ABSTRACT
In engineering applications, the data for training of neural networks may come from many sources, including experiments, field measurements, recordings during system operations, or computational simulations. In all these types of applications, the data are directly available for training of the neural networks model of the system itself. If we are interested in modeling of the components of the system with neural networks, and the only data available are the stimulus/response (input/output) of the system, then we need to solve an inverse problem. As discussed in the previous chapter, this is a type-two inverse problem; the input and output of the system are known and the model of the system, or model of its components, need to be determined. Autoprogressive Algorithm, also called Self-Learning Simulation, or SelfSim, was developed to solve this class of type-two inverse problems in engineering (Ghaboussi et al., 1998). In most of this class of problems, the data for directly training the neural network model of the components of system are not available; it is either impossible, or difficult to generate those data. However, the stimulus and response of the system itself can be directly measured, and it contains information about the behavior of the components of the system. Autoprogressive Algorithm is able to extract this information in the form of a trained neural network. We will present several examples in this chapter to illustrate the application of the method.
