基于一种改进的一维卷积神经网络电机故障诊断方法

故障诊断对于保障电机正常运行有着重要意义，卷积神经网络（CNN）对单一电机故障有着良好的诊断效果.然而传统CNN在处理不同尺寸的数据上存在局限性.针对这一问题，提出了一种基于空间金字塔池化和一维卷积神经网络相结合的故障诊断方法与参数优化策略.该方法不仅使网络可以处理不同尺寸的数据，还降低了网络结构的复杂性和所需运算量.所提出的参数优化策略从理论上解决了诊断过程中可能会发生的金字塔池化的尺度失配问题.仿真结果表明，与传统网络相比，所提出的方法提高了网络结构的鲁棒性与泛化能力，可以更加快速准确地实现电机的故障诊断. 

Motor Fault Diagnosis Method Based on an Improved One-Dimensional Convolutional Neural Network

Fault diagnosis is essential to ensure proper operation of the motor. Convolutional neural network (CNN) has showed a better performance on diagnosing single motor faults. However, traditional CNN has limitations in dealing with different sizes of data. To solve this problem, a fault diagnosis method was proposed based on spatial pyramid pooling (SPP), one-dimensional convolutional neural network and a parameter optimization strategy. The method was arranged to make not only the network be possible to process different sizes of data, but also reduce the complexity of the network structure and the amount of computation required. The parameter optimization strategy was designed to solve the scale mismatch problem in pyramid pooling during the diagnosis process. The simulation results show that, compared with the traditional network, the proposed method can improve the robustness and generalization ability of the network structure, making the fault diagnosis more quickly and accurately for motor.