E. Bradley and M. Easley. Reasoning about sensor data for automated system identi cation. Intelligent Data Analysis, 2(2):123-138, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....called phase portraits, or phase diagrams; when the axes of a map represent values of a single variable measured at different times, the maps are called delayed coordinate embeddings. Some previous work in AI and Cognitive Science that uses maps as representations includes Rosenstein, et al., 1997; Bradley and Easley, 1997; Campbell and Bobick, 1995; Thelen and Smith, 1994) The horizontal dimension is D(AB) the distance from A to B. The vertical dimension is VR, the relative velocities of A and B. The horizontal midline represents equal velocity, V(A) V(B) Above this midline, A is moving faster than B (or B is ....

Bradley, L and Easley, M. 1997. Reasoning about sensor data for automated system identification. In Advances in Intelligent Data Analysis. X. Liu, P. R. Cohen and M.

....5 and 10 volts, for example, can be dicult if one attempts to scan an ASCII text le, but it is trivial to see when the data is presented on an oscilloscope. Distilling available sensor information into this type of qualitative form is reasonably straightforward, as described in our IDA 97 paper[3], but reasoning about the information so derived is subtle and challenging. The second paradigm, qualitative bifurcation analysis, is based on a new construct called the qualitative state parameter space, an abstraction of regular state space with an added parameter axis, which is a useful way to ....

....16 seconds. This driver function operates via system calls to the Standard Instrument Control Library (SICL) 15] which contains high level procedures that control the data acquisition hardware. The Isaac tool then applies an array of intelligent sensor data analysis tools, described in detail in [3], to identify important qualitative properties of the time series. In the case of the data gathered from the thermistor, shown ISAAC multi DAQ voltageheat source controlled meter converter to analog digital thermistor Fig. 3. Experimental setup for automated input output modeling of a ....

E. Bradley and M. Easley. Reasoning about sensor data for automated system identi cation. Intelligent Data Analysis, 2(2):123-138, 1998.

....For example, an observation might inform Pret that the system to be modeled is autonomous; another observation could state that the state variable q oscillates and that this oscillation is damped. Observations can also be physical measurements made directly and automatically on the system (Bradley Easley 1998). Hypotheses about the physics involved, e.g. a hypothesis about friction, are supplied to Pret by the user; these may con ict and need not be mutually exclusive, whereas observations are always held to be true. Finally, speci cations indicate the quantities E E E E E ....

.... In order to achieve this behavior, Pret processes the observations curve tting, recognition of linear regions, and so on using Maple functions and simple phase portrait analysis techniques, both of which yield high level results that can then be used much as qualitative observations are (Bradley Easley 1998). Qualitative Simulation Before Pret resorts to the numerical level, it attempts to establish contradictions quickly and cheaply by reasoning about the qualitative states of the physical system (Kuipers 1992) Pret s qualitative envisioning module constrains the possible ranges of parameters in ....

[Article contains additional citation context not shown here]

Bradley, E., and Easley, M. 1998. Reasoning about sensor data for automated system identication. Intelligent Data Analysis. In press. Also in IDA-97.

....and asymptotes, and so on using Maple functions[23] and simple phase portrait analysis techniques[13] producing the type of abstract information that its inference engine can leverage to avoid expensive numerical checks. These methods, which are used primarily in the analysis of sensor data[15], are described in more detail in Sections 2.2 and 4. Pret does not currently reason about topology, but we are investigating how best to do so[71,72] 3.1.5 Parameter Estimation and Numerical Simulation Pret s nal check of any model requires a point by point comparison of a numerical ....

....a given state space portrait, allowing it to reason about these features at a much higher (and cheaper) abstraction level. These automated phase portrait analysis techniques, which combine ideas from dynamical systems, discrete mathematics, and arti cial intelligence, are covered in more detail in [15]. Raising the abstraction level of the analysis of individual sensor data sets, however, is only a very small part of the power of qualitative analysis of statespace portraits. Dynamical systems can be extremely complicated. Attempting to understand one by analyzing a single behavior ....

[Article contains additional citation context not shown here]

E. Bradley and M. Easley. Reasoning about sensor data for automated system identication. Intelligent Data Analysis, 2(2):123-138, 1998.

