The overall objective of this project is to demonstrate automation in the monitoring and control of the manufacture of solid rocket booster (srb) parts. Specifically, monitoring of structural quality via ultrasonic measurements and process control decisions will be implemented. This is an ideal application for the near-term migration to the manufacturing floor because the measurement data is already being taken, the control response time is well within the capabilities of typical neural network implementations, and tangible improvements in productivity and accuracy can be expected. The proposed technical approach utilizes a cooperative neural network/ qualitative physics monitoring module, which allows both experimental and casual knowledge to be used in `programming' the system. During the project, a qualitative physics model of the relevant parts of the srb manufacturing process will be constructed, the model will be integrated with a neural network to learn the specifics of monitoring, and a neural network for control decisions will be implemented. The resulting prototype will be tested with real data from the manufacturing floor. The feasibility and requirements for introduction to the manufacturing floor will be documented. Anticipated
Benefits: the proposed project will provide a practical monitoring and control technology integrating the experimental capabilities of neural networks with underlying process knowledge. In addition to srb manufacturing monitoring, the technology will be applicable to inspection of aircraft, bridges, pressure vessels, rails, pipelines, and other systems critical to the infrastructure of our society....
Keywords: Neural Networks; Artificial Intelligence; Ultrasonic Measurement; Qualitative Physics
