Sensors in an Aircraft generate huge volume of data and due to limitations on storage size only a fraction of this can be stored. Intelligent aggregation of time-series data is employed such that minimum amount of data is stored and off-boarded to help with further analysis. A use case of Cabin-Air quality is considered.
Real-time monitoring of aircraft cabin air quality is of paramount interest. The proposed system employs supervised ML techniques on simulated data to monitor air quality and suitably mitigate the anomaly as much as possible. The CO2 values and TPH parameters are monitored, and a custom built Adaptive Decision Block is used to classify anomalies to handle them effectively.
Data reduction using linear prediction
In this method local lines are fit to the time-series data. The points which are deviating from the line within a threshold are dropped such that the values do not differ drastically from actual values. This enables faithful reconstruction of data when required.
Error threshold selection:
A cost function decides the error threshold that needs to be used for compression. An optimum value of error threshold is chosen to achieve both good compression as well as keep errors under certain limit.