X-AMINOR: A Mobile Multi-Sensor Platform for Lifecycle-Monitoring

of Transformers

Ferdinand Fuhrmann

, Konrad Diwold

, Jakub Waikat

, Jakob Gaugl

, Fredi Belavic

and Franz Graf

JOANNEUM RESEARCH, Graz, Austria

Pro2Future, Graz, Austria

Siemens Energy Austria, Wien, Austria

Siemens Energy Austria, Weiz, Austria

Austrian Power Grid, Wien, Austria

fredi.belavic@apg.at

Keywords:

Transformer, Lifecycle-Monitoring, Transmission System.

Abstract:

The current developments in the energy system especially the increased integration of renewable and the

introduction of dynamic loads lead to a decoupling of consumption and production. Digitalization is one of

the major tools for grid operators to tackle these challenges, as it allows them to implement ﬂexible and

advanced operation and planning strategies, which reduce costs while increasing service security. In the

context of operation, transformers represent one of the most important assets in the transmission grid. Their

optimal utilization is of utmost importance to ensure the optimal operation for dependable feed-in. The article

presents the project X-AMINOR, which aims for the development of novel minimal invasive transformer

monitoring solutions designed to complement existing monitoring strategies to enable continuous monitoring

over a transformer’s entire life cycle to improve its operation.

1 INTRODUCTION

The increased integration of renewables and the in-

troduction of dynamic loads into the energy systems

creates new challenges for grid operators, as they lead

to a decoupling of consumption and production (Sin-

sel et al., 2020). Digitalization can be seen as one

of the major tools for grid operators to tackle these

challenges, as it allows to implement ﬂexible and ad-

vanced operation and planning strategies, reducing

costs while increasing service security. In the con-

text of medium and low voltage distribution grid op-

eration this led to a number of smart grid technolo-

gies (Tuballa and Abundo, 2016), allowing the oper-

ator to proactively control grids and provide ancillary

services. These transformations also impact the trans-

mission level, where the integration of renewables can

lead to instabilities (e.g., voltage ﬂuctuations), over-

loading of transformers and injection of harmonics

(Shaﬁullah, 2016).

Transformers constitute core infrastructure com-

ponents in energy grids on all voltage levels. Their

availability and longevity can thus be seen as integral

variables in the context of security of supply. Con-

tinuous monitoring and predictive maintenance are

therefore an important factor to increase the longevity

of these infrastructure elements and reduce unplanned

outages. Transmission systems utilize monitoring so-

lutions (Pudlo et al., 2002)(Al-Ali et al., 2004), tai-

lored towards speciﬁc power transformers, and moni-

tor transformers’ KPIs based on its operation param-

eters (Pudlo et al., 2002). These monitoring solutions

are integrated into a monitoring scheme via Super-

visory Control and Data Acquisition (SCADA) sys-

tems. The amount of information available from a

transformer during operation depends on the trans-

former itself. While new transformers provide a

multitude of information (winding hotspot sensors,

dissolved gas analyzers, etc.), operation parameters

available from legacy transformers are often limited.

Usually, the condition of a transformer is regu-

larly inspected by a human expert to guarantee fault-

free operation. Here, abnormal noise could indicate

the malfunctioning of an oil pump, or changes in its

thermal properties could indicate looming faults. Cur-

rently such inspection information is only available to

Fuhrmann, F., Diwold, K., Waikat, J., Gaugl, J., Belavic, F. and Graf, F.

X-AMINOR: A Mobile Multi-Sensor Platform for Lifecycle-Monitoring of Transformers.

DOI: 10.5220/0011968800003491

In Proceedings of the 12th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS 2023), pages 121-128

ISBN: 978-989-758-651-4; ISSN: 2184-4968

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

121

experts themselves. Additionally, the inspection pro-

cess is based on experience, which enables the human

expert to correctly identify, classify and interpret de-

viations.

The project X-AMINOR aims at developing an

automatic transformer lifecycle-monitoring solution

to detect and evaluate information, which previously

could only be captured by experts during inspection.

