Abstract: In the field of automobile assembly,the process is complicated with the rapid production cycle,large the output.Meanwhile more than 60% of the time is in the main and auxiliary materials handling.The transportation equipment will cause huge economic losses due to the shutdown caused by the failure.To evaluate the health status of production logistics equipment and make an early warning when a failure occurs,a warning method for vehicle assembly equipment failure based on growth-type neural gas clustering algorithm and improved Least Square-Support Vector Machines(LS-SVM) regression model was proposed.The feature extraction and dimension reduction processing were performed according to the historical signal data of the sensor to obtain the feature vector.The Growth state Neural Gas (GNG) algorithm was used to divide the normal state data into multiple working conditions,and several cluster centers were obtained.The feature vector obtained by current operational data and the Euclidean distance of cluster center were calculated to obtain a similarity trend.At the same time,the historical memory matrix was constructed,and the particle swarm optimization algorithm was used to optimize the parameters of LS-SVM regression model,calculate the residual value and obtain the risk coefficient by combining the residual value and the similarity trend,which evaluated and warned the equipment status.The method was applied to the automobile assembly equipment,and the Root Mean Square (RMS) of vibration value of reducer and bearing was input into the model to obtain the risk factor of equipment.The experiment proved the effectiveness of the method.

Key words: conveying equipment, fault warning, growing neural gas clustering algorithm, improved regression model least square-support vector machines, automobile assembly

摘要： 汽车总装工艺复杂、生产节拍快、产量大,百分之六十以上的时间都在进行主辅物料输送,因此输送装备由于故障引起的停机会造成巨大的经济损失。为了对生产物料输送装备的健康状态作出评估,并在可能发生故障时作出预警,提出一种基于生长型神经气聚类算法与改进最小二乘法支持向量机(LS-SVM)回归模型的汽车总装输送装备故障预警方法。首先根据传感器的历史信号数据进行特征提取和降维处理,获得特征向量;运用生长型神经气聚类算法,将正常状态数据划分为多种工况,得到若干聚类中心,并计算当前运行数据的特征向量与聚类中心的欧式距离从而得到相似度趋势;同时构建了历史记忆矩阵,并通过改进粒子群算法优化LS-SVM回归模型参数,计算残差值,并结合残差值与相似度趋势,得出风险系数,对装备状态进行评估和预警。将所提方法应用于汽车总装物料输送设备,将减速器与轴承的振动值的均方根输入模型,得出设备的风险因子,证明了该方法的有效性。

关键词: 输送装备, 故障预警, 生长型神经气聚类算法, 改进回归模型, 最小二乘法支持向量机, 汽车总装