Deep Learning Process Control via Non-Invasive Closed-Loop Exploration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing advanced process control methods, such as linear dynamic models, are inadequate for capturing the nonlinear behavior of certain chemical process units, limiting their ability to optimize process performance.
Innovation Solution
A method involving the creation of a Deep Learning model predictive controller, where a linear dynamic model is first used to control the process, collect data, and then a Deep Learning model is trained to create a more sophisticated controller capable of capturing nonlinear behavior, using techniques like recurrent neural networks and piecewise linear dynamic models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a linear dynamic model is used for process control, then the controller is simple to implement and maintain, but it cannot adequately capture nonlinear process behavior
Solution Approach 1:
The patent segments the modeling task into two phases: first using a simple linear dynamic model for initial control and data collection, then transitioning to a Deep Learning model for accurate nonlinear behavior capture. This segmentation allows the system to benefit from both simplicity and accuracy at different stages.
Solution Approach 2:
The linear model predictive controller serves as an intermediary tool that enables non-invasive closed-loop exploration to collect rich process data. This intermediary approach facilitates the transition from simple to sophisticated modeling without requiring invasive plant modifications.
2Reliability
If a Deep Learning model is trained to capture nonlinear behavior, then process modeling accuracy is improved, but significant amounts of rich process data are required which are not readily available from regular operation
Solution Approach 1:
The linear model predictive controller is deployed in advance to perform non-invasive closed-loop exploration and collect rich process data before training the Deep Learning model. This preliminary data collection action ensures sufficient training data is available without requiring invasive modifications to the running plant.
Solution Approach 2:
The linear controller acts as an intermediary that enables indirect data collection through closed-loop exploration, allowing the system to gather necessary training data without directly modifying the plant's normal operation or requiring invasive measurements.
3Quantity of substance
If invasive methods are used to collect process data for training, then sufficient data can be obtained, but plant operation is disrupted
Solution Approach 1:
The patent converts the limitation of regular operation data (insufficient quantity) into an opportunity by using the linear controller for non-invasive exploration. This approach collects rich training data while maintaining normal plant operation, turning a potential disruption into a beneficial data collection opportunity.
Solution Approach 2:
The linear model predictive controller performs self-service by automatically conducting closed-loop exploration and collecting necessary training data without requiring external invasive interventions or plant shutdowns, thereby maintaining continuous productive operation.
Data Source
AI summary
Deep Learning is a candidate for advanced process control, but requires a significant amount of process data not normally available from regular plant operation data. Embodiments disclosed herein are directed to solving this issue. One example embodiment is a method for creating a Deep Learning based model predictive controller for an industrial process. The example method includes creating a linear dynamic model of the industrial process, and based on the linear dynamic model, creating a linear model predictive controller to control and perturb the industrial process. The linear model predictive controller is employed in the industrial process and data is collected during execution of the industrial process. The example method further includes training a Deep Learning model of the industrial process based on the data collected using the linear model predictive controller, and based on the Deep Learning model, creating a Deep Learning model predictive controller to control the industrial process.