Data-driven process monitoring and diagnosis with support vector data description

This thesis targets the problem of fault diagnosis of industrial processes with data-drivenapproaches. In this context, a class of problems are considered in which the only informationabout the process is in the form of data and no model is available due to complexity of theprocess. Support vector data description is a kernel based method recently proposed in the fieldof pattern recognition and it is known for its powerful capabilities in nonlinear data classificationwhich can be exploited in fault diagnosis systems. The purpose of this study is to investigate SVDD applicability as a data-driven method in industrial process fault diagnosis. In this respect, a complete framework for fault diagnosis structure is proposed and studied. The results demonstrate that SVDD is a powerful method in process fault diagnosis.