Intelligent Control Systems with an Introduction to System of Systems Engineering

Thrishantha Nanayakkara, Ferat Sahin, Mo Jamshidi

• Publisher: CRC Press
• Year: 2018
• Print ISBN: 9781420079241
• eBook ISBN: 9781351834537
• Topic: Technology & Engineering
• Subtopic: Electrical Engineering & Telecommunications
• Index: Technology & Engineering

From aeronautics and manufacturing to healthcare and disaster management, systems engineering (SE) now focuses on designing applications that ensure performance optimization, robustness, and reliability while combining an emerging group of heterogeneous systems to realize a common goal.
Use SoS to Revolutionize Management of Large Organizations, Factories, and Systems Intelligent Control Systems with an Introduction to System of Systems Engineering integrates the fundamentals of artificial intelligence and systems control in a framework applicable to both simple dynamic systems and large-scale system of systems (SoS). For decades, NASA has used SoS methods, and major manufacturers—including Boeing, Lockheed-Martin, Northrop-Grumman, Raytheon, BAE Systems—now make large-scale systems integration and SoS a key part of their business strategies, dedicating entire business units to this remarkably efficient approach.
Simulate Novel Robotic Systems and ApplicationsTranscending theory, this book offers a complete and practical review of SoS and some of its fascinating applications, including:

• Manipulation of robots through neural-based network control
• Use of robotic swarms, based on ant colonies, to detect mines
• Other novel systems in which intelligent robots, trained animals, and humans cooperate to achieve humanitarian objectives

Training engineers to integrate traditional systems control theory with soft computing techniques further nourishes emerging SoS technology. With this in mind, the authors address the fundamental precepts at the core of SoS, which uses human heuristics to model complex systems, providing a scientific rationale for integrating independent, complex systems into a single coordinated, stabilized, and optimized one. They provide readers with MATLAB® code, which can be downloaded from the publisher's website to simulate presented results and projects that offer practical, hands-on experience using concepts discussed throughout the book.

1

Introduction

This book discusses the foundation for emerging intelligence in a loosely coupled system of systems (SoS) with a set of objectives. The term “intelligence” is a widely used term without a concrete definition. There are many schools of thought on what intelligence is, with arguments for and against each of them [1]. This book looks at intelligence as an emerging property of systems that interact among each other and with the outside world. Systems can be primitive computational elements such as neurons, mathematical equations, or electrical, electronic, and mechanical components, etc., or they can be complex entities such as a robot, a team of robots, a fleet of vessels, individuals in a group, etc. To understand the term “emergence,” let us look at a group of people listening to a speaker. As far as the audience is silent, the speaker goes on talking about the topic. Nothing emerges from the audience except for individual thoughts interacting with the speaker. Imagine a break is given for the audience members to talk to each other on the topic being discussed. Then, people go about talking to each other, exchanging ideas with agreements and disagreements. Over time, the individuals who contribute to the discussion change their original opinions and sometimes come up with new ideas triggered by new ways of thought introduced by others. Therefore, the group as a whole “emerges” concepts that were not there at the beginning. This book tries to argue that this fundamental idea of emergence holds powerful meaning in systems of any scale. The only requirement is to design a process that allows subsystems of a system to communicate among each other in order to emerge intelligent behavior envisaged by the designer.

There are several factors affecting the fruitfulness of the above interaction. If we go back to our example, the final outcomes can vary depending on how the discussion was coordinated. If there was an arbitrator who intervened in the way people met each other, for instance, by introducing one to another with some purpose, the type of ideas that would emerge would be different from those that would emerge if the group was allowed to interact randomly. In addition, the character of individuals also matters in the quality of the outcome. Suppose that there were very stubborn people who would not change and tolerate diverse ideas. They may make the discussion very boring. Therefore, internal characteristics of the participating subsystems, such as plasticity, compliance, adaptiveness, stability, and inertia to change, etc., affect the final behavior. The time span of interaction also decides how transient behaviors tend to settle in a stable behavior. Moreover, how information is interpreted by each individual also affects the final outcome. Therefore, it is difficult to pinpoint a single place where intelligence resides. In essence, it is decided by an array of factors, including how goals are set, the inherent characteristics of subsystems, how information is exchanged, how information is processed, the mechanisms available for adaptation, how adaptation is rewarded or punished, etc.

The book is more of a discussion with the reader than a formal presentation of theories. The chapters are organized in the following manner.

