Showing papers in "Neural Computing and Applications in 2000"
TL;DR: A non-linear model for an overhead crane system is derived which takes into account a combination of a trolley and a pendulum, and preliminary results are very encouraging, and indicate the feasibility of such a two rule base control strategy.
Abstract: A non-linear model for an overhead crane system is derived which takes into account a combination of a trolley and a pendulum. The overall mathematical model obtained is simulated using MATLAB-SIMULINK. Open-loop simulations run on cases depending on whether the air resistance is taken into account or not, and whether the angle of oscillation is smail or large, indicate the validity of such model, hence reflecting similar trends in industries which are concerned with material handling equipment. A hand-crafted fuzzy controller, which includes two rule bases, one for position control, the other for sway-angle control, was designed and successfully implemented on the above simulated model. Preliminary results are very encouraging, and indicate the feasibility of such a two rule base control strategy. The results obtained are presented, analysed and discussed.
86 citations
TL;DR: Differences in architecture, training parameter values and subsets of the data all deliver much the same impact rankings, which supports the notion that the technique ranks an inherent property of the available data rather than a property of any particular feedforward neural network.
Abstract: For a variety of reasons, the relative impacts of neural-net inputs on the output of a network's computation is valuable information to obtain. In particular, it is desirable to identify the significant features, or inputs, of a data-defined problem before the data is sufficiently preprocessed to enable high performance neural-net training. We have defined and tested a technique for assessing such input impacts, which will be compared with a method described in a paper published earlier in this journal. The new approach, known as the 'clamping' technique, offers efficient impact assessment of the input features of the problem. Results of the clamping technique prove to be robust under a variety of different network configurations. Differences in architecture, training parameter values and subsets of the data all deliver much the same impact rankings, which supports the notion that the technique ranks an inherent property of the available data rather than a property of any particular feedforward neural network. The success, stability and efficiency of the clamping technique are shown to hold for a number of different real-world problems. In addition, we subject the previously published technique, which we will call the 'weight product' technique, to the same tests in order to provide directly comparable information.
72 citations
TL;DR: The paper derives the necessary conditions for the assignability of eigenvalues to a region in the s-plane and the required conditions to guarantee the stability of adaptive fuzzy models via the Linear Matrix Inequalities (LMI) method.
Abstract: This paper demonstrates the application of a new fault-tolerant control scheme for a rail vehicle traction system using digital signal processing hardware and a representative induction motor testrig. The approach presented takes into account the stability and design of non-linear fuzzy inference systems based on Takagi-Sugeno (T-S) fuzzy models. The paper derives the necessary conditions for the assignability of eigenvalues to a region in the s-plane and the necessary conditions to guarantee the stability of adaptive fuzzy models. The problem is solved via the Linear Matrix Inequalities (LMI) method.
53 citations
TL;DR: A multi-net fault diagnosis system designed to provide an early warning of combustion-related faults in a diesel engine is presented and is shown to be effective when compared with the performance of the component nets from which it was assembled.
Abstract: A multi-net fault diagnosis system designed to provide an early warning of combustion-related faults in a diesel engine is presented. Two faults (a leaking exhaust valve and a leaking fuel injector nozzle) were physically induced (at separate times) in the engine. A pressure transducer was used to sense the in-cylinder pressure changes during engine cycles under both of these conditions, and during normal operation. Data corresponding to these measurements were used to train artificial neural nets to recognise the faults, and to discriminate between them and normal operation. Individually trained nets, some of which were trained on subtasks, were combined to form a multi-net system. The multi-net system is shown to be effective when compared with the performance of the component nets from which it was assembled. The system is also shown to outperform a decision-tree algorithm (C5.0), and a human expert; comparisons which show the complexity of the required discrimination. The results illustrate the improvements in performance that can come about from the effective use of both problem decomposition and redundancy in the construction of multi-net systems.
53 citations
TL;DR: An overview of the neural networks approach to user modelling and intelligent interface is presented and activities of various neural networks models are introduced to illustrate how user modelling problems can be solved by neural networks.
Abstract: This article presents an overview of the neural networks approach to user modelling and intelligent interface. We analyse and discuss activities in user modelling and intelligent interface. Activities of various neural networks models are introduced to illustrate how user modelling problems can be solved by neural networks. The practical utility of neural networks in supporting user modelling and intelligent interface is demonstrated by reviewing a selection of neural networks developed in this area. Structured summaries are provided for comparative purpose.
