关键词: 流程生产, 数字孪生, 过程质量自适应预测, 混合神经网络, 迁移学习

Abstract: Process production had characteristics of segmented equipment units and frequent changes in multi-working conditions,and the digital twin production line model established for specific scenarios lacked the ability to adapt to these changes,making it difficult to quickly and accurately perceive product quality changes. To improve the adaptability of the digital twin production line model,an adaptive prediction method for process production process quality based on InceptionDCNN-LSTM neural network and transfer learning was proposed (adaptive prediction models in twin environments). Firstly,an adaptive prediction overall framework consisting of a digital twin five-dimensional model and an adaptive prediction model was built. Secondly,according to the degree of change in working conditions,the corresponding adaptive migration strategy was proposed. The InceptionDCNN module was used to extract parameter features,the long short-term memory neural network was used to mine quality time series features,and transfer learning was introduced to address adaptability of the prediction model under variable conditions. Finally,taking the tobacco production line as an example,a twin workshop platform was built,and the feasibility of the adaptive prediction model in twin environment was verified. The results show that the prediction error of the adaptive updated prediction model was less than 1.7 % under different working conditions,and the stability advantage was significant,providing a new idea for improving the adaptive ability of the digital twin production line model under different working conditions.

Key words: process production, digital twins, adaptive prediction of process quality, hybrid neural networks, transfer learning