基于反向传播神经网络的机轮舱衬套加工质量预测方法

机轮舱衬套是一种高精度的薄壁零件,加工过程工艺参数的设置会影响到成品质量,关键尺寸不易保证。为了找到合适的加工工艺参数,根据已有的加工数据,提出一种基于多输出深度神经网络模型的加工质量预测方法。该方法首先对已有的机轮舱衬套加工数据进行数据处理、选取和相关性分析,明确影响加工质量的主要变量;然后以车削主轴速度、车削进给速度、磨削主轴速度、磨削进给速度、时效时间作为加工工艺参数,衬套内径、外径、圆度作为输出进行模型训练,并在150套机轮舱衬套数据上进行了试验验证。结果表明,提出的模型在衬套内径、外径、圆度的预测精度误差分别为0.202%、0.254%、0.274%,其训练的模型能快速预测成品加工质量,避免人工经验参数带来的误差,提高产品质量。

Wheel well bushing is a kind of high-precision thin-wall parts.A method of machining quality prediction based on multi-output deep neural network model is proposed according to existing machining data.Firstly,the data processing,selection and correlation analysis of the existing wheel well bushing processing data are carried out to identify the main variables affecting the machining quality.Then,the turning spindle speed,turning feed speed,grinding spindle speed,grinding feed speed and aging time were taken as the processing parameters,and the inner diameter,outer diameter and roundness of the bushing were taken as the output for the model training,and tested on 150 sets of wheel well bushing data.The results show that the accuracy errors of the proposed model are 0.202%,0.254% and 0.274% respectively in the inner diameter,outer diameter and roundness of the bushing.The trained model can quickly predict the machining quality of the finished product,avoid the errors caused by manual experience parameters,and improve the product quality.