Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Aircraft Fault Detection using Artificial Intelligence:

A Review
Ayeshmi Samaradivakara
School of Computing
Staffordshire University,UK.

Abstract:- Aircraft is one of the famous traveling mediums I. INTRODUCTION

to travel from one country to another. Aircraft can carry
more than 300 passengers on one trip. Many passengers Aircraft is a type of vehicle that is used to travel through
have died in airplane accidents in history. The leading the air by gaining power from the air. In the modern world,
cause of these accidents is the internal and external defects most people use aircraft to travel from one country to another
of the aircraft. Most errors can be early detected. With the country. There is a high capacity in an aircraft. Airbus A380
improvement of technology, there are many advanced has 525 passengers capacity [1]. There are many types of
techniques that people have invented to automate fault aircraft. They are amphibious, helicopters, cargo planes, flight
detection. Among them, artificial intelligence takes a jets, gliders, etc.
higher place than other techniques. Because these
techniques are straightforward to use and understand, Aircraft is a collection of many mechanical parts. The
they save time and cost. This research paper discusses power plant, empennage, landing gear, fuselage, and wings are
existing solutions for aircraft fault detection using AI and the main parts of aircraft. There are many faults that can be
the techniques that they used in their research papers. occurred in these mechanical parts. A small fault can be
Additionally, proposing a new solution for aircraft fire affected to a major breakdown of the system. Therefore, safety
detection. is a major concern of the aircraft. In 1985 [2], the crash of
Japan airlines flight 123 reported 520 deaths of passengers and
Keywords:- Artificial Intelligence, Deep Learning, Machine cabin crew members. The description of types of faults are as
Learning. follows,

Table 1 Description Of Fault Types

Fault Type Description
Design errors Manufacturing companies are responsible for these types of errors. Components of aircraft are
manufactured by correctly, but there are design errors. These small errors can be affected the
whole components of the aircraft.
Maintenance errors Almost 12% of aircraft accidents are caused by maintenance errors. Aeroperu Flight 603 [2],
Japan airline flight 123 and chalks flight 101 are some of the famous flight accidents that are
caused by maintenance errors.
Hydraulic failures Hydraulic failure means the pilots lost control of the airplane to turn left, right, and climb. This
is a rare fault because there are three or more hydraulic systems in bigger aircraft and small
aircraft has manual control systems.
Aircraft engine faults This is a rare fault because there are multiple engines in an aircraft. Engines are one of the most
complicated systems of aircraft. A 250-horsepower airplane engine lost power during flight as a
result of the propeller's excessive speed.
Manufacturing faults Manufacturing failures occurred when the aircraft is manufactured differently that its original
design.
Example: 1957 MCNbs [2] Island RCN Banshee crash.
Aircraft wing faults Wings are the heart of the aircraft. Because wings are very helpful to fly through the air. This is
an uncommon fault, but these types of failures can lead to major disasters.

Pilot error Tenerife disaster is one of the best examples of pilot error. All 583 [2] passengers and cabin
crew members died in this accident. The main reason for this accident is the pilot had begun the
takeoff run without getting air traffic control clearance.
Communication failure The appropriate connection between air traffic control units is a key requirement. If the flight
lost the connection, it will be a major issue for the entire process.

IJISRT23JAN072 www.ijisrt.com 593

Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Aircraft technicians play a major role to ensure aircraft II. AIRCRAFT AND ARTIFICAIL INTELLIGENCE
safety. Nowadays, most techniques that they use have manual
processes. Aircraft technicians detect these errors using their The aviation industry has been developing day by day
own knowledge. Aircraft skin inspection is one of the primary because of advanced technologies. In 1950, the introduction of
faults detection techniques. Sometimes this traditional the jet engine [3] is one of the turning points of advanced
prediction might be right or wrong. Therefore, if we can technologies. Among them, artificial intelligence is getting a
introduce a technological environment to detect these failures high rank because this technology is very advanced, easy to
before much damage will be helped to increase the accuracy use and understand, and saves money and time. From 1950 to
of the result and increase the performance of the aircraft. Small the present, there are many improvements that can be seen in
failures can be led to critical accidents, and they will be aircraft maintenance, aircraft manufacturing, traffic
directly affected human life. Aircraft Safety is a key management, etc. Most of the researchers mentioned that it is
requirement. very easy to use machine learning because of the abundance
of data in the aviation sector.

