基于工况识别的注塑过程产品质量预测方法

针对多工况注塑过程的在线质量预测问题,考虑了过程数据高维、耦合、非线性等特点,采用拉普拉斯特征映射(LE)方法实现过程数据的非线性降维;在低维特征空间中采用Mean Shift聚类算法完成样本的工况聚类,以便注塑过程的工况分析和知识挖掘;同时运用Mean Shift原理,提出一种新样本的在线工况识别方法;最后应用基于混合粒子群(PSO)参数寻优的偏最小二乘支持向量机(PLS-LSSVM)方法,建立了多工况注塑过程的产品质量软测量模型。实验结果表明,相较于PLS-LSSVM方法,本文方法的预测精度和泛化性能均有明显提高,可为实际注塑企业提供一种效果良好的多工况产品质量在线预测方法。

Product quality prediction method for injection molding process based on operating mode recognition

Taking into consideration the high-dimensional,correlated and nonlinear process data in the injection molding process with multiple operating modes,an online product quality prediction method was developed based on offline clustering and online recognition of the operating modes.Firstly,a nonlinear dimension reduction method,Laplacian Eigenmap (LE),was used to project the high-dimensional process data onto a low-dimensional feature space,where a Mean Shift based clustering algorithm was used to obtain the underlying operating patterns in the injection molding process.Meanwhile,an online operating mode recognition algorithm was proposed by making use of the principle of Mean Shift.After that,a PLS-LSSVM prediction method,where the particle swarm optimization (PSO)method was used for parameter determination,was used to develop the soft-sensor model of product quality for each operating mode.A multi-mode quality prediction method was finally developed by combining LE,Mean-Shift clustering and PLS-LSSVM algorithms.The experimental results showed that the proposed method outperformed the standard PLS-LSSVM method,and it could become a useful tool for online quality prediction for the industrial injection molding process with multiple operating modes.