Abstract
The paper presents a successful attempt to create a neural model for a casting process. Neural network, following the training process, is able to predict selected effects of casting cooling (as a number of metal crystals in certain point of the casting), based on the data referring to the cooling method (the casting is cooled with water, the flow of which is controlled according to a set time algorithm). Various structures of neural networks used for solving the problem in question have been illustrated and the results obtained have been discussed. After creating several successful versions of a straight model, an attempt was made to set up a reverse model, it is such model in which expected control result is supplied to the input of the neural network (here, the number of crystals in a casting), and the network is expected to generate output signal stating how the cooling of the casting should be controlled to accomplish that objective. This has proven impossible, and the relevant reasons and circumstances have been presented.