Bi-weekly seminar series on Information Technology Research |
DATE |
TITLE |
Presented By |
10-27-2000 |
Some Concepts of Networked and Decentralized Control Theory, Abstract, file |
Andrew Stubbs |
11-10-2000 |
Timing Problems in Real-Time Control Systems, Abstract, file |
Cheng Tang |
12-01-2000 |
An Efficient Thread Runtime System and its Application in a Scalable Java Server, Abstract, file1, file2 |
Fabian Breg |
02-09-2001 |
Analysis and Design of an Adaptive Virtual Queue Algorithm for Congestion Notification in the Internet, Abstract, file |
Srisankar Kunniyur |
02-23-2001 |
Petri Net Supervisors for Discrete Event Systems, Abstract, file |
Yingquan Wu |
03-23-2001 |
A framework for embedding real-time adaptability and QoS in non-real-time OSes, Abstract, file |
Dimitrios Nikolopoulos |
04-13-2001 |
Real-Time Control over the Internet, Abstract, file |
Benjamin Yoo |
05-11-2001 |
Control in stuctured interconnection and communication enviroments, Abstract |
Petros Voulgaris |
05-25-2001 |
Networked control system with communication constraints, Abstract, file |
Cheng Tang |
06-08-2001 |
Real-Time Processing Environments on Commodity Operating Systems, Abstract, file |
Dimitrios Nikolopoulos |
06-21-2001 |
Control with Random Delays: A Jump Systems Approach, Abstract, file |
Benjamin Yoo |
07-05-2001 |
CDMA Uplink Power Control as a Noncooperative Game, Abstract, file |
Tansu Alpcan |
07-31-2001 |
Rate-based flow controllers for communication networks in the presence of uncertain time-varying multiple time delays, Abstract |
Prof. Altug Iftar |
09-21-2001 |
Limited Data Rate in Contol Systems with Networks, Abstract, file |
Hideaki Ishii |
09-28-2001 |
Networked Control of Distributed Systems: A Testbed, Abstract, file |
Andrew Stubbs |
10-12-2001 |
Issues in Real Time Java, Abstract, file |
Fabien Breg |
11-09-2001 |
Control over a finite data rate channel, Abstract, file |
Hideaki Ishii |
11-30-2001 |
Network Control Design for Linear Systems, Abstract, file |
Cheng Tang |
01-18-2002 |
Coordinated Scheduling with Implicit Information in Distributed Environments, Abstract, file1, file2 |
Dimitrios Nikolopoulos |
02-01-2002 |
Visual Servo Control: Current Trends and New Directions, Abstract |
Seth Hutchinson |
02-15-2002 |
Rate Control for Communications Networks: shadow prices, proportional fairness, and stability, Abstract, file |
Ling Yei |
03-01-2002 |
Remote control of LTI systems over networks with state quantization, Abstract, file |
Hideaki Ishii |
03-27-2002 |
Global Stability of Congestion Controllers for the Internet, Abstract, file |
Supratim Deb |
04-05-2002 |
Non-Concurrent Fault Identification in Discrete Event Systems Using Encoded Petri Net States, Abstract |
Yingquan Wu |
04-09-2002 |
Control of Systems with Time Delays, Abstract |
Benjamin Yoo |
04-26-2002 |
Time Delays in Computer Control Networks for Control, Abstract |
Bjorn Wittenmark |
06-21-2002 |
Robust Control of Heterogeneous Networks, Abstract, file |
Ying Lei |
07-05-2002 |
Development of a vehicle network control testbed, Abstract |
Andrew Stubbs and Vladimeros Vladimerou |
08-30-2002 |
Stability Analysis of Quantized Sampled-Data Systems through Randomized Algorithms, Abstract, file |
Hikeaki Ishii |
09-20-2002 |
Distributed Scalable Java Operating System, Abstract, file |
Fabian Breg |
10-11-2002 |
Robot Motion Planning Under Visibility Constraints, Abstract, file1, file2 |
Rafael Murrieta Cid |
10-18-2002 |
Capacity of Rayleigh Fading Channels: A Team Problem Involving Fading Channels, Abstract, file |
Akshay Kashyap |
10-25-2002 |
HotDec Update, Abstract, file |
Andrew Stubbs and Vladimeros Vladimerou |
11-08-2002 |
Structured Controller and its Application to Network Control, Abstract |
Prof. Petros Voulgaris |
11-15-2002 |
Feedback Control over Imperfect Network Links, Abstract, file |
Rouzbeh Touri |
12-06-2002 |
Paper by: Sunil Kumar and P.R. Kumar, "Performance Bounds for Queuing Networks and Scheduling Policies," IEEE Trans. Automatic Control, August 1994, Abstract |
Andrew Stubbs |
02-28-2003 |
Building remote sensing and control on top of a distributed Java Platform, Abstract |
Prof. Polychronopoulos |
03-12-2003 |
A convex optimization-based nonlinear filtering with redundant observations and bounded disturbances, Abstract, file |
Ji-Woong Lee |
09-15-2003 |
Synthesis of switching rules for switched linear systems through randomized algorithms, Abstract, file |
Hideaki Ishii |
09-27-2003 |
Probabilistic Fault Detection in Networked DES, Abstract, file |
Prof. Christoforos Hadjicostis |
10-30-2003 |
Paper by: G.N. Nair and R.J. Evans, "Exponential Stabilizability of finite-dimensional linear systems with limited data rates," Automatica, v. 39, April 2003, pp. 585-593, Abstract, file |
Serdar Yuksel |
| Abstract:
This presentation reviews three papers dealing with networked and decentralized control theory. These papers cover the topics of a time invariant plant that is stabilized using a network. The resulting system is modeled using a periodically time varying system that uses decentralized controllers. The concept of decentralized controllers and "fixed modes" is then reviewed and a procedure for constructing a set of stabilizing decentralized feedback control laws is also presented. Finally a technique for "lifting" a periodically time varying system into a time invariant system is discussed. |
| Abstract: |
| Abstract:
In this presentation we introduce Nanothreads. NanoThreads is an efficient thread runtime system which uses lightweight data transfer between user space and OS kernel. In addition more efficient data structures like a non blocking run queue and restartable wait lists further increase thread performance. We will also show how incorporation of this thread runtime in a Java Virtual Machine can improve server networking performance under high loads. |
