What is unsupervised learning? Unsupervised Learning (see wikipedia) is used when no historical data is available that involves past faulty behaviour and/or ageing progression. Therefore,
Fuel consumption prediction
After each flight, a delta is observed between theoritical and actual fuel consumption levels.
Customer wished to understand reasons why a difference occured between the expected (according to the documentation) and the actual consumption level as measured by instruments during several flight phases. In order to get this understanding, customer asked Amiral Technologies to identify flight variables correlated to fuel consumption and to understand the weight of each in the observed shift.
Amiral Technologies was provided data from over 35 000 flights that included 50 variables covering 1) flight parameters 2) flight context 3) aircraft structure. Diagfit automatically identified variables that contributed to the difference between theoretical and observed consumption levels. Moreover these variables were ranked in order to create a linear model that explained and predicted the consumption delta.
Diagfit selected 5 variables among 50 and provided quantitative impact of each on the consumtion delta. Customer is now able to: 1) make a precise estimation of the consumption level. This is vital in order to compute the remaining autonomy of the aircraft and plan refueling. 2) understand flight, context and structure parameters impacting fuel consumption of the aircraft. This helps engineers to optimize design of some parts to make savings of fuel consumption. This leads to obvious gains in costs and savings in natural resources. The plot below shows the performance of the prediction model.