卷积神经网络与混沌理论在滚动轴承故障诊断中的应用

APPLICATION OF CONVOLUTIONAL NEURAL NETWORK AND CHAOS THEORY IN FAULT DIAGNOSIS OF ROLLING BEARINGS

为解决传统方法在判断轴承所处故障类型中因信号非线性强导致误判与错判,基于混沌理论,采用相空间重构法(Phase Space Reconstruction,PSR)还原系统动力学特性,通过卷积神经网络(Convolution Neural Network,CNN)学习并提取混沌序列中有效非线性信息,提出PSR-CNN智能故障诊断方法,并可视化吸引子轨迹,分析各故障信号非线性特性。以滚动轴承实验数据为研究对象,采用PSR-CNN方法进行轴承早期故障分析与诊断。结果表明:早期微弱故障信号因噪声干扰其吸引子轨迹不具备故障代表性;经CNN学习并提取有效非线性信息后,吸引子轨迹具有显著混沌特征,并呈故障可分状;采用PSR-CNN的故障诊断方法相比基于时域、频域所建立的CNN诊断模型具有更高的准确度与更好的泛化性能,且在收敛速度与稳定性方面均有较大优势。

In order to solve the traditional methods in the process of judging the fault type of the bearing caused by signal strongly nonlinear misjudgment wrongly, based on the chaos theory, the phase space reconstruction method(PSR) is used to restore the system dynamics characteristics and the convolution neural network(CNN) is used to learn and extract the effective nonlinear information from the chaotic sequence, proposes the PSR-CNN intelligent fault diagnosis method, visualizes the attractor trajectory, and analyzes the nonlinear characteristics of each fault signal. Taking the experimental data of rolling bearings as the research object, the PSR-CNN method is used to analyze and diagnose early bearing faults. The results show that the attractor trajectory of the early weak fault signal is not representative of the fault due to noise interference;after learning by CNN and extracting effective nonlinear information, the attractor trajectory has significant chaotic characteristics and shows a fault separable state. The fault diagnosis method using PSR-CNN has higher accuracy and better generalization performance than the CNN diagnosis model based on the time domain and frequency domain, and has greater advantages in convergence speed and stability.