Due to the heavy usage of Unmanned Aerial Vehicles (UAVs) and the co-evolution of modern technologies, a crucial introduction to fault diagnosis has taken place in recent studies in the avoidance of ravaging consequences. Machine Learning techniques are one of the other major fault-diagnosing approaches in the field of Artificial Intelligence. This review article delivers an elaborated overview of the latest studies concerning UAVs fault diagnosis utilizing Machine Learning and Deep Learning techniques. A summarized comparison of the diferent methods is distinguishably elaborated where the conclusion highlights that research on fault diagnosis systems is progressing and yet to end. Consideration should be given to a growing number of research and methodologies.
of variant applications have escalated rapidly in recent
years [
Fault diagnosis means diagnosing the event of defi- scenarios of security monitoring wildfires using machine ciencies within the utilitarian units of the process, which learning methods have demonstrated the efectiveness of leads to undesired or intolerable behavior of the com- fault detection in many aspects [22]. Another application plete framework. Studies and reviews on fault diagnosis is the detection of the disastrous citrus greening, where drones proved to be more eficient regarding inspections SYSYEM 2022: 8th Scholar’s Yearly Symposium of Technology, Engi- due to their wide coverage. Machine learning methods neering and Mathematics, Brunek, July 23, 2022 for citrus greening diagnosis were discussed, compared, " Luttfi.A.AlHaddad@uotechnology.edu.iq (L. A. Al-Haddad); and elaborated on, demonstrating their high accuracy in Ala0a0.A0 0.J-a0b0e0r1@-5u7o0t9e-c1h9n5Xolo(Agy..Jeadbue.riq) (A. Jaber) fault diagnosis [23, 24? ].
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g ACttEribUutRion W4.0oInrtekrnsahtioonpal (PCCroBYce4.0e).dings (CEUR-WS.org)
2.1.1. Convolutional Neural Network CNN
other neural networks, while the vibrations signals in the airframe were recorded experimentally and translated into codes using the fault diagnosis method. A fixedwing UAV fault diagnosis system based on five models, one of which was a long and short-term memory neural network [41]. Convenient predictions were provided based on numerical simulations.
In [39], the authors have developed and upgraded the used neural networks. Damaged blades in a quadcopter diagnosis based on a deep residual shrinkage network and an extra convolution layer have both emerged to produce an upgraded neural network algorithm named 1D-WIDRSN. The experimental statistical analysis has shown the efectiveness and accuracy of the hybrid algo2.1.3. Radial Basis Function Neural Network RBF rithm compared to normal neural networks. have used a
NN back propagation neural network (BPNN) as a machine
In [40], the authors have introduced a fault-tolerant con- learning method to diagnose faults in a sensor of a
quadtrol approach to detect actuator faults in a quadcopter. A copter. Then, it was optimized by a genetic algorithm
normal adaptive sliding mode control is combined with to fasten the convergence. The results are shown
exa radial basis function neural network, introducing a perimentally, supporting that enhanced BPNN is more
modified adaptive sliding mode control approach. An eficient in fault diagnosis. A strategy of scattered
faultexperimental, numerical comparison between the two is tolerant cooperative control to acquire a synchronized
elaborated, showing the significant role of a radial basis track control of UAVs was introduced in [47] by using
function network. The authors of [
ML Method CNN 17.4% K-NN 17.4% SVM 13%
ANN 8.7% LSTM NN 8.7% RBF NN 8.7%
DT 8.7%
SOM 4.3% Other NN 13%
UAV Type Fault Detection Part
Blades 32% Wing 16% Motor 12%
Actuators 8% Motor Bearing 8%
Sensors 8% Others 16% 2.2. K Nearest Neighbor K-NN assembling method to exhibit a model for diagnosing health status in a quadcopter UAV. Vibration features of three flight conditions (normal, motor base mount loose, unbalanced broken blades) were recorded and trained in a system that has assembled variant vibration patterns of fault situations. The experimental results have proved that the method can also predict the occurrence of the fault, not only diagnose it.
nosing and classification system for a fixed-wing UAV.
The system includes a description of data analysis from a
piezoelectric sensor system with independent component
analysis and machine learning methods. One of which
was the subspace K-nearest neighbor with the best results
and accuracy. In [34], variant faults in a quadcopter UAV
were examined in a fault diagnosis system. Damaged
parts were blades, armature eccentric, and motor base 2.4. Support Vector Machine SVM
loosening. Pulse and vibration signals were recorded and SVM is used in many diferent aspects [ 51]. In [42], the
analyzed using a machine learning method employing authors simulated an aircraft model and utilized it to
K-NN. Experimental results demonstrate the high efi- generate data and test some designed algorithms. The
ciency of the used method. Authors of [37] suggest an simulated measurements were collected from random
innovative system for fault diagnosis of fixed-wing UAVs lfight data. A supervised fault diagnosing method based
(FW-UAVs), where the procedure dynamics, operation on SVM was utilized to identify the faulty and nominal
conditions, changing data density, and procedure dis- flight states in loss of efectiveness in control surfaces
turbance are evaluated. A modified algorithm utilizing of a drone UAV. Results encourage the use of SVM in
Shared Nearest Neighbor based Distance (SNND) and fault diagnosis due to accurate and efective acquired
a K-Nearest Neighbor algorithm hiring SNND (SNND- accuracy. Furthermore, as discussed in subsection 2.2.
KNN) was proposed to realize ofline operation condition the authors of [
[30] and [44] have adopted the decision tree method in machine learning fault diagnosis where Gradient-based decision tree have showed better accuracy over normal decision tree machine learning methods.
on this review article, we have concluded the following: Adding up, the developing request for secure flights of unmanned aerial vehicles requires modern and worldlywise fault diagnosis methods not only for faults in blades and wings but also in other UAV subsystems. In this respect, a promising approach that appears to have captured the attention of researchers in recent years is the hybrid fault diagnosis methods that delicately address the undesired behavior of an unmanned aerial vehicle based on combined machine learning techniques or/with signal processing for important feature extraction.