ArticlePDF Available

Are Modern Market-Available Multi-Rotor Drones Ready to Automatically Inspect Industrial Facilities?

Authors:

Abstract

Industrial inspection is a well-known application area for unmanned aerial vehicles (UAVs), but are modern market-available drones fully suitable for inspections of larger-scale industrial facilities? This review summarizes the pros and cons of aerial large-scale facility inspection, distinguishing it from other inspection scenarios implemented with drones. Moreover, based on paper analysis and additionally performed experimental studies, it reveals specific issues related to modern commercial drone software and demonstrates that market-available UAVs (including DJI and Autel Robotics) more or less suffer from the same problems. The discovered issues include a Global Navigation Satellite System (GNSS) Real Time Kinematic (RTK) shift, an identification of multiple images captured from the same point, limitations of custom mission generation with external tools and mission length, an incorrect flight time prediction, an unpredictable time of reaching a waypoint with a small radius, deviation from the pre-planned route line between two waypoints, a high pitch angle during acceleration/deceleration, an automatic landing cancellation in a strong wind, and flight monitoring issues related to ground station software. Finally, on the basis of the paper review, we propose solutions to these issues, which helped us overcome them during the first autonomous inspection of a 2400 megawatts thermal power plant.
Citation: Gyrichidi, N.; Khalyasmaa, A.;
Eroshenko, S.; Romanov, A. Are
Modern Market-Available Multi-Rotor
Drones Ready to Automatically
Inspect Industrial Facilities? Drones
2024,8, 549. https://doi.org/
10.3390/drones8100549
Academic Editors: Dayi Zhang,
Gordon Dobie and Jesus Enrique
Sierra-Garcia
Received: 13 September 2024
Revised: 30 September 2024
Accepted: 1 October 2024
Published: 3 October 2024
Copyright: © 2024 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
drones
Review
Are Modern Market-Available Multi-Rotor Drones Ready
to Automatically Inspect Industrial Facilities?
Ntmitrii Gyrichidi 1, Alexandra Khalyasmaa 2, Stanislav Eroshenko 2and Alexey Romanov 1,*
1Institute of Artificial Intelligence, MIREA—Russian Technological University (RTU MIREA),
Moscow 119454, Russia; girikhidi0@gmail.com
2Ural Power Engineering Institute, Ural Federal University Named after the First President of Russia
B.N. Yeltsin (UrFU), Ekaterinburg 620002, Russia; a.i.khaliasmaa@urfu.ru (A.K.); s.a.eroshenko@urfu.ru (S.E.)
*Correspondence: romanov@mirea.ru
Abstract: Industrial inspection is a well-known application area for unmanned aerial vehicles (UAVs),
but are modern market-available drones fully suitable for inspections of larger-scale industrial facili-
ties? This review summarizes the pros and cons of aerial large-scale facility inspection, distinguishing
it from other inspection scenarios implemented with drones. Moreover, based on paper analysis and
additionally performed experimental studies, it reveals specific issues related to modern commercial
drone software and demonstrates that market-available UAVs (including DJI and Autel Robotics)
more or less suffer from the same problems. The discovered issues include a Global Navigation
Satellite System (GNSS) Real Time Kinematic (RTK) shift, an identification of multiple images cap-
tured from the same point, limitations of custom mission generation with external tools and mission
length, an incorrect flight time prediction, an unpredictable time of reaching a waypoint with a small
radius, deviation from the pre-planned route line between two waypoints, a high pitch angle during
acceleration/deceleration, an automatic landing cancellation in a strong wind, and flight monitoring
issues related to ground station software. Finally, on the basis of the paper review, we propose
solutions to these issues, which helped us overcome them during the first autonomous inspection of
a 2400 megawatts thermal power plant.
Keywords: industrial inspection; unmanned aerial vehicle; multi-rotor; drone; ground station
software; mission planning software; flight control
1. Introduction
Industrial inspection is a well-known application area for unmanned aerial vehicles
(UAVs) [
1
3
]. They are widely used for inspection of pipelines and overhead powerlines.
However, are modern market-available industrial drones fully suitable for inspections of
larger-scale industrial facilities? Let us discuss the power plant inspection task, to better
understand the challenges of larger-scale industrial facility inspection.
Industrial inspection with UAVs is a well-known approach. Moreover, large drone
manufacturers like DJI or Autel Robotics have drones specifically designed for this task
in their product range. Figure 1shows some examples of inspection flights performed on
large industrial facilities. As can be seen, these tasks require flying in an area with many
obstacles, very close to the equipment being inspected and sometimes even touching it with
the drone’s sensors. Not surprisingly, in most cases, the inspection drones are still used in
tele-operated mode. They serve as flying cameras, helping diagnostic engineers watch and
check the desired equipment from the best angle and a closer distance. However, manual
control has multiple limitations. First, images of the same equipment captured on different
days have low repeatability regarding angles and distance, making it difficult to process
them automatically. Second, in large industrial facilities with many wires, buildings, pipes,
and other obstacles, most pilots have to clearly see the drone with their eyes to guarantee
Drones 2024,8, 549. https://doi.org/10.3390/drones8100549 https://www.mdpi.com/journal/drones
Drones 2024,8, 549 2 of 34
safe flight, even if the drone is equipped with a first person view (FPV) camera. Thus,
each inspection flight is limited to a small area around the pilot, where the drone can
be clearly distinguished from obstacles. After each flight, the operator typically changes
locations, adding time to the process. In addition, the images captured during such a
manual inspection have low repeatability, making them difficult to interpret and compare
to data collected during previous inspections. As a result, it can take several days or even
weeks to inspect a large facility with a manually piloted drone.
(a)
(b)
(c)
Figure 1. Examples of aerial inspection on large-scale industrial facilities: (a) drone inspection tasks
in a cement production factory [
4
]; (b) a set of inspection routes on a large power plant [
5
] and an
industrial drone during non-contact inspection of insulators on the one of these routes [
5
]; (c) an
industrial drone during contact inspection on a refinery [6].
Excluding the smallest ones, modern industrial drones have the functionality to
automatically follow a route composed from waypoints with pre-defined geographical
coordinates. Moreover, at each waypoint, most of them can orient a camera and capture
images with it. Thanks to advances in Global Navigation Satellite System (GNSS) Real-
Time Kinematic (RTK) positioning, drones can determine their position with decimetre
Drones 2024,8, 549 3 of 34
accuracy [
7
]. The heading can be estimated using the so-called GNSS compass consisting
of 2 GNSS receivers. With a distance between antennas of 1 m, it provides an error below
0.2
[
8
] and is much less influenced by electromagnetic interference from high voltage
equipment. The automatic flights are currently used to inspect overhead power lines [
9
],
solar power plants [
10
], and even small power plants [
11
]. In all these cases, the drone flight
is either around the inspection area or at the altitude above the highest obstacle. Inspecting
large-scale industrial facilities, such as 2400 MW thermal power plants, requires flying
inside the inspection area between many obstacles, whose height is often above the drone’s
altitude. Moreover, in contrast to overhead line inspection, only a few areas are usually
suitable for safe takeoff and landing. This task can be solved by multi-criteria mission
planning, simultaneously considering obstacles’ position, radio communication availability,
air turbulence, and flight time [
5
]. Still, the drone should be able to accurately and safely
complete the planned missions.
The benefits of unmanned aerial inspection are undoubted. They have already been
published in many research and review papers. This paper aims to summarize the chal-
lenges related to aerial inspection performed using market-available drones and connect
them with specific solutions that will help overcome them or at least minimize related neg-
ative consequences. The review is specifically focused on market-ready products instead
of research prototypes, because insurance companies avoid insuring the latter or provide
insurance at much higher prices. At the same time, industrial facility owners usually strictly
require insurance to secure UAV damage and third-party liability risk when performing
aerial inspections.
The main contributions of this review are the following: (1) we summarized the pros
and cons of aerial large-scale facility inspection, distinguishing it from other inspection
scenarios implemented with drones; (2) we revealed specific issues related to modern
commercial drone software and demonstrated that even the market leaders’ products suffer
from them; (3) we proposed solutions to these issues, which helped us overcome them
during the first autonomous inspection of a 2400 MW thermal power plant.
This review shows how drone developers can improve their products, making them
more suitable for large-scale facility inspection. At the same time, it also demonstrates to
researchers and inspection engineers what issues they will probably face while inspecting
large-scale facilities using the industrial drones currently available on the market.
The rest of the paper is organized as follows. Section 2describes the paper selection
procedure. The analysis of the selected papers is provided in Section 3. Section 4discusses
the specific issues related to the market-available drone software and proposed solutions for
each of them. The review results are discussed in Section 5. Findings and recommendations
are summarized in Section 6.
2. Paper Selection
The paper selection procedure was performed using the Scopus database and the
“inspection AND drone” prompt without any other limitations. The initial search resulted in
2025 papers published from 1985. The results included 729 research papers, 1168 conference
papers and conference reviews, 65 reviews and short surveys, 49 books and book chapters,
and 13 documents of other types, including erratum, notes, and retracted papers. Figure 2
demonstrates distribution of these papers relative to the year of their publication. It can be
clearly seen that interest in inspection drones has constantly risen since 2014. Considering
that the initial selection already includes 65 review papers and short surveys, most of which
were published during the last ten years, it was decided to focus our analysis on these
surveys. Review papers marked as books and book chapters in the Scopus database were
also analyzed. We used two simple inclusion criteria: the paper should be dedicated to
multi-rotor drones and inspection of local industrial facilities and buildings. The review
is narrowed to multi-rotor systems because using other UAV types is significantly more
complicated due to the large number of obstacles in large-scale industrial facilities.
Drones 2024,8, 549 4 of 34
Figure 2. Distribution of drone-based inspection papers by year of publication: (a) all papers;
(b) journal research papers; (c) conference papers and conference reviews; (d) journal review papers;
(e) books and book chapters; (f) other documents, including erratum, notes, and retracted papers.
The linear facilities, like overhead powerlines, pipelines, railways, and roads, were
excluded because their monitoring requirements and complexity significantly differ from
the inspection of large-scale industrial facilities [
5
]. Moreover, considering the long distances
that should be traveled during such types of inspection, it is more promising to perform
them by fixed-wing UAVs [
12
]. Three specific types of local facilities were also excluded for
the same reason: agricultural fields, photovoltaic power plants, and bridges. Inspecting the
first two types of facilities is significantly simpler than buildings, factories, or thermal power
plants because it can be performed from a high altitude, making the problems of obstacles
and interference negligible. Thus, it can be performed by any drone suitable for remote
sensing [13].
On the contrary, the inspection of bridges is comparable and even sometimes more
complicated than the inspection of the other types of local industrial facilities. At the same
time, the set of problems in these cases differs from, e.g., power plant inspection. According
Drones 2024,8, 549 5 of 34
to a technical report of the Minnesota Department of Transportation [
14
], one of the main
challenges in bridge inspection is non-GPS navigation, which makes this case closer to
indoor applications of drones. However, indoor inspection is out of the scope of this paper.
Another exclusion criterion is that the application area is different from industrial
inspections. It was used to exclude papers on wild animal monitoring, military applications,
counter-drone systems, and aerial object manipulation. Also, a set of criteria was used to
focus our review on the capabilities of the drones as final market products. Thus, papers
reviewing drone components and algorithms without relating them to any specific drones
were also excluded. These papers were not used in the analysis of large-scale industrial
facilities monitoring pros and cons, but some of them were further used to propose solutions
to the revealed issues of the modern market-available UAVs.
