采用改进遗传神经网络的多载荷振动信号故障诊断

Multi-load Fault Diagnosis of Vibration Signal Based on Improved Genetic Neural Network

根据柴油机气阀机构运动规律,利用小波包分解提取缸盖振动信号的特征向量;针对多种载荷混合诊断的问题,采用二进制与实数混合编码的方式对使用遗传算法的误差反向传播(BP)神经网络的隐层结点数目、权值和阈值进行优化。通过实验检测,证明该方法在多种载荷混合振动信号诊断上,较一般方法学习、收敛速度快,检测准确率高。

According to the motion law of diesel engine valve, the characteristic vector of cylinder-cover＇ s vibration signal is extracted by wavelet packet decomposition. For multi-load fault diagnosis, the hidden layer node number, weights and threshold of the back propagation genetic algorithms are optimized by binary and real value hybrid coding. Experiment results show that the method has obvious advantages on multi-load vibration signal fault diagnosis. It is able to improve the network learning ability, convergence speed and accuracy of detection.