CESSNA Citation X Engine Model Identification using Neural Networks and Extended Great Deluge Algorithms

Abstract

Accurate numerical engine model is an enabling factor in aircraft performance evaluation and improvement. In this work, nonlinear engine input-output relationships are learned and predicted by two cascading multilayer feedforward neural networks. Machine learning approaches necessitate a great amount of data to achieve efficiency. To satisfy this operational requirement, 441,000 flight cases are designed for a Cessna Citation X turbofan engine using a Level D Research Aircraft Flight Simulator designed and manufactured by CAE Inc. For each flight case, cruise phase data comprising Mach number, altitude, throttle level angle, low-pressure compressor speed, high-pressure compressor speed, engine net thrust and engine fuel flow are recorded. These data are then organized into subsets for training and validation purposes. Each neural network configuration is obtained by means of the Extended Great Deluge algorithm. The latter is also responsible for coordinating neural network training and learning error computation. Analyses based on computer experiments showed a mean relative prediction error upper bound of 4% is achievable for engine output parameters for all cruise phase flight cases.

Keywords

• Engine
• performance prediction
• Cessna citation X
• metaheuristics
• feedforward neural networks
• extended great deluge