xMLC -- A Toolkit for Machine Learning Control

TL;DR

This paper introduces xMLC, an open-source toolkit based on genetic programming for automatic learning of control laws in fluid mechanics, aimed at both education and research.

Contribution

It presents a novel implementation of Machine Learning Control using linear genetic programming, with a focus on ease of use and practical application in fluid mechanics.

Findings

Demonstrates effectiveness of genetic programming for control law discovery

Provides a user-friendly software toolkit for MLC in fluid mechanics

Highlights the importance of exploration and exploitation balance in learning algorithms

Abstract

xMLC is the second book of this `Machine Learning Tools in Fluid Mechanics' Series and focuses on Machine Learning Control (MLC). The objectives of this book are two-fold: First, provide an introduction to MLC for students, researchers, and newcomers on the field; and second, share an open-source code, xMLC, to automatically learn open- and closed-loop control laws directly in the plant with only a few executable commands. This presented MLC algorithm is based on genetic programming and highlights the learning principles (exploration and exploitation). The need of balance between these two principles is illustrated with an extensive parametric study where the explorative and exploitative forces are gradually integrated in the optimization process. The provided software xMLC is an implementation of MLC. It builds on OpenMLC (Duriez et al., 2017) but replaces tree-based genetic programming but the linear genetic programming framework (Brameier and Banzhaf, 2006). The latter representation is preferred for its easier implementation of multiple-input multiple-output control laws and of the genetic operators (mutation and crossover). The handling of the software is facilitated by a step by step guide that shall help new practitioners to use the code within few minutes. We also provide detailed advice in using the code for other solvers and for experiments. The code is open-source and a GitHub version is available for future updates, options and add-ons.