37 citations
TL;DR: It is argued that rule weights have a negative effect on the linguistic interpretation of a fuzzy system, and thus remove one of the key advantages for applying fuzzy systems, and suggested that neuro-fuzzy learning should be better implemented by algorithms that modify the fuzzy sets directly without using rule weights.
Abstract: Neuro-fuzzy systems have recently gained a lot of interest in research and application. They are approaches that use learning techniques derived from neural networks to learn fuzzy systems from data. A very simple ad hoc approach to apply a learning algorithm to a fuzzy system is to use adaptive rule weights. In this paper, we argue that rule weights have a negative effect on the linguistic interpretation of a fuzzy system, and thus remove one of the key advantages for applying fuzzy systems. We show how rule weights can be equivalently replaced by modifying the fuzzy sets of a fuzzy system. If this is done, the actual effects that rule weights have on a fuzzy rule base become visible. We demonstrate at a simple example the problems of using rule weights. We suggest that neuro-fuzzy learning should be better implemented by algorithms that modify the fuzzy sets directly without using rule weights.
36 citations
TL;DR: A fast, efficient, and powerful non-linear input selection procedure using a combination of probabilistic neural networks and repeated bitwise gradient descent with resampling is discussed.
Abstract: Selection of input variables is a key stage in building
predictive models, and an important form of data mining. As exhaustive evaluation of potential input sets using full non-linear models is impractical, it is necessary to use simple fast-evaluating models and heuristic selection strategies. This paper discusses a fast, efficient, and powerful nonlinear input selection procedure using a combination of Probabilistic Neural Networks and repeated
bitwise gradient descent. The algorithm is compared
with forward elimination, backward elimination and genetic algorithms using a selection of real-world data sets. The algorithm has comparative performance and greatly reduced execution time with respect to these alternative approaches. It is demonstrated empirically that reliable results cannot be gained using any of these approaches without the use of resampling.
33 citations
TL;DR: COSIMIR (Cognitive Similarity Learning in Information Retrieval), an innovative model integrating human knowledge into the core of the retrieval process, is presented and applies backpropagation to information retrieval.
Abstract: Neural networks can learn from human decisions and preferences. Especially in human-computer interaction, adaptation to the behaviour and expectations of the user is necessary. In information retrieval, an important area within human-computer interaction, expectations are difficult to meet. The inherently vague nature of information retrieval has led to the application of vague processing techniques. Neural networks seem to have great potential to model the cognitive processes involved more appropriately. Current models based on neural networks and their implications for human-computer interaction are analysed. COSIMIR (Cognitive Similarity Learning in Information Retrieval), an innovative model integrating human knowledge into the core of the retrieval process, is presented. It applies backpropagation to information retrieval, integrating human-centred and soft and tolerant computing into the core of the retrieval process. A further backpropagation model, the transformation network for heterogeneous data sources, is discussed. Empirical evaluations have provided promising results.
24 citations
TL;DR: This paper discusses an approach of establishing system models of users’ task related characteristics, such as domain knowledge in human-computer interaction, that can be expected to overcome some limitations of user modelling approaches in terms of pattern recognition and classification.
Abstract: This paper discusses an approach of establishing system models of users' task related characteristics, such as domain knowledge in human-computer interaction. Several neural networks are tested for the modelling process. These networks function as associative memories that capture the causal relationships among assumptions about the users' characteristics. The outputs from the networks are considered as stereotypes assigned to individual users. It is suggested that this approach can be expected to overcome some limitations of user modelling approaches in terms of pattern recognition and classification.
24 citations
TL;DR: A control strategy that enhances a fuzzy controller with self-learning capability for achieving the control of a binary methanol-propanol distillation column using an Adaptive-Network-based Fuzzy Inference System extended to cope with multivarible systems is used.
Abstract: In this paper we use a control strategy that enhances a fuzzy controller with self-learning capability for achieving the control of a binary methanol-propanol distillation column. An Adaptive-Network-based Fuzzy Inference System (ANFIS) architecture extended to cope with multivariable systems has been used. This allows the tuning of parameters both of the membership functions and the consequents in a Sugeno-type inference system. To satisfy the control objectives the backpropagation gradient descent through the plant method is applied, hence identification of the plant dynamics is also needed. The performance of the resulting neuro-fuzzy controller under different reference settings for the concentration of methanol demonstrates the stabilisation of the concentration profiles in the column, leading to an effective methanol composition control.
