Neural Network Control for Dead-Time Transient Response
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing control systems using neural networks struggle to improve responsiveness in output response waveforms for step commands due to dead time in control targets, as they fail to effectively learn without being affected by dead time, leading to complex designs and difficulties in achieving both transient response characteristics and performance improvements.
Innovation Solution
A control device comprising a feedback controller, a reference model unit with a dead-time component, and a learning-based controller that minimizes the error between the control target's output and the reference model's output, allowing the neural network to learn without dead time effects, thereby improving transient characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a reference model including dead time is used for neural network learning, then learning delay is prevented, but the design becomes complex and model errors are introduced
Solution Approach 1:
The control system is divided into two separate controllers: a feedback controller that handles the dead-time component and a learning-based controller that handles the non-dead-time dynamics. This segmentation allows each controller to specialize in specific aspects, avoiding the complexity of designing a single comprehensive model while preventing learning delays.
Solution Approach 2:
The dead-time component is extracted and isolated into a dedicated feedback controller. By removing the dead-time element from the learning-based controller's responsibilities, the learning process can proceed without delay while the feedback controller independently manages the dead-time compensation.
2Measurement precision
If feedback error learning is used to make error zero, then control accuracy is improved, but responsiveness for step commands does not improve due to dead time
Solution Approach 1:
The control functions are segmented into accuracy-oriented feedback control and speed-oriented learning-based control. The feedback controller ensures control accuracy by eliminating errors, while the learning-based controller improves responsiveness by predicting and compensating for dead-time effects in advance.
Solution Approach 2:
The learning-based controller performs preliminary action by predicting the desired output and pre-compensating for dead-time effects before they actually occur. This allows the system to respond more quickly to step commands while the feedback controller maintains accuracy.
3Adaptability or versatility
If a single controller compensates for all targets including response, disturbance, and variation, then comprehensive control is achieved, but design and adjustment becomes difficult
Solution Approach 1:
The comprehensive control capability is achieved through segmentation into specialized controllers: feedback control for error elimination and learning-based control for predictive compensation. Each controller focuses on specific compensation targets, making design and adjustment easier while maintaining comprehensive control capability.
Solution Approach 2:
The dual-controller architecture provides universal control capability where the feedback controller handles deterministic error compensation and the learning-based controller handles dynamic prediction and disturbance rejection. Together they provide multi-functional comprehensive control without requiring a single complex controller.
Data Source
AI summary
To provide a control device that causes a neural network to perform learning without any effects of dead time even for a dead-time system and that has the capability of improving transient characteristics for a command input. A control device includes a feedback controller configured to control a control target including a dead-time component, a reference model unit including a dead-time component and configured to output a desired response waveform for an input. A learning based controller is configured to perform learning in a manner that a change in an output from the learning based controller minimizes an error between an output of the control target and an output of the reference model unit or causes the error to be a predetermined threshold or smaller, the output from the learning based controller being added to an output of the feedback controller and input to the control target.