Actuator Setpoint Control Using ML Explanation Value Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is no automatic way to utilize explanation values from machine learning algorithms for controlling actuators in industrial processes, making it difficult to validate and adjust the performance of these systems effectively.
Innovation Solution
A control arrangement is introduced that utilizes deviations between machine learning explanation values and normal explanation values to automatically adjust setpoint values for the controller, incorporating modules like P, I, and D for PID calculations, and input mapping using linguistic equations or fuzzy logic to refine the setpoint adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning algorithms are used for analyzing multivariable measurements, then the ability to automatically learn and improve from experience is improved, but the difficulty of interpreting how the algorithms have arrived at predictions increases
Solution Approach 1:
The patent introduces explanation values as an intermediary between the machine learning algorithm and the user. These explanation values (such as SHAP values, LIME method outputs, or DeepLIFT method outputs) serve as a mediator that translates the internal decision-making process of complex ML models into interpretable metrics that show how much each input parameter contributes to the predicted outcome, thereby resolving the contradiction between automation and interpretability
2Ease of operation
If explanation values are used to validate machine learning models, then the ease of validation is improved, but the automatic utilization of explanation values for controlling actuators remains unavailable
Solution Approach 1:
The patent implements a feedback mechanism where explanation values are not only used for validation but are fed back into the control system to automatically adjust setpoint values. The controller receives explanation values from the machine learning algorithm, processes them through a feedback loop, and uses them to modify control commands sent to actuators, thereby enabling automatic utilization of explanation values for controlling industrial processes
3Adaptability or versatility
If machine learning algorithms are used without explicit instructions, then the flexibility of the system is improved, but the lack of automatic control integration worsens
Solution Approach 1:
The patent merges the machine learning algorithm with the traditional control system by integrating explanation value generation directly into the control loop. The ML algorithm, which operates without explicit instructions for pattern recognition, is combined with the controller that uses these explanation values to automatically adjust setpoints, thereby merging adaptive learning capabilities with automated control integration
Data Source
AI summary
The invention provides a control arrangement where the controller is arranged to drive the actuator utilizing automatically the explanation values. The control arrangement has a controller, which is arranged to drive an actuator. The control arrangement comprises also a setpoint controller, which is arranged to utilize deviations between explanation values of machine learning and normal explanation values of machine learning. The setpoint controller forms a setpoint value for the controller.