Reuse of existing design information in the development of new electronic PTC devices via a neural network approach

Burning events and voltage endurance are two important aspects that need to be predicted during the design and development stage of a new series of electronic positive temperature coefficient (PTC) devices. In this paper, these problems are identified by experiments conducted on well-developed devices, and are resolved by improving the resistance-temperature characteristics of the PTC devices in order to overdamp, underdamp, or critically damp high-current/high-voltage surges. The use of neural networks is proposed, to learn the empirical or experimental design information that already exists, and then to predict the occurrence of burning events and the voltage endurance of new PTC devices at the design/development stage. Two predictive schemes are presented separately, for burning events and for voltage endurance, where the training patterns for the desired outputs are either generated from empirical formulae or collected from experiments on already-developed PTC devices. The predicted results are discussed against the experimental results that are available, and an overall concept is finally given for the integration of the neural predictive models into the computer-aided design/computer-aided engineering system used for the PTC devices