracy of 97.42% for the best single sensor. Compared with other methods， this method can adaptively fuse the multi-sensor data to meet the requirements of the diagnosis task， has good adaptability and robustness， and provides a feasible method for the fault diagnosis of rolling bearing.