Inadequate engine water wash can significantly reduce aerodynamic efficiency, resulting in deteriorating performance. Additionally, proper engine water washes can help improve the lifespan of components and reduce corrosion.
We developed a machine-learning algorithm that accurately predicted the effectiveness of an engine water wash and hence helped a partner aircraft maintenance company to avoid extra costs arising due to unplanned maintenance.
The major steps involved within this use case were the identification of various sensors that correlated well with the effectiveness of engine water wash and subsequently training a Random Forest algorithm on the identified feature set. The model triggered an alarm to highlight cases with low effectiveness probability immediately after an engine water wash. The solution was implemented for 3-4 families of engines and resulted in savings worth $70M.
Every minute spent by an airplane on the ground is lost aviation revenue. Predicting component failure in advance can help minimize unplanned shop visits and have a significant impact on costs and revenues.
For this use case, we built an anomaly detection algorithm to identify abnormalities in engine performance. Sensor data collected from various engine components were used as input to train a Gaussian Mixture Model. The trained model was able to detect abnormalities by combining multiple features (sensor readings). The model raised an alarm for high severity cases to direct preventative action. The solution resulted in overall cost savings from unplanned shop visits of $27M.