Hybrid Predictive Control for Cell Culture Feed Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Bioprocess control in cell cultures faces challenges due to the limitations of traditional PID control methods, which struggle with the complexity and variability of metabolic network pathways, leading to inaccurate predictions and insensitivity to feed strategies in fed-batch fermentation processes.
Innovation Solution
A hybrid predictive modeling approach that combines first-principle and data-driven models, using linear regressors or neural networks to predict metabolite concentrations and mass balance models to predict glucose concentration, allowing for more accurate and consistent predictions of cell culture attributes in Model Predictive Control (MPC) systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If first-principle models (Monod kinetics) are used for bioprocess control, then the control framework is established, but the models exhibit insensitivity to feed strategies and limited accuracy due to nonlinear state models with many unknown parameters
Solution Approach 1:
The patent combines first-principle models (mass balance equations) with data-driven models (neural networks) into a hybrid predictive model. The first-principle model provides the control framework and mass balance relationships, while the neural network compensates for parameter uncertainties and improves prediction accuracy by learning from historical process data.
Solution Approach 2:
The patent transforms the static parameter assumption in traditional Monod kinetics into dynamic parameter estimation by using neural networks to predict time-varying parameters such as maximum specific growth rate and yield coefficients, allowing the model to adapt to changing process conditions.
2Adaptability or versatility
If data-driven models (neural networks) are used for process monitoring and control, then model flexibility is increased, but accuracy is limited due to time-varying characteristics and limited training data
Solution Approach 1:
The patent merges data-driven neural network models with first-principle mass balance models. The neural network provides flexibility in capturing complex nonlinear relationships, while the mass balance model ensures physical consistency and provides constraints that guide the data-driven model, improving overall prediction accuracy despite limited training data.
Solution Approach 2:
The patent introduces an intermediary layer where the neural network predicts key parameters (such as specific growth rate and substrate consumption rate) that are then used in the mass balance equations. This intermediary approach allows the data-driven model to leverage physical constraints, improving accuracy while maintaining flexibility.
3Ease of operation
If simple PID control is used for bioprocess control, then ease of operation is maintained, but control performance is limited due to scarcity of measurements and process complexities
Solution Approach 1:
The patent implements model predictive control (MPC) with feedback mechanisms where the hybrid predictive model continuously estimates process states based on limited measurements and historical data. The controller uses this feedback to adjust feed strategies, maintaining reliability without requiring extensive measurements or complex manual intervention.
Solution Approach 2:
The patent uses the hybrid predictive model to perform preliminary actions by predicting future process states and optimizing feed strategies in advance. This allows the system to proactively adjust control actions based on predicted trajectories, improving control performance while maintaining operational simplicity through automated decision-making.
Data Source
AI summary
A method of controlling a cell culture process uses hybrid predictive modeling in a model predictive controller. The method includes, for multiple time intervals, obtaining current values of cell culture attributes associated, and generating a control value for a physical input to the cell culture process. Generating the control value includes predicting future values of the cell culture attributes based on the current values, by using one or more data-driven models to predict future values of a first one or more attributes of the cell culture attributes, and using one or more first principle models to predict future values of a second one or more attributes of the cell culture attributes. Generating the control value also includes determining the control value by optimizing an objective function subject to the predicted future values. The method also includes using the control value to control the physical input to the cell culture process.