Finally, we excluded papers focusing on the legal aspects of drone applications. Legal
issues usually differ depending on the laws of specific countries, while our review aims to
focus on engineering aspects of large-scale facility drone inspection.
The overall paper selection procedure is illustrated by the Figure 3. The initial selection
of 2025 papers was narrowed to 114 by limiting the document types to review, short survey,
book, and book chapter. Then, based on the chosen criteria and abstract analysis, another
75 papers were excluded from the list. Finally, after analyzing the remaining paper’s full
texts, we excluded 19 documents. The final selection includes 18 papers published from
2016 to 2024 [
1
,
15
31
], most of which were published in 2023. These papers were used to
summarize the pros and cons of aerial large-scale facility inspection, distinguishing it from
other inspection scenarios implemented with drones. The results of the performed analysis
are presented in the next section.
Scopus search promt: inspection AND drone
Limiters: published before September 2024
Special limiters by Scopus Document type:
Review, Book, Book Chapter, Short Survey
Titles found (n = 2025)
Titles limited (n = 114)
Excluded by abstract (n = 75)
Included for full-text analysis (n = 39)
Final selected papers (n = 18)
Excluded by full-text (n = 19)
Inclusion criteria:
- dedicated to multi-rotor drones
- dedicated to inspection of local industrial
facilities and buildings
Exclusion criteria:
- not a review paper (according to its content)
- not related to multi-rotor aerial drones
- dedicated to inspection of linear facilities
industrial facilities
- dedicated to inspection of agricultural fields,
photovoltaic power plants or bridges
- not related to industrial inspection
- dedicated to drone components (with no relation
to any specific drones)
- dedicated to algorithms (with no relation
to any specific drones)
- dedicated to legal aspects of drone applications
Figure 3. Paper selection procedure.
Drones 2024,8, 549 6 of 34
3. The Analysis of the Selected Papers
The analysis of the selected papers revealed the following key advantages of aerial
inspection performed with multi-rotor drones:
Cost and speed:
it is cheaper and faster than conventional inspection me-
thods [1,17,21,26,27,29].
Efficiency:
it provides more meaningful results, which are easier to interpret [
21
]. Moreover,
it allows the conducting of inspections more frequently, providing more accurate
failure predictions [20,26,30].
Safety: it provides a higher level of safety due to elimination or at least reduction of time
and personnel spending in the areas with increased danger, such as open distribution
devices or pipelines [15,25].
Enhanced analysis capabilities:
UAVs capture data from closer distances and in larger
quantities with better repeatability. Thus, these data become suitable for analysis using
neural networks and other machine-learning techniques [
1
,
16
,
27
,
28
]. Automated
intelligent analysis speeds up the process and eliminates human factors, such as
subjective opinions, providing better failure prediction accuracy.
Non-stop operation of the industrial facility:
in most cases, UAV inspection can be per-
formed on active equipment without stopping production processes on industrial
facilities [
31
]. Such capability indirectly affects the cost-efficiency of the aerial inspec-
tion by reducing losses caused by equipment downtime.
All the advantages can be summarized as better cost-efficiency, higher inspection
quality, and better safety. The discussion section provides a critical analysis of them.
The main challenges of the aerial inspection on large-scale industrial facilities, accord-
ing to the reviewed papers, are the following:
Low battery life:
modern multi-rotor drones’ maximum flight time rarely exceeds 1 h.
In most cases, the actual flight time with payload is below 30 min. This challenge
significantly increases the importance of optimal flight planning that minimizes the
number of battery replacements required in order to complete all the inspection
tasks [19,26,27,29].
Limited payload
is another challenge of aerial inspection with multi-rotor drones [
15
,
26
,
29
],
which limits the amount of inspection equipment installed on a drone in each fight.
Interference from surrounding objects in inspection data:
surrounding objects may in-
fluence the diagnostic equipment of the drone. For example, they can create reflections
that distort the images of thermal cameras. This challenge is mentioned in [
1
], where
the authors suggest recording data from different fields of view as a solution. At the
same time, it should be noted that this challenge is specific to the inspection of indus-
trial facilities in general, not only to aerial inspection. Moreover, because multi-rotor
drones can operate on short distances from the inspected equipment, the data they
collect are less influenced than those obtained from the ground.
High risk of collision
due to many obstacles and a short distance to the inspected equip-
ment during data acquisition [21].
Weather limitations:
all aerial vehicles have weather limitations, and UAVs are no excep-
tion. In some weather conditions, the inspection performed with the drone is not
allowed due to safety reasons, while in other cases, such weather phenomena as air
turbulence influence the inspection results [1,2225].
Operator competence:
the importance of eliminating human errors and increasing the
competence of UAV operators is declared in [18,22].
Regulation issues:
current regulations of airspace usage are not designed for frequent
UAV usage, especially on industrial facilities, which are usually located in restricted
areas with restricted airspaces, leading to a long legal process in order to gain flight
permission [15,18,22,28].
After analysis of the above challenges of large-scale industrial facility inspection, it can
be noted that all of them correspond to multi-rotor drones and industrial facility inspection
Drones 2024,8, 549 7 of 34
in general. Thus, the reviewed surveys may create an illusion that that a complete set of
the published technologies is already available on the market. The illusion is caused by
the fact that, while preparing a review, researchers usually assume that if there is at least
one paper that describes a solution to some problem, then the problem is solved, while the
path from the research paper to a market product is rarely straightforward. At the same
time, application engineers rarely use research prototypes in their solutions, preferring
market-available products to provide end-customer service, predictable reliability, insur-
ance availability, etc. To avoid such illusions in our review, we analyze issues related to
modern commercial drone software in the next section. We show that all of them have
a solution known from the state of the art, but in many cases, these solutions are not yet
implemented in the market products.
4. The Issues Related to Modern Commercial Drone Software
It is relatively easy to summarize modern science and technology advances because
they are the main focus of most research papers. It is much more difficult to reveal the
drawbacks of market-available products because their manufacturers often try to mask
them. Thus, in this part of our review, we mostly rely on the results of our own experimental
studies performed on different drones. Whenever possible, we support our findings with
results that can be found in the papers of other authors.
During the first autonomous inspection of a 2400 MW thermal power plant using
a commercial UAV [
5
], we discovered multiple issues that were negligible for general
remote sensing and monitoring applications but became critical in the case of large-scale
industrial facility inspection. Further, we tested the market-available drones from other
manufacturers and found that even the market leaders’ products suffer from the same
problems, more or less. Thus, we understood that it is essential to focus the attention of
manufacturers and application engineers on these issues. The results of our analysis are
presented below.
The experimental studies reported in [
5
] were performed using a R.A.L. X6 hexacoper
(RusAeroLab LLC, Moscow, Russia, https://rusaerolab.ru/products_ral_x6.html, accessed
on 6 September 2024) (Figure 4). According to the datasheet provided by the manufac-
turer, this drone has following parameters: maximum flight distance—30 km; maximum
reachable altitude—7 km; effective flight altitude—1 km; maximum flight time—1 h (no
load)/45 min (1 kg load)/20 min (5 kg load); maximum wind gusts—25 m/s; temperature
range—
30–60
C; acceptable weather conditions—wind, snow, rain. The distance be-
tween opposite motors is 940 mm. The drone is equipped with Trimble MB-Two receivers
that estimate position and heading. For this purpose, two GNSS antennas are mounted
above the motors on opposite sides of the drone. Another Trimble MB-Two is connected to
the ground station and serves as an RTK base station. The overall GNSS navigation system
is similar to the one used on the mobile robot presented in [32].
The drone and ground station communication is performed using three R.A.L. 01-002
mesh radio modems. One was installed on the ground station, the other was mounted
on the roof of the main machinery building, and the third was manufactured in a light
version integrated with the drone’s control system. According to their datasheet, these radio
modems manage to download data from the drone with a 1.2–12 Mbps rate and upload
with a data rate of 0.115–1 Mbps in a range of 20 km (line of sight). The radio modem
mounted on the roof was used as a retransmitter to enlarge radio availability zones [5].
The R.A.L. X6 in the current research was equipped with a dual-specter RGB/IR cam-
era (GIT UR-640D) mounted on a three-axis gyro-stabilized gimbal. This camera is connected
to the drone’s control system with the Universal Asynchronous Receiver/Transmitter (UART)
interface, making it possible to trigger image and video capturing. The camera has a resolution
of 4608
×
3456 and 640
×
480 pixels for RGB and IR channels, respectively. The IR channel’s
field of view is 32
×
24
, which defines the effective image capturing distance in a range of
15–30 m.
Drones 2024,8, 549 8 of 34
Figure 4. R.A.L. X6 equipped with GIT UR-640D camera.
The R.A.L X6 drone is based on proprietary software and hardware designed by
its manufacturer. It is incompatible with open-source mission-planning software, like
QGroundControl (http://qgroundcontrol.com/). Instead, it uses the proprietary RALTool
application for configuration, mission planning, and control during the flight.
It is worth mentioning that the analysis performed before the research reported in [
5
]
showed that the R.A.L. X6 should be suitable to perform missions on the reference power
plant: it can fly in strong winds, supports automatic missions with gimbal and camera
control commands, is equipped with precise GNSS RTK position and heading estimation
hardware, and can communicate with the ground station through distributed mesh network,
making it possible to maximize radio coverage in the power plant area. (Without radio
coverage, the drone cannot receive GNSS RTK corrections from the base station, which are
required for precise positioning). Moreover, it was successfully operated on a substation and
used for overhead powerline inspection, which supported the manufacturer’s statement
that the drone’s navigation system is resistant to electromagnetic interference from high-
voltage equipment. At the same time, as mentioned above, the conditions in large-scale
power plants are much more demanding than in the mentioned usage cases.
Figure 5demonstrates other multi-rotor aerial systems discussed in this section. They
include those produced by DJI and Autel Robotics and those based on ArduPilot, PX4, and
INAV open autopilots.
Figure 5. Examples of multi-rotor aerial systems discussed in current research: (a) DJI Matrice 300 RTK
with DJI Zenmuse L1 laser scanner [
33
]; (b) Autel Robotics EVO II Pro [
34
]; (c) Voljet M690 PRO
drone, which is equipped with Pixhawk 2.1 ORANGE flight controller, compatible with ArduPilot/PX4
firmware [35]; (d) a 24” heavy drone equipped with a controller running INAV [36].
Drones 2024,8, 549 9 of 34
DJI is a market leader [
37
] in both industrial and entertainment drones. Their drones
are widely used in many industrial inspection applications [
9
,
38
,
39
]. The drones of the
Mavic and Phantom series are usually operated manually with partial automation. In
contrast, the Matrice series is often used automatically when a drone flies along the pre-
programmed route. Also, for Matrice 30, DJI Mavic 3E, DJI Mavic 3T, Matrice 300 RTK,
and Matrice 350 RTK drones, DJI provides advanced FlightHub 2 [
40
] mission planning
and ground station software (FlightHub 2 2024-9-24 release version, SZ DJI Technology
Co., Ltd, Shenzhen, China). It also provides DJI Fly software that is installed on DJI RC
controllers (DJI Fly V 1.13.10 version, SZ DJI Technology Co., Ltd, Shenzhen, China).
