Experiment-in-the-Loop Tuning of Active Vibration Controllers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AVC systems require time-consuming and complex numerical modeling for tuning control algorithms, which is not user-friendly and can lead to instability and diminished performance.
Innovation Solution
An experiment-in-the-loop (EITL) system autonomously generates optimal control algorithms through iterative experimental tests, eliminating the need for numerical models and utilizing straightforward control theory and optimization methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If numerical modeling and numerical design of control algorithms are used, then control algorithm optimization is achieved, but the tuning process becomes time-consuming and complex
Solution Approach 1:
The patent replaces the traditional numerical modeling and simulation approach with a physics-informed machine learning model that directly processes sensor data from the physical system. This substitution eliminates the need for creating and refining numerical models, significantly reducing the tuning time while maintaining control effectiveness through experimental data-driven optimization.
Solution Approach 2:
The system enables autonomous self-tuning of control algorithms through automated experimental procedures. The machine learning model automatically processes sensor data, identifies optimal control parameters, and updates the control algorithm without requiring manual intervention or iterative refinement by operators, thereby reducing both time and complexity.
2Measurement precision
If numerical modeling is required for tuning, then control algorithm precision can be optimized, but the system becomes less user-friendly and requires complex mathematical knowledge
Solution Approach 1:
The patent replaces complex numerical modeling procedures with a machine learning-based system that automatically extracts system dynamics from physical experiments. This substitution maintains high precision in control parameter identification while eliminating the need for users to possess advanced mathematical knowledge, as the system handles all computations autonomously through data-driven methods.
Solution Approach 2:
The system creates a virtual representation of the physical system's dynamics through machine learning models trained on experimental data. This copied dynamic model enables precise control parameter optimization without requiring users to develop or understand complex numerical models, making the process accessible to users with basic engineering knowledge.
3Reliability
If iterative refinements of numerical models are performed, then validation accuracy improves, but the complexity of the tuning process increases
Solution Approach 1:
The patent replaces the iterative numerical model refinement process with a single-step physics-informed machine learning approach. The system trains the ML model on experimental data to capture system dynamics, then uses this trained model for direct control parameter optimization. This substitution achieves equivalent or superior validation accuracy while dramatically reducing process complexity by eliminating multiple iterative refinement cycles.
Solution Approach 2:
The system performs preliminary system identification and dynamics extraction through initial experimental tests before control optimization begins. By pre-training the machine learning model on system behavior data, the approach establishes an accurate dynamic model upfront, eliminating the need for subsequent iterative refinements during the control tuning process.
Data Source
AI summary
An experiment-in-the-loop system for determining one or more control parameters for an active controller includes a first processor and a second processor. The first processor determines one or more parameters for a first control algorithm during a series of experiments, where one or more external stimuli are applied to a device under test (DUT). The second processor derives a cost of a physical response to the one or more external stimuli based on measurements from one or more sensors. In response to an option parameter received by the second processor, a switch can select a first, second, or third controller to connect to the DUT. The first, second, or third controller produces an actuation voltage that is routed to a first stimulator device. The first stimulator device applies a first external stimulus to the DUT.