MDEL900提梁机卷扬系统多神经网络故障诊断研究

Fault diagnosis of the winch system for MDEL900 girder crane involving multiple neural networks

针对MDEL900提梁机卷扬系统液压故障多样性问题,讨论了传统反向传播(BP)神经网络、粒子群算法优化的反向传播(PSO-BP)神经网络及遗传算法优化的反向传播(GA-BP)神经网络这3种常见诊断方法的优劣性。在AMESim软件中建立了卷扬系统的机电液模型,设置了变量马达、变量泵和阀两两之间的混合故障模型,以此获得故障诊断样本。将样本数据导入MATLAB软件中训练神经网络,使其在不确定工况的情况下,能根据系统最终输出数据判断系统内部损坏位置,并进一步研究不同诊断方法对故障诊断率的影响,为故障诊断的改良提供参考。研究表明,遗传算法和粒子群算法均可提高BP神经网络的诊断率。其中,遗传算法参数更多,可塑性强,但寻优算法较复杂,所需时间较长;粒子群算法参数更少,寻优方式简单,能更快速地找到最优解。

Aimed at the diversity of MDEL900 girder crane winch system hydraulic fault,the advantages of several representative hydraulic fault diagnosis methods including Back Propagation(BP)neural network,Particle Swarm Optimization-BP(PSO-BP)neural network and Genetic Algorithm-BP(GA-BP)neural network were investigated.The electromechanical hydraulic model of the winch system was established by using AMESim software and the hybrid fault models between motor,pump and valve were set up to obtain fault diagnosis samples.After that,the sample data was imported into MATLAB software to train the neural network and the damage position of the system according to the final output data under uncertain conditions was revealed.Furthermore,fault diagnosis accuracy of different diagnosis methods was presented to provide reference for the improvement of fault diagnosis.The research revealed that both GA and PSO algorithm could improve the diagnosis accuracy of BP neural network.GA algorithm has more parameters and strong flexibility,while the complex optimization algorithm is time-consuming.PSO algorithm has fewer parameters,simple optimization method,and can find the optimal solution more quickly.