Autel Robotics is another well-known manufacturer of industrial drones, whose prod-
ucts (especially EVO II, which was used in experimental studies, performed in current research,
and its modifications) are also a popular choice for aerial inspection missions [38,41,42].
ArduPilot and PX4 are two well-known open-source firmware for multi-rotors [
43
].
They both are designed for the same hardware evolved from PixHawk flight controller [
44
].
Initially, this was designed for hobby projects. However, currently, CUAV and CubePilot
produce industrial-grade controllers compatible with ArduPilot and PX4, which are used
in many market-available drones manufactured by mid-size companies [45,46].
INAV Multicopter is one of the most recent open-source drone firmwares [
47
]. Having
the same ancestor, it is close to BetaFlight but has improved navigation functionalities and
ground station software supporting mission planning [
48
]. INAV firmware is widely used
for fixed-wing and vertical takeoff and landing (VTOL) aircraft [
49
,
50
]. At the same time, it
supports mulitcopters [
51
] and is used by several small companies around the world to
manufacture low-cost thermal inspection industrial drones for local markets.
UgCS from SPH Engineering, a universal ground station software. In our research,
we analyze its capabilities in addition to the described above multi-rotor systems be-
cause it is compatible with multiple drones from different vendors (DJI, Autel Robotics,
ArduPilot/PX4-based, etc.) and widely used for industrial inspection [46,5254].
The rest of the section is dedicated to the description of the revealed issues and
corresponding solutions.
4.1. An RTK Shift
4.1.1. An Issue Description
GNSS RTK is widely used in industrial inspection to provide centimeter-range posi-
tioning precision. It requires a base receiver with a static antenna with known geographic
coordinates to operate [
55
]. The receiver itself can determine the position of this antenna
following a so-called “survey-in” procedure or a geodetic survey. The second approach is
generally preferable because it provides the drone’s position in the same coordinate system
used in construction plans and geo-referenced 3D models. Thus, it is possible to plan the
routes offline using those plans and models, and then the drone will fly to the right points.
During the experimental studies, it was noticed that when the RTK base antenna po-
sition was defined using the geodetic survey results, the drone performed undesired
maneuvers in case of radio connection loss. Specifically, it changed its altitude and
shifted to the west by several meters. Then, when the connection was restored, the
drone shifted back. Moreover, the maneuvers were very similar regardless of the place of
communication failure.
We assumed that the absence of RTK correction caused the problem. At the same time,
it needed to be clarified why the drone constantly shifts in the same direction and at the
same distance. Everything became apparent after the position of the RTK base antenna
was determined by a survey-in procedure. (The survey-in procedure was performed until
0.2 m accuracy was reached. The estimated position of the antenna lay approximately
2.5 m to the east relative to the coordinates provided by a geodetic engineer. We repeated
the survey-in procedure multiple times in the following days, and all the results included
similar additive errors (Figure 6).
Drones 2024,8, 549 10 of 34
This error describes the undesired behavior after connection loss: the drone stops
receiving RTK corrections from the base receiver, and its reference system shifts 2.5m,
which causes movement aimed to compensate for this shift. Then, when communication
is restored, the process repeats backward. At the same time, it is essential to understand
the cause of the additive error to avoid it. Such a shift between geodetic and survey-in
coordinates is hard to describe with known sources of the GNSS errors [
56
] as it stays
close to constant during long periods and appears even if the survey-in is performed
several minutes after the geodetic survey. The clue was found when the geodetic engineer
was asked to complete a survey without receiving corrections from the local Networked
Transport of RTCM via Internet Protocol (NTRIP) server: the difference from the receiver
of the drone’s static base reduced to a centimeter level. Thus, the observed RTK shift was
introduced by the NTRIP corrections during the geodetic survey.
0 0.5 1 1.5 2 2.5 3
-0.5
0
0.5
1
1.5
x, m
y, m
Figure 6. Position of the RTK base antenna determent by the receiver itself during survey-in proce-
dures relative to the position provided by geodetic survey results: red circles—results of multiple
survey-in procedures performed in different days; blue star—mean positions evaluated based on all
the survey-in results.
As it recovered due to federal law, companies owning NTRIP stations in our coun-
try have to configure their coordinates with the ones defined using ellipsoids different
from WGS84. The reason is that the results of geodetic surveys performed with these
corrections should match existing geodetic plans. Due to historical reasons, in ex-USSR
countries, geodetic plans were created in the SK-42 reference system based on Krasovsky
1940 ellipsoid. This ellipsoid is still actively used in many cartographic projections in
Russia and ex-USSR countries. At the same time, most modern GNSS receivers use the
WGS84 ellipsoid, which has different parameters from Krasovsky 1940. Thus, latitude and
longitude parameters obtained using different ellipsoids should be appropriately converted
into one coordinate system before use. This problem was mentioned multiple times in
scientific papers on geography, geology, and geophysics [
57
60
]. Nevertheless, these papers
were published in specialized journals, which drone or even geodetic engineers rarely read.
Thus, there is no tool for RTK compensation in market-available inspection drones.
4.1.2. A Solution
The straightforward solution for the issue is configuring the RTK base location on
the ground station using the survey-in procedure result or the geodetic survey performed
without NTRIP corrections. However, in this case, the reference system of the drone will
constantly shift relative to the existing construction plans and geo-referenced 3D models,
making it impossible to use them while planning the routes. Meanwhile, the constant
shift of the RTK base leads to the same shift of the rover position [
61
]. Thus, it can be
compensated during the mission planning by adding difference between latitude and
longitude determent by the receiver during the survey-in procedure and the ones provided
as the result of the geodetic survey to the corresponding coordinates of each inspection
route’s point. This approach is easy to implement for automatic mission generation but
can be complicated during manual planning. Due to the RTK shift problem being relevant
for most of the ex-USSR countries described above, we convinced the manufacturer to
integrate the shift compensation into RALTool software. Thus, the mission can be planned
Drones 2024,8, 549 11 of 34
according to standard geo-referenced models, and the additive corrections to the drone
reference system are configured as two parameters on the ground station.
4.2. An Identification of Multiple Images Captured from the Same Point
4.2.1. An Issue Description
An industrial drone can take over a hundred images in a single flight. Moreover, these
images often contain similar equipment that is difficult to distinguish from each other
during post-processing (Figure 7). The base version of the R.A.L. X6 firmware was naming
all the images sequentially. It was also able to store a timestamp (in seconds) and GNSS
coordinates of the drone in the image metadata.
Figure 7. Similar IR images (raw) of different insulators captured during an inspection mission.
It was soon discovered that the camera may fail to make an image for different
reasons, even if triggered. In such rare cases, the camera proceeded to name files from the
identifier of the last successful capture, making it hard to match the planned and actually
recorded data.
A typical task during the inspection of a large-scale facility is capturing many parts
of equipment from a single point [
5
]. At the same time, as noted above, these equipment
parts are very similar. Thus, GNSS metadata only partially solved the problem of robustly
distinguishing images from each other. Finally, the drone takes multiple shots per second
to minimize downtime during missions while its gimbal rotates at an angular speed of
0.7 rad/s. As a result, the 1-second accuracy is not sufficient to distinguish images of two
closely aligned pieces of equipment from the multiple shots of a single object.
4.2.2. A Solution
The R.A.L. X6 drone had three previously reserved bytes in its camera triggering
command. We suggested that the manufacturer use them to encode the unique identifier
of the captured equipment and shooting angle among all the routes. This identifier is
transferred to the camera and further used as a part of a filename. Thus, even if the
missions were re-planned to fit new weather conditions [
5
], all the images will save their
names, making it easier to process overall inspection results. This solution is currently
available for the serial R.A.L. X6 drones delivered with a GIT UR-640D camera.
4.3. Limitations of Custom Mission Generation with External Tools
4.3.1. An Issue Description
The initial version of RALTool used the proprietary mission file format. It allowed the
saving and loading of missions created using this software but did not provide any function-
ality to import sets of waypoints generated by external tools. At the same time, inspection
planning heavily relies on digital models of the industrial facility, which cannot be imported
into RALTool or many other mission planning programs. Also, general mission planning
tools provided by drone manufacturers rarely provide extended route optimization func-
tionality, which is required to minimize the number of batteries required to inspect all the
equipment. Thus, the ability to load missions created using external application-specific
software becomes critical for the inspection of complex industrial facilities.
Drones 2024,8, 549 12 of 34
4.3.2. A Solution
As a partial solution, the drone’s manufacturer provided us with the structure of the
binary mission file used by RALTool (Table 1). Unfortunately, they decided not to make it
open source and reserved the right to change it in the future, as well as for other versions
of their drones. In the research presented in [
5
], this file format was used to generate the
routes by mission planning and optimization software created by our team. These routes
were then loaded into the drone and executed using RALTool.
Table 1. RALTool binary mission format.
Bytes Preamble Datatype
0–3 Fixed preamble 32-bit unsigned integer
4–7 Number of waypoints (N) 32-bit unsigned integer
(8+40 ·(n1))(8+40 ·(n1) + 3)Target latitude of n-th waypoint,32-bit single float
(8+40 ·(n1) + 4)(8+40 ·(n1) + 7)Target longitude of n-th waypoint,32-bit single float
(8+40 ·(n1) + 8)(8+40 ·(n1) + 11)Target altitude of n-th waypoint, m 32-bit single float
(8+40 ·(n1) + 12)n-th waypoint radius, m 8-bit unsigned integer
(8+40 ·(n1) + 13)Reserved field 8-bit unsigned integer
(8+40 ·(n1) + 14)(8+40 ·(n1) + 15)Identifier of the image captured in the n-th
waypoint (added on request of the authors) 16-bit unsigned integer
(8+40 ·(n1) + 16)(8+40 ·(n1) + 19)Delay in the n-th waypoint, ms 32-bit unsigned integer
(8+40 ·(n1) + 20)Maximum horizontal speed used to reach n-th
waypoint, 0.1 m/s 8-bit unsigned integer
(8+40 ·(n1) + 21)Maximum vertical speed used to reach n-th
waypoint, 0.1 m/s 8-bit unsigned integer
(8+40 ·(n1) + 22)RTK shift latitude correction specific for n-th
waypoint, cm 8-bit signed integer
(8+40 ·(n1) + 23)RTK shift longitude correction specific for n-th
waypoint, cm 8-bit signed integer
(8+40 ·(n1) + 24)(8+40 ·(n1) + 27)Reserved field 32-bit unsigned integer
(8+40 ·(n1) + 28)(8+40 ·(n1) + 31)Target heading angle at n-th waypoint,32-bit single float
(8+40 ·(n1) + 32)(8+40 ·(n1) + 33)Target gimbal pitch angle at n-th waypoint,16-bit signed integer
(8+40 ·(n1) + 34)(8+40 ·(n1) + 35)Target gimbal yaw angle at n-th waypoint,16-bit signed integer
(8+40 ·(n1) + 36)(8+40 ·(n1) + 37)Waypoint index 16-bit unsigned integer
(8+40 ·(n1) + 38)additional actions at n-th waypoint (e.g.,
capture an image or start/stop a video) 8-bit unsigned integer
(8+40 ·(n1) + 39)n-th waypoint command type (turn on/off
motors, land, fly, etc.) 8-bit unsigned integer
··· Other waypoints ···
(8+40 ·(N))(8+40 ·(N)) + 71 Pre-defined end sequence 72-byte unsigned integer array
As it is seen, all commands in the mission file (Table 1) have the same format, which
includes spatial position, heading angle, camera targeting parameters, image identification,
and commands to start capturing. Such a unified approach is more beneficial for inspecting
complex industrial facilities in terms of fitting into waypoint limits because such missions
have much more camera targeting and capturing commands than general waypoints.