Artificial intelligence is the intelligence provided by

machines [4]. Those machines have been trained to think like
a human by using artificial intelligence techniques. There are
many AI techniques in the world today. The description of
techniques of artificial intelligence is as follows.

Table 2 Description of Techniques

Technique Description
Machine Learning Machine learning is a key technique in AI. Machine learning is an application that provides
the ability to learn and improve system requirements by using data and algorithms. Batta
Mahesh [6] defined machine learning as a technique that is used to give some guidance to
handle big data in a useful manner.
Natural Language Processing Natural language is an interaction between computer and human language which is created to
process natural language. Most people use NLP for text conversation and audio conversation.
This technique uses algorithms to identify the exact hidden patterns of the data and convert
those data into a format that is helped to understand by the computers.
Automation and Robotics Automation is a process that is used software and other technology to process human work.
Robotics is a field that combines computer science and engineering to build robots to
perform human work. There are three types of automation. They are industrial automation,
software automation, and process automation.
Machine Vision Machines can detect visual information and analyze those for future work. Cameras can
capture visual information and convert those data to digital data. Voice recognition is the best
example of computer vision. This technique can capture complex data and use it effective
way.

There are many existing solutions that are already using like Airbus are already using and implementing intelligent AI-
AI techniques. Self-piloted airplanes are one of the famous based maintenance solutions for their aircraft.
improvements of airplanes using aircraft intelligence. This
product uses a vision-based automated system to process the The digital twins’ approach is deemed a wonderful fit for
aircraft during landing, taxi, and take-off. This system is air traffic control applications. By automation routine actions
developed by artificial intelligence algorithms using 500 flight during peak hours, controllers may concentrate on moves that
details dataset. Natural language processing is used for virtual are crucial for safety. The ultimate objectives of an AI-based
assistance. Aircraft researchers mentioned that 70% of aircraft air traffic management solution would be increased operational
accidents [5] are caused by human factors errors. Self-piloted effectiveness, weak separation trajectory prediction, and
artificial intelligence can eliminate these errors because AI has automation of air traffic control. According to the experts, AI
advanced machine learning algorithms. will make it easier for air traffic controllers to handle the
complexity of adding new vehicles as well as the expected
It is a well-known fact that improper or neglected airplane increase in air traffic. With better predictive capabilities, it will
maintenance has contributed to some of the worst aviation assure secure, safe, and highly effective airspace management,
accidents. Even last year, aircraft maintenance was the biggest leading to fewer delays and fewer gas emissions because less
cause of domestic flight delays. The revenue streams of airline fuel will be spent.
operators are significantly impacted by several existing
approaches to aircraft maintenance. For instance, in 2017,
airlines spent a combined $76 billion on maintenance[5],
repair, and overhaul, or around 11% of all operating expenses.
Based on information from various types of sensors, businesses