....This input output (I O) approach to dynamical system modeling, which distinguishes pret from other AI modeling tools, is very powerful and also extremely dicult. Distilling available sensor information into qualitative form is reasonably straightforward, as described in our IDA 97 paper[3], but reasoning about the information so derived is subtle and challenging. Dealing with actuators is even harder because of the nonlinear control theory that is involved. Among other things, determining what experiments one can perform from the system s present state involves complicated ....

....The input output modeling strategies that are the topic of this paper play important roles in both the generate and the test phase. The input half of pret s intelligent sensor actuator analysis and control module which is reviewed brie y in the following sections and covered in detail in [3] uses geometric reasoning and delay coordinate embedding to distill abstract, useful qualitative information from a highly speci c numeric sensor data set. The output part, described in the following sections, reasons about multiple sets of observations about a given system using a new ....

[Article contains additional citation context not shown here]

E. Bradley and M. Easley. Reasoning about sensor data for automated system identication. Intelligent Data Analysis, 2(2), 1998.

....continuous time nonlinear dynamic systems. It first uses domain knowledge to combine model fragments into ODEs, then observes the target system using sensors, and finally tests those ODEs against the sensor data using a body of mathematical knowledge encoded in first order logic(Stolle Bradley 1998). In order to interact with the target system, pret makes use of sensors and actuators, as shown in Fig. 4. Distilling available sensor information into qualitative form is reasonably straightforward (Bradley Easley 1998) but reasoning about the information so derived is subtle and di#cult. If ....

....data using a body of mathematical knowledge encoded in first order logic(Stolle Bradley 1998) In order to interact with the target system, pret makes use of sensors and actuators, as shown in Fig. 4. Distilling available sensor information into qualitative form is reasonably straightforward (Bradley Easley 1998), but reasoning about the information so derived is subtle and di#cult. If the target system has 34 state variables, for example, and one can only measure one of those 34 signals, it would appear that the conclusions that one can draw from the sensor data are fundamentally limited. This is control ....

[Article contains additional citation context not shown here]

Bradley, E., and Easley, M. 1998. Reasoning about sensor data for automated system identification. Intelligent Data Analysis 2(2).

.... Declarative Meta Control to Work Apollo Hogan and Reinhard Stolle and Elizabeth Bradley # Department of Computer Science University of Colorado at Boulder Boulder, Colorado 80309 0430 hogan,stolle,lizb cs.colorado.edu December 20, 1998 University of Colorado Technical Report CU CS 856 98 In review for publication in the IEEE Transactions on Systems, Man, and Cybernetics Abstract We present a logic programming system that accomplishes three important goals: equivalence of declarative and operational semantics, declarative ....

....a good model. The model checker makes use of several non logic based modules, e.g. the commercial symbolic algebra package Maple [CGG 91] a simple qualitative envisioning module, a nonlinear numerical parameter estimator [BOR98] and a geometric reasoner for intelligent data analysis [BE98] Calls to these modules require knowledge to be passed to them explicitly. This knowledge is made available by the inference engine by declaring the necessary predicates as relevant. Di#erent reasoning techniques vary considerably in their cost. Symbolic techniques are usually quick and cheap; ....

E. Bradley and M. Easley. Reasoning about sensor data for automated system identification. Intelligent Data Analysis, 2(2), 1998. Also in Second International Symposium on Intelligent Data Analysis (IDA-97).

....position coordinates q 1 and q 2 , and finally specifies resolutions and ranges. respect to the state variable q; another observation could state that q oscillates and that this oscillation is damped. Observations can also be physical measurements made directly and automatically on the system [ Bradley and Easley, 1997 ] Hypotheses about the physics involved, e.g. a hypothesis about friction, are supplied by the user; these may conflict and need not be mutually exclusive, whereas observations are always held to be true. Finally, specifications indicate the quantities of interest and their resolutions a ....

....roots 0 linear should satisfy x i = 0 constant should satisfy x i = 0 conservative r Delta f = 0 damped oscillation and autonomous r Delta f 0 Figure 3: Some observations and the corresponding inferences drawn by the logic system. that can then be used much as qualitative observations are [ Bradley and Easley, 1997 ] In this abstraction hierarchy, ODE rules that concern numeric information are considered concrete, whereas ODE rules that deal with qualitative information are considered abstract. Information between these two levels is also available. For example, if a physical system oscillates, the ....

[Article contains additional citation context not shown here]

E. Bradley and M. Easley. Reasoning about sensor data for automated system identification. In Proceedings of the Second International Symposium on Intelligent Data Analysis (IDA97) , 1997. London, U.K.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC