基于注意力机制的互特征融合旋转机械故障检测技术

Rotating Machinery Fault Detection Technology for Mutual Feature Fusion Based on Attention Mechanism

旋转机制在生产生活中的应用愈加广泛;但旋转机械存在应用环境较为复杂,生产环境恶劣,各部件相互影响,单一信号无法完整表现故障特征等问题;针对此问题,研究根据注意力机制构建卷积神经网络,在网络结构中引入自注意力模块,采用多信号源进行数据提取,将不同信号特征互补融合并构建旋转机械故障检测模型,同时使用傅里叶变化进行数据优化;实验结果表明,构建模型的故障分类准确率为99.92%,比第二优的算法高出1.89%,故障检测精度达到了99.64%,数据进行傅里叶变换后的检测精度平均提升了17.32%;由此可得,构建的故障检测模型能够有效提取并融合不同数据采集的故障特征,大幅提升旋转机械的故障检测精度,且将数据特征融合模块加入模型中能够有效减少单独计算的运行成本,提高运算速度,减少因机械故障产生的生产安全事故。

Rotating mechanism is widely applied in production and life.However,the application environment of rotating machinery has the problems of complex application environment,harsh production environment,mutual influence of various components,and inability of a single signal to fully show fault characteristics.To solve this problem,a convolutional neural network is constructed according to the attention mechanism,self-attention module is introduced into the network structure,multiple signal sources are used for data extraction,the complementary fusion of different signal features is carried out to build a rotating machinery fault detection model,and Fourier transform is used for data optimization.Experimental results show that the fault classification accuracy of the constructed model is 99.92%,1.89%higher than that of the second best algorithm,with a fault detection accuracy of 99.64%.The the detection accuracy of the data is improved by an average of 17.32%after Fourier transform.It can be concluded that the fault detection model constructed can effectively extract and fuse fault features of different data acquisition,greatly improving the fault detection accuracy of rotating machinery.Moreover,the data feature fusion module is added in the model to effectively reduce the operating cost of separate calculation,improve the calculation speed,and reduce the production safety accidents caused by mechanical faults.