Embedded Tuning of Model-Less Controllers Under Changing Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Model-less controllers in industrial automation systems, such as PID controllers, face challenges in adapting to changes in operating conditions and system behavior due to their inability to account for future system responses, leading to inefficiencies and reduced robustness, especially in complex systems where step tests can disrupt desired operating conditions.
Innovation Solution
A method is introduced that dynamically tunes model-less controllers using a Bayesian optimization algorithm to generate an initial model based on normal operation data, minimizing the need for step tests and ensuring the system operates within acceptable bounds by analyzing data in real-time and applying targeted excitations to enrich information content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If model-less controllers (e.g., PID controllers) are used to control industrial automation systems, then the control system is simple to implement and operate, but the controller cannot adapt to changes in operating conditions and system behavior
Solution Approach 1:
The patent implements dynamic tuning of controller parameters by continuously adapting the controller gains based on real-time system behavior analysis. The system transitions from static PID parameters to dynamic parameter adjustment, allowing the controller to adapt to changing operating conditions while maintaining simplicity of implementation through automated adaptation mechanisms.
Solution Approach 2:
The controller performs self-tuning by automatically analyzing its own performance data and adjusting its parameters without external intervention. The system uses its operational data to generate models and refine controller parameters autonomously, eliminating the need for manual retuning while preserving ease of operation.
2Manufacturing precision
If step tests are performed to tune model-less controllers, then the controller parameters can be optimized, but the desired operating conditions are disrupted
Solution Approach 1:
The system performs preliminary modeling using historical operational data before actual tuning is needed. By continuously building and updating system models during normal operation, the controller is prepared for optimal tuning without requiring disruptive step tests, thus maintaining both parameter optimization and operational continuity.
Solution Approach 2:
The patent replaces physical step tests with virtual modeling and simulation approaches. Instead of mechanically disrupting the system to gather tuning data, the system uses computational models based on normal operational data to achieve parameter optimization, eliminating the need to interrupt production.
3Ease of manufacture
If traditional tuning methods are used for model-less controllers, then the tuning process is simple, but the performance improvement is limited
Solution Approach 1:
The patent introduces an intermediate computational layer that bridges simple tuning methods and complex optimization goals. This intermediary system uses automated model generation and analysis to enhance traditional tuning approaches, providing robust performance improvement while maintaining ease of implementation through the automated intermediary process.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method may include receiving data representative of one or more commands generated by a model-less controller to control operations of devices within a system and output parameters associated with the devices of the system. The method may also include determining whether the data is indicative of a change in operational characteristics of the system and generating a model representative of the operational characteristics of the system as a function of the data based on a Bayesian optimization algorithm in response to the data being indicative of the change. The method may also involve transmitting an excitation input to the devices in response to the data not being indicative of the change, receiving updated output parameters associated with the devices of the system after the excitation input is transmitted, and generating the model based on the updated output parameters and the excitation input.