For example, the popular ArduPilot firmware uses three separate commands to move
the drone, target the camera using a gimbal, and initiate image capturing. Thus, despite
ArduPilot controllers supporting up to 1312 commands in one mission [
62
], the maximal
number of the inspected parts of the equipment will be smaller than by R.A.L. X6, the
missions of which can include up to 1000 commands.
4.4. An Incorrect Flight Time Prediction
4.4.1. An Issue Description
Our experimental studies discovered that a flight time predicted by RALTool software
can be more than 1.5 times shorter than an actual flight time. Moreover, the estimations
provided by popular UgCS (https://www.ugcs.com/) universal ground station software
were even more incorrect [
5
]. As we assume, the main reason is that the models used
in these software products to predict flight duration do not include the time required
for heading change and gimbal rotation in their estimations, considering it incomparable
Drones 2024,8, 549 13 of 34
to overall flight time. At the same time, while inspecting large-scale industrial facilities,
the drone spends much time at each waypoint, rotating and targeting multiple parts of
equipment with its camera.
4.4.2. A Solution
We proposed a new, more accurate model to predict flight time during industrial
inspection missions (1).
Tf=
K
k=1
(max(Thk,Tvk,Tgk) + Tyk+Tdk), (1)
where
Tf
—predicted flight duration,
K
—the number of waypoints in the route;
Thk
—the
time required to perform horizontal flight from the
k
1 to the
k
-th point;
Tvk
—the time
required to perform vertical flight from the
k
1 to the
k
-th point;
Tgk
—the time required
to rotate the gimbal to the orientation corresponding to the
k
-th point from the one it was
in the
k
1-th point;
Tyk
—the time required to change the drone’s heading to the direction
corresponding to the
k
-th point from the one in the
k
1-th point;
Tdk
—the time delay
configured for k-th point of the route.
Thk
,
Tvk
,
Tgk
, and
Tyk
are evaluated using the
fr
function defined in
(2)
. Generally, it
evaluates the time required to change the value of the drone’s corresponding parameter
d
following a trapezoidal ramp with
v
maximum speed setpoint,
aa
acceleration, and
ad
deceleration desired rates.
fr(d,v,aa,ad) =
d
v+v
2aa+v
2ad,dv2
2aav2
2ad
>0
r2aaadd
a2
a(aa+ad)+r2aaadd
a2
d(aa+ad),dv2
2aav2
2ad0
(2)
The distribution between the time estimates used as arguments of maximum function
in
(1)
and those placed outside represents the behavior of the R.A.L. X6 drone, which first
reaches the desired heading and only then performs the flight and gimbal orientation.
At the same time, it can be easily adapted for another drone. For example, if the drone
changes heading simultaneously with the flight to the next waypoint,
Tyk
should be used
as another argument for the maximum function.
The inspection flights performed on the reference power plant showed that the pro-
posed model provides a forecasting error of less than 10% in most cases and significantly
outperforms those implemented in commercial mission planning software. A detailed
description of this model and experimental studies demonstrating its accuracy are pre-
sented in [
5
]. According to the drone’s manufacturer statements, the proposed flight time
estimation model will be used in future versions of RALTool software.
4.5. An Unpredictable Time of Reaching Waypoint with a Small Radius
4.5.1. An Issue Description
Multiple inspection missions performed in strong wind conditions showed that in-
spection time could vary in a wide range and significantly deviate from the flight-time
prediction model described above. These deviations are caused by the unpredictably long
time the drone requires to reach waypoints with a relatively small radius. This issue most
often occurs at waypoints in so-called Venturi zones near the edges of the buildings where
air speed significantly rises and the air becomes turbulent.
Figure 8demonstrated the distance to a waypoint with a 2 m radius during capturing
of 12 photos in moderate (6 m/s, Figure 8a) and strong (11 m/s, Figure 8b) wind. The
process takes around 120 s in moderate conditions, while the overall duration rises to 319 s
in strong wind. As seen from Figure 8, additional time is required to return the drone inside
the waypoint’s radius. It should be noticed that gimbal orientation and camera trigger
Drones 2024,8, 549 14 of 34
commands in the R.A.L. X6 mission also have waypoint position and radius parameters,
forcing the drone to reach the desired location within the radius before their execution.
The above issue may cause a situation in which the drone’s battery capacity is not
enough to complete the mission and safely return to the landing area.
0 20 4 0 60 8 0 10 0 1 20
0
0.5
1
1.5
2
2.5
3
Relative time, s
Dis tance to
wa ypoint, m
(a)
0 50 100 1 50 2 00 2 50 300
0
0.5
1
1.5
2
2.5
3
Relative time, s
Dis tance to
wa ypoint, m
(b)
Figure 8. The distance to a waypoint with 2 m radius during capturing of 12 photos in (a) mod-
erated (6 m/s) and (b) strong (11 m/s) wind. The dashed lines represent the moment the drone’s
camera captured the frame.
4.5.2. A Solution
The issue is generally caused by the absence of the integral action in the drone’s
position controller. The proportional and proportional-derivative controllers have a static
error, which raises within the rise of the disturbance. Thus, in strong wind, the drone quickly
reaches a 4 m radius, requires a lot of time to reach a 2m zone, and may never get into a 1 m
circle around the waypoint coordinates. Multiple papers propose the position controllers
for multi-rotor drones, including integral action [
63
65
]. At the same time, tuning such
controllers to guarantee overall control loop stability is a complex task. As a result, the
proportional position controller remains one of the most popular solutions [
66
], widely
used in many industrial drones, including the ones based on ArduPilot and
PX4 firmware.
As a partial solution, [
5
] proposes to use a local turbulence map to avoid Venturi
and air turbulence zones during mission planning, where the issue was most commonly
observed. At the same time, this is inadequate from a safety point of view.
The significant changes in the position controller’s structure required much additional
testing from the drone’s manufacturer. Considering that the R.A.L. X6 faced the issue of
reaching small-radius waypoints only in strong winds (unsuitable for many other drones
on the market), we proposed a simple, safety-oriented solution. The drone firmware was
programmed to monitor an additional safety-timeout radius
rst
around each waypoint
(Figure 9).
This radius should be big enough so the drone could reach it in any acceptable
wind conditions in the inspection area and, at the same time, small enough so that after
reaching it, the drone can proceed with the mission to the next waypoint without hitting
any obstacle. Generally, rst should be lower than rssafety distances defined in [5].
The proposed safety-related algorithm (Algorithm 1) is described below. The safety
timer starts when the drones reach the
rst
radius. If the drone cannot reach the programmed
waypoint radius before the timer exceeds
Tst
safety timeout, the mission will switch to
the next waypoint, which is distant from the current waypoint, at least on
rst
. The last
condition is used to skip all the commands related to multiple camera shots performed
from one spatial area, assuming they all suffer from the same issue. In our research,
rst
was
configured as 4 m and Tst as 10 s.
Drones 2024,8, 549 15 of 34
Wind
Wind
Initial approach Proposed approach
Waypoint
radius Waypoint
radius
Safety-timeout
radius rst
Figure 9. The timeout radius is larger than the waypoint radius, and if the drone stays inside of it for
a certain amount of time without entering the waypoint radius, the point is skipped.
Algorithm 1 Proposed safety-related algorithm
1: Start
2: while True do
3: if rst is reached then
4: Start timer
5: while True do
6: if Tst is reached then
7: Skip Waypoints closer than rst
8: Exit
9: else if Waypoint is reached then
10: Exit
11: end if
12: end while
13: else
14: Continue checking if rst is reached
15: end if
16: end while
The proposed solution is based on the assumption that, from a safety point of view, it
is better to skip the waypoints whose radius cannot be reached in a reasonable time, and
proceed with the mission instead of wasting energy and risking ending it with an accident.
It is currently available in the firmware of the serial R.A.L. X6 drones.
4.6. Deviation from the Pre-planned Route Line between Two Waypoints
4.6.1. An Issue Description
Another issue found during the initial stages of aerial inspection with the R.A.L. X6
was that the drone had high deviations from the route line if the distance between waypoints
was more than 50 m. Figure 10 shows the planned and actual path of the drone along the
Π-shaped flight profile with the distance between waypoints equal to 782 m.
0
50
0
10 0
15 0
20 0
25 0
30 0
-50
-10 0
-15 0
-20 0
-25 0
-30 0
-35 0 7 00
60 0
50 0
40 0
30 0
20 0
10 0
0 0 10 20 30 40 5 0
0
50
10 0
15 0
20 0
25 0
30 0
0 10 0 2 00 300 40 0 5 00 60 0 70 0
-35 0
-30 0
-25 0
-20 0
-15 0
-10 0
-50
0
Realtive altitude, m
Realtive altitude, m
x, m x, m
y, m
x, m
y, m
to north to east to east
wind
wind
to east
wind
to north
3D view Top view Side view
Figure 10. The planned (blue) and actual (red) path of the drone along the
Π
-shaped flight profile
with a distance between waypoints equal to 782 m.
Drones 2024,8, 549 16 of 34
As seen, the deviation between the desired and actual locations of the drone was up
to several dozen meters, which is unacceptable in terms of obstacle avoidance and flight
safety in general. Such issues are rare in modern drones but can still be met in the ones
based on INAV firmware [67].
4.6.2. A Solution
The absence of an interpolator in the trajectory planner of the initial R.A.L. X6 firmware
caused the issue. The next waypoint coordinates were directly loaded into the position
controller each time the drone reached the previous one. When any external factor shifted
the drone (e.g., strong wind), it was trying to reach the waypoint following the shortest
way in the direction of waypoint (Figure 11a) instead of following the initial line between
the previous and current waypoints. Thus, the drone significantly shifted from the route
line, as was observed in the flight demonstrated in Figure 10. The issue became more
significant with the increase in distance between waypoints. The deviation is in direct ratio
with the distance between waypoints so that it may grow from a couple of meters on a
short distance to dozens of meters.
Wind Wind
Route Line
The shortest
path
Current waypoint Current waypoint
Previous waypoint Previous waypoint
Virtual
waypoint
Route Line
Deviation caused by
the wind
(a) (b)
Figure 11. Drone flight to the waypoint: (a) following the shortest path to the waypoint; (b) using the
carrot-chasing algorithm.
As a solution requiring minimal changes to the R.A.L X6 firmware, we proposed
implementing the carrot-chasing algorithm [
68
]. This algorithm assumes that the drone’s
position controller is dynamically loaded with a virtual waypoint, which follows the straight
line between the consequent waypoints in the mission with the desired speed (Figure 11a).
Thus, the drone “chases” the virtual waypoint and returns to the route line even in case of
any deviations.
The “carrot-chasing” algorithm is well-known. The version of it described in [
68
] is
implemented by two modules. The first one is the Path Generator. It interpolates a set
of waypoints into the path, consisting of a more dense list of waypoints, which can then
be approximated into a curve. The second module is the Path Follower. It receives the
lookahead distance, desired velocity, and interpolated set of waypoints. The algorithm
calculates the virtual target position that the drone is trying to reach. The Path Follower
algorithm is straightforward and consists of 3 stages.
1.
