TITLE: Applying Optimized Entropy Network To Signal Processing Of Composite Nondestructive Evaluation

Abstract:
Quality assurance is a concern to all involved in the production of engineering components and is a necessity to the further advancement of automation in the composite manufacturing industry. Further advancement of automated manufacturing is a mute point unless it is coupled with equivalent advancements in inspection capabilities. Similarly, enhancements and the development of new NDE techniques are restricted by signal data processing capabilities.

Artificial Neural Systems ANS are known to be excellent signal data processing tools. AJI has developed a new algorithm for training neural networks that has been shown to overcome all of the known problems with traditional back propagation methods. This algorithm was developed through the use of a thermodynamic model of the network operation that included both the delta energy, the kinetics of the network and optimization functions. The technique, known as the Optimized Entropy Network (O-E Net), has been used on several applications ranging from high speed signal processing to vision systems. The O E Net has achieved learning rates greater than 1000 times that of the standard back propagation method while also preventing the network from falling into learning instabilities.

Composite NDE presents a unique set of signal data processing challenges which the O-E net can successfully overcome. To complete an efficient in-line NDE of composite manufacturing the integration of new NDE techniques and effective signal data processing will be required. The signal data processing capability of the O-E net is a small but important part of the required improvements needed for a successful composite in-line NDE.

Source: Proceedings of the 'NDE applied to Process Control of Composite Fabrication' - Conference, 4-5 Oct 1994 St. Louis, Missouri. Publisher and Organizer: Nondestructive Testing Information Analysis Center (NTIAC) Texas Research Institute Austin, Inc. Austin, Texas [http://www.ntiac.com] [Buying the Proceedings]