The project develops a mobile multi-sensor platform

(audio, video, thermal) providing additional informa-

tion to complement existing monitoring. X-AMINOR

will utilize sensor data to build data models, forming

the basis for predictive maintenance and continuous

product improvements. The developed monitoring so-

lution is applicable in the entire lifecycle of a trans-

former, thus providing additional information from

its cradle to grave. The developments are therefore

driven by two application scenarios in the life cycle

of a transformer: ﬁnal (end of line) testing after pro-

duction as well as operation.

The technical result of the project constitutes a

functional monitoring prototype, which will be de-

ployed and tested in the context of transmission grid

operation as well as end-of-line tests. Long term eval-

uation scenarios will be used for an in-depth quantita-

tive and qualitative evaluation of the system and will

allow the quality assessment of the developed meth-

ods as well as its beneﬁts in the context of end of line

testing and automated condition monitoring.

The remainder of the article is structured as fol-

lows: Section 2 depicts the general challenges X-

AMINOR addresses and outlines potential monitor-

ing use-cases. Section 3 discusses the overall system

design and introduces the utilized sensing technolo-

gies. Finally, Section 4 concludes the article and gives

an outlook on the next steps towards the envisioned

monitoring solution.

2 CHALLENGES

Climate protection, the energy system transformation

and the required market efﬁciency create major de-

mands for today’s energy suppliers. The increasingly

volatile utilization of the grid operation components

is particularly challenging. This is grounded in the

growing availability of renewable forms of energy, the

European exit from coal and nuclear power, as well as

new ”energy-hungry” industrial sectors with dynamic

consumption patterns (Shi et al., 2016). However, the

digitalization of the energy system offers new oppor-

tunities to increase efﬁciency. In the following, the

challenges addressed by X-AMINOR are listed to-

gether with the associated, demand-oriented solution

approaches.

Network Utilization and Redispatch. The trans-

mission and distribution grid are particularly chal-

lenged by the volatile performance of sources (Janda

et al., 2017) – e.g., photovoltaics or windfarms. In or-

der to guarantee reliable supply, a redispatch is carried

out whenever grid bottlenecks are detected (Hedman

et al., 2011). With this redispatch power plant oper-

ators are instructed to reschedule their performance

to avoid possible network bottlenecks, causing sig-

niﬁcant costs for grid operation and energy trading.

To avoid future bottlenecks and corresponding redis-

patches, specialized monitoring and control solutions

can be used to continuously supervise and control the

grids and their assets. In contrast to distribution grids,

digitalization is already well advanced in the trans-

mission grid. By using SCADA (Supervisory Control

and Data Acquisition) systems (Pliatsios et al., 2020),

the grid can be monitored, analyzed, and optimally

controlled over a large area. In addition to informa-

tion on the current utilization of a resource, its status

as well as information on its past activities, SCADA

systems offer the possibility of calculating load sce-

narios more precisely. This applies to transformers,

which - in addition to the network itself - represent

the most important asset of a transmission network.

Transformer Utilization and Fault-Related Out-

ages. Optimal operation of transformers is required

to ensure continuous reliability of supply. The oper-

ator currently calculates the possible utilization of a

transformer on the basis of selected overload param-

eters (nominal power, oil and winding temperature,

etc.), which are continuously recorded. While these

parameters enable an estimation of the overload, ex-

panding existing models with additional information

would enable a better assessment of the current load

and the associated possible time-limited overload of

transformers or substations. Expanding existing mod-

els requires additional information, such as an accu-

rate assessment of the transformer’s current thermal

state. A possible consequence of an inaccurate as-

sessment of the overload capacities of transformers is

a shortened service life or premature failure. Here,

the failure distribution of a transformer during its life-

time follows a so-called bathtub curve (Klutke et al.,

2003), showing an increased probability of fault at the

beginning and end of its service life (Figure 1).

Faults at the beginning of the transformer’s life are

caused by incorrect design, manufacturing errors or

defects in production. At the end of its life, the failure

probability increases due to age and fatigue. Hence

we identiﬁed a number of use-cases to be addressed

SMARTGREENS 2023 - 12th International Conference on Smart Cities and Green ICT Systems

122

Figure 1: Fault rate distribution of transformers over its life-

time.

by the X-AMINOR monitoring system, ranging from

transformer development to operation. These use-

cases thus encompass encompass the whole life-cycle

of transformers. In the following we outline exem-

plary monitoring use-cases at different stages of a

transformer’s life-cycle, which X-AMINOR aims to

automate and optimize.