At the first step, the argument
λp(t)
is found as
λp(t) = arg minλ[0,L]|p(t)γ(λ)|
,
where
λ
—set of waypoints,
γ
—waypoint position,
p(t)
—drone’s position,
λp(t)—closest waypoint to the current drone’s position.
2.
At the second step, the lookahead distance is added, and the virtual target pose is
found as pt(t) = γ(λp(t) + d).
3. A drone is navigated to the virtual target position.
Drones 2024,8, 549 17 of 34
This issue was discovered during our first flights and reported to the manufacturer.
The suggested solution was rapidly implemented and integrated into the mainline of
the R.A.L. X6 firmware. Thus, all the flights reported in [
5
] were performed using the
abovementioned modification.
4.7. A High Pitch Angle during Acceleration/Deceleration
4.7.1. An Issue Description
The R.A.L. X6, being a drone designed to fly in strong winds, can perform pitch and
roll movements with high amplitude and acceleration to compensate for air turbulence.
Initially, the same maneuvers were used to perform accelerations and decelerations while
flying along the route.
Figure 12 shows the transient processes of pitch/roll angles recorded in one of the
inspection missions performed in moderate wind (5–7 m/s) conditions. The demonstrated
part of the mission included the inspection of 6 pylons on the roof of the power plant,
which were divided into groups of 2 and 4 entities. The distance between pylons in each
group was 12 m, and the distance between the groups was approximately 55 m. The small
peaks on the horizontal speed plot (Figure 12) represent the transitions between the pylons
in one group, the high peak represents the transition between the groups, and the gaps
when speed was close to zero represent the periods when the drone was targeting camera
and captured the inspected equipment.
Figure 12. The horizontal speed and pitch/roll angles during the part of the inspection flight in
moderate wind conditions.
As seen from Figure 12, to perform accelerations/decelerations before/after the way-
points, the drone’s pitch and roll angles reached and even exceeded 30
. Such amplitudes
can be considered high for a drone with a 0.94 m frame. If the drone encounters high air
turbulence during such accelerations/decelerations, it may reach critical roll/pitch angles,
resulting in undesirable maneuvers accompanied by significant loss of altitude. Moreover,
during such maneuvers, the drone will suffer from significant overloads, which may result
in motor overheating and damage to mechanical components. Thus, from a safety point of
view, high roll/pitch angles should be used only to compensate for external disturbances,
but not to perform rapid acceleration/deceleration in waypoints of the inspection mission.
4.7.2. A Solution
According to our recommendation, the virtual waypoint in a carrot-chasing algorithm
was programmed to accelerate/decelerate with a pre-configured rate significantly lower
than the maximum values defined by the drone’s capabilities. In terms of control theory,
Drones 2024,8, 549 18 of 34
this solution is similar to adding a second degree of freedom to the position controller [
69
],
with the only difference being that less dynamic reaction on setpoint is provided by re-
duced acceleration rates in the path planner instead of introducing an additional reference
controller to the design.
It is worth mentioning that lowering acceleration/deceleration rates does not sig-
nificantly influence the overall flight duration, as the acceleration/deceleration phase is
much shorter then the time required for long transitions, camera targeting, and image
capturing. The proposed solution is currently integrated into the mainline version of the
R.A.L. X6 firmware.
4.8. An Automatic Landing Cancellation in a Strong Wind
4.8.1. An Issue Description
Many research papers [
70
,
71
] are focused on different equipment and techniques to
perform safe automatic landings. Moreover, most commercial drone platforms have in-built
functions for automatic landing and take-off [
72
], and the R.A.L. X6 is no exception. Being
different in complexity and relying on various sensors, most of those solutions focus on
landing at a specific point with minimum tolerance. As a result, while landing in a strong
wind gust, even at a low altitude above the ground, the drone starts to compensate for
displacement, performing quite aggressive roll and pitch movements. These maneuvers
create a significant risk of hitting the ground or landing zone infrastructure (RTK and
communication antennas, lights, etc.). We observed such behavior multiple times, with
the R.A.L. X6 during the power plant inspection reported in [
5
] and with drones based on
ArduPilot and DJI flight controllers in other projects. To save the drone from a crash, the
operator may cancel the landing sequence and wait until the wind gusts are over. At the
same time, after the cancellation, the drone proceeds hovering around its last position.
This low-altitude hovering has similar risks of hitting the ground, so many professional
operators decide to land the drone manually in strong wind conditions.
Compared to many other drones, the R.A.L. X6 is often delivered without manual re-
mote control, assuming it will be used only for automatic flying along the pre-programmed
missions. Thus, the landing cancellation sequence becomes much more critical as it is the
only reasonable action the operator can initiate if something goes wrong. In the initial
version of the firmware, which was pre-loaded at the time of delivery, the stop command
executed during the landing, similarly to many other commercial multicopters, switched
the drone’s controller to the loiter mode. Moreover, it was soon revealed that if the stop
command was received at the exact moment when the drone touched the ground, the
controller proceeded to spin the motors without the ability to take off or proceed with the
landing procedure.
4.8.2. A Solution
The issue was reported to the drone manufacturer with the recommendation to start
climbing automatically to 5 m higher in case of landing cancellation. The reason for this
recommendation was to give the drone enough volume to safely perform wind compensa-
tion maneuvers without the risk of hitting the ground. Hovering at a safe altitude gives
the operator additional time to make the right decision: to repeat the landing sequence or
change the landing zone to one more appropriate for the current weather conditions.
The manufacturer both fixed the bug related to the landing abort when touching the
ground and implemented our recommendation to climb 5 m higher if a stop command was
received during a landing procedure.
Another recommendation was to limit the roll and pitch angles during the landing
procedure, assuming that in most cases during inspection missions, it is better to be blown
by the wind and land several meters from the desired location than to hit the ground during
the aggressive compensation maneuver. At the same time, following this recommendation
required additional analysis of the influence of the roll/pitch limitation during the landing
Drones 2024,8, 549 19 of 34
on the drone’s controller stability. Thus, it has yet to be implemented but has already been
added to the manufacturer’s backlog.
4.9. Flight Monitoring Issues Related to Ground Station Software
4.9.1. An Issue Description
The last set of issues revealed during our experimental studies is dedicated to the
user experience of the operator controlling the drone during the inspection mission. Many
commercial products (e.g., DJI industrial drones) include remote control, which must be
used to take over the drone in case of any abnormal situation during autonomous missions.
Generally, the operator performs manual control using the drone’s cameras or just looking
at the drone when it is in sight. This is why, in these products, the image from the camera
takes the central part of the operator’s screen.
In the case of inspections of large industrial facilities, this approach does not provide
operators with comfort or flight safety. Let us discuss an example of a power plant in-
spection reported in [
5
]. Due to the lack of suitable landing areas at the power plant, the
distance between the operator and the drone most of the flight time ranged from 300 m
to 850 m, making it impossible for the operator to estimate the drone’s orientation by his
own eyes, even when the drone was not closed off by the buildings. At the same time,
FPV piloting in a power plant is significantly more complicated because the inspection
drone has to fly between the wires and buildings at distances below 20 m. Moreover, to
safely control the drone in FPV mode, the last one should be equipped with a low-latency
wide-angle camera in addition to an inspection one. Without performing an FPV flight, the
only safe maneuver the operator can perform in an emergency is to gain altitude above all
the buildings and fly back to the landing zone.
The complexity of drone manual control conditions in large industrial facilities crit-
ically influences the ground station interface requirements. The operator is no longer a
pilot who flies the drone using the remote control. The main task of the inspection drone
operator is to identify the potential accident and initiate returning to landing as soon as
possible. The main dangerous situations that should be monitored are the following: (1) not
enough battery to finish the mission; (2) hardware malfunctions; (3) significant change in
weather conditions; (4) significant deviations from the pre-planned mission route.
The first two dangerous situations are controlled automatically by the R.A.L. X6
firmware and reported to the operator by sound notifications. Moreover, the drone can
be programmed to initiate return-to-home sequences in case of them automatically. The
change in the weather conditions is generally controlled by the operator and his assistants
(if present), but the R.A.L. X6 ground station software can show the measurements of the
remote meteo station on the main screen (Figure 13). The main dangerous situation the
operator should visually control during the flight is the last one. Significant deviations
from the pre-planned mission route can be caused by local wind gusts, air turbulence,
navigation errors, mechanical damage to the drone parts, etc. The earlier the operator
notices the deviations, the more chance the drone will be able to return to the landing safely.
Thus, during the inspection flight, the operator is focused on the drone and a zone on a
map around it (Figure 13, zone fenced with a blue line).
As seen from Figure 13, RALTool ground station software demonstrated the drone
position on the map (yellow circle), the line of the route (yellow line), the virtual way-
point (white circle with the red crosshair), the horizontal distance from the drone to the
virtual waypoint (green distance label), and the information about the drone position and
altitude (yellow label beneath the drone circle).
The first ground station-related issue of the R.A.L. X6 is that in the initial version of
RALTool software, there was no option to hide the yellow text with the drone’s coordinates.
The operator cannot analyze these data in any way during the flight. At the same time, this
text often overlaps the label with the distance to the virtual waypoint (Figure 14), which is
essential for detecting the situations when the drone significantly deviates from the route.
Drones 2024,8, 549 20 of 34
The second issue is that RALTool initially did not provide any information about the
vertical distance between the drone and the virtual waypoint. Thus, the only way for
the operator to notice any vertical deviation on the three-dimensional inspection mission
is to analyze the current altitude of the drone and compare it with the next waypoint’s
altitude. This comparison process significantly overloads the operator on routes with
dozens of waypoints. Finally, in case the yellow label with the drone’s coordinates is
hidden, the operator will have to look at the altitude on the sidebar with other telemetry
parameters (Figure 13, zone fenced with a red line), which is generally far from its primary
focus of view. Thus, it was almost impossible to analyze vertical deviations from the route
during the flight using the initial version of RALTool.
Figure 13. The main screen of the R.A.L X6 ground station software (RALTool). The main parameters
received from the drone’s telemetry are fenced red. The widget corresponding to the meteo station
data is marked with the green fence (the meteo station connection was not established at the moment
of the screenshot). The blue line fences the drone position and a virtual waypoint on the map.
(a) (b)
Figure 14. Examples of the situations when distance is (a) non-overlapped and (b) overlapped by the
drone’s coordinates in the RALTool interface.
Drones 2024,8, 549 21 of 34
The third, but not the least important, issue is that in the ground station software
delivered with the R.A.L. X6 drone, a mouse click on the map near the waypoint could
move it. Moreover, such route modifications were automatically transferred to the drone
during the flight without any additional confirmation. Thus, when the operator tried to pan
the map, it could lead to an accident because the modified route went through the obstacle.
The constant risk of such a critical mistake undoubtedly was a source of significant stress
on the operator.
4.9.2. A Solution
All the issues were reported to the R.A.L. X6 manufacturer and were fixed in the
following way: (1) an additional button was added to lock mission modification by any
mouse clicks on the map (Figure 15a); (2) the horizontal distance to the virtual waypoint
was accompanied with the vertical one (Figure 15b); (3) a checkbox was added in the
RALTool main screen configuration to hide the drone’s coordinates (Figure 15b).
(a) (b)
Lock mission modification button
Figure 15. Modifications introduced into RALTool to overcome flight monitoring issues related to
ground station software.