| Abstract:
Our goal is to design a low-delay service for elastic users in the Internet using Active Queue Management (AQM) schemes. AQM schemes in Internet routers provide congestion information to end sources by marking dropping packets. We will propose and analyze a class of AQM schemes that, in conjunction with TCP and other congestion controllers, allow the router to achieve a high level of utilization while keeping the queue length small. We will then propose a specific AQM scheme called the Adaptive Virtual Queue Algorithm that can be easily implemented in routers and performs well compared to other AQM schemes. We will discuss its implementation and its performance relative to other AQM schemes. |
| Abstract:
The talk will describe a particular methodology for constructing Petri net supervisors. This approach, which was developed in [1,2], is based on place invariants and can be used to enforce logical and algebraic constraints between elements in a discrete event system. In addition, these techniques can handle constraints that arise due to uncontrollable and unobservable transitions. We are interested in extending these ideas into distributed/decentralized settings and also in relating them with our techniques for Petri net monitoring. [1] K. Yamalidou, J. O. Moody, M. D. Lemmon and P. J. Antsaklis,
"Feedback control of Petri Nets based on place invariants," Automatica,
vol. 32,
|
| Abstract:
Contemporary operating systems are designed to improve throughput and responsiveness, based on a classification of jobs into CPU-bound, I/O-bound, interactive, and batch classes. This classification has driven workload-based evaluation of computer systems since the appearance of the first multitasking computers in the 60s. Unfortunately, throughput and response time are not the only criteria for the quality of service delivered to the users of modern computer systems. For numerous important applications, mostly from the areas of control and multimedia, the quality of service is measured as the ability of the system to provide resource availability guarantees during specified time intervals and meet time constraints imposed by the jobs themselves. These applications are usually referred to with the term real-time applications. The needs of real-time applications have motivated intensive efforts on constructing schedulers that strive for satisfying resource reservations and deadlines imposed by real-time tasks. Research on real-time scheduling has been active for at decades and real-time operating systems have already reached a mature point, being actively employed in several industrial settings. On the other hand, the most popular computer systems in use today, namely desktop PCs and enterprise servers, are far from efficient in supporting real-time applications. Modern operating systems are heavily based on priority decaying time-sharing, a technique that has proven to be robust for traditional resource-bound workloads. Real time applications are either ignored or treated as a class of jobs with priority higher than any other job in the system. This treatment is sufficient for handling simple single-task real-time tasks, such as playing a video, but incapable of providing any quality of service for -possibly multitasked or multithreaded-real-time jobs that require resource reservations and deadline satisfaction.The treatment is common to most operating systems in use today, including Windows NT, Linux, and Solaris. Integrating real-time support and quality of service in non real-time operating systems is a non-trivial task, since any solution employed must enable the symbiosis of real-time and non real-time jobs in a non-intrusive manner. We present an operating system architecture that promises to meet this requirement. Our architecture is structured around a shared arena, an efficient communication substrate that lets user jobs and the operating system communicate with minimal, and most importantly predictable overhead. The shared arena bypasses several system calls and other time-consuming kernel control paths, by having the programs and the kernel communicate through loads and stores in shared memory. This bi-directional communication path can be used to implement a complete interface for negotiating resources between the jobs and the operating system. The shared arena is ideal for implementing reservations and allocation of resources such as CPU, memory, and bandwidth, with an overhead bound by the memory accesses required to exchange the required information. This information is directly deposited to shared-memory by the user program or the kernel and retrieved with a bounded number of memory accesses. The strongest advantages of the shared arena are high performance and the ability to export critical resource management decisions to the applications. Letting applications manage their own resources is the key feature of operating system extensibility. Extensibility is necessary for improving the performance of jobs for which the traditional fair-share policies of operating system managers fail to meet the requirements. The benefit of extensibility has been already demonstrated in several important application domains, such as databases, web servers, and parallel computing. It stems from the ability of applications to utilize better the resources available to them at runtime, by prioritizing their tasks according to their criticality. The concept of extensibility has driven operating system research in the last 15 years, leading to several new kernel designs such as micorkernels and exokernels. The shared arena is ideal for embedding extensibility in already deployed operating systems such as Linux, since it does not require changes of the monolithic kernel architecture. |
