Tools for Knowledge-Based Signal Processing with Applications to System Identification

This thesis contains three parts: firstly there is an introduction to the field of knowledge-based signal processing - which denotes a combination of symbolic and numeric software capable of implementing algorithms of both algorithmic and heuristic character. Secondly suitable tools for knowledge-based signal processing are discussed - this includes the tools developed in our research. Finally we present our applications of knowledge-based signal processing.The research in computer science and especially in the area of artificial intelligence (AI) has provided useful techniques and tools - e.g. expert systems - for "implementing" heuristic knowledge. This is hard to express in conventional programming languages like Fortran, Pascal and C.To make it easy to implement knowledge-based signal processing applications, tools integrating numerical and symbolical computations (including logic inference) are needed: The tools developed in the undertaken research contains integrated building blocks: Common Lisp with YAPS (or Scheme) - the main programming language of the developed tool extended with an expert system shell, MATLAB - standard numeric software, and MACSYMA - a computer algebra system.The applications are directed toward intelligent support system for system identification software. Two different help systems are discussed and implemented, one for help in black-box modeling, and one for help with physical modeling. The latter system uses bond graphs to describe systems. Bond graphs give a graphical description of physical systems which allow for modeling multiple domains in a form which makes a good compromise between a high level description of physical systems and easy computation of the underlying differential equations.