Knowledge-Enhanced CO-monitoring in Coal-Mines

Detection of underground res is an important security task in hard-coal mining. Automated re detection systems are usually based on the monitoring of carbon monoxide (CO). Systems using conventional technology based on threshold and tendency observations, however, generate a large number of false alarms. We show how CO-concentrations can be forecast by appropriate models of the physical and chemical processes. We furthermore describe a rule-based speciication system utilizing forecasting for CO-monitoring. The improvement of this approach over the conventional is threefold. First, the number of false alarms is reduced by 50%, at least. Simultaneously, the thresholds for warnings and alarms can be reduced so that, second, the detection of real res becomes both quicker and more reliable. Third, heuristic rules for re detection and suppression of false alarms as well as the control of the forecasting can be described in a declarative way. While our system is still in a prototypic stage, the three major German hard-coal mining companies decided to use our approach in their CO-monitoring systems.