Information fusion using Granular Computing Neural-Fuzzy Networks and expert knowledge

Using Granular Computing (GrC) structures to describe or model processes adds to the system´s transparency, interpretability and allows the user to interact with the system with greater understanding of the process. Furthermore, such structures can potentially be used in combination with expert knowledge to provide an automated human-machine interactive system. The work presented in this paper involves a GrC-Neural-Fuzzy (NF) structure that is used as the core system for providing process information. Additional process information is provided by further evaluating the local performance of the GrC-NF model and by using an expert knowledge database in the form of a linguistic fuzzy rule-base. All sources of information (GrC-NF model, local performance evaluation and expert knowledge) are combined together by means of a Fuzzy-Fusion algorithm to provide a single piece of information. An experimental study is used to validate the structure´s performance. The process under investigation is the prediction of mechanical properties of heat treated steel, which involves a multi-dimensional, non-linear and complex data space. Real industrial data are used for the experimental study.