Neural Observer for the Hot Isostatic Pressing

Hot Isostatic Pressing (HIP'ing) is one of several processes used for consolidating porous materials (such as metal powders) to high density. The equations describing the HIP'ing system dynamics are immensely complicated and highly nonlinear, thus complicating any controller design for the system. Further complications arise due to the fact that one of the system's states, grain size, cannot be measured directly as a system output. Instead, a method for estimating grain size must be formulated and implemented. To achieve this goal, it is rst necessary to show that it is indeed possible to reconstruct grain size from direct measurements of the system's other parameters (relative density, applied temperature , applied pressure). Then, a working methodology must be found in order to accomplish the reconstruction task. Recent results by the author show that the system is observable, and therefore grain size may be estimated during the HIP'ing process without directly measuring it as an output of the system. To accomplish the reconstruction task, we investigate the use of a backpropagation trained neural observer. Results of simulations are extremely promising. Findings indicate that use of a backpropagation trained neural observer yields an accurate estimate of grain size for use in the feedback control loop.