4.10. The Analysis of Other Market-Available Solutions
Table 2demonstrates the presence of the revealed issues in different market available
solutions. The table was compiled based on an analysis of scientific papers, drone manuals,
and datasheets, as well as the results of experiments on real products. The first two
columns represent the initial version of the R.A.L. X6 drone before our research and the one
improved by the manufacturer according to our recommendations. The rest of the columns
are dedicated to widespread market-available solutions. The details of their analysis are
presented in the following subsections. Before discussing these solutions in detail, it is
worth mentioning that, to the best of our knowledge, none of them provide any in-built
solution for the RTK shift issue. Most likely, the absence of such functionality is caused
by the widespread use of WGS84 worldwide, while the issue becomes critical only for
countries that have historically used another reference system for a long time.
Drones 2024,8, 549 22 of 34
Table 2. The presence of the revealed issues in different market-available solutions *.
Issue Initial
R.A.L. X6
Improved
R.A.L. X6 DJI Autel
Robotics
Ardu-
Copter/ PX4 INAV UgCS
An RTK shift x x x x x x
An identification of multiple images
captured from the same point x x x x x
Limitations of custom mission generation
with external tools and mission length x x/ x/x/x
An incorrect flight time prediction x x x n/a x
An unpredictable time of reaching
waypoint with a small radius xn/a n/a x x n/a
Deviation from the route line x x n/a
A high pitch angle during
acceleration/deceleration xx x x
An automatic landing cancellation in a
strong wind x x x n/a
Uploading custom missions x/x/x
Flight monitoring issues related to
ground station software x x x x
* x—the issue is present;
—the issue is not observed (overcome); x/
—the issue is generally resolved, but with
some limitations; n/a—the issue is not applicable.
4.10.1. DJI
DJI drones, the leading solutions on the market, generally do not suffer from most of
the issues we have faced with the R.A.L. X6. For example, the full name of each picture
taken can be programmed during the mission planning through FlightHub 2, and the
landing cancellation results in automatic altitude control by the obstacle avoidance sensors.
Moreover, modern DJI drones have trajectory controller, which allows them to follow the
exact pre-planned trajectory. Finally, in the older versions of DJI ground station software,
such as DJI GS PRO, the flight time prediction algorithm was inaccurate and did not take
into account hovering time in a waypoint, but in newer versions of DJI software, this
drawback was eliminated, and now time accuracy of flight time prediction is 5–8%.
Finally, the FlightHub 2 software allows online control of the drone’s position in 3D
anywhere on the route, and during mission planning, estimated flight time matches the
real one with a tolerance within a few dozen seconds.
The main drawback of the DJI software is that neither FlightHub 2 nor DJI Fly software
allow manually changing acceleration and deceleration at waypoints (Figure 16). Both of
these software packages allow the change in only velocity along the trajectory. Also, there is
no possibility of changing the waypoint radius, which makes the accuracy of the waypoint
following unclear, especially when the drone is affected by strong wind or turbulence.
At the same time, the targeting of narrow-angle inspection cameras is heavily influenced
by any positioning errors [62].
For a long time, the mission planning capabilities of DJI drones were limited. For ex-
ample, in GS PRO, the maximum number of waypoints in a single mission was 99, which
is insufficient for large-scale industrial facility inspection. However, this number is signifi-
cantly increased for modern DJI drones, reaching 950 waypoints on the Matrice 350 drone.
However, even older versions of DJI software supported the import of Keyhole Markup
Language (KML) files [
73
] as a template for the drone’s mission, making it easy to generate
missions from external geographic information systems. At the same time, e.g., for the
Matrice 100 drone, it was still impossible to define such parameters as flight speed or trigger
camera using KML tags. Thus, after the KML was imported, a significant amount of work
was required to turn it into the drone’s mission, especially in the large-scale inspection
Drones 2024,8, 549 23 of 34
tasks, which include a lot of commands related to gimbal control in image capturing. The
situation changed after DJI introduced WayPoint Markup Language (WPML)—an exten-
sion to KML designed to fully describe the drone’s mission, including heading, gimbal
orientation, and camera parameters. Currently, WPML is actively used by researchers to
create a mission with external planning software (e.g., [74]).
Figure 16. DJI Fly waypoint configuration screen. The software provides the possibility to change
only the waypoint speed (red fence), but not accelerations or waypoint radius.
4.10.2. Autel Robotics
Two mainly used Autel software solutions for mission planning are cloud-based
Autel SkyCommand Center and Autel Explorer, installed on Autel Start Controllers. Their
maturity is very high, similar to DJI products, but even they have some issues previously
revealed in our research for the R.A.L X6.
First of all, similarly to DJI, neither in Autel Explorer nor in Autel SkyCommand
Center is it possible to introduce RTK shift and change acceleration and deceleration along
the trajectory.
Second, in both of these Autel software packages, it is impossible to pre-determine
the names of the pictures taken in the trajectory waypoints. The only way to distinguish
the captured data is to use Exchangeable Image File Format (EFIX) data from the picture
with GNSS coordinates. However, this approach is not a solution when multiple images
are captured from a single waypoint.
Third, Autel software makes inaccurate flight time predictions. After six consecutive
flights, the average difference between predicted and real flight time was 30%, which is
significant, especially in comparison with modern DJI software flight time estimation or
the model proposed in [5].
Finally, in Autel Explorer software, there is no possibility to change the accelerations
of the drone or inclination angles (Figure 17). These parameters are set automatically by the
autopilot software without considering local external factors, such as weather conditions.
Similarly to DJI, modern Autel drones support WMPL, which allows the import of
missions created by external tools. The maximum number of waypoints, which can be
loaded into the drone using Autel software development kit (SDK), is 500, enough for
many inspection tasks. At the same time, for large facilities, it can lead to dividing missions
into smaller ones to fit the limit, which will require more time to complete the inspection.
However, if the mission is loaded using Autel Explorer without SDK, the maximum number
of waypoints will be reduced to 99, which can be considered as a significant limitation.
Drones 2024,8, 549 24 of 34
Figure 17. Autel Explorer waypoint configuration screen. As can be seen, it is only possible to change
the velocity of the drone in the point (red fence), but not accelerations of inclination angles.
4.10.3. ArduPilot/PX4
ArduPilot and PX4 firmware differ in terms of underlying real-time operation and
software architecture. However, they implement very similar functionality, controller
structure, interface with ground stations using MAVLink protocol, and are compatible with
both Mission Planner and QGroundControl mission planning applications [
75
]. Finally,
they share the same issues (Table 2).
Compared to the R.A.L. X6, ArduPilot/PX4-based drones follow the line of the route
in a strong wind and can be configured to use acceleration/deceleration rates below the
limits while flying between waypoints. At the same time, they suffer from the rest of
the issues discussed in this paper. Their mission cannot store any specific identifiers for
captured images or videos. The Mission Planner software does not provide any tools for
flight time estimation. At the same time, according to our experiments, QGroundControl
does not consider the time required for gimbal rotations, altitude, and heading changes,
which makes its estimations inaccurate for typical inspection missions. Due to using a
proportional position controller [
66
], ArduPilot and PX4 firmware have unpredictable
times of reaching waypoints with a small radius in strong wind conditions. At the same
time, in deference to the R.A.L. X6, their ground station software has a special command
that allows them to skip part of the route and proceed with the mission from a specific
waypoint, which can be considered as a partial solution for situations when the operator
has a robust communication channel with the drone during all of the flight. According
to our experiments and the description of the landing procedures for ArduPilot [
76
] and
PX4 [
77
] (including the advanced ones), this firmware suffers from the same issue in case
of landing cancellation as the initial version of the R.A.L. X6.
Finally, both Mission Planner and QGroundControl applications support only 2D
demonstration of the mission plan. Thus, they do not provide information for the operator
to monitor 3D deviations between the drone and the planned route.
Because ArduPilot and PX4 are open source firmware, their waypoint file format is
well-known. At the same time, these firmwares have non-obvious limitations regarding
waypoint numbers. Generally, a modern ArduPilot/PX4-compatible controller supports
up to around 1300 commands in a single mission [
62
]. This number is comparable to
waypoint limits of the R.A.L. X6 or enterprise DJI drones. At the same time, in deference
to these drones, ArduPilot/PX4 firmware includes all waypoints, gimbal, and camera
control commands in this limit. Thus, the maximum number of actual waypoints becomes
comparable to or even smaller than that of Autel drones.
4.10.4. INAV
INAV drones are the least advanced compared to the other analyzed aerial systems
(Table 2).
Drones 2024,8, 549 25 of 34
The most critical drawback of INAV firmware is the deviation from the planned route
line, similar to the one revealed in the R.A.L. X6. As it can be seen from Figure 18, the
actual trajectory (red) deviated from the planned (blue) line along all routes. Such behavior
is caused by the INAV Multicopter navigation algorithm, which does not follow the line
between two waypoints. Instead, it constantly recalculates bearing to the next waypoint
and uses it for navigation.
The next drawback is the inability to choose the name of the captured image. The
INAV cannot control any camera parameters. Instead, it just sends the shutter trigger signal
through the servo output. Thus, the naming of the captured pictures is determined only
by the camera being used. However, this camera cannot obtain any information about the
inspection mission from the flight controller.
Figure 18. The planned (blue) and actual (red) path of the drone, controlled by INAV. As can be seen,
the drone deviates from the planned route line.
Since INAV Multicopter uses a proportional controller for position control, it has
the problem of reaching the waypoint with a small radius, similar to the R.A.L. X6 and
ArduPilot/PX4. Also, in the case of landing abortion, the drone stays at the same height
on which the landing was canceled, which may lead to undesired contact with the ground
during the wind compensation movements if the altitude is small enough. However, the
INAV Multicopter supports using a sonar or an optical flow sensor, which allows setting the
minimum altitude above the terrain. This altitude will be gained if the landing is canceled.
The INAV has several issues relative to flight monitoring ergonomics. The main
drawback is that INAV is designed to be monitored using On Screen Display (OSD) and
FPV goggles. Thus, the INAV ground station lacks visual information about the current
state of the drone and is mainly used for setting the drone up and mission uploading.
The mission control screen shows only a limited amount of information about the drone’s
behavior, for example, the drone’s position, waypoints, heading, ground speed, and altitude.
These data are marked with a red fence on Figure 19. Also, the operator can see (Figure 19
the position of waypoints and the drone itself (blue fence) on a 2D map, and the mission
height profile is marked with a green fence (Figure 19). Still, as in the initial version of
RALTool, it is not enough to provide flight safety and operator comfort during an inspection
of a large-scale industrial facility. There is not even information about battery usage, making
the ground station unusable for constant flight monitoring without FPV goggles or a screen,
where this information is shown via OSD.
Finally, INAV has a limit of only 120 commands for most of the supported flight
controllers. Similar to ArduPilot/PX4, this limit is shared by waypoints and other actions.
Moreover, the mission file format lacks commands to control gimbal or camera behavior.
Thus, INAV functionality is minimal in terms of automated large-scale industrial facility
inspection requirements.
Drones 2024,8, 549 26 of 34
Figure 19. INAV ground station mission control tab.
4.10.5. UgCS
Due to UgCS being only ground station firmware, all the issues related to the drone’s
firmware are generally not applicable to it (Table 2).