23 citations
TL;DR: A neural networks based method and a system for colour measurements on printed halftone multicoloured pictures and halft one multi-coloured bars in newspapers witnesses its usefulness through the improved quality of multicolours pictures, the reduced consumption of inks and, therefore, less severe problems of smearing and printing through.
Abstract: This paper presents a neural networks based method and a system for colour measurements on printed halftone multicoloured pictures and halftone multicoloured bars in newspapers. The measured values ...
TL;DR: A recurrent neuro-fuzzy network-based nonlinear long range model predictive control strategy that avoids the time consuming numerical optimisation procedure, and the uncertainty in convergence to the global optimum which are typically seen in conventional nonlinear model-based predictive control strategies.
Abstract: A recurrent neuro-fuzzy network-based nonlinear long range model predictive control strategy is proposed in this paper. The process operation is partitioned into several fuzzy operating regions. Within each region, a local linear model is used to model the process. The global model output is obtained through the centre of gravity defuzzification. Based upon a neuro-fuzzy network model, a nonlinear model-based predictive controller can be developed by combining several local linear model-based predictive controllers which usually have analytical solutions. This strategy avoids the time consuming numerical optimisation procedure, and the uncertainty in convergence to the global optimum which are typically seen in conventional nonlinear modelbased predictive control strategies, Furthermore, control actions obtained based on local incremental models contain integration actions which can naturally eliminate static control offsets. The technique is demonstrated by an application to the modelling and control of liquid level in a water tank.
TL;DR: This paper presents an approach to determining the colours of specks in an image of a pulp being recycled through colour classification by an artificial neural network trained using fuzzy possibilistic target values.
Abstract: This paper presents an approach to determining the colours of specks in an image of a pulp being recycled. The task is solved through colour classification by an artificial neural network. The network is trained using fuzzy possibilistic target values. The number of colour classes found in the images is determined through the self-organising process in the two-dimensional self-organising map. The experiments performed have shown that the colour classification results correspond well with human perception of the colours of the specks.
TL;DR: A multi-layered perceptron designed to classify the two types of circulations – nontor-nadic and tornadic – based on various attributes of the circulations is developed.
Abstract: There exist radar-based algorithms designed to detect circulations in the atmosphere. Not all detected circulations, however, are associated with tornados on the ground. Outlined herein is the development of a multi-layered perceptron designed to classify the two types of circulations - nontornadic and tornadic - based on various attributes of the circulations. Special emphasis is placed on the role of local minima in determining the optimal architecture via bootstrapping, and on the performance of the network in terms of probabilistic measures.
TL;DR: It is shown that a modified SOM can be used to approximately Input/Output (I/O) linearise and to control nonlinear systems using a combination of the SOM learning algorithm, and a biologically inspired optimisation algorithm known as chemotaxis.
Abstract: Two applications of Self Organising Map (SOM) networks in the context of nonlinear control are introduced, one in approximate feedback linearisation and the second in optimal control. It is shown that a modified SOM can be used to approximately Input/Output (I/O) linearise and to control nonlinear systems using a combination of the SOM learning algorithm, and a biologically inspired optimisation algorithm known as chemotaxis. A proof to guarantee the stability of the closed loop during the training of the network and the operation of the whole system is included. The results are illustrated with simulations of a single link manipulator.
TL;DR: Experiments show that the neural networks can be applied to the quality testing problem in the semiconductor industry efficiently and the performance of ART algorithms is better than that of the other architectures.
Abstract: In this paper, we present the application of supervised neural algorithms based on Adaptive Resonance Theory (Fuzzy ARTMAP, ART-EMAP and distributed ARTMAP), as well as some feedforward networks (counter-propagation, backpropagation, Radial Basis Function algorithm) to the quality testing problem in the semiconductor industry. The aim is to recognise and classify deviations in the results of functional and Process-Control-Monitoring (PCM) tests of chips as soon as they are available so that technological corrections can be implemented more quickly. This goal can be divided in two tasks that are treated in this paper: the classification of faulty wafers on the basis of topological information extracted from functional tests; and forecasting the yield of chips using the results of PCM tests. Experiments show that the neural networks can be applied to this problem efficiently, and the performance of ART algorithms is better than that of the other architectures.
