Sparse Attention-driven Quality Prediction for Production Process Optimization in Digital Twins

TL;DR

The paper tackles real-time production process optimization in complex process industries by deploying a data-driven digital twin that encodes operational logic with a Probabilistic Sparse Self-Attention enhanced Temporal Convolutional Network for quality prediction. It introduces a cloud-edge DT workflow with a multi-objective optimization module driven by Archimedes Optimization Algorithm, incorporating Sobol sensitivity to identify controllable parameters. Validated on a tobacco shredding line, the framework achieves high online prediction accuracy and quality acceptance, demonstrating seamless virtual-real integration and tangible gains in efficiency. This work provides a scalable, data-centric approach for prescriptive control in tightly coupled production processes and outlines pathways for broader industrial adoption.

Abstract

In the process industry, long-term and efficient optimization of production lines requires real-time monitoring and analysis of operational states to fine-tune production line parameters. However, complexity in operational logic and intricate coupling of production process parameters make it difficult to develop an accurate mathematical model for the entire process, thus hindering the deployment of efficient optimization mechanisms. In view of these difficulties, we propose to deploy a digital twin of the production line by encoding its operational logic in a data-driven approach. By iteratively mapping the real-world data reflecting equipment operation status and product quality indicators in the digital twin, we adopt a quality prediction model for production process based on self-attention-enabled temporal convolutional neural networks. This model enables the data-driven state evolution of the digital twin. The digital twin takes a role of aggregating the information of actual operating conditions and the results of quality-sensitive analysis, which facilitates the optimization of process production with virtual-reality evolution. Leveraging the digital twin as an information-flow carrier, we extract temporal features from key process indicators and establish a production process quality prediction model based on the proposed deep neural network. Our operation experiments on a specific tobacco shredding line demonstrate that the proposed digital twin-based production process optimization method fosters seamless integration between virtual and real production lines. This integration achieves an average operating status prediction accuracy of over 98% and a product quality acceptance rate of over 96%.

Figures (11)

• Figure 1: The framework of DT for process production optimization. The system is composed of 4 major modules: the physical entities of the production line, the data and information processing module, the prediction and optimization service, and the client application.
• Figure 2: Data processing flow of the DT production line.
• Figure 3: The overview of the proposed NN model. It is composed of three major parts: input representation, stacked TCN with ProbSparse self-attention, and prediction head. The feature dimensions are determined according to the data of the production line considered in Section \ref{['experiments']}.
• Figure 4: Re-organization of the raw input data sequence, where different colors represent different types of data chunks.
• Figure 5: The structure of the proposed NN model composed of TCN and PSA, where the attention score is computed based on (\ref{['eq_qkv']})-(\ref{['eq_attention']}).