黄焯晖 宋卓明 晋刚 梁成就

摘要：针对注塑成型生产过程工况多而难于质量监测的问题，利用统计模量（statistics pattern， SP）简化数据结构、避免复杂的数据预处理，采用支持向量数据描述（support vector data description， SVDD）算法解决多模态问题，并提出一种基于SP-SVDD的多工况注塑成型质量监测方法。以螺杆位移和模腔压力为样本数据的采集对象，提取过程数据的SP并建立SVDD模型。将基于SP-SVDD的监测方法与传统统计模量分析（statistics pattern analysis， SPA）监测方法进行对比，结果表明：基于SP-SVDD监测方法的准确率远高于传统SPA监测方法，SP-SVDD监测方法的故障集监测准确率为96.67%，且能同时监测不同工况的故障集，可以为多工况过程的质量监测提供参考。

关键词：

注塑成型; 统计模量; 质量监测; 支持向量数据描述; 故障监测

中图分类号：TQ320.66;TP391.92

文献标志码：B

Quality monitoring method of injection molding based on

statistics pattern and support vector data description

HUANG Zhuohui1， SONG Zhuoming1， JIN Gang1， LIANG Chengjiu2

（1.National Engineering Research Center for New Polymer Forming Equipment; Key Laboratory of Polymer

Processing Engineering（Ministry of Education）， South China University of Technology， Guangzhou 510640， China;

2.Datong Machinery Group Donghua Machinery Co.， Ltd.， Dongguan 523000， Guangdong， China）

Abstract：

As to the issue that it is difficult to monitor the quality of injection molding because of multiple working conditions， the statistical pattern（SP） is used to simplify the data structure and avoid the complicated data preprocessing， and the support vector data description（SVDD） algorithm is used to solve multimodal problems. An injection molding quality monitoring method based on SP-SVDD is proposed. Taking the screw displacement and the cavity pressure as the sample data collection objects， the SP of process data is extracted and the SVDD model is established. The monitoring method based on SP-SVDD is compared with the traditional statistical pattern analysis（SPA） monitoring method. The results show that the accuracy of the SP-SVDD monitoring method is much higher than that of the traditional SPA monitoring method， and the monitoring accuracy of SP-SVDD on the fault set is 96.67%. At the same time， the fault sets of different working conditions can be monitoredby SP-SVDD， which can provide a reference for the quality monitoring of multiple working conditions.

Key words：

injection molding; statistics pattern; quality monitoring; support vector data description; fault monitoring

0 引 言

注塑成型是塑料行業最重要的加工方法之一，也是一种典型的间歇加工过程。[1]间歇加工过程对应的工况往往比较复杂，注塑成型的工况是指当前注塑机的生产条件，如原材料、模具和工艺参数等。当注塑过程存在原材料变更、产品变换或外部环境变化等情况时，为满足生产需求，需要对注塑机进行工艺参数调整、模具更换等操作，即注塑成型的工况发生变化。

在生产加工过程中，质量监测与产品性能密切相关。近年来，数据驱动过程监测方法发展迅速，该方法依靠过程数据识别过程异常，不依赖于先验知识，可用于注塑过程监测。[1]多工况过程具有数据多模态[2]、数据不等长 [3]、三维结构多[4]等特点。……