| Abstract:
This presentation gives a broad overview of networked control systems. Networked control systems involve the controller, sensor, and actuator signals being transmitted through a communications network such as Ethernet. Advantages are easier wiring and maintenance of the control system because point-to-point connections are eliminated. The disadvantage is the time delay that is caused by transferring feedback and control signals through the network. The presentation goes over networking protocols and describes a simple experiment involving speed control of a DC motor. The experiment uses Visual Basic and C++ code to perform the necessary TCP/IP communications. Results from the experiment are described as well. |
| Abstract:
We present a list of optimal disturbance rejection problems in systems in which the overall control scheme is required to have a certain structure. These structures correspond to various classes of controlled systems which include what we refer to as nested, chained, hierarchical, delayed interaction and communication, and, symmetric systems. These classes can be associated with several practical applications in integrated flight propulsion systems, platoons of vehicles, networked control, MEMS, production lines, chemical processes, etc. The common thread in all of these classes is that by taking an input-output point of view we can characterize all stabilizing controllers in terms of convex constraints in the Youla-Kucera parameter. The disturbance rejection problem can therefore be casted as a convex, yet nonstandard, model matching problem. Approaches that solve this problem are presented for various optimality criteria. |
| Abstract: |
| Abstract:
This talk is based on some recent papers from the IEEE Real-time technology symposia. It focuses on the design, implementation and performance of real-time processing environments built with minimal extensions to commodity operating systems such as Linux and windows NT. The topics discussed are resource kernels, OS support for fine-grain distributed real-time processing, and real-time extensions to the Java threading and synchronization infrastructure. The talk outlines advantages and limitations of these environments for different types of real-time applications, with respect to extensibility, portability, implementation effort and performance. |
| Abstract:
A review of a recent conference paper authored by Xiao, Hassibi, and How is the subject of this talk. The team studies the networked control problem by formulating the problem as a jump linear system with jumping parameters determined by a Markov process. The formulation is applicable to both static and dynamic controller cases. The authors propose using V-K iteration, a heuristic algorithm, as a way to synthesize feedback controllers that will stabilize the system with delays. The stability condition for the problem is mean square stability. |
| Abstract:
We present a game-theoretic treatment of distributed power control in CDMA wireless systems. We make use of the conceptual framework of noncooperative game theory to obtain a distributed and market-based control mechanism. Thus, we address not only the power control problem, but also pricing and allocation of a single resource among several users. A cost function is introduced as the difference between the pricing and utility functions, and the existence of a unique Nash equilibrium is established. In addition, two update algorithms, namely parallel update and random update, are shown to be globally stable under reasonable conditions. Convergence properties and robustness of each algorithm are studied through extensive simulations. |
| Abstract:
This presentation gives an overview on data rate issues that arise when networks are used in control systems. First, a summary is given on how such networks are used in practice and on what advantages and limitations they can bring in control. Then two problems are described, one on distributed control over networks and the other on quantized control. |
| Abstract:
The concept and design of a Network Control Test Bed is discussed. This includes the uses for the completed bed and how the bed functions. |
| Abstract:
The original design goals of the Java programming platform did not include performance, much less real-time performance. In fact, the most important original design goal, portability of the Java platform, seems to conflict the requirements needed by real-time applications. As the Java platform grew more popular, and its performance steadily improved, Java also found its way into real-time systems programming. Through the Java Community Process, the Real-Time for Java Experts Group has drafted a specification outlining extensions to the Java API that can be used in the development of real-time applications. After providing a short introduction into the relative aspects of the Java platform, his talk will outline these real-time extensions to the Java API. Besides extending the Java API, in order to obtain real-time performance, the JVM, the heart of the Java platform, must offer real-time performance. The third part will discuss issues relating to real-time JVM performance. |
| Abstract:
Communication over a digital channel involves sampling and quantization of signals and time delay in their transmission. In this talk, these three issues are considered in a controller design problem of stabilizing a linear time-invariant system. This method gives a way to calculate the data rate necessary for control explicitly. |