IJISRT23JAN072 www.ijisrt.com 594

Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. RELATED WORKS and strikes a reasonable balance between speed and detection
accuracy, YOLO4 is more useful for engineering applications.
Many studies have been conducted at aircraft fault
detection using various AI approaches and machine learning Murat Bronz et al [11] recognized the real-time
algorithms. The following work represents the existing works challenges of small-scale fixed-wing UAVs. It has been shown
of aircraft fault detection. that real-time fault detection is feasible under conditions of
noisy measurements, communication barriers, and wrapped
Xuyum F. et al [7] identified that there are large noise and wing structures. They used eleven flight logos for this fault
complex data in Aircraft Communication Addressing and detection. This team proposed a data-driven supportive vector
Reporting System (ACARS) data. Therefore, it is very hard to machine algorithm for classifying the behaviors of aircraft.
detect faults effectively. They introduced a novel method for They faced an overfitting problem because of limited data. The
aircraft engine fault detection. These researchers used a highest accuracy that they achieved is 95%.
convolutional neural network (CNN) and supportive vector
machine (SVM) algorithms. And CFM56-7B26 engine data is Maren D. Dangust et al [12] introduced a new technique
used to validate the effectiveness of data. This team clearly for reducing the volume of logs of aircraft warning and failure
mentioned that their proposed solution requires less time and messages that are kept by the central maintenance system. The
fewer parameters more than existing solutions. The highest suggested model integrated a bidirectional gated recurrent
accuracy that they achieved is 96.16%. Their future target is to network auto-encoder that cooperated to offer the right
use Quick Access Recorder (QAR) data for aircraft failure connection failure or warning signs associated with removing
detection. that aircraft LRU aiding in the recognition of aberrant patterns.
Aircraft central maintenance system (ACMS) data was used for
Caizhi L. et al [8] mentioned that aircraft skin is very their prediction. AE-CNN-BGRU model is indicated in the
important to recognize the flight and safety performance of the evaluation result. Their future goal is to study this AE-CNN-
aircraft. They identified existing solutions have very low BGRU architecture for translating time data to graphical
accuracy levels. This research team used the convolutional displays that make use of recurrence graphs.
neural network to introduce a signal classification model for
aircraft skin detection. They used four optimization algorithms. Qin Liu et al [13] identified that nowadays researchers are
They are, SGD, momentum, RMSprop, and Adam. But Adam's giving their focus on making aircraft more energy efficient,
optimization algorithm showed the best accuracy levels for this reliable, environment friendly, and low maintenance costs.
signal classification. This method offers a fresh approach to the They defined More Electric Aircraft (MEA) as the best solution
detection of aircraft skin because it can recognize the probe for this focus. They proposed real-time failure detection and
picture, ultrasonic A-scan signal, and, track and assess the kind isolation system for MEA. They used an FPGA-based neural
of the change’s region. network method for this prediction. The highest accuracy that
they achieved is 99.5%. Their future goal is to introduce
Brian M. et al [9], proposed a data-driven algorithm for multiple fault detection systems for aircraft.
full automated aircraft sensor failure detection. A time series of
typical behavior is used in the proposed machine learning P A Mallikarjuna et al [14] recognized that the gearbox
method to mimic the evolution of the relevant measures by a is one of the important parts of aircraft engines. Any slight
linear time-invariant system. A Kalman observer is used to gearbox damage can result in the aircraft engine breaking
keep a distinct real-time estimate of the measurement of down. Consequently, it is important to research fault diagnosis
interest given additional information from related sensors. An in the gearbox system. In order to categorize the status of the
anomaly is identified by the sustained difference between the gearbox as excellent or bad, two deep learning models Long
measurements and the estimate. The amount of variation Short-Term Memory (LSTM) and Bi-Directional Long Short-
required to identify a sensor defect is calculated using a Term Memory (BLSTM) are proposed in this study. These
decision tree, which is informed by integrating additional models are applied to time and frequency-domain data on
sensor measurement information. By using it to three test aircraft gearbox vibration. The effectiveness of the suggested
systems demonstrating different forms of sensor defects, models is assessed using a dataset on airplane gearbox
including data from actual commercial flights, an unstable vibration that is available to the public. The findings
aerodynamics model with dynamic stall, and a model for demonstrated that precision produced by LSTM and BLSTM
longitudinal flight dynamics induced by air turbulence, they is very reliable and applicable in the time domain as compared
validated the methodology. to the frequency domain in health monitoring of aircraft
gearbox systems.
Changrui et al [10], identified existing aircraft skin fault
detection as a low-efficiency level. Because of that, they Renee S. & Douglas L. [15] diagnosed that the pitot-static
proposed a YOLO4 object detection algorithm for aircraft skin system, which consists of two ports outside of the aircraft and
failure detection. Lightweight design with YOLO4 improved provides crucial airspeed information, is susceptible to
10% of detection speed. MobileNetV3 is used for the feature interference and malfunctions because of its location. An
extraction method of the network. From the basic YOLO airplane can be taught to autonomously identify mistakes in
algorithm to the sophisticated YOLO4, the YOLO series has faulty sensors and learn to repair them if it has access to the
been created. Because it can train and detect on a single GPU redundant sensor output. In this work, they created a unique
machine learning method for identifying sensor failures in

