Sofiane Guessasma

COMBINATION OF INVERSE AND NEURAL NETWORK METHODS TO ESTIMATE HEAT FLUX

ABSTRACT This article deals with the estimation of the heat flux transmitted to a workpiece from a flame gun during the preheating process that is implemented quite often in thermal spraying. Internal temperature measurements permit determination of the temperature distribution in the workpiece. The heat flux is then determined by solving the inverse problem and the correlations among some processing parameters and the heat flux are recognized by implementing an artificial neural network.

Designing expert system using neural computation in view of the control of plasma spray processes

Abstract This paper aims at integrating the artificial intelligence methodologies in a quality control of ceramic coating fabrication using the atmospheric plasma spray (APS) process. In such a way, the average velocity, temperature and diameter of thermally sprayed Al 2 O 3 -13 wt.% TiO 2 particles before impinging the work piece and forming a deposit are monitored. Then, as these particle characteristics represent the most pertinent indicators of the coating properties and characteristics reproducibility, they are chosen as the output of an expert system based on neural computation. The model is built also considering at the system input the plasma and particle powder injection-processing parameters. After an optimisation procedure, the predicted results are compared to the results of experimental data resulting from a non-intrusive sensor conventionally used by industrials to control the coating quality. The good agreement found between these results permits to establish the overall effect of each processing parameter on the in-flight particle characteristics.

On the implementation of the fractal concept to quantify thermal spray deposit surface characteristics

A methodology is suggested to accurately describe the surface state of thermally sprayed coatings, which is mostly related to the process operating conditions and to the feedstock characteristics. Up to now, standard methodologies and their related descriptors are not sufficient to give simultaneously global and local information about deposit roughness. An approach is proposed, taking into account the multi-scale complexity of the profile. This approach is based on a description of the surface in terms of fractal geometry. Several aspects of the fractal methodology are discussed and some results related to the substrates and upper surface deposit profiles are presented. Comparisons with conventional methodologies do not permit a direct correlation: the fractal methodology proves to be sensitive to the measurement scale and to the calculation protocol.

Neural Computation to Predict In-Flight Particle Characteristic Dependences From Processing Parameters in the APS Process

In-flight particle sensors for thermal spraying are used for real-time monitoring of coating manufacture. However, such tools do not offer facilities to tune the processing parameters when the monitoring reveals fluctuations or instabilities in the thermal jet. To complete the process control, any diagnostic sensors need to be coupled with a predictive system to separate the effect of each processing parameter on the in-flight particle characteristics. In this work, a nonlinear dynamic system based on an artificial neural network (ANN) model is proposed to play this role. It consists of a method that relates the processing parameters to the particle emitted signal characteristics recorded with a DPV2000 (TECNAR Automation, St-Bruno, QC, Canada) optical sensing device. In such a way, a database was built to train and optimize an ANN structure. The in-flight particle average velocity, temperature, and diameter of an alumina-13wt.%titania feedstock were correlated to the injection and power parameters. Correlations are discussed on the basis of these predictive results.

On the Implementation of Neural Network Concept to Optimize Thermal Spray Deposition Process

Numerous processing parameters, up to fifty, characterize the plasma spray deposition process. A better quality control of the resulting deposits induces a better understanding of their effects on coating formation mechanisms. Numerical models can help to provide such an understanding. From a mathematical point of view, d.c. plasma spray deposition process is assimilated to a nonlinear problem in regards to its variables (operating parameters, environment, etc.). This paper develops a global approach based on an implicit describing of the mechanisms implementing Artificial Neural Networks (ANNs). The global concept and the protocols to implement are presented and developed for an example related to d.c. plasma spray process.

Neural Computation to Estimate Heat Flux in an Atmospheric Plasma Spray Process

Temperature is a key parameter in the thermal spray process and is a consequence of the heat flux experienced by the workpiece. This paper deals with the estimation of the heat flux transmitted to a workpiece from an atmospheric plasma spray torch during the preheating process often implemented in thermal spraying. An inverse heat conduction problem solution using a conjugate gradient method was considered to determine the heat flux starting from a known temperature distribution. Results from the later method were used to train an artificial neural network to discover correlations between selected processing parameters and heat flux.

Analysis of the influence of atmospheric plasma spray (APS) parameters on adhesion properties of alumina-titania coatings

Alumina-13 wt% titania wear resistant coatings were deposited using the Atmospheric Plasma Spray (APS) process under several processing conditions. Coating adhesion was then measured locally on cross sections by the indentation test and results were correlated with process variables. In order to identify the most influential factors on adhesion, artificial intelligence was used. The analysis was based on an Artificial Neural Network (ANN) taking into account training and test procedures to predict the dependences of measured property on experimental conditions. This study pointed out primarily that adhesion was largely sensitive to parameters that modified the in-flight particle characteristics (i.e. velocity and temperature). These effects were quantitatively demonstrated and predicted with an optimized neural network structure.