UgCS provides an advanced 3D dimensional interface for mission planning and flight
monitoring (Figure 20). Moreover, it imports missions in the easy-to-understand JavaScript
Object Notation (JSON) format [
78
]. At the same time, it also suffers from several issues
discussed in this paper. First, it does not provide any tool to define any identifier for
collected camera data, even for DJI drones, which support assigning specific file names for
captured images. Second, it has low accuracy of flight time prediction [
5
,
79
]. Finally, it does
not allow the setting of desired accelerations and declarations of the drone, controlling its
tilting angle (Figure 20).
Figure 20. UgCS ground station interface. The red fence marks the waypoint configuration window,
the green fence marks the 3D mission view, and the blue fence marks the mission elevation profile.
Drones 2024,8, 549 27 of 34
5. Discussion
Aerial inspection using multi-rotor drones is a fast-growing technology with multiple
advantages compared to traditional non-destructive inspection approaches. However, the
attainability of these advantages varies between various application fields. Inspecting
large-scale industrial facilities is one of the most complex tasks in aerial inspection. This
section will focus on trade-offs that should be resolved in this application area.
All the advantages listed in Section 3can be summarized as better cost-efficiency,
higher inspection quality, and better safety. The cost-efficiency is generally achieved by
performing inspections without stopping production processes, shorter inspection times
compared to other methods, and reduction of work in hazardous areas, which are also
quite expensive. At the same time, the estimates of aerial inspection costs rarely include all
the indirect expenses, such as drone maintenance and diagnostics between flights, infras-
tructure (landing areas, equipment storage rooms, workshops, etc.), transportation costs,
etc. They can look negligibly small, but this often does not seem right. For example, if the
outsourcing company performs the inspection, this company can minimize costs on drone
maintenance and infrastructure by sharing them between multiple clients. Simultaneously,
the transportation cost for such a company will be higher because large industrial facilities
are usually distributed around the region and situated in different towns. Moreover, the
outsourcing company will also suffer from weather conditions limits because it should
cover the salary and travel expenses of the operator, who waits for suitable weather in a
remote location. Another example is when the employees of the industrial facilities perform
an inspection. In this case, transportation costs and the effects of weather conditions will be
minimal. However, the infrastructure and maintenance costs will significantly rise because
the owners of the industrial facility should employ a qualified operator and prepare all
necessary infrastructure for one or two drones, which will operate only a few days per
month. Finally, even with qualified operators and all the precautions in both described-
above cases, the risk of drone crash is not zero. At the same time, the crash can result in a
stop of production and lead to losses incomparable to the inspection cost. Insurance can
cover such risks, but according to our experience, its price is also relatively high and makes
sense only if a single drone is used to inspect multiple facilities.
The higher quality of inspection data is a more straightforward benefit than cost-
efficiency. Despite the reasons published in [
1
], the influence of the reflections and noises
caused by surrounding objects on the data collected by multi-rotor drone is generally less
compared to the other types of inspections. A comparison presented at the end of [
5
] clearly
shows that thermal images captured from the air have better quality than the images of the
same equipment shot from the ground (Figure 21). At the same time, to reach such quality
it is not enough to make a shot from a short distance, which is alone not an easy task in
automated mission planning. For accurate inspection, the images should be captured from
different fields of view and out of the turbulent zones [
1
,
5
,
26
]. Moreover, during the angle
of shot selection, one should consider the position of the sun and the presence of other
interference that could influence the quality (like hot pipes on the image of the insulators
during thermal inspections) [
1
,
21
,
23
,
62
]. Thus, efficient multi-criteria mission planning is
crucial to achieving high inspection quality on large-scale industrial facilities.
The higher level of safety is another advantage of the inspection using multi-rotors
that can lead to unnecessary illusions. From the point of view of the diagnostic engineer,
it is much safer to capture the equipment located on top of a factory pipe using a drone
than doing the same using manual inspection equipment after climbing that pipe. From
another point of view, the drone itself can be considered a source of hazard, especially in
the case of large industrial drones like the R.A.L. X6 or DJI Matrice 300. Moreover, all the
areas beneath the drone’s route can be considered hazardous, and should be cleared of any
other personnel. Thus, we can stress one more time the importance of the multi-criteria
mission planning, which should not only consider factors influencing inspection data
quality but also trace the inspection routes in such a way as to minimize flying above the
Drones 2024,8, 549 28 of 34
areas critical for facility non-stop operation (from where personnel cannot be removed
during the inspection).
(c) (d)
(g) (h)
°С
°С
°С
°С
(a) (b)
(e) (f)
°С
°С
°
С
°
С
Figure 21. Quality comparison between aerial (a,b,e,f) and ground (c,d,g,h) inspection [
5
]. The
top row contains raw images and the bottom row contains post-processed images of the insulators
cropped from the raw images.
After discussing the advantages, it is also worth discussing the challenges listed
in Section 3. The limited battery life of modern drones is partially compensated by the
relatively affordable prices of battery units for most industrial drones. Today, even small
facilities are fully inspected only using several batteries. Moreover, the ability to charge
multiple batteries simultaneously makes it possible to perform non-stop flights using a big
enough battery pack. At the same time, while inspecting large facilities, the problem of low
battery life becomes more complex. Due to the lack of landing area and large distances the
drone has to fly between the equipment, it is necessary to always have enough spare battery
time to return back to the landing in case of emergency or weather conditions changes [
5
],
once again highlighting the importance of effective mission planning.
The limited payload, in many cases, is another aspect of low battery life. Modern
industrial drones can simultaneously handle multiple cameras [
80
], but this option is rarely
used because it is more effective to divide all the inspected equipment parts by the classes
according to the drone payload that should be used for their diagnostic and then create a
set of routes to inspect each class of equipment separately using only the necessary payload.
Thus, all the main technical challenges and trade-offs of large-scale industrial facility aerial
inspection using multi-rotor drones can be fully or partially resolved by tuning the criteria
of the mission planning.
The rest of the listed challenges are operator competence and regulation issues. It is
often assumed that more automated drones will require less competent operators. At the
same time, it is true only if the complexity of the task stays constant. During industrial
inspection using a manually controlled drone, the operator should be professional enough
to keep the drone in the air, avoid collisions, and be able to reach the desired point of
view to collect the inspection data. At the same time, such flights are usually performed
in the area near equipment, where the drone is always kept in the operator’s line of sight.
Moreover, only a few equipment parts are inspected in a single flight. Performing such
flights in automated or fully automatic mode will require additional efforts related to
mission planning, which make sense only if done for a relatively large area. In this case,
flights are performed far from the operator in areas crowded with obstacles with a high risk
of collisions [
5
,
21
]. Simple switching to manual control will not save the situation in case of
emergency. Moreover, industrial drones, like the R.A.L. X6, can be purchased even without
any manual control. At the same time, the competence of the operator for such automated
flights should be even higher than that of manual ones. The operator should know the area
deeply and constantly control for changes in weather conditions. Also, pre-flight checks
and drone maintenance have a greater influence on safety than short manual flights. Thus,
the difference between the operator of manually controlled drones and the automated one
Drones 2024,8, 549 29 of 34
is similar to the difference between pilots of a small single-seater aircraft and of a large
plane like the Airbus A380. The first one literally controls the plane on the tip of the fingers,
while the second one relies on many automatic tools but should constantly analyze the
situation and be ready to perform necessary actions in case of an emergency. There is no
doubt that the minimal competence of the second one is usually higher than that of the
first one.
As stated in the paper selection procedure description, the legal and regulation issues
are out of the scope of this review. At the same time, despite it being one of the exclusion
criteria, the papers from the final selection still mentioned them alongside other challenges.
Modern airspace control rules in many countries were written decades ago and, after that,
were only slightly adapted for the recent technology changes. At that time, people only
imagined that one day, hundreds and thousands of UAVs would fly in the air indoors
and outdoors. These rules were designed to control a relatively small amount of aerial
vehicles controlled by professional pilots and guided by high-qualified air traffic controllers.
Currently, in many countries, drone flights are either not properly integrated into the
existing air traffic control mechanisms or integrated using the same rules as for regular
human-controlled aircraft. There are currently many projects in this area [
81
85
]. However,
until these projects are integrated into the legal system, the regulation issues will be a
significant limiter for the implementation of any unmanned aerial inspection, especially
inspection of large-scale industrial facilities.
After discussing the general pros and cons of aerial inspection of large scale industrial
facilities, we now switch to the technical issues related to the market available multi-rotor
drones. This topic is also very important, because if the drone lacks of any necessary
capabilities, there will be no guarantee of safe and efficient inspection even in the case of
the best mission planning.
As expected, the DJI and Autel Robotics products are the most mature of the analyzed
ones. Furthermore, it is even more surprising that they share some of the issues we observed
on the R.A.L. X6 during our research. At the same time, if the manufacturers pay attention
to the discussed drawbacks with their resources, it will not take much time to overcome
these issues.
ArduPilot and PX4 are extremely popular worldwide among mid- and small-size
companies that produce drones. Thus, it is essential to show that both of these firmwares
share key issues we revealed while using the R.A.L. X6 in the large-scale facility inspection
application. Moreover, most of these issues can be considered safety related. At the same
time, ArduPilot and PX4 are open-source projects with large communities worldwide,
which makes it possible to rapidly improve the firmware based on solutions we proposed
for the R.A.L. X6 in current research.
INAV is the least functional solution among the ones analyzed in this paper. It shares
most of the issues we faced with the R.A.L. X6. Moreover, its overall functionality is much
lower compared to open-source solutions such as ArduPilot and PX4. It is currently
used mostly on manually controlled drones, but the presence of functions related to
automated flights along the mission may create an illusion that it is suitable for more
complex inspections, but, according to our review, it is not correct. Moreover, whether it
requires any improvements or should just be considered unsuitable for complex inspection
applications in large industrial facilities is questionable.
UgCS is universal and advanced mission planning software compatible with many
industrial drones from different manufacturers. However, to be fully suitable for automatic
inspection of industrial facilities, it requires improvement of its timing estimation algo-
rithms, provision of functionality to specificity identifiers of the data collected during the
flight, and implementation of the ability to configure lower acceleration/deceleration rates
at waypoints.
Initially, the R.A.L. X6 drone was comparable to the ones running INAV Multicopter
firmware, which made it unsuitable for large-scale industrial facilities inspection both
in terms of safety and functionality. At the same time, as clearly shown above, all of its
Drones 2024,8, 549 30 of 34
issues can be solved using state-of-the-art methods. We translated this information to the
manufacturer, who released special patches for the drone’s firmware. After all the fixes, the
drone successfully performed the first autonomous inspection of a 2400 MW thermal power
plant [
5
]. Currently, the only claim left to this product is that its ground station software
does not support to import missions in widely used file formats, like WPML. Opening the
current version of their proprietary mission file format can be considered only as partial
solution, especially in the case that they do not guarantee to support it in the future. We
hope the manufacturer will change its opinion and will add WPML support in the next
versions of the RALTool.
However, the modification of the industrial drone following the feedback of the
research community can be considered a success story, and we hope it can motivate other
drone manufacturers, especially those using flight controllers with open firmware, to adapt
their products to fulfill the requirements of large-scale industrial facility inspection.
6. Conclusions
This review summarizes the pros and cons of aerial large-scale facility inspection,
distinguishing it from other inspection scenarios implemented with drones. Based on their
analysis, we reveal the main trade-offs and demonstrate the importance of highly efficient
multi-criteria mission planning. At the same time, even the best planning will not result in
safe and accurate inspection if the drone is not capable of completing the mission.
