Application of expert system for marine diesel engine diagnosis

W artykule przedstawiono koncepcję systemu diagnostycznego okrętowego silnika tłokowego opartą na modelu systemu ekspertowego. Zrealizowano pozyskiwanie wiedzy diagnostycznej, opracowano bazę wiedzy oraz zaproponowano ogólną strukturę systemu. Wiedza dla ekspertowego systemu diagnozowania silnika okrętowego została pozyskana od ekspertów (specjalistów w dziedzinie eksploatacji) oraz z diagnostycznych baz danych. Do pozyskiwania wiedzy od ekspertów zastosowano wywiad kwestionariuszowy. Podjęto próbę pozyskania podstawowej wiedzy z dziedziny eksploatacji silników umożliwiającą ocenę ich stanu technicznego. Pozyskiwanie wiedzy z baz danych przeprowadzono z wykorzystaniem indukcyjnych metod uczenia maszynowego. Dane uczące dla algorytmów indukcji zostały zgromadzone w wyniku realizacji eksperymentu czynnego na silniku Sulzer 3Al 25/30.The development of diagnostic systems for marine diesel engines is vital for both ship safety and economic reasons. Nowadays, many diagnostic systems have been created by both research laboratories and engine producers. Typical disadvantage of most systems is their completeness. This means that diagnostic algorithms of technical conditions, adopted during system creation, cannot be updated or modified during later operation. The solution to the problem could be an expert system in ship engine diagnosis. Module system structure, and above all, the separation of database from remaining program, enables creation of diagnostic system of open type, where diagnostic knowledge can be updated and cumulated. This paper presents diagnostic system concept for marine diesel engine, basing on expert system model. The relevant knowledge database was created with the use of collected diagnostic data. Diagnostic data were collected from experts (ship engine professionals) and diagnostic databases. The paper questionnaire was used to the knowledge acquisition from experts. Basic knowledge related to the marine diesel exploitation was undertaken. The expert knowledge covers the weakness point of engine, the kind of faults and diagnostic relation between faults and their symptoms. The group of experts contained the experienced merchant navy officers. The selected machine learning methods was used to obtain the relationship in the form of diagnostic rules from data base. The results obtained with the algorithms LEM2, MODLEM, and EXPLORE was compared. MODLEM algorithm allows the use of numerical data directly without having to prediscretization. Learning examples, stored in the diagnostic database, were obtained as a result of the active experiment, carried out on laboratory Sulzer engine 3AL 25/30. During the experiment, damages of the turbocharging system, fuel injection system, and combustion chamber were simulated. Only the elementary states (single damage in the same time) in a variable load were included. Tenfold cross validation technique was used for evaluation of the obtained rules classifiers. The obtained diagnostic rules have also been assessed in substantive terms, including an analysis of the relationship between disability states and received symptoms. Complex diagnostic systems for marine diesel engine diagnosis face limited application in ships, particularly due to their high cost. Ship engines are fitted with assorted indicators and measurement tools enabling control of many operational parameters, as well as, storing such measurements in databases. Technical condition verdict is, however, still the responsibility of the engine operator, and here comes the room for IT systems, which could facilitate such processes. The expert system application may substantially enhance abilities of monitoring systems presently existent in power rooms, in respect of ship engine diagnosis. Such a system enables saving valuable, operational knowledge for later use. Additional advantage represents the opportunity of automatic collection of diagnostic information with machine learning methods. The usefulness of such methods for creation of diagnostic rules was proved on the basis of examples stored in database. The expert system enables integration within a single frame of both information collected from experts and automatically collected one. A doubtless advantage of expert system is the opportunity of updating and developing the content recorded in the database. Due to this feature, the effectiveness of the system may grow during engine operation and facilitate gaining new experience