TL;DR: The preliminary results on the use of neural networks to forecast SO2 concentration levels in the industrial area of Ravenna are presented, which show the high levels of SO2 occurring during relatively rare episodes.
Abstract: 2 concentration levels in the industrial area of Ravenna. Ground level concentrations of pollutants were analysed in the area of Ravenna, in particular the high levels of SO2 occurring during relatively rare episodes. These events are typically correlated with many different aspects, like complex local meteorology, topography, and industrial emissions parameters. In many cases, during these episodes, the deterministic models (e.g. Gaussian models) fail to explain the high ground level concentrations. The neural networks are trained with a Bayesian learning scheme.
TL;DR: The problem of choosing the number of hidden neurons and the numberof taps and delays in the FIR and IIR network synapses is formalised as an optimisation problem, whose cost function to be minimised is the network error calculated on a validation data set.
Abstract: A general purpose implementation of the tabu search metaheuristic, called Universal Tabu Search, is used to optimally design a locally recurrent neural network architecture. The design of a neural network is a tedious and time consuming trial and error operation that leads to structures whose optimality is not guaranteed. In this paper, the problem of choosing the number of hidden neurons and the number of taps and delays in the FIR and IIR network synapses is formalised as an optimisation problem, whose cost function to be minimised is the network error calculated on a validation data set. The performance of the proposed approach has been tested on the problem of modelling the dynamics of a non-isothermal, continuously stirred tank reactor, in two different operating conditions: when a first order exothermic reaction is occurring; and when two consecutive first order reactions lead to a chaotic behaviour. Comparisons with alternative neural approaches are reported, showing the usefulness of the proposed method.
TL;DR: A simple model could be identified which, together with a valve model, can be further applied for the purposes of accurate flow control and linear dynamics and non-linearities due to the pump, valve and the pulp consistency are identified.
Abstract: Pulp and paper mills can be seen as big pumping plants, where mass is pumped from one step to another. The proper operation of the process in its different stages does not allow large deviations from given operation conditions, which makes it essential to monitor and control the flow rate and the consistency of the pulp. A structure is suggested for the modelling of pulp flow rate, and possibilities for using the pump-valve system as a flow meter are examined. The overall model structure consists of a Wiener model for pressure difference, a non-linear dynamic valve model, and a static mapping for flow rate. A full-scale pilot plant of pulp flow through a valve in a pipeline is used for experimentation. Linear dynamics and non-linearities due to the pump, valve and the pulp consistency are identified based on online measurements obtained from calibration tests. The results show that a simple model could be identified which, together with a valve model, can be further applied for the purposes of accurate flow control.
TL;DR: Three ANN methods were applied to the helicopter low airspeed problem: a linear network, a Radial Basis Function network, and a Multi-Layer Perceptron (MLP), trained using an implementation of the Levenberg–Marquardt (L–M) algorithm.
Abstract: A helicopter's airspeed and sideslip angle is difficult to measure at speeds below 50 knots. This paper describes the application of Artificial Neural Network (ANN) techniques to the helicopter low airspeed problem. Three ANN methods were applied to the problem: a linear network, a Radial Basis Function (RBF) network, and a Multi-Layer Perceptron (MLP), trained using an implementation of the Levenberg-Marquardt (L-M) algorithm. Internally available measurements, such as control positions and body attitudes and rates, were generated using a realistic simulation model of a Lynx helicopter. These measurements formed the inputs to the ANN methods. The MLP was found to be the superior method. Further testing, including a Taguchi analysis, indicated the validity of the method. It is concluded that ANN techniques present a promising solution to the helicopter low airspeed problem.
TL;DR: The results of preliminary evaluation showed that NSSOM is capable of enhancing precision without sacrificing recall, and a user-friendly browsing facility has been developed which helps users predict the desired components by providing an intelligible search space.
Abstract: This paper presents an approach to self-structuring software libraries. The authors developed a representation scheme to construct a feature space over a collection of software assets. The feature space is represented and classified by a variety of the self-organising map, called the Nested Software Self-Organising Map (NSSOM), consisting of a top map and a set of sub-maps nested in the top map. The clustering on the top map provides general improvements in retrieval recall, while the lower-level nested maps further elaborate the clusters into more specific groups enhancing retrieval precision. The results of preliminary evaluation showed that NSSOM is capable of enhancing precision without sacrificing recall. In addition, a user-friendly browsing facility has also been developed which helps users predict the desired components by providing an intelligible search space. The present approach attempts to achieve an optimal combination of efficiency, accuracy and user-friendliness, which is not offered by the existing software retrieval systems.