Industrial facility owners frequently require insurance to secure UAV damage and
third-party liability risk when performing aerial inspections. In turn, insurance companies
prefer to insure the flights of market-available drones rather than the customized ones
created by the research teams. Most companies producing industrial drones claim their
products are entirely suitable for industrial inspection. At the same time, during our
research, which resulted in the first autonomous inspection of a large-scale thermal power
plant, we had to overcome multiple issues with the R.A.L. X6 drone, influencing its safety
and efficiency. Moreover, the analysis of the other popular market-available solutions
showed that most of them suffer from the same problems, more or less. Thus, our research
shows how drone manufacturers can improve their products. At the same time, it also
demonstrates to the researchers and inspection engineers what issues they will probably
face while inspecting large-scale facilities using the industrial drones currently available on
the market.
Author Contributions: conceptualization, N.G., A.K. and A.R.; methodology, N.G., A.R. and A.K.;
software, N.G.; validation, N.G., A.R. and S.E.; formal analysis, N.G. and A.K.; investigation, N.G.;
resources, A.K. and S.E.; data curation, N.G. and A.R.; writing—original draft preparation, N.G.;
writing—review and editing, N.G., S.E., A.K. and A.R.; visualization, N.G.; supervision, A.R.; project
administration, A.K.; funding acquisition, A.K. All authors have read and agreed to the published
version of the manuscript.
Funding: The research funding from the Ministry of Science and Higher Education of the Russian
Federation (Ural Federal University Program of Development within the Priority-2030 Program) is
gratefully acknowledged.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data available on request due to restrictions. Due to the fact that the
field data were collected at the operating power plant, the authors can only provide it in anonymized
form without raw geospatial data. The raw data can be provided upon reasonable request only after
obtaining permission from the company operating the power plant.
Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design
of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or
in the decision to publish the results.
Drones 2024,8, 549 31 of 34
Abbreviations
The following abbreviations are used in this manuscript:
GNSS Global Navigation Satellite System
RTK Real Time Kinematic
UAV Unmanned Aerial Vehicle
FPV First Person View
RGB Red, Green, and Blue
IR Infrared
RTCM Radio Technical Commission for Maritime Services
NTRIP Networked Transport of RTCM via Internet Protocol
WGS World Geodetic System
KML Keyhole Markup Language
SDK Software Development Kit
JSON JavaScript Object Notation
USSR Union of Soviet Socialist Republics
EFIX Exchangeable Image File Format
OSD On Screen Display
WPML WayPoint Markup Language
References
1.
Nooralishahi, P.; Ibarra-Castanedo, C.; Deane, S.; López, F.; Pant, S.; Genest, M.; Avdelidis, N.P.; Maldague, X.P. Drone-based non-
destructive inspection of industrial sites: A review and case studies. Drones 2021,5, 106. https://doi.org/10.3390/drones5040106.
2.
Wang, Z.; Gao, Q.; Xu, J.; Li, D. A review of UAV power line inspection. In Proceedings of the 2020 International Conference on
Guidance, Navigation and Control (ICGNC 2020), Tianjin, China, 23–25 October 2020; Springer: Singapore, 2022, pp. 3147–3159.
https://doi.org/10.1007/978-981-15-8155-7_263.
3. Khalyasmaa, A.I. Synthesis of Models and Methods for Automated Diagnostics of High-Voltage Equip-Ment of Power Stations
and Substations. Habilitation Thesis, SSUWT, Novosibirsk, Russia, 2024. https://doi.org/10.13140/RG.2.2.17622.05445.
4.
Alsayed, A.; Yunusa-Kaltungo, A.; Quinn, M.K.; Arvin, F.; Nabawy, M.R. Drone-assisted confined space inspection and stockpile
volume estimation. Remote Sens. 2021,13, 3356. https://doi.org/10.3390/rs13173356.
5.
Romanov, A.M.; Gyrichidi, N.; Volkova, M.A.; Eroshenko, S.A.; Matrenin, P.V.; Khalyasmaa, A.I. Automated Mission Planning for
Aerial Large-Scale Power Plant Thermal Inspection. J. Field Robot. 2024,41, 1313–1348.https://doi.org/10.1002/rob.22320.
6.
Trujillo, M.A.; Martinez-de Dios, J.R.; Marton, C.; Viguria, A.; Ollero, A. Novel Aerial Manipulator for Accurate and Robust
Industrial NDT Contact Inspection: A New Tool for the Oil and Gas Inspection Industry. Sensors 2019,19, 1305. https:
//doi.org/10.3390/s19061305.
7.
Ekaso, D.; Nex, F.; Kerle, N. Accuracy assessment of real-time kinematics (RTK) measurements on unmanned aerial vehicles
(UAV) for direct geo-referencing. Geo-Spat. Inf. Sci. 2020,23, 165–181. https://doi.org/10.1080/10095020.2019.1710437.
8.
Tran, T.D.Q.; Nguyen, V.H. Heading Estimation for Autonomous Robot Using Dual-Antenna GPS. Int. J. Mech. Eng. Robot. Res.
2020,9, 1566–1572. https://doi.org/10.18178/ijmerr.9.12.1566-1572.
9.
Park, J.Y.; Kim, S.T.; Lee, J.K.; Ham, J.W.; Oh, K.Y. Method of operating a GIS-based autopilot drone to inspect ultrahigh voltage
power lines and its field tests. J. Field Robot. 2020,37, 345–361. https://doi.org/10.1002/rob.21916.
10.
Zefri, Y.; ElKettani, A.; Sebari, I.; Ait Lamallam, S. Thermal infrared and visual inspection of photovoltaic installations by UAV
photogrammetry—Application case: Morocco. Drones 2018,2, 41. https://doi.org/10.3390/drones2040041.
11.
Lekidis, A.; Anastasiadis, A.G.; Vokas, G.A. Electricity infrastructure inspection using AI and edge platform-based UAVs. Energy
Rep. 2022,8, 1394–1411. https://doi.org/10.1016/j.egyr.2022.07.115.
12.
Mohsan, S.A.H.; Khan, M.A.; Noor, F.; Ullah, I.; Alsharif, M.H. Towards the unmanned aerial vehicles (UAVs): A comprehensive
review. Drones 2022,6, 147. https://doi.org/10.3390/drones6060147.
13.
Zhang, Z.; Zhu, L. A review on unmanned aerial vehicle remote sensing: Platforms, sensors, data processing methods, and
applications. Drones 2023,7, 398. https://doi.org/10.3390/drones7060398.
14.
Dorafshan, S.; Maguire, M.; Hoffer, N.V.; Coopmans, C. Challenges in bridge inspection using small unmanned aerial systems:
Results and lessons learned. In Proceedings of the IEEE 2017 International Conference on Unmanned Aircraft Systems (ICUAS),
Miami, FL, USA, 13–16 June 2017; pp. 1722–1730. https://doi.org/10.1109/ICUAS.2017.7991459.
15.
Lescuyer, T. Technologies: Drones facilitate inspection of networks; [Les drones facilitent l’inspection des réseaux]. Tech. Sci.
Methodes 2018,2018, 3–4.
16. Warren, C. Drones and AI converge for power delivery inspections. EPRI J. 2019,2019, 20–21.
17.
Nonami, K. Drone technology, cutting-edge drone business, and future prospects. J. Robot. Mechatron. 2016,28, 262–272.
https://doi.org/10.20965/jrm.2016.p0262.
18.
Fontaine, O.; Martinetti, A.; Michaelides-Mateou, S. Remote pilot aircraft system (RPAS): Just culture, human factors and learnt
lessons. Chem. Eng. Trans. 2016,53, 205–210. https://doi.org/10.3303/CET1653035.
Drones 2024,8, 549 32 of 34
19.
Chen, S.; Laefer, D.; Mangina, E. State of technology review of civilian UAVS. Recent Patents Eng. 2016,10, 160–174. https:
//doi.org/10.2174/1872212110666160712230039.
20.
Jordan, S.; Moore, J.; Hovet, S.; Box, J.; Perry, J.; Kirsche, K.; Lewis, D.; Tse, Z. State-of-the-art technologies for UAV inspections.
IET Radar Sonar Navig. 2018,12, 151–164. https://doi.org/10.1049/iet-rsn.2017.0251.
21.
Rakha, T.; Gorodetsky, A. Review of Unmanned Aerial System (UAS) applications in the built environment: Towards automated
building inspection procedures using drones. Autom. Constr. 2018,93, 252–264. https://doi.org/10.1016/j.autcon.2018.05.002.
22.
Agnisarman, S.; Lopes, S.; Chalil Madathil, K.; Piratla, K.; Gramopadhye, A. A survey of automation-enabled human-in-the-loop
systems for infrastructure visual inspection. Autom. Constr. 2019,97, 52–76. https://doi.org/10.1016/j.autcon.2018.10.019.
23.
Liang, H.; Lee, S.C.; Bae, W.; Kim, J.; Seo, S. Towards UAVs in Construction: Advancements, Challenges, and Future Directions
for Monitoring and Inspection. Drones 2023,7, 202. https://doi.org/10.3390/drones7030202.
24.
Muhmad Kamarulzaman, A.; Wan Mohd Jaafar, W.; Mohd Said, M.; Saad, S.; Mohan, M. UAV Implementations in Urban
Planning and Related Sectors of Rapidly Developing Nations: A Review and Future Perspectives for Malaysia. Remote. Sens.
2023,15, 2845. https://doi.org/10.3390/rs15112845.
25.
Choi, H.W.; Kim, H.J.; Kim, S.K.; Na, W. An Overview of Drone Applications in the Construction Industry. Drones 2023,7, 515.
https://doi.org/10.3390/drones7080515.
26.
Fun Sang Cepeda, M.; Freitas Machado, M.; Sousa Barbosa, F.; Santana Souza Moreira, D.; Legaz Almansa, M.; Lourenço de
Souza, M.; Caprace, J.D. Exploring Autonomous and Remotely Operated Vehicles in Offshore Structure Inspections. J. Mar. Sci.
Eng. 2023,11, 2172. https://doi.org/10.3390/jmse11112172.
27.
Rodríguez, D.; Lozano Tafur, C.; Melo Daza, P.; Villalba Vidales, J.; Daza Rincón, J. Inspection of aircrafts and airports using UAS:
A review. Results Eng. 2024,22, 102330. https://doi.org/10.1016/j.rineng.2024.102330.
28.
Slama, D. Drone-based facade inspection (TUEV SUED). In The Digital Playbook: A Practitioner’s Guide to Smart, Connected Products
and Solutions with AIoT; Springer International Publishing: Cham, Switzerland, 2023; pp. 365–370. https://doi.org/10.1007/978-
3-030-88221-1_29.
29.
Satyarthi, D.; Arya, K.; Dixit, M. Drone Technologies: Aviation Strategies, Challenges, and Applications; Scrivener Publishing LLC:
Austin, TX, USA, 2023; pp. 117–152. https://doi.org/10.1002/9781394168002.ch6.
30.
Gowroju, S.; Santhosh Ramchander, N. Applications of Drones—A Review; Scrivener Publishing LLC: Austin, TX, USA, 2023;
pp. 183–206. https://doi.org/10.1002/9781394168002.ch8.
31.
Moreno-Jacobo, D.; Toledo-Nin, G.; Ochoa-Zezzatti, A.; Torres, V.; Estrada-Otero, F. Evaluation