EVALUATION OF AUTOMATICALLY LEARNED INTELLIGENT ALARM SYSTEMS

In this contribution it is investigated whether a combination of mathe matical simulation and inductive machine learning can replace the usua l knowledge elicitation techniques. To test this a domain was selected for which knowledge based systems had a high performance: intelligent alarm systems. A mathematical model of a breathing circuit and ventil ated patient was implemented in PSpice(R). Airway pressure, gas flows and CO2 concentration were simulated with this model, during normal fu nctioning of the breathing circuit and during several mishaps, for a w ide range of simulated patients. With an inductive machine learning pr ogram, classification trees were created from the simulated patient da ta. The classification trees described each breathing circuit mishap i n terms of changes in signal feature values with respect to the normal situation and were implemented as alarm system knowledge bases. The a larm systems were tested with data measured at 17 mechanically ventila ted animals. During ventilation of the animals several mishaps were in troduced. For each animal, 93-100% of all mishaps could be detected co rrectly by the alarm systems. The false alarm rate ranged on average f rom one false alarm per h to one false alarm every 2.5 h. It was concl uded that the suggested approach to knowledge elicitation was successf ul. (C) 1997 Elsevier Science Ireland Ltd.