Search

Inverse Heat Transfer Problems (IHTP) are characterized by estimation of unknown quantities by utilizing any given information of the system. In this study, the inverse problem of estimation of boundary heat flux for a given temperature distribution on the walls of a two dimensional square cavity with a finite wall thickness is considered. A non-iterative method is applied utilizing Artificial Neural Network (ANN) and Principal Component Analysis (PCA) to estimate the parameters that define the boundary heat flux. The forward model is numerically solved with Fluent 6.3 for known values of a linearly varying boundary heat flux and the temperature distribution thus obtained is utilized to train the ANN for the inverse model. A parametric study is carried out to determine the effect of the thermal conductivity of the top and bottom walls on the flow and temperature distribution in the cavity. PCA analysis is carried out to reduce the dimensions of the input data set for the inverse model. These reduced dimensions are used to train the network and due to low dimensionality of the input, the effort required to train the network is considerably less. The trained networks are finally used to estimate boundary heat flux for any desired temperature distribution on the top and bottom walls. Additionally, covariance analysis is carried out in order to estimate the required number of temperatures during an experiment, on the top and bottom walls for the prediction of heat flux with a reasonable accuracy. The inverse model with covariance analysis is compared with the inverse model with PCA and both the methods are found to be equally potent. � 2010 Elsevier Ltd.

We investigate a semiparametric regression model where one gets noisy non linear non invertible functions of the observations. We focus on the application to bearings-only tracking. We first investigate the least squares estimator and prove its consistency and asymptotic normality under mild assumptions. We study the semiparametric likelihood process and prove local asymptotic normality of the model. This allows to define the efficient Fisher information as a lower bound for the asymptotic variance of regular estimators, and to prove that the parametric likelihood estimator is regular and asymptotically efficient. Simulations are presented to illustrate our results.

A robust throttle valve control has always been an attractive problem since throttle by wire systems were established in the mid-nineties. Often in control strategy, a feedforward controller is adopted in which an inverse model is used. Mathematical inversions of models imply a high order of differentiation of the state variables and consequently noise effects. In general, neural networks are a very effective and popular tool mostly used for modeling. The inversion of a neural network produces real possibilities to involve the networks in the control problem schemes. This paper presents a control strategy based upon an inversion of a feed forward trained local linear model tree. The local linear model tree is realized through a fuzzy neural network. Simulated results from real data measurements are presented in which two control loops are explicitly compared. A robust throttle valve control has always been an attractive problem since throttle by wire systems were established in the mid-nineties. Often in control strategy, a feedforward controller is adopted in which an inverse model is used. Mathematical inversions of models imply a high order of differentiation of the state variables and consequently noise effects. In general, neural networks are a very effective and popular tool mostly used for modeling. The inversion of a neural network produces real possibilities to involve the networks in the control problem schemes. This paper presents a control strategy based upon an inversion of a feed forward trained local linear model tree. The local linear model tree is realized through a fuzzy neural network. Simulated results from real data measurements are presented in which two control loops are explicitly compared.

A present study addresses some cornerstones in designing high performance optical instruments. Very often the practical work starts via feasibility studies on whether something can be measured or not. Either lab or process instruments are used to find response in a given spectral range. Other specifications address signal-to-noise ratio, sensitivity, selectivity, application-related linearity as well as instrument related photometric linearity, accuracy, and repeatability. Much of the work focuses on establishing a correlation between primary lab method and secondary spectral method. Less attention is often paid to scattering and absorption properties of samples. At some later point in application development, all method developers pay attention to these issues at latest when calibrating spectral instruments with chemometric tools, either knowingly or unknowingly. In many industries, both near and mid infrared spectroscopy rely on sample chemistry and specific absorptions while down-weighting the physical interferences caused by packing density variations, particle size distribution differences, surface topography variations, and/or other optical path length variations. Chemometric tools such as standard normal variate (SNV), spectral interference subtraction (SIS), inverted signal correction (ISC), multiplicative signal correction (MSC), and their extensions (EISC and EMSC) correct for these physical interferences and greatly simplify calibrations. While these methods solve the problem for calibration, there are areas where this is not enough, namely, the design processes of optical instruments. This work illustrates some tools developed and used routinely in VTT Optical Instrumentation Center prior to designing new optical instruments. Particularly, these tools address the estimation of optical properties of sample materials representative to real processes. This work illustrates how experimental data can be collected using a specially designed hardware accessory and high performance double-beam lab UV-vis-NIR spectrometer and how tailored software tools can be used to study these data and compute optical properties of sample materials. Though these optical properties may not be exact estimates, they are very important and accurate enough for the design of tailored optical instruments for performance-critical customer applications. The study shown here presents results from near infrared (NIR) measurement made on industrial applications. The results and methods shown here are directly applicable both for studying sample properties and for designing appropriate sampling optics for existing commercial or tailor-made optical instruments whether the working range is in ultraviolet (UV), visible, fluorescence, Raman, NIR, or mid infrared (MIR). This study shows how the above methods can bring along significant benefits in the design phase of optical instruments. To conclude this presentation, we present also general guidelines how to design a high performance optical instrument