In Chapter 2, we discuss the elements of a classical control system. It is important to discuss the elements of a control system because any system can be viewed as a control system with or without a feedback loop [2,3]. A system changes its state when an external stimulus is applied. There are systems that respond to the stimuli without worrying about how the states are being changed. These are called open loop systems. An example of an open loop system is a water tap. The more you open it, the faster the water flow will be. There are other systems that try to add a corrective feedback action on the system by further processing the states being changed by the external input. These are called feedback control systems. An example is an air conditioner. When you set a desired temperature, the air conditioner tries to keep the room temperature at that level when the outside temperature goes through variations. Knowledge of the essential components needed to build a system that renders a desired primitive behavior is important to design more complex systems with more features such as learning and adaptation. Here, we take examples from robotics to understand how these fundamentals can be used to design stable and optimal controllers for dynamical systems found in the industry [4].

Chapter 3 introduces the concept of SoS and the challenges ahead to extend systems engineering to SoS engineering. The birth of a new engineering field may be on the horizon—System-of-Systems Engineering. An SoS is a collection of individual, possibly heterogeneous, but functional systems integrated together to enhance the overall robustness, lower the cost of operation, and increase the reliability of the overall complex (SoS) system. Having said that, the field has a large vacuum from basic definition, to theory, to management and implementation. Many key issues, such as architecture, modeling, simulation, identification, emergence, standards, net-centricity, control (see Chapter 8), etc., are all begging for attention. In this chapter, we will be going briefly through all these issues and bring the challenges to the attention of interested readers [11, 12, 13, 14, 15, 16, 17, 18].

This growing interest in SoS as a new generation of complex systems has opened a great many new challenges for systems engineers. Performance optimization, robustness, and reliability among an emerging group of heterogeneous systems in order to realize a common goal has become the focus of various applications, including those in military, security, aerospace, space, manufacturing, service industry, environmental systems, and disaster management.

Chapter 4 discusses the mathematical tools available to observe the internal states of a system not directly measured by physical sensors, and to optimally estimate the real state of the system when the measurements and states themselves are contaminated with noise. Consider, for instance, a pendulum that swings back and forth under gravity. You have an angle sensor attached to the pivotal point of the pendulum around which it swings. therefore, you get a direct measurement of the angle of the pendulum at any given time. yet, if you see the voltage signal coming from the angle sensor on an oscilloscope, you will notice that the angle signal is contaminated with noise. The level of noise depends on the quality of the sensor on one hand and the firmness of the mechanical assembly of the pendulum on the other. We call The former type of noise the measurement noise, and the latter the process noise. The Next question is how to estimate the state of true angle, and the other higher-order states such as angular velocity of the pendulum using this noisy measurement. To understand how to solve engineering problems like this, we first discuss the basic concepts of observing based on our knowledge about state space modeling and feedback control systems. then, we go on to discuss the mathematical derivation and application of kalman filters and particle filters that are extensively used in various industrial applications [9,10]. Knowledge about these filtering and state observing techniques is important in the systems integration stage in order to insulate one system from noise injected from another system.

Chapter 5 discusses how linguistic information can be modeled to improve the performance of a system or a combination of independent systems. Here, we introduce the concept of “membership” of a measurement in a given set. For instance, if we see a 5.7-ft-tall man, how do we categorize him in the classes of “short” and “tall”? This is the type of classification where we use the notion of memberships in a class. In this particular example, we may assign a membership of 0.7 in the “tall” class and 0.3 in the “short” class. All our subsequent decisions about this person will depend on this interpretation of his height. This conversion of a measurement to membership in linguistic classes is important to exploit linguistic rules being used by people to express their experience on controlling complex systems. For instance, a linguistic rule such as “if the road is slippery and the traffic is high, then keep your speed low.” This type of inferencing based on linguistic labels is known as fuzzy inferencing [11,12]. This rule may help numerous sensors such as speedometers, cameras, inertial navigation systems, brake sensors, engine power sensors, etc., mounted on intelligent cars to coordinate among each other to do a meaningful job.

