Showing papers presented at "International Symposium on Intelligent Control in 2016"

Augmented reality-enhanced structural inspection using aerial robots

Abstract: This paper investigates the arising potential when automated path planning for aerial robotic structural inspection is combined with an Augmented Reality interface that provides live feed of stereo views fused with real-time 3D reconstruction data of the environment, while allowing seamless on-the-fly adaptation of the next robot viewpoints using intuitive head motions. The proposed solution aims to address the problem of accurate inspection and mapping of structures and environments for which a prior model exists but is not accurate, potentially outdated, or does not encode important features and semantics such as human-readable indications and other texture information. To approach the problem, the robot computes an optimized inspection path given any prior knowledge of the environment, while the human operator utilizes the live camera views and the real-time derived 3D map data to locally adjust the reference trajectory of the robot, such that it visits an updated set of viewpoints which provides the desired coverage of the real environment and sufficient focus on certain features and details. An autonomous aerial robot capable of navigation and mapping in GPS-denied environments is employed and combined with the Augmented Reality interface to experimentally demonstrate the potential of the approach in structural inspection applications.

Discretized optimal control approach for dynamic multi-agent decentralized coverage

Abstract: This paper presents a novel discrete-time decentralized control law for the Voronoi-based self-deployment of a Multi-Agent dynamical system. The basic control objective is to let the agents deploy into a bounded convex polyhedral region and maximize the coverage quality by computing locally the control action for each agent. The Voronoi tessellation algorithm is employed to partition dynamically the deployed region and to allocate each agent to a corresponding bounded functioning zone at each time instant. The control synthesis is then locally computed based on an optimal formulation framework related to the Lloyd's algorithm but according to the discrete-time agent's dynamics equation. The performance of the discretized optimal solution will be demonstrated via an illustrative example.

Convex relaxations and Gramian rank constraints for sensor and actuator selection in networks

Abstract: We propose a convex relaxation heuristic for sensor and actuator selection problems in dynamical networks using Gramian metrics. We also propose heuristic algorithms to enforce a rank constraint on the Gramian that can be used in conjunction with combinatorial greedy algorithms and the convex relaxation. This allows selection of sensor or actuator sets that optimize an objective function while preserving a certain amount of observability or controllability throughout the state space, combining previous methods that focus exclusively on either rank or Gramian metrics. We illustrate and compare the greedy and convex relaxation heuristics in several numerical examples involving random and regular networks.

On enlarging backward reachable sets via Zonotopic set membership

Abstract: While a number of efficient methods have been proposed for approximating backward reachable sets, no synthesis method via backward reachable sets has been developed for estimating and enlarging the region of attraction (RA). This paper shows how to use backward reachable sets to enlarge the estimate of the RA of linear discrete-time systems, by using an optimal static feedback controller. Two controller design methods are provided: the first method enlarges the estimate of the RA via invariant sets, whose existence is ensured by zonotope containment; the second method provides the optimal control input by using Lyapunov stability and quadratic stabilization. The backward reachable set is represented by zonotopes which give a good compromise between accuracy and efficiency. The effectiveness of both methods is illustrated by a numerical example.

Design and fabrication of an intention based upper-limb exo-skeleton

Abstract: Exoskeletons are a special type of collaborative robot that can be thought of as a humanoid robot clung to a human body (or a pilot). The proposed powered exoskeleton/ robot is a 3 Degree of Freedom (DoF) robot and it relies on DC motors to apply the necessary joint torques which in turn require an input to drive these motors. The idea here is to read the pilot's intention using an array of Force Sensing Resistors (FSRs) and then parsing it through a Dynamic Intention Filter (DIF), that yields a meaningful output, necessary to drive the exoskeleton. The DIF proposed here is a novel approach that serves as an intelligent element in the system; based on various inputs, the DIF intelligently controls the motors within the predefined and dynamically-changing safe limits of a typical human arm. Nested PID controllers are used to drive the exoskeleton in "torque-control" and "position-control" modes, and a comparison is made regarding their advantages and disadvantages.

Fractional-order approximations to implicitly-defined operators for modeling and control of networked mechanical systems

Abstract: In many modern engineered systems, including Cyber-Physical Systems, there may be many interacting components with coupled dynamics. Our prior work has shown that such systems exhibit fractional-order dynamics, which may serve as a concise model for such very high-order systems. This paper extends that prior work to a more general class of systems in which the operator describing the dynamics of the system can only be determined implicitly. A system-identification procedure for a specific set of example cases indicates that these systems are also predominantly fractional-order as well. In such cases, then, given the central role of system models in control theory, development and awareness of such fractional-order dynamics in CPS are essential for controls engineers.

MPC for linear parameter-varying systems in input-output representation

Abstract: In this paper, we propose a method for model predictive control of linear parameter-varying (LPV) systems described in an input-output (IO) representation and subject to input- and output constraints. By assuming exact knowledge of the future trajectory of the scheduling variable, the on-line computations reduce to the solution of a nominal predictive control problem. An incremental non-minimal state-space representation is used as a prediction model, giving a controller with integral action suitable for tracking piecewise-constant reference signals. Closed-loop asymptotic stability is guaranteed by a terminal cost and terminal set constraint, and the computation of an ellipsoidal terminal set is discussed. Numerical examples demonstrate the properties of the proposed approach. When exact future knowledge of the scheduling variable is not available, we argue and show that good practical performance can be obtained by a scheduling prediction strategy.

