A Trend Monitoring Diagnostic Algorithm for Automatic Pre-identification of Turbofan Engines Anomaly

Each day a medium range aircraft schedules almost ten flights. During those flights each aircraft may send data to the ground via a satellite communication channel. With more than 35000 CFM engines flying every day Safran Aircraft Engines receives a big amount of such messages. However, all messages do not imply an immediate maintenance action. Most of them are already managed by preventive maintenance operations. On 1000 messages managed each day by fleet operators, only 10 of them lead to airline reports. To select the 1% important messages, a fleet manager plots some data curves collected during the last flights, if the operator sees a specific pattern he may decide to transfer the alert and write a report. As we have more and more engines in operation it becomes necessary to helps the operators with some diagnostic tool. Hence we build a detection algorithm that automatically analyses the multidimensional data curves and uses a supervised learning process to select a good combinations of tests giving more than 85% of good pre-identification. As the expert operator may also have some degree of liberty in his decision, such performance is considered as good enough for a pilot implementation.