Safety critical systems using artificial neural networks

The use of Artificial Neural Networks (ANN) as embedded software modules in systems is growing. To be used in Safety Critical Systems (SCS) it is necessary the development of a certification process for the ANN modules that can be inserted in the global SCS certification process. The Verification and Validation (V and V) process of expert systems is widely addressed in the literature, while ANN has little attention. For the application of a V and V process it is necessary the existence of a well defined development life cycle for the system. The knowledge in the ANN is implicit in the architecture and weights of the net resulting from the training process. Thus, the aiming focus of a certification process must be over the net architecture, the training type, the quality and quantity of the learning data used, and other specific particular aspects of the application. The emphasis of this work is directed to the development of ANN modules for SCS. (author)