Discrete-time indirect adaptive control of a class of single state systems using concurrent learning for parameter adaptation

Abstract: We have previously showed that it is possible to achieve parameter identification of discrete-time structured uncertainties without requiring persistency of excitation when using Concurrent Learning. Instead, granted a less restrictive condition compared to that of persistency of excitation is verified, exponential convergence of parameter estimates to their true values ensues. The present study applies the previously developed discrete-time Concurrent Learning adaptation law within a control loop for discrete-time adaptive control of a discrete-time single-state plant containing structured uncertainties. Provided that the same condition as for the standalone estimation is met, we can prove exponential convergence of the tracking error and parameter error to zero without necessitating persistency of excitation.

Data-driven FDI for wind farms using W-SVM

Abstract: The adoption of clean, renewable energy has brought to the forefront an increase in the studies and research around their reliable and efficient implementation. The increasing demand of wind-turbine generated power has led to the construction of larger turbines which require higher reliability guarantees in order to operate with reduced down-times and moderate repair costs. The use of advanced techniques for fault detection and isolation (FDI), and the subsequent fault tolerant control implementation in wind turbines is one of the proposed solutions to reduce losses in efficiency and ensure their continued operation. Although the implementation of FDI strategies in wind turbines have been developed greatly in the last decade, little work has been done at the wind farm level; this approach can solve the problems of detecting certain faults that have proven to be difficult to detect at the wind turbine level (e.g. those caused by mechanical wear on the internal structure of the wind turbine). This article presents the results of designing a data-driven FDI strategy for a wind turbine farm system via Weighted Support Vector Machines (W-SVM), achieving a fast and reliable way to detect faults with reduced missed detections, low number of false positive and fast enough detection rates.

Model-based parameter optimization of a fused deposition modelling process

Abstract: The aim of this research was to develop mathematical models of a fused deposition modelling process by two different approaches. These models were used to predict the process behavior, to derive optimal process parameters and also to compare these different modelling approaches in terms of model quality. Four subordinate targets were defined. For each, two different modelling approaches were applied. First, black box models were established using design of experiments (DoE). Second, white box models were derived from a theoretical analysis of the process. Afterwards, three different parameter optimizations were applied for the considered fused deposition modelling system to evaluate and compare the prediction accuracy of the models.

Comparative analysis of MIMO adaptive torque control for 7-DOF redundant robotic arm

Abstract: In this paper, a multi-input multi-output (MIMO) direct adaptive torque controller is presented that uses a conventional fuzzy system to provide asymptotic end-effector tracking of a reference path for a 7-DOF redundant robotic arm. In order to find a control strategy that is both robust and efficient with respect to disturbances, sensor noise and poorly understood dynamics, we compare this controller with two other dynamic controllers: single-input, single-output (SISO) PID controller and multi-input, multi-output (MIMO) feedback linearization controller. It is illustrated via simulations that the MIMO adaptive controller, which drives the torque of each joint to control end-effector dynamic variables, can highly improve the robotic performance considering both its kinematics and dynamics while executing motion control or tracking a reference in work space. The efficacy of our control algorithm affects the accuracy, stability and robustness of both motion control and path tracking.

Non-uniform 1D antenna array optimal configuration via extremum seeking control

Abstract: Methods to find the optimum non-uniform onedimensional (1D) antenna array geometry for maximizing directivity are presented. These methods develop a novel objective function to incorporate with the extremum seeking control methods considered. A comparison is performed between the major extremum seeking control approaches for solving the resulting optimization problem: (a) Perturbation-based Extremum Seeking Control (PESC), (b) Numerical Optimizationbased Extremum Seeking Control using Particle Swarm Optimization, to confirm the validity and the effectiveness of the proposed methods.

Identification of PEM fuel cells based on support vector regression and orthonormal bases

Abstract: Polymer Electrolyte Membrane Fuel Cells (PEMFC) are efficient devices that convert the chemical energy of the reactants in electricity. In this type of fuel cells, the performance of the air supply system is fundamental to improve their efficiency. An accurate mathematical model representing the air filling dynamics for a wide range of operating points is then necessary for control design and analysis. In this paper, a new Wiener model identification method based on Support Vector (SV) Regression and orthonormal bases is introduced and used to estimate a nonlinear dynamical model for the air supply system of a laboratory PEMFC from experimental data. The method is experimentally validated using a PEMFC system based on a ZB 8-cell stack with Nafion 115 membrane electrode assemblies.

Sufficient conditions for static-output feedback control of fractional-order linear systems

Abstract: A convex-optimization procedure is proposed for static-output feedback (SOF) control of commensurate fractional-order systems (1 < α < 2). The proposed sufficient conditions ensure the pole placement in the conic stability domain with a prescribed degree of stability. Numerical simulations are given to validate and approve the efficacy of the stabilization algorithm.

Recursive least square approximation using monotonic fuzzy systems

Abstract: Recursive least squares algorithms for approximation of a multivariate nonlinear function by a fuzzy system guaranteeing monotonicity of the corresponding mapping with respect to individual inputs are presented in this paper. Since the exact solution suffers from high computationally complexity two approximating solutions are presented as well. Two illustrative examples are given to compare the algorithms and to demonstrate the benefit of fuzzy systems preserving monotonicity.