People with severe motor disabilities (spinal cord injury (SCI), amyotrophic lateral sclerosis (ALS), etc.) but with intact brain functions are somehow prisoners of their own body. They need alternative ways of communication and control to interact with their environment in their everyday life. These new tools are supposed to increase their quality of life by giving these people the opportunity to recover part of their independence. Therefore, these alternative ways have to be reliable and ergonomic to be successfully used by disabled people. Over the past two decades, numerous studies proposed electroencephalogram (EEG) activity for direct brain-computer interaction. EEG-based brain-computer interfaces (BCIs) provide disabled people with new tools for control and communication and are promising alternatives to invasive methods. However, as any other interaction modality based on physiological signals and body channels (muscular activity, speech and gestures, etc.), BCIs are prone to errors in the recognition of subject's intent, and those errors can be frequent. Indeed, even well-trained subjects rarely reach 100% of success. In contrast to other interaction modalities, a unique feature of the brain channel is that it conveys both information from which we can derive mental control commands to operate a brain-actuated device as well as information about cognitive states that are crucial for a purposeful interaction, all this on the millisecond range. One of these states is the awareness of erroneous responses, which a number of groups have recently proposed as a way to improve the performance of BCIs. However, most of these studies propose the use of error-related potentials (ErrP) following an error made by the subject himself. This thesis first describes a new kind of ErrP, the so-called interaction ErrP, that are present in the ongoing EEG following an error of the interface and no longer errors of the subject himself. More importantly, these ErrP are satisfactorily detected no more in grand averages but at the level of single trials. Indeed, the classification rates of both error and correct single trials based on error-potentials detection are on average 80%. At this level it becomes possible to introduce a kind of automatic verification procedure in the BCI: after translating the subject's intention into a control command, the BCI provides a feedback of that command, but will not transfer it to the device if ErrP follow the feedback. Experimental results presented in this thesis confirm that this new protocol greatly increases the reliability of the BCI. Furthermore, this tool turns out to be of great benefit especially for beginners who normally reach moderate performances. Indeed, filtering out wrong responses increases the user's confidence in the interface and thus accelerates mastering the control of the brain-actuated device. The second issue explored in this thesis is the practical integration of ErrP detection in a BCI. Indeed, providing a first feedback of the subject's intent, as recognized by the BCI, before eventually sending the command to the controlled device, induces additional information to process by the subject and may considerably slow down the interaction since the introduction of an automatic response rejection strongly interferes with the BCI. However, this study shows the feasibility of simultaneously and satisfactorily detecting erroneous responses of the interface and classifying motor imagination for device control at the level of single trials. The integration of an automatic error detection procedure leads to great improvements of the BCI performance. Another aspect of this thesis is to investigate the potential benefit of using neurocognitive knowledge to increase the classification rate of ErrP, and more generally the performance of the BCI. Recent findings have uncovered that ErrP are most probably generated in a deep fronto-central brain area called anterior cingulate cortex (ACC). This hypothesis is verified using a well-known inverse model called sLORETA. Indeed, the localization provided for ErrP shows clear foci of activity both in the ACC and the pre-supplementary motor area (pre-SMA). The localization results using the cortical current density (CCD) model are very similar and more importantly, this model outperforms EEG for ErrP classification. Thanks to its stability, this model is likely to be successfully used in a BCI framework. The ELECTRA model for estimating local field potentials is also tested, but classification and localization results using this method are not so encouraging. More generally, the work described here suggests that it could be possible to recognize in real time high-level cognitive and emotional states from EEG (as opposed, and in addition, to motor commands) such as alarm, fatigue, frustration, confusion, or attention that are crucial for an effective and purposeful interaction. Indeed, the rapid recognition of these states will lead to truly adaptive interfaces that customize dynamically in response to changes of the cognitive and emotional/affective states of the user.