Chapter 6 discusses some techniques available to approximate nonlinear static and dynamic systems using a set of primitive functions. This is inspired by the manner in which biological brains approximate the function of complex systems. If we take a simple example, let us consider the nonlinear function given by y = exp(x³ + 2x² + x + 5). If we are only concerned about doing what this function does in terms of mapping the value of the variable x to the corresponding value of y, do we really have to know the exact equation y = exp(x³ + 2x² + x + 5)? The answer is no. It is enough to know the landscape created by this function in the x-y coordinate system to do this mapping. Similarly, our brain receives sensory information from five sensors, and the processed information is often mapped to actions through muscle commands. This mapping should take place to suit the task being done. For instance, if we are required to learn how to ride a bicycle, the brain has to construct a model of the dynamics of the bicycle in order to perform smooth maneuvering. The brain does this by constructing a neural network that approximates the dynamics of the bicycle. This can be done by combining a set of primitive mathematical landscapes found in special cells knows as neurons. Therefore, in this chapter, we demonstrate how such primitive mathematical functions known as artificial neurons can be used to approximate complex systems. This type of approximation can be useful to complement crude mathematical models of subsystems of a large system or to understand complex dynamics emerging from a system of simpler subsystems of which the dynamics are fully or partially known.

Chapter 7 discusses an SoS [5,6], simulation framework, related case studies, and agent-in-the-loop simulation for SoS test and evaluation. First, it introduces the SoS concepts and recent work on SoS simulation and modeling. Then, it explores an SoS simulation framework based on discrete event specification tools and Extensible Markup Language (XML) [7]. Three case studies (scenarios) are simulated with the SoS simulation framework on robust threat detection and data aggregation using heterogeneous systems of rovers (systems). Finally, a real-time SoS simulation framework is introduced in agent-in-the-loop setting for testing and evaluating systems in an SoS. With the agent-in-the-loop simulation, a robust threat detection scenario is simulated with four virtual (simulated) robots and one real robot. Continuity models for the real and virtual robots are presented in addition to a communication interface between the real robot and the virtual ones [8].

Chapter 8 discusses one of the key theoretical open problems of SoS—control design. Among all open questions in engineering of SoS, control and sensing are among the most important ones. From the control design viewpoint, the difficulty arises that each system’s control strategy cannot solely depend on its own onboard sensory information, but also due to communication links among all the neighboring systems or between sensor, controllers, and actuators. The main challenge in the design of a controller for SoS is the difficulty or impossibility of developing a comprehensive SoS model, either analytically or through simulation, by and large; SoS control remains an open problem and is, of course, different for each application domain. Should a mathematical model be available, several control paradigms are available, which will be the focus of this chapter. Moreover, real-time control—which is required in almost all application domains—of interdependent systems poses an especially difficult problem. Nevertheless, several potential control paradigms are briefly considered in this chapter. The control paradigms discussed in Chapter 8 are hierarchical, decentralized, consensus-based, cooperative, and networked. Simulation results and design algorithms are presented in this chapter.

Chapter 9 discusses how adaptive systems can optimize their behaviors using external rewards and punishments. This is an important area of machine learning. The basic idea is to use qualitative feedback as to whether the performance is good or bad to retune various system and control parameters. For instance, let us consider a robotic head that is supposed to look at a human and react with facial expressions like the Kismet robot at Massachusetts Institute of Technology. Here, there are no quantitative measures such as position or velocity error to correct the reaction of the robot. The human can say whether the robot’s reaction is close to being natural or not. Perhaps the human can give a score between 0 and 100. Then, the robot can use a strategy to change the behavior to obtain higher scores. This strategy is called a reinforcement-based learning strategy [19]. This chapter discusses the fundamentals of how such a learning system can be designed. Reward-based learning systems are very important to future large-scale systems because the engineered systems undergo many internal and external changes (e.g., friction, electrical impedance, etc.). It is very difficult to design controllers to be optimal and stable for a wide class of environments. If...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Table of Contents
• Preface
• About the Authors
• 1 Introduction
• 2 Elements of a Classical Control System
• 3 Introduction to System of Systems
• 4 Observer Design and Kalman Filtering
• 5 Fuzzy Systems—Sets, Logic, and Control
• 6 Neural Network-Based Control
• 7 System of Systems Simulation
• 8 Control of System of Systems
• 9 Reward-Based Behavior Adaptation
• 10 An Automated System to Induce and Innovate Advanced Skills in a Group of Networked Machine Operators
• 11 A System of Intelligent Robots–Trained Animals–Humans in a Humanitarian Demining Application
• 12 Robotic Swarms for Mine Detection System of Systems Approach
• Index