When large amounts of data are not available from the actual manufacturing process, simulated data can be substituted. Simulated data includes data prepared from statistical models (such as the normal distribution) and computational simulations (such as finite-element analysis methods). Networks trained using this type of data can be used to capture information from a process model and to replace the model equations. The trained network might then be used to estimate process parameters by conducting a grid search on the region of interest. There are advantages of using simulated data. First, simulated data can be noise-free. The random uncontrollable variation, which effects real process data, can be eliminated. Hence it should be easier to train the neural network with this data. Second, it is usually easier, cheaper and faster to train a network, and run an experiment, using simulated data. However, the simulated data may have a serious drawback, that is the statistical or computational models providing the data might not be a good fit to the real production process, which might lead to inaccurate results from the NN when operating with the real process data. As Table 1 shows, several researchers have combined simulated data with actual process data.