TL;DR: This paper deals with the potential application of neural networks to the multivariable control of a solvent extraction pilot plant, which presents a non-linear behaviour and time-varying dynamics.
Abstract: Modelling and control of chemical process systems are usual applications of artificial neural networks that have been explored so far with success. This paper deals with the potential application of neural networks to the multivariable control of a solvent extraction pilot plant. The pilot plant to be controlled is a pulsed liquid-liquid extraction column, which presents a non-linear behaviour and time-varying dynamics. Previous works have shown that the column could be maintained in its optimal behaviour by means of the control of conductivity by action on the pulse frequency. A given product specification can be obtained by the control of the product concentration in the outlet stream by acting on the solvent feed-flow rate. Owing to interactions between one variable and the other, a two input-two output control scheme has been developed and implemented. Promising experimental results have been obtained by using neural networks as an alternative tool for online control of chemical plant with dynamic changes.
TL;DR: The proposed technique uses Incremental Tree Induction (ITI) as the learning algorithm, and incorporates fuzzy logic to deal with uncertainties apparent in the process to be controlled, in contrast to the manual implementation of traditional FAMs.
Abstract: This paper proposes a new multi-strategy (hybrid) intelligent control technique whose concept is applicable to the control of a wide range of processes The proposed technique uses Incremental Tree Induction (ITI) as the learning algorithm, and incor porates fuzzy logic to deal with uncertainties apparent in the process to be controlled. ITI operates solely on symbolic fuzzy knowledge, as both the input features (data obtained from the process to be controlled) and the output decisions of the intelligent controller are described by fuzzy linguistic variables. Fuzzy associative memories (FAMs) are employed to store and manage the fuzzy knowledge, and are simulated operationally by fuzzy binary decision trees which encode the input-output space. The main novelty of this approach is the automatic synthesis (both off-line and on-line) of multi-dimensional FAMs from inception, in contrast to the manual implementation of traditional FAMs. A second important advantage is the self-explanatory nature of the FAMs (and their underlying control laws) generated by this approach. The new technique is demonstrated in its application to the intelligent navigation of a mobile robot.
TL;DR: A novel adaptive EC algorithm is proposed to conceal the error for block-based image coding systems by using neural network techniques in the spatial domain and results show that the visual quality and the PSNR evaluation of a reconstructed image are significantly improved using the proposed EC algorithm.
Abstract: Image coding algorithms such as Vector Quantisation (VQ), JPEG and MPEG have been widely used for encoding image and video, These compression systems utilise block-based coding techniques to achieve a higher compression ratio. However, a cell loss or a random bit error during network transmission will permeate into the whole block, and then generate several damaged blocks. Therefore, an efficient Error Concealment (EC) scheme is essential for diminishing the impact of damaged blocks in a compressed image. In this paper, a novel adaptive EC algorithm is proposed to conceal the error for block-based image coding systems by using neural network techniques in the spatial domain. In the proposed algorithm, only the intra-frame information is used for reconstructing the image with damaged blocks. The information of pixels surrounding a damaged block is used to recover the errors using the neural network models. Computer simulation results show that the visual quality and the PSNR evaluation of a reconstructed image are significantly improved using the proposed EC algorithm.
TL;DR: A new technique of data coding and an associated set of homogenous processing tools for the development of Human Computer Interactions (HCI) facilitates the fusion of different sensorial modalities and simplifies the implementations.
Abstract: This paper presents a new technique of data coding and an associated set of homogenous processing tools for the development of Human Computer Interactions (HCI). The proposed technique facilitates the fusion of different sensorial modalities and simplifies the implementations. The coding takes into account the spatio-temporal nature of the signals to be processed in the framework of a sparse representation of data. Neural networks adapted to such a representation of data are proposed to perform the recognition tasks. Their development is illustrated by two examples: one of on-line handwritten character recognition; and the other of visual speech recognition.
TL;DR: This improvement is another step in a deep study of the time-changing conditions of the heat transmission due to ageing, but it will also make possible the introduction of new parameters in the physical formulation whose relation is unknown.
