Concurrent Learning Neural Network Adaptive Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing adaptive control systems face challenges in guaranteeing weight convergence and tracking error convergence without persistent excitation, particularly in applications like flight control where persistent excitation can be restrictive and infeasible to monitor online.
Innovation Solution
The use of concurrent learning methods that combine recorded and current data within the framework of Model Reference Adaptive Control (MRAC) to adaptively control neural networks, ensuring weight convergence and tracking error convergence by utilizing stored data that is sufficiently rich and linearly independent.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If persistent excitation is used to guarantee weight convergence, then parameter convergence is achieved, but control effort increases and monitoring becomes infeasible online
Solution Approach 1:
The system performs preliminary action by recording and storing system state data during periods when persistent excitation conditions are met. These pre-recorded data sets are then used for weight convergence guarantee without requiring continuous persistent excitation, thereby eliminating the need for real-time monitoring while maintaining convergence reliability
Solution Approach 2:
The invention creates copies of system state data from periods when persistent excitation occurred. Instead of requiring live persistent excitation signals, the system uses stored copies of excitation data that were recorded previously, allowing weight convergence without current persistent excitation requirements
2Reliability
If persistent excitation reference inputs are used, then parameter convergence is guaranteed, but fuel consumption increases and stress on aircraft increases
Solution Approach 1:
The system captures and stores excitation data during naturally occurring exciting periods in flight operations. By performing the data collection action in advance during normal operations, the system eliminates the need for additional fuel-consuming persistent excitation maneuvers while still gathering sufficient data for parameter convergence
Solution Approach 2:
The invention converts naturally occurring flight maneuvers and disturbances (which would otherwise be routine operations) into beneficial excitation events. By recording data during these normal operations, the system transforms ordinary flight activities into useful learning opportunities without requiring extra fuel consumption or creating additional stress on the aircraft
3Productivity
If conventional adaptive control methods are used, then instantaneous uncertainty suppression is achieved, but weight convergence is not guaranteed
Solution Approach 1:
The system merges two previously separate functions into a unified approach: instantaneous uncertainty suppression through adaptive control and weight convergence through concurrent use of recorded data. By combining real-time adaptive control with offline learned models, the system achieves both rapid uncertainty suppression and guaranteed weight convergence simultaneously
Data Source
AI summary
Various embodiments of the invention are neural network adaptive control systems and methods configured to concurrently consider both recorded and current data, so that persistent excitation is not required. A neural network adaptive control system of the present invention can specifically select and record data that has as many linearly independent elements as the dimension of the basis of the uncertainty. Using this recorded data along with current data, the neural network adaptive control system can guarantee global exponential parameter convergence in adaptive parameter estimation problems. Other embodiments of the neural network adaptive control system are also disclosed.