基于动态学习率与BP神经网络的变速箱故障诊断研究

Transmission Fault Diagnosis Based on Dynamic Vector and BP Neural Network

变速箱是汽车传动系的重要组成部分,因此对变速箱常见故障类型进行诊断研究很有必要。以四种变速箱常见故障类型为研究对象,利用动态学习率对BP神经网络进行改良,建立了变速箱故障诊断网络模型。通过测量并提取已发生故障的信号特征参数,收集大量信息数据作为已知样本来训练某状态下的神经网络,再用其它转速下变速箱故障数据对网络进行验证。对于学习率的恰当改变可以提升网络的速度和稳态性。诊断结果表明,网络模型通过对已知故障数据样本的学习,实现了对变速箱未知故障的诊断。

The gearbox is an important component of the automobile drive train, so common fault types of gearbox diagnosis research is necessary. In this paper, four types of common fault gearbox as the research object, using the vector dynamic to improved Back Propagation neural network, transmission network fault diagnosis model is established. Through the measurement and extract has the characteristic parameters of breakdown, collect a lot of information as the known sample data to train the neural network under some condition, then use other speed gearbox failure data to verify this network. More appropriate change can accelerate the training speed of network, guarantee the stability of the network. Diagnostic results show that the network model by learning from samples of known fault data, the implementation of the unknown fault diagnosis of gearbox.