Abstract: The process of electricity generation in a power plant involves several phases for heat exchange generally between hot fluid (water, fumes or steam) and one of the input fluids (cool water, cook oven gas, fuel, blast furnace gas, etc.). The process of exchange evolves in time at the same time as installation itself. This work presents a study to create a model for a regenerative rotative heat exchanger that considers the effects of the time dependent variations. This improvement is another step in a deep study of the time-changing conditions of the heat transmission due to ageing, but it will also make possible the introduction of new parameters in the physical formulation whose relation is unknown. Modelling is done using feedforward neural networks after a very extensive pre-processing step.
TL;DR: A fuzzy rule-based hybrid controller is proposed in this paper, which incorporates conventional linear voltage control along with the fuzzyRule-based supplementary power damping control to form a unified global controller for Statcom.
Abstract: Static synchronous compensator (Statcom) is a powerful new device for power systems, which can be used for various purposes. The multi-objective demands are quite different in nature, e.g. continuous linear control for voltage maintaining, and discrete bang-bang control for oscillation damping. Unfortunately, they often conflicr with each other. In this respect, a supplementary damping control together with an independent voltage control is normally used. However, inevitable small disturbance and uncertainties will cause problems in the coordination of the two functions. To overcome such difficulties, a fuzzy rule-based hybrid controller is proposed in this paper, which incorporates conventional linear voltage control along with the fuzzy rule based supplementary power damping control to form a unified global controller for Statcom. Because only simple fuzzy rules and a few input signals are involved, it is very easy to implement in a practical power system. The simulation performed on a Sin gle-Machine-Infinite-Bus (SMIB) power system and an actual large power system demonstrates the effectiveness of the proposed approach.
TL;DR: A three-layer hierarchical clustering neural network is developed to build fuzzy rule-based models from numerical data and both structure identification and parameter optimisation of the fuzzy model can be carried out automatically.
Abstract: A simple and effective fuzzy modelling approach is presented in this paper. A three-layer hierarchical clustering neural network is developed to build fuzzy rule-based models from numerical data. Differing from existing clustering-based methods, in this approach the structure identification of the fuzzy model is implemented on the basis of a class of subclusters created by a self-organising network instead of on raw data. By combined use of unsupervised and supervised learning, both structure identification and parameter optimisation of the fuzzy model can be carried out automatically. The simulation results show that the proposed method can provide good model structure for fuzzy modelling and has high computing efficiency.
TL;DR: The concept of orthogonal fuzzy rule-based systems are proposed as a judgment as to whether the optimal rules are selected and an illustrative example to create a model for the solder paste printing stage of surface mount tech-nology.
Abstract: In this paper, the concept of orthogonal fuzzy rule-based systems is introduced Orthogonal rules are an extension to the definition of orthogonal vectors when the vectors are vectors of membership func tions in the antecedent part of rules The number and combination of rules in a fuzzy rule-based system will be optimised by applying orthogonal rules The number of rules, and subsequently the complexity of the fuzzy rule-based systems, are directly associated with the number of input variables and distinguishable membership functions for each individual input variable A subset of rules can be used if it is known which subset provides closer behaviour to the case when all rules are used Orthogonal fuzzy rule-based systems are proposed as a fudgment as to whether the optimal rules are selected The application of orthogonal fuzzy rules becomes essential when fuzzy rule-based systems containing many inputs are used An illustrative example is presented to create a model for the solder paste printing stage of surface mount technology
TL;DR: A hybrid networks model of the reactor is presented and global software is elaborated to achieve film thickness control in an experimental LPCVD reactor pilot plant, in order to get a defined and uniform deposition thickness on the wafers all along the reactor.
Abstract: In this paper, a new approach of LPCVD reactor modelling and control is presented, based on the use of neural networks. We first present the development of a hybrid networks model of the reactor. The objective is to provide a simulation model which can be used to compute online the film thickness on each wafer. In the second section, the thermal control of a LPCVD reactor is studied. The objective is to develop a multivariable controller to control a space- and time-varying temperature profile inside the reactor. A neural network is designed using a methodology based on process inverse dynamics modelling. Good control results have been obtained when tracking space-time temperature profiles inside the LPCVD reactor pilot plant. Finally, global software is elaborated to achieve film thickness control in an experimental LPCVD reactor pilot plant, in order to get a defined and uniform deposition thickness on the wafers all along the reactor. Experimental results are presented which confirm the efficiency of the optimal control strategy.