IJISRT23JAN072 www.ijisrt.com 595

Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
aircraft and forecasting using an online paradigm, to gather However, it is quite challenging to apply wave scattering
data. The status of the pitot-static system is categorized using theories to this section due to the geometrical and material
the autocorrelation of the incoming pressure data. complexity. This research team used a convolutional neural
network and a 1D-CNN algorithm for their application. Based
Ranting et al [16] discovered that the ultrasonic-guided on the training data that has been learned, the DL approach
wave transducer arrays can be used for the detection and automatically chooses the most sensitive wave characteristics.
localization of structural damage in a stiffened skin-to-stringer Additionally, the network's generalization skills enable the
composite panel typical of modern aircraft construction. detection of damage that may differ from the training instances.

Table 3 Comparative Study Of Reseracher Works

Research paper Author Type of Technique Algorithms and
name and year Type of dataset failure accuracy

Aircraft [7] engine Xuyum Fu Aircraft Feature vectors SVM: Error

fault detection Hui Luo Communication techniques with DAE+SVM: 76.86%
based on grouped Shisheng Zhong Addressing and supportive vector Partial dimension
convolutional Lin Lin Reporting System Aircraft machine and CDAE: 89.82%
denoising (ACARS) data of engine convolutional Full dimension
autoencoders (2018) CFM56-7B26 failures neural network CDAE: 95.18%
(CNN) algorithm Novel
are mainly used. method(propose):
96.16%
Recognition [8] of Caizhi Li 500 pictures are Internal Optimization Improved SSD:
the internal Xiaolong Wei collected of different failures algorithms and 96.5%
situation of aircraft Hanyi Guo shooting distances and using convolutional SSD: 95.1%
skin based on deep Weifeng He different shooting aircraft neural networks YOLOv3: 95.7%
learning (2021) Xin Wu angles. skin- Faster R-CNN: 97.7%
Hanjun Xu based
Xinyu Liu detection
Physics-informed Brain M. De 21 fights sensor data Machine learning This team recognized
machine [9] silva techniques with five types of faults.
learning for sensor Jared Callaham decision tree The accuracy levels
fault detection with Jonathan Jonker algorithm that they achieved for
flight test data Nicholas Goebel these failures are as
(2020) Jennifier Aircraft follows,
Klemisch sensor Catastrophic failure:
Darren Mc fault 97.32%
Donald detection Slow oscillation:
Nathan Kutz 96.72%
Steven L. Increased noise:
Brunton 97.30%
Alexsandra Slow drift: 97.84%
Aravlin All: 98.18%
Application [10] of Changrui Nong Not mentioned YOLOV4 object Mean average
light weight Jing Zhang detection algorithm precision
Aircraft
YOLOv4 in aircraft Zhenyu Liu MobileNetV3 YOLOV4: 56.75%
skin fault
skin fault detection Qungsong Zeng feature extraction YOLOV4-ML:
detection
(2022) Tianqui Zhang method 55.26%

Real-time [11] fault Murat Bronz Flight test logs Link: Supportive vector SVM: 95%
detection on small, Elgiz Baskaya https://github.com/mrtb Aircraft machine (SVM)
fixed wing UAVs Daniel Delehaye rnz/fault_detection wing algorithm
using machine Stephane faults
learning (2020) Puechmorel

