Dynamic projection network for associative memory and pattern classification and its utility for mechanical diagnosis

This study proposed the utility of a type of dynamic networks known as a projection network in the realm of mechanical diagnosis. First, the projection network was studied for the relationship between its parameters and its properties including type, location and stability condition of equilibrium points and their interactions. A special case of stable axis equilibrium point or axis attractor was further studied and employed to formulate proper structure and parameter learning methods in each application area. The utility of the projection network was then established for associative memory and pattern classification.; In associative memory, guidelines and detailed algorithms for structure formulation and parameter learning were outlined. The application of projection network associative memory in mechanical diagnosis was then demonstrated for the diagnosis of a High Pressure Air Compressor (HPAC).; In pattern classification, this study developed and demonstrated the utility of the projection network for both supervised and unsupervised classification. First, it outlined the structure formulation and parameter learning for the projection network supervised classification system. It then incorporated the outlier elimination and clustering algorithm into the supervised classification system, thus making it capable of unsupervised classification tasks. The projection network supervised classification system was then evaluated and compared to the existing algorithms with three benchmark data including the Fisher's iris data, the heart disease data and the credit screening data. The projection network unsupervised classification system was also evaluated and compared to its supervised counterpart.; Finally this study demonstrated the utility of both the supervised and unsupervised classification system in mechanical diagnosis. First it demonstrated the utility of the supervised classification system in the diagnosis of two mechanical systems: a High Pressure Air Compressor and a jet engine. The utility of the unsupervised classification system was also demonstrated for the diagnosis of a High Pressure Air Compressor.