| Abstract: |
| Abstract:
This talk covers the fundamental concepts of co-scheduling, a recently proposed technique which aims at ensuring coordinated allocation of resources to communicating processes. The talk motivates coscheduling and discusses the most important details of relevant coscheduling algorithms, including implicit and dynamic coscheduling. The second part presents some recent work done at CSRD on combining coscheduling with thrashing prevention to improve the throughput of clustered computational farms with limited memory resources. This is based on a recent paper accepted for presentation at the 2nd IEEE/ACM International Symposium on Cluster Computing and the Grid. The paper is the winner of the Best Paper Award at the conference. |
| Abstract:
Visual servo control is a maturing method for controlling robots using real-time vision feedback. It can be considered the fusion of computer vision, robotics and control, and it has been a distinct field for over 10 years, though the earliest work dates back close to 20 years. Over this period several major, and well understood, approaches have evolved and been demonstrated in many laboratories around the world. Historically there have been two paradigms for visual servo control, Position Based and Image Based, each with its own shortcomings. We will briefly describe some recently discovered performance problems faced by these systems, and provide geometric explanations for them. We then describe some recent hybrid approaches that address these problems. Finally, we introduce a new hybrid approach that employs a novel decoupling of the degrees of freedom of camera motion. The new approach improves performance while simultaneously ensuring that field of view constraints are enforced. |
| Abstract: |
| Abstract:
When a finite data rate channel is used in a feedback control system, quantization/coding is one important aspect to guarantee control objectives. In this paper, we consider a system in a remote control setup, where the plant and the controller are connected by a network cable, and study the problem of designing quantizers for its stabilization. It is assumed that the computation available on the plant side in the sensor/actuator is limited and also that broadcast of messagesis allowed over the channel. Motivated by this setup, we first study a state quantization scheme and then apply this to the remote control problem. |
| Abstract:
We consider a single link accessed by a single source which responds to congestion signals from the network. The design of controllers for such sources in the presence of feedback delay has received much attention recently. Here we present conditions for the global, asymptotic stability and semi-global exponential stability of congestion controllers which are natural extensions of earlier linearized analysis of such systems. Our result on exponential stability provides the missing link in the proof of how one obtains a single deterministic congestion control equation from a system with many congestion-controlled sources and random disturbances. Using numerical examples, we compare the conditions on the congestion-control parameters obtained using local and global stability analysis. |
| Abstract:
In this paper we extend a previously developed, coding-based methodology for monitoring faults in discrete event systems that are described by Petri nets. We consider fault models that capture faults in both Petri net transitions and places, and present a systematic design that enables the detection and identification of multiple (and mixed) such faults. More specifically, using an encoded Petri net model with $2k$ redundant places (and the connections and tokens associated with them), we can non-concurrently detect and identify up to a maximum of $2k-1$ transition faults and up to a maximum of $k$ place faults that may occur at various instants during the operation of the Petri net. The worst-case complexity of the detection and identification procedure is $O(k^2m)$ operations, where $m$ is the number of transitions in the given Petri net model. The monitoring mechanism, which receives and stores information about the activity in the original system in a way that enables non-concurrent detection and identification, is a Petri net with $2k$ places and can be obtained from the encoded Petri net model in a straightforward manner. The proposed fault detection and identification approach does not need to explicitly track or reconstruct the system state evolution and is, thus, well-suited for non-concurrent diagnosis. |
| Abstract:
This talk will discuss two approaches in synthesizing controllers to compensate for communication delays in a control system. One approach is to cast the problem into an LPV framework, allowing the delays to be parameters for plant and controller. Then an H-infinity controller canbe synthesized using well-known techniques. Another approach is to formulate the problem into a stochastic setting, specifically using jump systems with a parameter governed by a Markov chain. A recent paper will be reviewed that gives conditions for synthesizing an H-infinity controller in the stochastic setting. |
| Abstract:
