Core Concepts
The authors present a new algorithm that improves the convergence speed of iterative adaptation of feedforward controller parameters, enabling better performance in electromechanical switching devices.
Abstract
The paper addresses the challenge of enhancing the response time and control accuracy of feedforward control systems, which can be hindered by modeling errors or disturbances. The authors propose a new algorithm that combines several techniques to improve the convergence speed of the iterative adaptation of feedforward controller parameters.
Key highlights:
The algorithm uses a basis change technique based on the sensitivity of the feedforward law to decompose the initial high-dimensional problem into separable one-dimensional problems.
It employs a search method that combines Pattern Search and sign gradient descent philosophies to efficiently find the descent coordinate and perform line searches.
A subgradient-based learning rate is introduced to help the algorithm escape local minima and reach the global optimum, especially when the initial parameter estimates are significantly different from the optimal values.
Simulation results on a well-known electromechanical switching device control problem demonstrate the effectiveness of the proposed algorithm, showing faster convergence and better performance compared to previous approaches.
The authors conclude that the new algorithm addresses the questions raised in their previous work regarding the periodicity of updating the basis and the number of dimensions to reduce, while also improving the algorithm's ability to handle larger initial parameter errors.
Stats
The system dynamics depend on 9 uncertain parameters, which can be grouped in the parameter vector p.
p = [ ks zs m κ1 κ2 κ3 κ4 κ5 κ6 ]⊺
The desired position trajectory is designed as a 5th-degree polynomial with specific boundary conditions.