Purpose

The purpose of this study is to make artificial neural network (ANN)-based prediction about thrust using the flight control parameters of aircrafts.

Design/methodology/approach

In today’s transportation, airplanes have an important place because of their safety, quality and speed. One of the most important parameters affecting the secure flying of aircrafts is the thrust value of aircraft engines. Determining the optimum thrust value should be investigated. If thrust value is less than optimum level, the flight safety runs a risk. Otherwise, fuel consumption goes high and some unwanted vibrations occur that cause uncomfortable flight. In this study, multi-layer perceptron ANNs, which are one of the intelligent optimization methods and frequently used in the literature, are preferred to predict the optimum thrust value during take-off, cruise and landing. The actual flight data, which is taken from the black box of an Airbus A319 aircraft, is used to train ANN models using back propagation algorithms. Velocity, altitude and ambient temperature values of the aircraft are selected as inputs and the thrust value is selected as output. During the training process of ANN, eight different training algorithms with different structures are used to figure out optimum ANN model with minimum error.

Findings

Different ANN models were trained using eight different training algorithms. The ANN model with minimum error has multi-layer perceptron structure, which is trained using Levenberg–Marquardt (LM) algorithm.

Research limitations/implications

To obtain the ANN structure with minimum error training, process takes more than a day depending on the capacity of a computer for LM training algorithm. But after training process, the trained ANN model produces sufficient output in a few milliseconds.

Practical implications

Totally 15,670 input-output data sets are obtained from an Airbus A319 aircraft. 12,889 of them are used as training data and the rest of the data sets, selected randomly are used as test data. Test data sets are never used in training phase, and the obtained results show that the ANN model successfully predicts thrust value using unseen input data.

Social implications

The ANN could be used as an alternative method to predict other flight control parameters of aircrafts.

Originality/value

To the best of authors’ knowledge, this study is the first example in literature to predict the thrust value of the aircraft using ANN.

中文翻译：

---

使用人工神经网络预测飞机的推力

目的

这项研究的目的是使用飞机的飞行控制参数对推力进行基于人工神经网络（ANN）的预测。

设计/方法/方法

在当今的交通运输中，飞机因其安全性，质量和速度而占有重要地位。影响飞机安全飞行的最重要参数之一是飞机发动机的推力值。应该研究确定最佳推力值。如果推力值小于最佳水平，则飞行安全有风险。否则，燃油消耗会很高，并且会发生一些不必要的振动，从而导致飞行不舒服。在这项研究中，多层感知器人工神经网络是一种智能优化方法，并且在文献中经常使用，它是预测起飞，巡航和着陆过程中最佳推力值的首选方法。从空中客车A319飞机的黑匣子中获取的实际飞行数据用于使用反向传播算法训练ANN模型。速度，选择飞机的高度和环境温度值作为输入，并选择推力值作为输出。在人工神经网络的训练过程中，采用了八种结构不同的训练算法，以求出误差最小的最优人工神经网络模型。

发现

使用八种不同的训练算法训练了不同的人工神经网络模型。具有最小误差的ANN模型具有多层感知器结构，该结构使用Levenberg-Marquardt（LM）算法进行训练。

研究局限/意义

为了以最少的错误训练获得ANN结构，过程需要一天以上的时间，具体取决于计算机用于LM训练算法的能力。但是在训练过程之后，经过训练的ANN模型会在几毫秒内产生足够的输出。

实际影响

从空客A319飞机上总共获得了15,670个输入输出数据集。其中12,889个用作训练数据，其余的随机选择的数据集用作测试数据。测试数据集从未在训练阶段使用，获得的结果表明，ANN模型使用看不见的输入数据成功地预测了推力值。

社会影响

人工神经网络可以用作预测飞机其他飞行控制参数的替代方法。

创意/价值

据作者所知，该研究是文献中使用ANN预测飞机推力值的第一个例子。