基于主元分析与小波神经网络的脱粒滚筒故障诊断

Fault diagnosis of threshing cylinder based on PCA and wavelet neural network

针对联合收割机脱粒装置受地形与作物姿态影响易发生滚筒堵塞故障的问题,提出了一种基于主元分析与小波神经网络模型技术的多参数故障监测与诊断方法。首先,选取脱粒滚筒的多个工作参数作为故障诊断研究对象;其次,通过主元分析方法预处理脱粒滚筒工作参数的相关性,将影响脱粒滚筒的四个参数减少为无相关性的三个指标变量,新的特征向量实际包含了收割机滚筒作业时的吞入量、饱和量、堵塞量三个信息;最后,以新的特征向量建立脱粒滚筒堵塞故障小波神经网络诊断模型,并与前向反馈神经网络模型、传统的额定转速监测诊断方法相比较。实验分析表明,与前向反馈神经网络相比,在网络结构与计算复杂度方面,小波神经网络更适合脱粒滚筒故障诊断,收敛速度更快;与传统的设置脱粒滚筒轴额定转速的方法相比,多参数监测与诊断方法可靠性更高。实验结果表明,基于主元分析及小波神经网络的故障诊断能有效避免脱粒滚筒堵塞故障诊断分析的片面性。

Since the combine harvester threshing device is affected by the topography and crop postures and prone to lead to jam fault,a Principal Component Analysis（ PCA） and wavelet neural network technology based on multi-parameter fault monitoring and diagnosis method was put forward.First,several main working parameters of the threshing cylinder were selected as the fault diagnosis research basis.Secondly,the PCA method was used to pre-process the correlation between working parameters of threshing.The four working parameters were reduced to the three index variables which had no correlation between them.New feature vector contained the harvester threshing operation information of swallowed volume,saturation capacity,and blockage magnitude.Finally,threshing jam fault wavelet neural network diagnosis model was established with the new feature vector,and separately compared with BP neural network and the traditional diagnostic methods of speed monitoring.In the comparison experiments with BP neural network,from the network structure and the computation complexity,wavelet neural network had been more suitable for fault diagnosis of the threshing with faster convergence speed;compared with the traditional threshing roller shaft of the rated speed,multi-parameter monitoring measurement and diagnosis has higher reliability.Experimental results show that fault diagnosis based on PCA and wavelet neural network can effectively avoid the sidedness of the threshing jam fault diagnosis analysis.