Biochemical Process Control Using Simulated Data for Quality Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing processes for producing biochemical products face challenges in achieving targeted quality and quantity due to sparse historical data, leading to inefficient training of artificial intelligence models and increased waste and time delays.
Innovation Solution
A method utilizing a predictive model trained with a combination of historical and simulated data, where a physical model generates simulated quality attributes to enhance training, reducing waste and time, and a process control system adjusts actuation parameters to meet quality targets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If historical data is used to train the artificial intelligence model, then the model can learn from real process outcomes, but the sparse historical data is insufficient to train the model accurately
Solution Approach 1:
The patent creates virtual copies of historical data by training a physical model on available historical data, then using this trained model to generate synthetic training datasets. These synthetic datasets are copies that mimic the characteristics of real data but are generated in abundance, allowing the AI model to be trained on sufficient data without requiring extensive real historical data
Solution Approach 2:
The patent performs preliminary action by first training the physical model on the limited historical data before generating the training dataset for the AI model. This preliminary training of the physical model enables the generation of realistic synthetic data that preserves the underlying process relationships, solving the data scarcity problem before the main AI model training begins
2Measurement precision
If physical experiments are conducted to gather training data, then real quality attributes can be obtained, but this increases waste of reactants and goods
Solution Approach 1:
The patent replaces physical experiments with virtual experiments by using a trained physical model to generate synthetic training data. This copying approach creates virtual representations of experimental outcomes without requiring actual reactants or goods, eliminating the waste associated with physical experimentation while maintaining the ability to learn quality attributes
Solution Approach 2:
The patent substitutes the mechanical/physical experimentation system with a computational model system. Instead of physically running experiments that consume materials, the system uses a trained physical model to simulate and generate training data, replacing the material-intensive approach with a computation-intensive approach that produces no physical waste
3Measurement precision
If physical experiments are conducted to gather training data, then real quality attributes can be obtained, but this increases time delays
Solution Approach 1:
The patent uses virtual copying of experimental outcomes through the trained physical model to generate training data instantly, rather than waiting for physical experiments to complete. This eliminates the time delays associated with setting up and executing physical experiments while still providing realistic quality attribute data for training
Solution Approach 2:
The patent performs the computationally intensive physical model training as a preliminary action before data generation. Once the physical model is trained, it can rapidly generate training datasets without the time delays of physical experimentation, enabling fast iteration and training of the AI model
4Manufacturing precision
If the predictive model is deployed to provide correction actuation parameters, then the quality attribute can be maintained within targeted interval, but this requires accurate predictions
Solution Approach 1:
The patent implements feedback by deploying the trained predictive model to continuously monitor process data and provide correction actuation parameters when quality attributes deviate from target intervals. The model learns from historical and synthetic data to predict quality outcomes and provides real-time feedback for process adjustment, creating a closed-loop control system that maintains manufacturing precision
Solution Approach 2:
The patent performs preliminary action by training the predictive model extensively on combined historical and synthetic data before deployment. This preliminary training phase ensures the model achieves high prediction accuracy, and once trained, it can reliably provide correction actuation parameters to maintain quality attributes within target intervals during actual production
Data Source
AI summary
A method for optimizing a process (PROC) to produce a biochemical product (P) defined by a quality attribute, the process being controlled by an actuation parameter (C) and being monitored to get a measured value (T). The method includes training a predictive model (PRED) on a training database; and deploying the trained predictive model (PRED) to provide a correction actuation parameter (dC) when a predicted quality attribute (pQA) is out of a targeted quality attribute interval (QAmin, QAmax). The method also includes a step of designing a physical model of the process (PROC) able to provide a simulated quality attribute, the training database comprising simulated quality attributes computed from the physical model and experimental quality attributes computed from biochemical products (P) previously produced.