IJISRT23JAN072 www.ijisrt.com 596

Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
A rare [12] failure Maren David Two datasets are used, Deep learning Accuracy for
detection model for Dangut 1. Aircraft central techniques using FIN_4000HA using
aircraft predictive Lan K. Jennions maintenance system auto-encoder and AE–BGRU
maintenance using a Steve King’ (ACMS) data. bidirectional gated and AE–CNN–
deep hybrid Zakwan Skaf 2.Aircraft maintenance recurrent unit BGRU model:
Aircraft
learning approach activities data (A330, network 99.74%
rare
(2022) A320) Accuracy for
faults
FIN_11HB using
AE–BGRU
and AE–CNN–
BGRU model:
99.93%
Real [13] time Qin Liu Not mentioned LSTM based FPGA-based neural
FPGA-based Tian Liang network with network: 99.5%
hardware neural Zhen Huang Electric different hyper
network for fault Venkata aircraft parameters.
detection and Dinavahi fault FPGA-based neural
isolation in more detection network.
electric aircraft
(2019)
Aircraft [14] PB 24 aircraft gearbox Deep learning LSTM in time
gearbox fault Mallikarjuna vibration datasets techniques-based domain: 98.38%
diagnosis system: Sreenatha M Aircraft LSTM and BLSTM BLSTM in time
An approach based Manjunath S gearbox based models domain: 99.75%
on deep learning Niranjan C fault LSTM in frequency
techniques (2020). Kundur detection domain: 93.25%
BLSTM in frequency
domain: 93.75%
A machine learning Renee Swischuk NASA aircraft dataset K-nearest neighbors Not mentioned.
Pitot
[15] approach to Douglas L. Steepest descent
static
aircraft sensor error Allaire algorithm
system
detection and Sammon’s mapping
failures
correction (2018)
Damage [16] Ranting Cui PZT excitation signal Deep learning Pristine-skin-
imaging in skin- Guillermo dataset convolutional stringer_Flage model:
stringer composite Azuara neural networks 99.86%
aircraft panel by Francesco Lanza Aircraft Stringer_cap model:
ultrasonic-guided di Scalea skin- 99.97%
waves using deep Eduardo Barrera based
learning with failures
convolutional
neural network
(2021)

After considering the above table, most researchers gave AI techniques for aircraft fault detection. Because aircraft
their focus on aircraft skin fault detection. Because aircraft skin systems are very difficult to understand, it is very hard to find
fault inspection is the main fault detection type that is used by datasets and there is high noise data in a dataset.
aircraft technicians. Aircraft engine fault detection is the
second most focused topic. Because the engine is the heart of IV. FUTURE WORK
an aircraft. If there is any failure in the engine, it will be a
reason to fail the whole system. Aircraft sensor fault detection After reviewing the past researchers' works and existing
is another important topic that most researchers gave their solutions, there are few drawbacks are identified. The main
attention. Aircraft sensors are very helpful for pilots to give a problem that was identified is that there is low attention given
report of aircraft conditions. Other focused areas are aircraft by researchers to fire detection systems of aircraft. Fire can
gearbox, wings, and rare failure detection. occur at any time because of internal and external faults.
Therefore, ensuring the aircraft and passengers' safety is a key
Most researchers used neural networks for their responsibility. There are five areas where fire and overheat
predictions. Because there is a high complexity of aircraft real- detectors are located. They are, wing and fuselage bleed air
time data. After considering past researchers' works and ducting, support pylons, wheel wells, APU, and cargo
existing solutions, there is a low level of attention to the use of compartments. This system works separately to detect fire.

