Former investigations of thermal spraying processes have shown, that the relations between process parameters and the objective measurements are very complex. An approved approach to control complex processes is the usage of Neural Networks (NN).

Thus neural networks have been designed to control the process of APS and HVOF spraying. So called MLP (Multi Layer Perceptron) networks are used. They are able to control a process.

The way to train the Neural Network is to conduct as much experiments as possible. This is the major difficulty for the industrial use of NN. To save time and money DoE is used to create an optimal experimental plan for the training.

For testing the implementation of neural networks coatings were sprayed with both APS and HVOF (liquid fuel), using DoE for the experimental design. The NN were combined with the particle flux imaging tool.

In future the combination of the PFI with a Neural Network based control unit will be able to provide an open loop control for thermal spray processes. The NN will be integrated with the software of the PFI-unit in order to create an easy to handle and affordable process control device.

First experiments have been done with the APS process by spraying ZrO2 onto steel substrates. Afterwards the porosity of the coating was correlated to the recorded images and to the process parameters.

The presentation on the ITSC 2006 will show the way from the design and the training of neural networks to the industrial implementation and verification.