Machine Learning Control Research Articles

Machine Learning for Control [About this Issue

Machine Learning for Control [About this Issue

Large language model-based interpretable machine learning control in building energy systems

The potential of Machine Learning Control (MLC) in HVAC systems is hindered by its opaque nature and inference mechanisms, which is challenging for users and modelers to fully comprehend, ultimately leading to a lack of trust in MLC-based decision-making. To address this challenge, this paper investigates and explores Interpretable Machine Learning (IML), a branch of Machine Learning (ML) that enhances transparency and understanding of models and their inferences, to improve the credibility of MLC and its industrial application in HVAC systems. Specifically, we developed an innovative framework that combines the principles of Shapley values and the in-context learning feature of Large Language Models (LLMs). While the Shapley values are instrumental in dissecting the contributions of various features in ML models, LLM provides an in-depth understanding of the non-data-driven or rule-based elements in MLC; combining them, LLM further packages these insights into a coherent, human-understandable narrative. The paper presents a case study to demonstrate the feasibility of the developed IML framework for model predictive control-based precooling under demand response events in a virtual testbed. The results indicate that the developed framework generates and explains the control signals in accordance with the rule-based rationale.

Machine Learning Control for Floating Offshore Wind Turbine Individual Blade Pitch Control

Offshore wind energy is a promising option for emission-free power generationbecause its potential can satisfy the entire US energy demand. However, deep offshore wind energy is mostly left untapped due to the high levelized cost of energy (LCOE) of floating offshore wind turbines (FOWTs). To address the challenge, individual blade pitch control (IPC) of each blade is necessary to reduce fatigue due to the nonlinear dynamics involving unbalanced nonstationary wind/wave loading. In this paper, a machine learning control (MLC) method is proposed that utilizes a genetic program to selectively evolve promising control law candidates from simulated FOWT sensor data. The proposed method utilizes unique MLC features including an efficient selection of design load cases (DLCs) to accelerate the evolution, DLC-based elitism to identify promising candidates, and the use of internal state feedback to learn effective controller properties across all the target DLCs. These features are used to reduce evaluation time on the complex nonlinear FOWT model. The feasibility of the method is demonstrated by reducing fatigue and ultimate loading by 41% under the representative DLCs provided by the Aerodynamic Turbines with Load Attenuation Systems (ATLAS) competition hosted by the Advanced Research Projects Agency–Energy (ARPA-E). Unlike the methods founded upon black-box models [e.g., artificial neural networks (ANNs)], the proposed method provides interpretable results; thus, it can contribute to learning important design features to guide the future controller design. For example, the best individual driven during the feasibility study exhibits a nonlinear proportional controller to a sigmoidlike platform pitch signal, which can be a starting point for more advanced IPC design for FOWTs. The proposed method will improve the cost-effectiveness of FOWTs and be applied to similar nonlinear control problems, such as unmanned aerial vehicle (UAV) control.

Recent Advances in Wake Dynamics and Active Drag Reduction of Simple Automotive Bodies

AbstractThis is a compendium of recent progresses in the development of wake dynamics and active drag reduction (DR) of three-dimensional (3D) simple automotive models, largely focused on the generic Ahmed body. It covers our new understanding of involved instabilities, predominant frequencies, pressure distribution, and unsteady flow structures in the high- (12.5 deg < φ < 30 deg) and low-drag (φ > 30 deg) bodies and the square-back body (φ = 0 deg), where φ is the rear slant angle of the body. Various DR methods and their performances are reviewed, including open- and closed-loop controls along with machine-learning control (MLC). The involving DR mechanisms, net saving, and efficiencies are discussed. Comments are made for the areas that deserve more attention and future investigation.

Optimization and sensitivity analysis of active drag reduction of a square-back Ahmed body using machine learning control

A machine learning control (MLC) is proposed based on the explorative gradient method (EGM) for the optimization and sensitivity analysis of actuation parameters. This technique is applied to reduce the drag of a square-back Ahmed body at a Reynolds number Re = 1.7 × 105. The MLC system consists of pulsed blowing along the periphery of the base, 25 pressure taps distributed on the vertical base of the body, and an EGM controller for unsupervised searching for the best control law. The parameter search space contains the excitation frequency fe, duty cycle α, and flow rate blowing coefficient Cm. It is demonstrated that the MLC may cut short the searching process significantly, requiring only about 100 test runs and achieving 13% base pressure recovery with a drag reduction of 11%. Extensive flow measurements are performed with and without control to understand the underlying flow physics. The converged control law achieves fluidic boat tailing and, meanwhile, eliminates the wake bistability. Such simultaneous achievements have never been reported before. A machine-learned response model is proposed to link the control parameters with the cost function. A sensitivity analysis based on this model unveils that the control performance is sensitive to fe and α but less so to Cm. The result suggests that a small sacrifice on performance will give a huge return on actuation power saving, which may provide important guidance on future drag reduction studies as well as engineering applications.

