Fuzzy logic in anaesthesia machine fault diagnosis

Abstract

Patient safety has always been a very important factor in hospital equipment design. The basic principle in the design of modern life supporting devices is that no single fault can cause a risk of injury to the patient. The AS/3 Anaesthetic Delivery Unit (AS/3 ADU) is a new generation anaesthetic machine developed by Datex-Engström Division Instrumentarium Corporation. The AS/3 ADU has built in fault diagnosis system as a safety backup for almost all of the electro-pneumatic components. The fault diagnosis is based on the fault tree and the failure mode and effects analysis. In the AS/3 ADU, the faults are detected with limit checking. One of the problems in the current fault diagnosis system is that proper fault detection limits are hard or even impossible to find. New technical fault diagnosis methods have been developed during the last couple years, and therefore there was an interest to bring more intelligence to the current fault diagnosis system of the AS/3 ADU.In this thesis, an expert system was designed and implemented for developing and testing sophisticated fault detection methods on the AS/3 ADU. An external fault detection system was implemented on a PC microcomputer. A serial adapter was built to connect a personal computer and the AS/3 ADU. The system is capable of real time fault diagnosis. It can also be used to acquire and store data for off-line analysis. In addition, tools for data visualisation were created to get a grasp of ADU signals under incorrect operation.The fault detection knowledge base of the expert system was implemented as a fuzzy logic rule-base. Fuzzy logic fits particularly well for developing fault diagnosis expert systems because it can model uncertain process states. Expert knowledge can be expressed with linguistic terms which can be incorporated into fault diagnosis system straight forward. Fuzzy logic makes decision support systems modular and easy to maintain compared with the traditional expert systems. Digital signal processing algorithms were used for feature extraction expert systems. Digital signal processing algorithms were used for feature extraction before the actual fault detection. It was discovered that the design and the implementation of feature extraction algorithms were much more time consuming than that of the actual fuzzy interference system

Diplomityössä suunniteltiin ja toteutettiin asiantuntijajärjestelmä AS/3 anestesiakoneen vikadiagnostiikkaan. Viantunnistuksen asiantuntijajärjestelmän tietämyskanta toteutettiin sumeallasääntökannalla. Järjestelmää kokeiltiin seitsemällä uudella vikaluokalla, joista kolme oli anturivikaaja neljä sähköistä vikaa. Jokaiselle vikaluokalle luotiin oma sumea sääntökanta. /Kir10