IJISRT23JAN072 www.ijisrt.com 597

Volume 8, Issue 1, January – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
These systems diagnose fire by using simple concepts. Fire [8]. Caizhi Li, Xiaolong W., Hanyi G., Weifeng He, Xin Wu,
detection systems detect fire when the temperature is higher Haojun Xu, and Xinyu Liu. Recognition of the internal
than a specific temperature. This temperature is measured by situation of aircraft skin based on deep learning, 22
the manufacturing company and different aircraft have October 2021.
different temperature levels. Gas detectors are detected by [9]. P B Mallikarjuna, Sreenatha M., Manjunath S., and
using cameras. Sometimes, fire alarms are switch-on, on Niranjan C Kundur. Aircraft gearbox fault diagnosis
without having a fire. This simple concept is not good to system: an approach based on deep learning techniques,
diagnose fire. 21 May 2020.
[10]. Changrui N, Jing Z., Zhenyu L., Qingsong Z. and Tianqi
After considering the existing drawbacks, the proposed Zhang. Application of lighwight YOLOv4 in aircraft skin
solution is to introduce IoT and machine learning-based fault detection, 15 August 2022.
systems to detect fire at the early stage. The special feature is [11]. Murat B., Elgiz B., Daniel D., Stephane P., Real-time
this system can diagnose fire at any part of the aircraft with less fault detection on small fixed-wing UAVs using machine
error percentage. There are five places to detect the fire of an learning, 02 December 2020.
aircraft. This IoT-based device is attached to these five places [12]. Qin L., Tian L., Zhen H. and Venkata D.. R. A rare
and connects with the system. This proposed fire detection failure deteaction model for aircraft predictive
system works together with these five places and measures the maintenance using a deep hybrid learning approach, 2
required inputs using a sensor of an IoT device and diagnoses March 2022.
the fire at any time. The main advantages of this system are less [13]. Qin L., Tian L., Zhen H. and Venkata D.. R. Real-time
time is required to predict the result, accuracy is higher than the FPGA- based hardware neural network for fault
normal process and less money is required to manufacture this detection and isolation in more electric aircraft, 1
IoT-based device. November 2019.
[14]. Qin L., Tian L., Zhen H. and Venkata D.. R., Aircraft
V. CONCLUSION gearbox fault diagnosis system: an approach based on
deep learning techniques. 27 May 2020.
This review paper's main goal is to identify the AI [15]. Amare D., Valentina Z. and Konstations K.. Aircraft
strategies of the aviation industry. After considering past engine performance monitoring and diagnostics based on
researchers' work and existing solutions, the main thing that I deep convolutional netural networks. [online], 2021
noted is that the aviation industry is getting 50% of the [16]. Ranting C, Guillermo A., Francesco L. and Eduardo B.,
advantages of AI. But they are not getting 100% of 2021. Damage imaging in skin-stringer composite
effectiveness. There are few fields that are still available to use aircraft panel by ultrasonic-guided waves using deep
AI techniques. In my point of view, the aviation industry can learning with convolutional neural network. [online],
get fast develop by using AI. 2021

REFERENCES

[1]. Airbus, “A380, unique passenger experience”, 2022

[online], Available: A380 | Airbus
[2]. Popular Mechanics, “13 infamous plane crashes that
changes aviation forever”, [online], Available: Airplane
Accidents: 13 Famous Plane Crashes That Changed
Aviation (popularmechanics.com)
[3]. Glenn Research center, “Jet Engines Mature”, [online],
Available: Jet Engines Mature - Glenn Research Center |
NASA
[4]. IBM(2020). Artificial Intelligence. [online] Available
from:< What is Artificial Intelligence (AI)? | IBM>
[Accessed 03 June 2020]
[5]. Aeroclass.org, “Artificial intelligence- AI in Aviation”,
[online], Available: Jet Engines Mature - Glenn Research
Center | NASA
[6]. Batta Mahesh (2018). Machine learning Algorithms- A
Review. [online]researchgate.com . Accessed 01 January
2020
[7]. Xuyun FU, Hui LUO, Shisheng ZHONG and Ling LIN.
Aircraft engine fault detection based on grouped
convolutional denoising autoencoders, 05 January 2019.

IJISRT23JAN072 www.ijisrt.com 598