A knowledge-based adaptive clustering technique for neurocomputing modeling of structures
Z. Peter Szewczyk · 39th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference and Exhibit · 1998
This paper presents an adaptive clustering technique to increase the accuracy of neurocomputing modeling and reduce the time required to develop neurocomputing applications. Both objectives are attained by (i) partitioning the training data so that the structural response in each contiguous subsets belongs to a similar class, and (ii) designating different neural networks to fitting partitioned data. Partitioning, or domain decomposition is performed based on quantificatio n of the response characteristic by an automated clustering routine. The routine uses a rule-based module with a fuzzy logic inference engine to adapt data partitioning to the changing response characteristic. The adaptive clustering is illustrated with an example involving modeling of the nonlinear deformation of an upper skin of a wing-box. It is also shown that neurocomputing models compress data that are produced by finite element analysis. An example involving storing natural modes of a solar panel is used to demonstrate the feasibility of using neurocomputing models for data compression.