We used our technology to determine with greater accuracy the future demand for aeroplane parts, helping the carrier to manage stock more efficiently. By reducing the amount of time aeroplanes are unable to fly, we helped the carrier to save money, operate flights on time and improve customer satisfaction.
A leading short-haul airline operating in more than 30 countries. It employs over 3,000 pilots and 10,000 cabin crew, and flies more than 80 million passengers per year.
Modern aeroplanes comprise 10,000 parts, 500 of which are critical. Should those parts break down, the aeroplane cannot fly. As a major carrier with hundreds of planes flying more than 800 routes, the airline keeps a stock of some critical parts at various airports to speed up the repair process, if required.
As a result, it is critical that the airline is able to predict the demand for parts If demand exceeds stock, plane repairs will take longer than necessary as spare parts will have to be flown in; if demand is lower than stock, the airline is not operating efficiently and the spare parts should be stocked at the airports where they are more likely to be needed.
Delayed or cancelled flights bring a number of consequences, from dissatisfied customers to the need for compensation payments (which can run to tens of thousands of pounds per flight), lost revenue and a cascade of further disruptions as other flights are affected. Faculty was asked to develop a model that could more accurately predict which parts would be needed, and where.
The project used data about the daily demand: the number of units of each part used each day in each airport. With the historical data in place, we implemented a random forest regressor to forecast demand and determine the quantity of each part to stock at each airport.
Against the backdrop of regular patterns in the data, there were significant anomalies. These sudden, unlikely shifts were often the result of changes out of the airline’s control, such as factors relating to the airport or the manufacturer. The model was able to identify these shifts in demand, highlight the anomalies and prevent them from skewing the forecast.
Once trained, the model predicts (orange) the actual demand (black) very well
The model was able to determine with greater accuracy the future demand for parts, helping the carrier to manage stock more efficiently. By reducing the amount of time aeroplanes are unable to fly, we helped the carrier to save money, operate flights on time and improve customer satisfaction.