Multivariate statistical process control for continuous monitoring of networked early warning fire detection (EWFD) systems

Sensor array networks provide much information concerning an environment if the data is adequately used. However, monitoring large networks of sensor arrays can be data intensive. Multivariate statistical process control (MSPC) methods allow monitoring of an entire system at a supervisory level. These methods are demonstrated for fire detection using the early warning fire detection (EWFD) system. The EWFD system comprised of four sensors that included photoelectric and ionization smoke detectors, as well as carbon monoxide and carbon dioxide sensors. Fourteen sensor arrays were distributed in 10 compartments and passageways throughout the ex-USS SHADWELL, the advanced damage control fire research platform of the Naval Research Laboratory. Data was collected for a series of smoldering and flaming fire sources as well as nuisance sources in different compartments over two decks. The network of sensor responses, location and temporal data are used to identify events, determine source location and monitor fire rate of growth. Hotelling’s statistic and the Q-statistic are employed initially for event detection. Subsequently, contribution plots are used to determine source location, rate of growth and to discriminate between actual fires and their byproducts in adjacent compartments. The response times of the MSPC method are compared to the commercial smoke detectors co-located with the EWFD systems. MSPC is shown to be an efficient method for continuous monitoring of EWFD systems.