Manufacturing. Currently, various quality aspects

of the individual manufacturing steps are checked

manually, requiring a training phase and many years

of experience. With X-AMINOR, parts of the quality

check could be carried out automatically via the mon-

itoring system. For instance, a 3D variance analysis

for manually isolated transformer parts could yield vi-

tal information to identify potential error sources dur-

ing operation.

Factory Acceptance Test. During the end of line

test, manual noise measurements are currently per-

formed. Since the transformer is under load during

the test, this method poses a safety risk for employees.

With the help of a mobile monitoring solution this test

can be automated. In addition, the system allows the

generation of a 3D sound model of the transformer,

which can be used as a ﬁngerprint for reference in

later analyses.

Operation. During operation, transformer infor-

mation is continuously collected via SCADA sys-

tems and used in the context of condition monitor-

ing systems (CMS). Such systems can utilize trans-

former parameters which are actively collected via the

SCADA system (e.g., voltages, temperature, etc.). X-

AMINOR will enable a continuous external inspec-

tion of a transformer (acoustically and visually) and

thus the detection of errors which start to manifest

themselves (e.g., small leaks or the beginning of in-

sulation defects), which currently can only be discov-

ered in later stages. Moreover, peripheral systems of

the transformer (e.g., fans) can also be monitored in

this way.

Post Mortem Analysis. Currently, the planning of

new transformers does not involve data gained from

a post mortem analysis nor any other historical data.

Making such data available via X-AMINOR can fur-

ther reduce errors caused by uncertainties in the di-

mensioning of new transformers, as well as using this

information for optimizing the state of the grid.

3 X-AMINOR SYSTEM DESIGN

AND ASSOCIATED

TECHNOLOGIES

X-AMINOR is designed to collect and analyze vi-

sual and acoustic information in addition to already

available operating parameters in order to enable a

more precise assessment of the transformer’s condi-

tion. For an autonomous operation it requires suitable

transportation means (mobile platform) and IT tech-

nologies (middleware and edge computing). The next

section outlines the envisioned system design and the

underlying technologies that will be applied in the

context of the cross-sensor lifecycle-monitoring plat-

form.

3.1 System Design

The overall X-AMINOR system design is sketched in

Figure 2. The system contains a scalable (on-premise

cloud) backend, which performs acoustic and visual

data analytics, as well as X-AMINOR middleware

nodes, which are installed at speciﬁc transformer lo-

cations to monitor their behavior during operation.

Figure 2: X-AMINOR system design.

In this context, X-AMINOR utilizes recent devel-

opments in the area of the Internet of Things (IoT),

where a range of protocols and architectures for in-

stantiating intelligent monitoring systems have been

proposed (Wang et al., 2013). To guarantee scalabil-

ity, edge computing will be utilized via GPU-enabled

computing nodes. This will allow nodes to prepro-

X-AMINOR: A Mobile Multi-Sensor Platform for Lifecycle-Monitoring of Transformers

123

cess data streams and perform ﬁrst analyses, while

the backend performs computation-heavy analytics as

well as model training and development. In regard to

its system design X-AMINOR takes inspiration from

a number of middleware systems and solutions for

distributed monitoring, which have been developed

in the context of smart grid operation (Cejka et al.,

2018)(Diwold et al., 2018). To increase the integrity

of the node applications, they will be isolated in run-

time containers, which have almost negligible inﬂu-

ence on the performance of programs compared to

their native execution (Morabito, 2017).

The core methodology of X-AMINOR concerns

the visual and acoustic assessment of transformers.

State of the art of acoustic and visual technologies

as well as the contribution in these areas from X-

AMINOR, together with a ﬁrst glimpse into applying

such technologies in the context of transformer mon-

itoring are outlined in the following subsections.

3.2 Visual Diagnostics

Visual diagnostics in X-AMINOR rely on three pil-

lars: a reasonable accurate mapping of the trans-

former and localization of detected conditions and

ﬁndings, the visual detection of speciﬁc defect types

known in advance, as well as the detection of generic

changes on the transformer surface.

3.2.1 Mapping and Localization

Acquiring an as-built 3D model is a prerequisite for

the mapping and localization of conditions.

