基于改进LSTM的数码雷管模组印刷质量预测

Quality Prediction for Digital Detonator Module Printing Based on Improved LSTM

由于数码雷管模组印刷过程中生产工艺复杂、强时序性等特点,其质量的精准预测已成为提高产品质量管理水平的关键。基于此提出一种改进长短期记忆(long short-term memory,LSTM)网络的数码雷管模组印刷质量预测模型。首先根据数码雷管模组印刷过程提炼机器运行参数、环境参数与检测参数作为印刷产品质量的原始特征,并对关键检测参数进行时序特征重构以增强特征表达能力;其次基于改进的LSTM网络建立数码雷管模组印刷特征提取框架,采用卷积神经网络提取空间特征避免LSTM挖掘高维印刷特征时隐含关系的不足,通过全局注意力机制自适应学习不同时刻印刷特征对印刷产品质量的贡献度,为LSTM提取的深层时序特征分配不同权值;最后以深层特征作为输入,通过全连接网络实现数码雷管模组印刷产品的质量预测。实验结果表明,相较于BP神经网络、门控循环单元网络、LSTM等预测方法,改进的LSTM网络有效提高了数码雷管模组印刷产品质量的预测精度。

Due to the complex production process and strong timing characteristics of the digital detonator module printing process,accurate prediction for its quality has become a key to improving product quality management.For this reason,a quality prediction model for digital detonator module printing that improves the long short-term memory network is proposed.First,based on the digital detonator module printing process,the parameters of machine operating,environment and detection are extracted as the original features of the printed product quality,and the key detection parameters are reconstructed in time series to enhance the feature expression ability.Secondly,a digital detonator module printing feature extraction framework is established based on the improved long shortterm memory network.The convolutional neural network is designed to extract spatial features to avoid the shortcomings of implicit relationships when LSTM mines high-dimensional printing features.The global attention mechanism is used to adaptively learn the contribution of printing features at different moments to the quality of printed products,and assign different weights to the deep temporal features extracted by LSTM.Finally,deep features are used as input to achieve quality prediction for the digital detonator module printing through a fully connected network.Experimental results show that compared with prediction methods such as BP neural network,gated recurrent unit network,and LSTM,the improved long short-term memory network effectively improves the accuracy of quality prediction for digital detonator module printing.