关键词: 产品质量, 关联规则挖掘, 时间序列预测模型, 机器学习 

Abstract: The dimensional accuracy of injection products was related to the injection processing parameters, and the real-time conditions of each stage in the injection processes and the changes of real-time working conditions. A workpiece quality diagnosis model was developed herein based on data mining. The real-time data collected by high-frequency sensors such as temperature, pressure, and displacement etc. in mold were used to construct the high-dimensional time series feature set. A three-stage feature selection method was used to determine the key feature subset, which was used to train the online quality detection model based on LightGBM classifier. The future values of each features were predicted based on the CNN-LSTM temporal prediction model, and the product quality was forecasted in advance with the classifier. The results show that the average recall rate of the macros is as 89.1%, and the average recall rate of the macros is as 81.6%.

Key words:  , product quality, association rule mining, time series forecasting model, machine learning