Related Work: Numerous methods based on

RGB (Taketomi et al., 2017) and RGB-D (K

ahler

et al., 2016) sensors are known for 3D mapping. Al-

though these offer low hardware costs, the accuracy

is usually limited to a range of several millimeters.

Higher accuracies can be achieved with other sensors,

in particular with terrestrial LIDAR scanning. Com-

mercial solutions (e.g., Leica

) are routinely used for

such tasks. RGB or RGB-D cameras are also suitable

for localizing aspects relative to 3D models (Sattler

et al., 2019). Here, efﬁcient transfer learning for new

scenes is currently the main challenge (Balntas et al.,

2018).

Contributions: Within X-AMINOR we will ap-

ply state of the art methods for 3D model data genera-

tion like LIDAR scanning, RGB-D sensors and Struc-

ture from Motion (SfM). We will apply associated lo-

calization techniques, both for the initial scanning of

https://leica-geosystems.com

the transformer and the periodic scanning with a mo-

bile device later on. Figure 3 and 4 depict ﬁrst re-

sults of localization experiments performed within the

project.

Figure 3: As-built 3D transformer model acquired with a

Leica BLK360 scanner.

Figure 4: View on a transformer with an RGB-D sensor

(Microsoft Azure Kinect).

3.2.2 Visual Inspection of Known Defect Types

This task deals with the visual detection of speciﬁc

defect types known in advance.

Related Work: The literature on visual detection

of surface defects is yet extensive, an up-to-date

overview can be found here (Czimmermann et al.,

2020). In most approaches, classiﬁers are trained

with existing training data using machine learning.

More recent works (see (Tabernik et al., 2019)) try

to minimize the number of examples required for

suitable training. Also, thermal infrared information

is used to create 3D temperature models of the

measured surfaces using LIDAR data (Borrmann

et al., 2013), through 3D reconstruction with a depth

camera (Vidas et al., 2013) or using CAD models

(Sels et al., 2019). Moreover, in existing transformers

analog measuring devices are often used, from which

SMARTGREENS 2023 - 12th International Conference on Smart Cities and Green ICT Systems

124

digital values cannot be read out easily. (Lee et al.,

2018) proposes an approach for realtime automatic

instrument status monitoring using deep neural

networks.

Contributions: In the literature, a majority of ap-

proaches focuses on the detection of defects in indi-

vidual 2D images. In the X-AMINOR project indi-

vidual 2D detection results are integrated and consol-

idated in a 3D model of the transformer. This way, the

reliability of detections can be improved by correcting

individual false positives via the results from other

images. Based on the given 3D transformer model

and the localization information of the sensor system

relative to the device, a holistic detection model can

be established, linking the different modalities. Ad-

ditionally, the system reads analog displays from im-

ages acquired via the sensor system. The ﬁrst chal-

lenge here is the reliable detection and identiﬁcation

of the display elements and furthermore a reading at

unfavorable viewing angles which can falsify the re-

sult. In X-AMINOR, the reading of thermometers as

well as the control of pump ﬂow indicators are con-

sidered.

3.2.3 Change Detection with Mobile Sensors

In contrast to speciﬁc visual inspection tasks change

detection focuses on observing a scene over a longer

period and register all types of expected or unex-

pected changes.

Related Work: For mobile inspection one could

detect purely geometric changes compared to a

reference model (Palazzolo and Stachniss, 2017).

However, if changes such as contamination or

corrosion need to be found the paradigm of visual

change detection must be adapted to mobile cameras.

A ﬁrst work in this direction was published with

the aim of ﬁnding changes in street scenes which

can lead to map updates (Alcantarilla et al., 2018).

For inspection tasks, comparable methods for the

surface assessment of concrete pipes were examined

(Stent et al., 2015). Here, the (semi-) automatic

generation of sufﬁcient training data plays a central

role (Sakurada et al., 2020). More recent works even

aim to ﬁnd a textual interpretation of the changes

(Park et al., 2019).

