MPC Model Switching for Online Adaptation Without Downtime
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional model predictive controllers face performance degradation and require costly downtime when process conditions change, as they rely on linear dynamic models that fail to match nonlinear processes, necessitating offline manual intervention and resulting in poor performance during model changes.
Innovation Solution
A method for adapting process models in model predictive controllers through a switching mechanism that replaces active elements with background elements online, allowing seamless transitions without taking the system offline, using a multivariable model-based predictive control approach with SISO relationships and adjusting internal states to maintain prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If linear dynamic models are used in model predictive controllers, then the controller structure is simple and easy to implement, but the controller performance degrades when process conditions change significantly
Solution Approach 1:
The patent implements dynamic model adaptation by allowing the controller to switch between multiple linear models based on current operating conditions. Each linear model is optimized for specific ranges of process variables, and the controller dynamically selects or blends models to maintain accuracy across varying conditions, thus making the otherwise static linear models adaptive to changing processes.
Solution Approach 2:
The patent changes model parameters by maintaining multiple pre-identified linear models with different parameter sets, each valid for specific operating ranges. The controller monitors process conditions and switches between models or interpolates between them based on current operating point, effectively adapting the model parameters to match current process conditions without requiring complex nonlinear modeling.
2Reliability
If new model functions are identified and replaced manually, then the model can be updated to match current process conditions, but the system requires costly downtime and manual intervention
Solution Approach 1:
The patent applies preliminary action by pre-identifying multiple linear models covering different operating conditions before the controller is deployed. These models are prepared in advance and stored in the controller, so when process conditions change, the controller can immediately switch to the appropriate pre-prepared model without requiring real-time model identification or system shutdown, thus eliminating downtime during model transitions.
Solution Approach 2:
The controller is designed to autonomously select and switch between multiple pre-identified models based on real-time process conditions without requiring manual intervention. The system self-manages model selection by monitoring operating variables and automatically transitioning between models or blending them, eliminating the need for costly manual model updates and system downtime.
3Adaptability or versatility
If the application is taken offline for model changes, then new model functions can be implemented, but future prediction capability is lost temporarily
Solution Approach 1:
The patent implements dynamic model adaptation by allowing the controller to switch between multiple linear models based on current operating conditions. Each linear model is optimized for specific ranges of process variables, and the controller dynamically selects or blends models to maintain accuracy across varying conditions, thus making the otherwise static linear models adaptive to changing processes.
Solution Approach 2:
The controller maintains continuous operation and future prediction capability by switching between multiple pre-identified models without taking the system offline. The model switching occurs seamlessly in real-time, ensuring that the controller continues to provide accurate predictions and control actions throughout the transition, thus maintaining uninterrupted useful action.
4Adaptability or versatility
If multiple linear models are maintained for different operating conditions, then the controller can adapt to nonlinear processes, but the device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the operating range into multiple segments, each with its own optimized linear model. The controller segments the nonlinear process behavior into manageable linear approximations valid for specific operating ranges, and switches between these segmented models based on current conditions, thus handling nonlinear processes through piecewise linear modeling rather than requiring a single complex nonlinear model.
Solution Approach 2:
The controller is designed with multi-functionality to handle both linear and nonlinear process behaviors using a unified architecture. By incorporating multiple linear models and a switching mechanism, the single controller system can universally adapt to different operating conditions and nonlinear process dynamics, eliminating the need for separate controllers for different operating modes and thus managing complexity through consolidation.
Data Source
Figure 1
Figure 2~4
Figure 5
AI summary
It is advantageous to switch an active element with a background element. By using a switching mechanism, a background element can be made active while the application remains active. That is, the application does not need to be taken offline to switch the active element with a previously loaded background element. Single input/single output (SISO) relationships are defined for background elements and an active element which provides one or many functions between one input variable and one output variable. The switching mechanism acts on a given SISO relationship to determine the background element to use to replace the active element.