Applying machine learning to detector data

Abstract

Machine learning algorithms are used to analyse data obtained from detectors 120 e.g. fire, smoke or carbon monoxide detectors in fire protection systems. A time at which maintenance is required for one or more of the detectors is predicted based on identified trends. This time may be a present time (i.e. maintenance of a detector is required immediately) or a future time (i.e. maintenance of a detector is going to be required). The data may be grouped into clusters and the time at which maintenance is required may be based on comparing a rate of change to a threshold. The rate of change may be the rate of change of a distance from a centroid of the cluster. The machine learning algorithms may be one or more of a support vector machine, k-nearest neighbours or an autoencoder. The method ensures that the fire detection system receives timely maintenance, thereby maximizing system uptime and reliability. The detectors may be a smoke detectors with a labyrinth structure where increase in dust can lead to an increased level of obscuration.

Images