The industrial trend in embedded systems is than an increasing amount of control applications are implemented in distributed computer systems. Additional problems then face the designers such as significant and potentially varying end-to-end time-delays, the fact that data may be lost on the network, and that one or more nodes, or even parts of a node, may experience errors. Control systems often rely on that actions in the control system are done synchronized. In a distributed control system this may not be the case. If no schedule is used for the network connecting different parts of a control system, the communication of sensor and actuator signals may be delayed a random time due to varying system load. Similar effects can occur in one processor systems with dynamic scheduling. In the case with a schedule for network transmissions the schedule has to be synchronized with sampling periods in the control system to avoid jitter effects. In a system with many different sampling periods this may not be possible, leading to delay variations. A system using dynamic scheduling of communication services is also more flexible. For instance, new units can be added without redesigning the schedule. The seminar describes some of the analysis and control problems that occur in distributed control systems. Analysis tools and an optimal LQG controller is described for systems with time delays. Time delays due to the multi-rate nature of the distributed control system is also described. |
| Abstract:
The presentation is based on Prof. Glenn Vinnicombe "Robust Control of Heterogeneous Networks". It mainly talks about how to build stable and robust large scale feedback interconnections of linear dynamical systems and how to get the stability of the network using decentralized condition. |
| Abstract:
The talk will be an update on the design and development of HoTDeC, a testbed for network control of distributed vehicle networks. The developed system utilizes communication via a combination of Bluetooth and ethernet links. |
| Abstract:
We present a novel development of randomized algorithms for quadratic stability analysis of sampled-data systems with memoryless quantizers. The specific randomized algorithm employed generates a quadratic Lyapunov function and leads to realistic bounds on the performance of such systems. A key feature of this method is that the characteristics of quantizers are exploited to make the algorithm computationally efficient. |
| Abstract:
This presentation will introduce the concept of distributed shared memory, and possible approaches to introduce that concept in the Java programming language. After introducing a number of existing projects that implement this concept, some extensions to this concept will be proposed. These extensions will be implemented in the Distributed Scalable Java Operating System. This presentation will continue by providing an overview of the design of DSJOS, outlining how this design facilitates the goals stated for this project. To conclude, the current status of our implementation will be presented, followed by some of the major research goals still to be addressed in the future. |
| Abstract:
The goal of this work is to develop techniques that allow one or multiple robotic observers to operate with full or partial autonomy while accomplishing the tasks of model building and target tracking. The planning algorithms we develop operate using certain simple but flexible models of the observer sensor and actuator abilities. We provide techniques that allow us to implement these sensor models on top of the capabilities of the actual sensors we have. It is worth keeping the following points in mind regarding our goals: - even with completely idealized sensing and mobility capabilities, the algorithmic task of model building and target tracking are quite challenging and areas of current and future research. - computational techniques can be used to approximate and implement these idealized sensors on the top of actual sensors - the quality and success of the generated plans depends significantly on the observer capabilities, study of this dependency terms of high-level parameters describing the sensors (e.g., max. distance sensed, viewing frustum) is part of the this work. One characteristic feature of this study is the need to satisfy visibility constraints while planning motions. This work focuses on the fundamental motion planning problem considering information provided by sensors. Some of the questions that this work tries to answer are: Which locations must be visited by a robot to efficiently map a building? How must a robot move to explore an environment? What motions will keep a target in view? To answer these questions we propose randomized motion planning techniques. Two main tasks regarding motion planning under visibility constraints are discussed in this talk A) Map Building and B) Target Tracking. |
| Abstract:
The talk is a tutorial on the basic information theory involved in the study of fading channels. Material in the first part of the talk is from I. E. Telatar's paper on "The capacity of Multi Antenna Gaussian Channels". (1999, Eur. Trans on Telecomm -- won the IEEE Information Theory Society Best Paper Award.) Rayleigh fading channels arise in wireless communication whenever there is no line of sight between the transmitter and receiver. We study the Rayleigh fading channel as a matrix channel, and derive the capacity of an ergodic channel. For the non-ergodic channel, we discuss the outage capacity formulation. We show the capacity advantage of using multiple antennas at both the transmitter and receiver. After thus establishing the limits of communication on fading channels, we discuss a team problem involving communication on a fading channel. We pose the problem as the design of a linear encoder and a corresponding decoder for a given channel, with the objective of minimizing the mean square error between the transmitted and the reconstructed vectors. We then show that the solution of this problem is very similar to the computation of capacity. We compare the performance of this extremely simple coding-decoding scheme with the optimal performance theoretically attainable. |
