Hybrid ML Feedback Control with Limited Corrective Output
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Control devices using machine learning controllers face reliability issues due to the possibility of generating abnormal control commands when encountering situations significantly different from the learning data, particularly in large-scale deep neural networks where parameter adjustments are difficult to grasp.
Innovation Solution
A feedback control device that combines a first control unit using PID control and a second control unit with a learning unit based on machine learning, where the output from the second unit is limited by an adder and an output limiter, ensuring the controlled object receives a manipulated variable that suppresses abnormal outputs from the machine learning unit.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a controller based on machine learning is used to improve control performance, then control accuracy can be enhanced, but reliability deteriorates due to the possibility of generating abnormal control commands for situations different from learning data
Solution Approach 1:
A determination unit is introduced as an intermediary component that evaluates whether the input situation belongs to the distribution of learning data before the machine learning controller processes it. This mediator prevents abnormal situations from reaching the machine learning controller, thereby maintaining reliability while preserving control accuracy for valid inputs.
Solution Approach 2:
The determination unit performs preliminary evaluation of the input situation before the machine learning controller executes its control function. By pre-checking whether the input falls within the learned distribution, the system prepares and filters inputs in advance, preventing abnormal commands from being generated in the first place.
2Adaptability or versatility
If a large-scale deep neural network is used to improve control capability, then control performance can be enhanced, but understanding and grasping parameter responses becomes difficult
Solution Approach 1:
The control system is segmented into distinct functional components: a determination unit for input validation, a machine learning controller for control decision-making, and a limiter for output constraint. This segmentation allows each component to have a specific, understandable function, making the overall system more transparent despite using complex machine learning models.
Solution Approach 2:
The determination unit provides feedback about whether the input situation is within the learned distribution. This feedback mechanism creates a closed-loop validation system that enhances understanding of system behavior by clearly indicating when the machine learning controller is operating within its validated parameters.
3Device complexity
If only a machine learning controller is used to reduce device complexity, then system structure can be simplified, but reliability deteriorates due to lack of constraint mechanisms
Solution Approach 1:
Multiple control functions are merged into a unified control device structure: the determination unit, machine learning controller, and limiter are integrated as coordinated components within a single system. This merging maintains reliability through functional combination while presenting a unified, manageable system interface.
Solution Approach 2:
The system dynamically changes operational parameters based on input validation results. When input is within the learned distribution, the machine learning controller operates with full capability; when outside, the determination unit changes the operational state to reject or limit processing, thereby maintaining reliability without requiring complete system redesign.
Data Source
AI summary
A feedback control device that takes information regarding a control deviation between a measured value and a desired value of a controlled object as input, and outputs a manipulated variable for the controlled object, includes: a first control unit that takes information regarding the control deviation as input, and outputs a manipulated variable for the controlled object; a second control unit that takes information regarding the control deviation as input, and that includes a learning control unit in which a parameter for outputting a manipulated variable for the controlled object is determined by machine learning; and an adder that adds a first manipulated variable output from the first control unit and a second manipulated variable output from the second control unit. A manipulated variable from the adder is output to the controlled object, and the second control unit includes a limiter that limits the second manipulated variable.