A Probability Neural Network (PNN) Estimator for Improved Reliability of "Noisy" Sensor Data

A self-directed, real-time, material processing system known as Qualita tive Process Automation (QPA) employs a knowledge base for autoclave curing of compos ites which requires information related to the "velocity" (d/dt) and "acceleration" (d ² /dt ² ) of temperature during the process of curing. The problem with using such information is the susceptibility of sensed data to process noise which results in sporadic, inconsistent and thus unreliable process data. To improve the reliability of "noisy" sensor data and therein the performance of a self-directed, real-time material processing system, an esti mator based on a Probability Neural Net (PNN) was developed. The paper is organized to: first, address the pertinent and distinguishing properties of PNNs, as well as reasons for their use in the QPA autoclave curing application, second, to suggest a methodology for coping with the practical difficulties of PNNs, and finally, to present the computational results of this study. A conclusion section focuses on the identification of other PNN appli cations, and some future PNN research.