Contributions: On the one hand, a system for

change detection must be sufﬁciently robust in terms

of recording conditions and localization accuracy, but

on the other hand, the high level of generalization

must be achieved with a low training effort. X-

AMINOR aims to build on the robustness of cur-

rent methods from the literature, which fundamen-

tally seems promising but has to be evaluated in detail

for the present scenario. The effects of small amounts

of training data as well as transfer learning methods

are examined. A further approach to improve the per-

formance is the aforementioned consolidation the de-

tected changes from several individual images in the

3D model of the transformer. Last but not least, the

creation of an evaluation data set, in which the tar-

geted irrelevant changes (lighting, seasons, weather)

as well as relevant changes (contamination, leakage,

corrosion etc.) are adequately covered is another con-

tribution.

3.3 Acoustic Condition Assessment and

Evaluation

In X-AMINOR, we analyze the emitted airborne

sound via acoustic monitoring. Our aim is to

reconstruct a complete 3D representation of the

transformer’s sound radiation (acoustic heat map).

We will utilize microphone arrays to record the

transformer’s location dependent sound radiation. By

analyzing different recording positions all around the

transformer, we construct a complete representation

of the emitted sound pressure levels, which we then

use for further analyses. Figure 5 shows an example

sound radiation pattern recorded during the ﬁrst

measurement sessions of the project.

Related Work: Acoustic monitoring refers to the

automatic detection of abnormal events through the

analysis of acoustic signals. In acoustic condition

monitoring, a statistical model is formed from the

acoustic signals of a predeﬁned “normal” condition

of the process. The task of this model is the automatic

detection of deviations from this normal condition

given new input data. Acoustic event monitoring

is restricted to the detection of predeﬁned acoustic

events, which are often associated with certain error

states. Using pre-recorded error-state examples

and machine learning methods, classiﬁers for the

automatic detection of these events are developed

(Heittola et al., 2018). Literature reports systems

for acoustic monitoring of trafﬁc ﬂows (Graf and

Gruber, 2018), production machines (Siebald et al.,

2017), industrial plants (Koester et al., 2018) as well

as urban spaces (Bello et al., 2019).

Contributions: To create an acoustic heat map,

we transform the sensory data into a spatial 3D repre-

sentation and adapt them to the 3D model of the trans-

former. We use the heat map for consistency valida-

tion in condition monitoring during the transformer’s

X-AMINOR: A Mobile Multi-Sensor Platform for Lifecycle-Monitoring of Transformers

125

life-cycle. We ﬁrst construct an initial sound radiation

model during factory acceptance testing. Once the 3D

representation is constructed, we use it as a reference

for later comparisons. During the transformer’s oper-

ation, we record new data which resulting 3D repre-

sentation is compared to the reference. If we detect

certain deviations, we trigger an alarm, and a manual

check is initiated. Moreover, we derive a failure de-

tection analysis from the sound radiation model. Cer-

tain failure states such as broken cooling fans or cool-

ing units produce characteristic sounds, which we use

to construct speciﬁc failure event classiﬁers. During

data acquisition, we record data from these, mostly

simulated, failure states and construct specialized de-

tectors. In order to obtain meaningful results from

the acoustic modeling, sound source localization, sig-

nal characteristics and the position of the sensor ar-

ray must be analyzed together. The merging of these

dimensions into a single, closed model leads to new

possibilities for the statistical modeling of acoustic

signals.

Figure 5: Sound radiation pattern for frequency band 90-

140Hz of a transformer.

3.4 Transformer Modelling

Within X-AMINOR the sensor data will be used for

modelling a transformer’s aging process during oper-

ation. A multitude of transformer models have been

proposed, focusing on different aspects such as capac-

ity estimation and prediction (Alvarez et al., 2019) or

thermo-hydraulic aging (Seitlinger, 2000). It has been

shown that a multitude of factors must be considered

to achieve accurate modelling (Raith et al., 2020), in-

cluding the transformer’s thermal behavior, insulation

moisture as well as the dynamic variation of load-

ing and temperatures. Modelling typically utilizes as-

sumed initial conditions and operation assumptions

for the evolution of the simulation. The data es-

tablished with the X-AMINOR monitoring system

will allow to validate and tune existing aging mod-

els based on operational data. Moreover, such data

can be used as additional model input to individually

model a speciﬁc target transformer. Additionally, this

information allows the development of new models

which relate acoustic, visual, and thermal informa-

tion established during end-of-line testing with oper-

ational data, allowing for a continuous evaluation of

the transformer.