A present study addresses some cornerstones in designing high performance optical instruments. Very often the practical work starts via feasibility studies on whether something can be measured or not. Either lab or process instruments are used to find response in a given spectral range. Other specifications address signal-to-noise ratio, sensitivity, selectivity, application-related linearity as well as instrument related photometric linearity, accuracy, and repeatability. Much of the work focuses on establishing a correlation between primary lab method and secondary spectral method. Less attention is often paid to scattering and absorption properties of samples. At some later point in application development, all method developers pay attention to these issues at latest when calibrating spectral instruments with chemometric tools, either knowingly or unknowingly. In many industries, both near and mid infrared spectroscopy rely on sample chemistry and specific absorptions while down-weighting the physical interferences caused by packing density variations, particle size distribution differences, surface topography variations, and/or other optical path length variations. Chemometric tools such as standard normal variate (SNV), spectral interference subtraction (SIS), inverted signal correction (ISC), multiplicative signal correction (MSC), and their extensions (EISC and EMSC) correct for these physical interferences and greatly simplify calibrations. While these methods solve the problem for calibration, there are areas where this is not enough, namely, the design processes of optical instruments. This work illustrates some tools developed and used routinely in VTT Optical Instrumentation Center prior to designing new optical instruments. Particularly, these tools address the estimation of optical properties of sample materials representative to real processes. This work illustrates how experimental data can be collected using a specially designed hardware accessory and high performance double-beam lab UV-vis-NIR spectrometer and how tailored software tools can be used to study these data and compute optical properties of sample materials. Though these optical properties may not be exact estimates, they are very important and accurate enough for the design of tailored optical instruments for performance-critical customer applications. The study shown here presents results from near infrared (NIR) measurement made on industrial applications. The results and methods shown here are directly applicable both for studying sample properties and for designing appropriate sampling optics for existing commercial or tailor-made optical instruments whether the working range is in ultraviolet (UV), visible, fluorescence, Raman, NIR, or mid infrared (MIR). This study shows how the above methods can bring along significant benefits in the design phase of optical instruments. To conclude this presentation, we present also general guidelines how to design a high performance optical instrument

A general technique to design advanced controllers for non-linear systems is described, using component oriented modeling and symbolic algorithms as used for Modelica models. Starting point are linear design techniques that use linear inverse models as a core part of the controller structure. Starting from such a structure, the approach is to replace the linear inverse model with a nonlinear one, resulting in a controller that is applicable over the full operating range of the (nonlinear) plant. It is shown that nonlinear inverse models may be automatically generated from the plant model in Modelica.

In this paper a model component library for developing multi-disciplinary aircraft flight dynamics models is presented, named FlightDynLib. This library is based on the object-oriented modelling language Modelica that has been designed for modelling of large scale multi-physics systems. The flight dynamics library allows for graphical construction of comlex rigid as well as flexible aircraft dynamics models and is fully compatible with other available libraries for electronics, thermodynamics, control systems, etc.

Inverse models to determine the permeability are generally based on existing forward models for the pressure. The permeabilities are adapted in such a way that the calculated pressures match the specified pressures in a number of points. To assimilate a priori knowledge about the flux, we introduce the 'flux assimilation method', which is based on the 'vector potential-pressure formulation' of Darcy's law. Thanks to an unconventional discretization technique - the 'edge-based face element method' - not only the specified pressures, but also specified information about the flux density can easily be assimilated. A relatively simple, but insightful analytical example illustrates the potential of this method. © 2004 Kluwer Academic Publishers.

Given a random vector X, we address the question of linear separability of X, that is, the task of finding a linear operator W such that we have (S1,…,SM)=(WX) with statistically independent random vectors Si. As this requirement alone is already fulfilled trivially by X being independent of the empty rest, we require that the components be not further decomposable. We show that if X has finite covariance, such a representation is unique up to trivial indeterminacies. We propose an algorithm based on this proof and demonstrate its applicability. Related algorithms, however with fixed dimensionality of the subspaces, have already been successfully employed in biomedical applications, such as separation of fMRI recorded data. Based on the presented uniqueness result, it is now clear that also subspace dimensions can be determined in a unique and therefore meaningful fashion, which shows the advantages of independent subspace analysis in contrast to methods like principal component analysis.

A new procedure to formulate nonlinear empirical models of a dynamical system is presented. This nonlinear modeling technique generalizes the Markovian techniques used to build linear empirical models, but incorporates a quadratic nonlinearity. The model fit is accomplished using a genetic algorithm.The nonlinear empirical model is applied to two low order model test cases demonstrating different forms of nonlinearity. The two equation predator/prey model (Lotka-Volterra equations) is modeled in the regime of a stable limit cycle. The nonlinear empirical model is able to capture the general shape of the limit cycle, but does not display the long time stability. The second example is the three dimensional Lorenz system forced in the chaotic regime. The general shape and location in phase space of the chaotic attractor is reproduced by the nonlinear empirical model.The results presented here demonstrate that nonlinear empirical models may be able to reproduce some of the nonlinear behaviors of dynamical systems.