叶片生产过程的非线性轮廓控制方法研究

Nonlinear Profile Control Method for Blade Production Process

针对叶身型线轮廓特征参数间相关关系复杂、历史批次数据不足等原因造成的生产过程型面几何轮廓质量控制困难的问题,提出一种叶片生产过程非线性轮廓控制方法。该方法以叶身型线三坐标值为基础数据进行分析,首先构建叶身型线轮廓差异度量指标,并基于测量数据进行指标参数提取及标准化处理,作为非线性轮廓控制对象;其次以受控状态下的差异度量指标分类准确率为优化目标,通过自适应粒子群算法(APSO)优化高斯核函数及惩罚系数;然后联合优化参数及受控指标数据训练支持向量数据描述(SVDD)模型,获得超球体半径作为控制限,构建基于超球体核距离的非线性轮廓控制图;最后计算待测轮廓指标数据点到超球体中心的内核距离,得到控制图标绘点,进而判断叶片生产过程是否异常。仿真结果表明,相比于传统方法,该方法可有效表征型线不同区域质量特征对轮廓形状的综合影响,且对不同均值偏移都具有更强的异常波动检测力,并能有效解决中小批量生产过程控制中数据量不足的问题。

A nonlinear profile control method for blade production process is proposed to deal with the problem that the quality control of the surface contour of production process is difficult due to the complex relationship between the profile parameters of blade body and lack of historical data.According to three-coordinate data of blade profile,the blade body profile difference metrics are constructed,and the index parameter extraction and standardization processing are performed based on the measured data,which are taken as the nonlinear contour control objects.Then the classification accuracy of the difference metrics under controlled state is considered as the optimization goal,and the parameters of Gaussian kernel function are optimized by the adaptive particle swarm optimization.The SVDD model is trained by combining the optimized parameters and controlled index data to obtain the radius of the hypersphere as the control limit,and the nonlinear profile control chart based on the hypersphere kernel-distance is constructed.The distance between the measured contour index data point and the center of the hypersphere is calculated to obtain the statistical point of the control chart,and then it is determine whether the blade production process is abnormal.Compared with the traditional methods,the proposed method can effectively characterize the comprehensive influence of the quality characteristics of different regions of the profile on the contour shape,and has a stronger detection ability for abnormal fluctuation of different mean deviation to solve the difficulty of insufficient data in the process control for small and medium-sized batch production.