| Abstract:
The presentation is an update on the ongoing HoTDeC (HOvercraft Testbed for Decentralized Control) project. We will briefly describe the project and provide information on recent developments. |
| Abstract: |
| Abstract:
The increasing availability of network connectivity (such as the Internet) opens up possibilities for computer-based control methodologies that are implemented centrally or distributively using existing network infrastructure as the communications backbone. Before such network control methodologies become widely employed, however, a number of questions need to be addressed. Among other issues, we need a better understanding of the role of communication delays, the effect of packet drops in the network, and the constraints these impose on different control methodologies. Recent work has looked at the stability of networked control systems (i.e., control systems in which the feedback loops are closed through networks) and focused on performance guarantees under networked induced delays and various data packet rates. In this work, these ideas are used in a probabilistic setting by modeling a packet dropping network as an erasure channel and by developing bounded variance stabilization schemes. We also present extensions of these ideas to continuous-time LTI systems and show how to optimally choose the sampling interval in a way that results in minimal variance. |
| Abstract:
This paper discusses how to find performance bounds for queuing networks in steady state using linear programming. I will start with the detailed examination of open networks and then show how the framework can be expanded as it is in the paper for networks with buffer policies, closed systems, Bernoulli splitting, machine failures, E2/M/1 networks using the method of stages. |
| Abstract:
In this informal session, we will explore ways in which to implement remote sensing and control capabilities using a distributed Java based platform that hides the complexities of network communication from the developer. We will consider issues like ease of use (how well does this platform fit the target application) and performance. The goal is to understand the needs of the application and the nature of the proposed platform and their possible interaction. |
| Abstract:
In the field of semiconductor manufacturing, real-time estimation and control of patterned wafer topography is becoming important as pattern dimensions of modern integrated circuits continuously shrink. The key difficulties to overcome have been that the underlying dynamics are often unknown, and that the computational complexity of existing observation models is too high. This talk will introduce a convex optimization-based nonlinear filtering algorithm for unknown dynamics and high-dimensional, high-complexity observations. The algorithm will be shown to yield bounded state estimation errors if the disturbances are small and bounded, and if the observations are redundant in a certain sense. An experimental result will be presented to demonstrate the algorithm's capability in real-time estimation of patterned wafer parameters in a plasma etching process with optical observation. |
| Abstract:
In this talk, the design of a stabilizing switching rule for a switched linear system is considered. We first propose a probabilistic algorithm for a known nonconvex condition that employs a multiple Lyapunov function. We prove a probability-one convergence of the algorithm under a new notion of convergence. Then, to reduce its complexity, modified versions of the algorithm are developed. We also present a class of more general nonconvex problems (including stabilization of switched nonlinear systems) to which this approach can be applied. The results are illustrated using two- and three-dimensional systems with multiple switch states. |
| Abstract:
This talk discusses a probabilistic methodology for detecting permanent or transient functional changes in the state-transition mechanism of a deterministic finite-state machine (FSM). The associated detector observes the frequencies with which states are occupied and detects faults by analyzing the deviation between the expected frequencies and the actual measurements. In addition to state occupancy measurements, the detector requires a statistical characterization of the input, but does not need to know the order with which states appear or the input sequence that is applied to the FSM. These features can be useful in distributed settings where the input/state order may not be known due to synchronization or communication constraints. The discussion in this talk initially focuses on the case of a single fault and on the performance of the fault detector in terms of the probabilities of detection and false alarm, the number of states, and the length of the observation window. We then discuss how these ideas can be extended to specific applications, including monitoring and fault detection/diagnosis in distributed systems, and testing of digital sequential circuits. |
| Abstract:
In this paper, exponential moment stability of a multi-dimensional state with an unknown initial state is analyzed. Necessity conditions for the rate requirements are evaluated using asymptotic quantization theory. A construction with a rate greater than the bound is presented. In the remaining parts of the talk, a brief overview of relevant papers is discussed. |