253 Forum on Fuzziness Multilevel Failure Detection System P. Vaija and M. JiirvelAinen Laboratory of Chemical Engineering, Helsinki University of Technology, Kemistintie 1 M, SF-02150 ESPO0 15, Finland M. Dohnal Department of Chemical Engineering, Technical University of Brno, Kravi Hora 2, CS-60200 BRNO, Czechoslovakia A fuzzy multilevel failure detection system is developed for complex processes with ill-known mathematical model and poor measuring accuracy. The decomposition of the problem is made with the aid of fuzzy switches and diagnostic algorithms which are specified by a set of conditional statements. The diagnosis system is demonstrated by a real low density polyeth- ylene reactor in full detail. Keywords: failure detection, failure diagnosis, expert systems, fuzzy decision, fuzzy expert system, fuzzy logic. Pirjo Valja was born in 1950, in Fin- land. She received her M.Sc. in 1976 and Tech. Lic. in 1979 in Chemical Engineering from Helsinki University of Technology. Her main research in- terests include safety and reliability aspects of chemical processes, mathe- matical modelling and the application of fuzzy mathematics to different in- dustrial problems. She has published several articles and congress papers in these fields. Mmlli Ji~rvel~nen was born in Fin- land in 1935. He received his M.Sc. in Engineering in 1962 and Doctorate of Technology in 1974 from Helsinki University of Technology. He is now an Associate Professor of Chemical Engineering (Plant Design) at Helsinki University of Technology. Formerly, he was employed as a Process Design Engineer and as a Project Manager with Neste Oil Co. and Rintekno Co. His research interests include ration- alization of plant design, cost evalua- tion, application of fuzzy mathematics and computers in chem- ical engineering. He has written some 50 papers on this subject. North-Holland Comvuters in Industry 6 (1985) 253-263 1, Introduction In process industries, safety studies and design have been traditionally made on the bases of expe- rience by analysing previous incidents [1]. In some companies this analysis has been used to create a data base containing cause-effect data of failures for a particular process [2,3,4]. The main purpose of these data bases is to assist the operator in fault diagnosis and administration. In this paper, a fuzzy based expert system is applied to develop a failure detection system for complex chemical processes where there are many process variables causing quite similar symptoms in a disturbance and confusing the operator. A fuzzy based expert system is used because fuzzy mathematics can put experience-based and subjec- tive knowledge in a form acceptable to a com- puter. It can also take into account inaccuracy in measuring the results of process variables. Preliminary studies for application of fuzzy based single level systems to failure detection and accident analysis is presented in [5,6,7]. Practical problems require, however, extensive modelling with many mutually dependent variables. There- fore, a fuzzy multilevel algorithm for failure diag- nosis is proposed in this article. If the decomposition of the system under study is optimally chosen, it results in a considerable M. Dohnal was born in Brno, Czecho- slovakia in 1943. He received his Ing. and CSc. degrees in Mechanical En- ~neering from the Technical Univer- sity of Brno in 1965 and 1971, respec- tively, and his DrSc. degree in System Engineering from the Technical Uni- versity of Leuna-Merseburg, GDR. In 1968 he joined the Technical Univer- sity of Brno, where he is currently an Associate Professor of Control En- gineering. His research interests in- clude application of fuzzy mathe- maac~, ~y~tem and chemical engineering. He has written more than 100 papers on these subjects. He was formerly employed as a Process Engineer with Chemoprojekt and Chepos, perfor- ming CAD and economic evaluation of chemical and food plants. 0166-3615/85/$3.30 © 1985 Elsevier Science Publishers B.V. (North-Holland)