Identification of rotation anomaly of fan disks using random forest model

• Main Authors: Htike@Muhammad Yusof, Zaw Zaw, Nyein Naing, Wai Yan
• Format: Conference or Workshop Item
• Language: English; English; English
• Published: 2016
• Subjects: AC Collections. Series. Collected works
• Online Access: http://irep.iium.edu.my/54869/1/Fault%20Monitoring%20of%20a%20turbine%20Engine%20Disk.pdf; http://irep.iium.edu.my/54869/2/Programme_schedule.pdf; http://irep.iium.edu.my/54869/3/List_of_publiction_with_ID_title.pdf; http://irep.iium.edu.my/54869/

A fan disk is one of the most crucial parts of a modern jet engine. A fan disk failure is usually very difficult to predict and has disastrous outcomes. United Airlines Flight 232 crashed due to fan disk failure, which caused hydraulic systems to be ruptured. Since then, there have been numerous attempts by researchers to come up with systems to identify anomaly rotation in fan disks. The bottleneck in non-destructive fault monitoring lies in data analysis. State-of-the-art systems are not accurate due to high dimensionality of sensory data. This paper proposes a two-layered framework to perform anomaly detection. The first layer performs dimensionality reduction using autoencoder neural networks that compress input data onto a lower dimensional manifold. The second layer classifiers whether the lower dimensional data contains any anomaly using an ensemble technique called random forest. Satisfactory preliminary results on existing datasets are quite promising and encourage us to develop a full model.