Digital Twin-Empowered Intelligent Monitoring and Decision Support for Food Fermentation Using SCSO-Optimized CNN-BiLSTM with Attention Mechanism
Main article
Abstract
The deep convergence of Industrial Internet of Things (IIoT) and artificial intelligence is driving the digital transformation of food fermentation industry, yet the multi-parameter dynamic coupling nature of fermentation bioprocesses presents fundamental challenges that existing monitoring systems cannot adequately address. Conventional monitoring architectures suffer from fragmented sensor integration, static modeling assumptions, and inability to capture the nonlinear time-series interdependencies among biological, chemical, and physical process variables, resulting in lagging control responses and suboptimal product quality outcomes. This paper proposes a Digital Twin (DT)-empowered intelligent monitoring and decision support framework that addresses these limitations through two integrated innovations. First, a multimodal IIoT sensor network provides comprehensive real-time data collection across seven process variables: pH, temperature, dissolved oxygen, turbidity, CO2 off-gas, humidity, and biomass concentration. Second, a self-optimizing hybrid deep learning model, SCSO-CNN-BiLSTM-AT, serves as the framework intelligent core. The model synergistically integrates the Sand Cat Swarm Optimization (SCSO) algorithm for adaptive hyperparameter optimization with a deep learning architecture that combines Convolutional Neural Network (CNN) spatial feature extraction, Bidirectional Long Short-Term Memory (BiLSTM) temporal modeling, and an attention mechanism for dynamic feature weighting. The digital twin engine maintains real-time synchronization between physical fermentation processes and virtual process models, enabling predictive intervention before quality deviations manifest in physical measurements. Validation on solid-state and submerged fermentation datasets demonstrates that SCSO-CNN-BiLSTM-AT achieves RMSE = 0.0184, MAE = 0.0152, and R2 = 0.972 for multi-parameter prediction, representing improvements of 32.2%, 34.2%, and 11.1% over CNN-BiLSTM-AT without SCSO optimization. Ablation analysis across five fermentation scenarios confirms the independent contribution of each framework component. The proposed framework advances the digital transformation of food fermentation manufacturing by enabling data-driven process optimization at a level of accuracy and real-time responsiveness unavailable to conventional monitoring approaches.