Artificial Bee Colony Optimization Algorithm Incorporated With Fuzzy Theory for Real-Time Machine Learning Control of Articulated Robotic Manipulators

This paper presents a real-time machine learning control (MLC) of articulated robotic manipulators using artificial bee colony optimization (ABC) algorithm incorporated with fuzzy theory. The modified ABC with dynamic weight is used to optimize the fuzzy structure and fractional order. The fractional parameters, fuzzy membership functions and rule base are determined by means of the ABC computation. This ABC-fuzzy hybrid learning algorithm is applied to real-time MLC of robotic manipulators by including fractional order proportional-integral-derivative (FOPID) control strategy. The MLC's control gain parameters are online tuned via the ABC-fuzzy optimization. With the kinematics analysis of a six-degree-of-freedom (DOF) articulated arm via reverse coordinates approach, an ABC-fuzzy MLC is developed to achieve motion control. A real-time operating system (RTOS) on a microprocessor collaborates with the ABC-fuzzy MLC to meet critical timing constraint by considering the dynamics of actuators. Finally, the mechatronic design and experimental setup of a six-DOF articulated robotic manipulator are constructed. Experimental results and comparative works are provided to demonstrate the merit of the proposed methods. Compared with the conventional control schemes, the proposed ABC-fuzzy MLC has theoretical and practice significance in term of real-time capability, online parameter tuning, convergent behavior and hybrid MLC. The proposed MLC methodologies are applicable to designing real-time modern controllers in both industry and academia.

Jet mixing optimization using machine learning control

We experimentally optimize mixing of a turbulent round jet using machine learning control (MLC) following Li et al (2017). The jet is manipulated with one unsteady minijet blowing in wall-normal direction close to the nozzle exit. The flow is monitored with two hotwire sensors. The first sensor is positioned on the centerline 5 jet diameters downstream of the nozzle exit, i.e. the end of the potential core, while the second is located 3 jet diameters downstream and displaced towards the shear-layer. The mixing performance is monitored with mean velocity at the first sensor. A reduction of this velocity correlates with increased entrainment near the potential core. Machine Learning Control (MLC) is employed to optimize sensor feedback, a general open-loop broadband frequency actuation and combinations of both. MLC has identified the optimal periodic forcing with a small duty cycle as the best control policy employing only 400 actuation measurements, each lasting for 5 seconds. This learning rate is comparable if not faster than typical optimization of periodic forcing with two free parameters (frequency and duty cycle). In addition, MLC results indicate that neither new frequencies nor sensor feedback improves mixing further-contrary to many of other turbulence control experiments. The optimality of pure periodic actuation may be attributed to the simple jet flapping mechanism in the minijet plane. The performance of sensor feedback is shown to face a challenge for small duty cycles. The jet mixing results demonstrate the untapped potential of MLC in quickly learning optimal general control policies, even deciding between open- and closed-loop control.

Mixing Layer Manipulation Experiment

Open- and closed-loop control of a turbulent mixing layer is experimentally performed in a dedicated large scale, low speed wind-tunnel facility. The flow is manipulated by an array of fluidic micro-valve actuators integrated into the trailing edge of a splitter plate. Sensing is performed using a rake of hot-wire probes downstream of the splitter plate in the mixing layer. The control goal is the manipulation of the local fluctuating energy level. The mixing layer’s response to the control is tested with open-loop forcing with a wide range of actuation frequencies. Results are discussed for different closed-loop control approaches, such as: adaptive extremum-seeking and in-time POD mode feedback control. In addition, we propose Machine Learning Control (MLC) as a model-free closed-loop control method. MLC arrives reproducibly at the near-optimal in-time control.

Machine learning in control of functional electrical stimulation systems for locomotion

Two machine learning techniques were evaluated for automatic design of a rule-based control of functional electrical stimulation (FES) for locomotion of spinal cord injured humans. The task was to learn the invariant characteristics of the relationship between sensory information and the FES-control signal by using off-line supervised training. Sensory signals were recorded using pressure sensors installed in the insoles of a subject's shoes and goniometers attached across the joints of the affected leg. The FES-control consisted of pulses corresponding to time intervals when the subject pressed on the manual push-button to deliver the stimulation during FES-assisted ambulation. The machine learning techniques used were the adaptive logic network (ALN) [1] and the inductive learning algorithm (IL) [2]. Results to date suggest that, given the same training data, the IL learned faster than the ALN, while both performed the test rapidly. The generalization was estimated by measuring the test errors and it was better with an ALN, especially if past points were used to reflect the time dimension. Both techniques were able to predict future stimulation events. An advantage of the ALN over the IL was that ALN's can be retrained with new data without losing previously collected knowledge. The advantages of the IL over the ALN were that the IL produces small, explicit, comprehensible trees and that the relative importance of each sensory contribution can be quantified.