SBIR-STTR Award

The objective is to increase the productivity of materials research by automating the infrastructure of such efforts. Global competition in manufacturing, limitations in fiscal and physical resources, ecological constraints, and technically challenging mission goals require that productivity be increased in the design, fabrication, and utilization of new or improved materials. The technical approach is based on unified learning and interpretation of combined process and product composition data in terms of resulting product properties. The approach builds on the results of two successful prior technical efforts. One is the CAD/Chem methodology by AI WARE, Inc. for exploring relationships between product formulation and product properties. CAD/Chem is based on neural-net computing and evolutionary programming. Another contributing methodology is the QPA work (US Patent #5032525) relating process sensor data, process control actions and product properties. The result of this work will be a methodology for building material models for interpolation and extrapolation from existing experimental data. The models will be accurate and useful because they will realistically incorporate process information instead of just nominal compositions; structure will be included in the model implicitly and explicitly if feasible. The models can be used to predict and discover. OPTION TASK: None submitted.

Keywords:
PROCESSING OPTIMIZATION DISCOVERY MODELING NEURAL NETS EVOLUTIONARY PROGRAMMING QPA CONTROL

The objective is to establish and demonstrate process discovery methods for increasing the productivity of materials research. This will be done in the context of research on pattern recognition algorithms applied to sensed process data for accurate (repeatable) growth of thin film structures with Molecular Beam Expitaxy (MBE), ceramic fiber coating by Chemical Vapor Deposition (CVD). Limitations in resources, ecological constraints, and increasingly challenging technical goals require that productivity be increased in the design, fabrication and utilization of new and improved materials and material structures. This work addresses this need by automating the process of discovering and understanding associative and causal relationships between process variables, materials composition, and end-product materials characteristics. This effort builds on three enabling technologies. AI WARE's proprietary neural net and evolutionary programming methodologies for modeling and optimal control of processes are incorporated in CAD/Chem and FLN Control Toolkit products. A third component is AI WARE's Visual EAM, providing a visual spreadsheet interface to an intelligent Episodal Associative Memory, and enabling real-time and off-line imaging of sensed data. The result of this work will be a powerful automation tool for use by researchers to enhance the productivity of their effort.