4 OUTLOOK AND

CONCLUSIONS

In this article we have outlined the vision of the X-

AMINOR project, which aims to study and demon-

strate a novel cross sensor platform monitoring so-

lution for transformers. X-AMINOR is designed to

complement existing monitoring solutions and enable

continuous monitoring of a transformer to improve its

operation. The project will address use-cases in the

context of manufacturing (end-of-line tests) and oper-

ation to ensure the applicability of the solution across

the whole lifecycle of a transformer. Besides demon-

strating such holistic monitoring concepts, the project

will advance methods of visual, thermal and acous-

tic condition monitoring and demonstrate their appli-

cation in the context of grid installations. Currently

the project is at a very early stage, as of now initial

data acquisition campaigns have started to enable the

development of data-driven methods as well as reﬁn-

ing existing transformer models. In parallel the target

hardware and overall system is currently under de-

sign.

ACKNOWLEDGMENTS

This research is funded by the Austrian Research

Promotion Agency (FFG) within the project X-

AMINOR (881186) and the FFG-COMET-K1 Cen-

ter ”Pro2Future” (Products and Production Systems

of the Future), Contract No. 881844.

REFERENCES

Al-Ali, A.-R., Khaliq, A., and Arshad, M. (2004). Gsm-

based distribution transformer monitoring system. In

Proceedings of the 12th IEEE mediterranean elec-

trotechnical conference (IEEE Cat. No. 04CH37521),

volume 3, pages 999–1002. IEEE.

Alcantarilla, P. F., Stent, S., Ros, G., Arroyo, R., and

Gherardi, R. (2018). Street-view change detection

with deconvolutional networks. Autonomous Robots,

42(7):1301–1322.

Alvarez, D. L., Rivera, S. R., and Mombello, E. E. (2019).

Transformer thermal capacity estimation and predic-

tion using dynamic rating monitoring. IEEE Transac-

tions on Power Delivery, 34(4):1695–1705.

SMARTGREENS 2023 - 12th International Conference on Smart Cities and Green ICT Systems

126

Balntas, V., Li, S., and Prisacariu, V. (2018). Relocnet:

Continuous metric learning relocalisation using neu-

ral nets. In Proceedings of the European Conference

on Computer Vision (ECCV), pages 751–767.

Bello, J. P., Silva, C., Nov, O., Dubois, R. L., Arora, A.,

Salamon, J., Mydlarz, C., and Doraiswamy, H. (2019).

Sonyc: A system for monitoring, analyzing, and mit-

igating urban noise pollution. Communications of the

ACM, 62(2):68–77.

Borrmann, D., Elseberg, J., and N

uchter, A. (2013). Ther-

mal 3d mapping of building fac¸ades. In Intelligent

autonomous systems 12, pages 173–182. Springer.

Cejka, S., Frischenschlager, A., Faschang, M., Stefan,

M., and Diwold, K. (2018). Operation of modu-

lar smart grid applications interacting through a dis-

tributed middleware. Open Journal of Big Data

(OJBD), 4(1):14–29.

Czimmermann, T., Ciuti, G., Milazzo, M., Chiurazzi, M.,

Roccella, S., Oddo, C. M., and Dario, P. (2020).

Visual-based defect detection and classiﬁcation ap-

proaches for industrial applications—a survey. Sen-

sors, 20(5):1459.

Diwold, K., Mayer, S., Einfalt, A., Parreira, J. X.,

Hodges, J., Anicic, D., and Mosshammer, R. (2018).

Grid watch dog: A stream reasoning approach for

lightweight scada functionality in low-voltage grids.

In Proceedings of the 8th International Conference on

the Internet of Things, pages 1–8.

Graf, F. and Gruber, M. (2018). Rapid incident detection in

tunnels through acoustic monitoring–operating expe-

riences in austrian road tunnels.

Hedman, K. W., Oren, S. S., and O’Neill, R. P. (2011). A re-

view of transmission switching and network topology

optimization. In 2011 IEEE power and energy society

general meeting, pages 1–7. IEEE.

Heittola, T., C¸ akır, E., and Virtanen, T. (2018). The machine

learning approach for analysis of sound scenes and

events. In Computational Analysis of Sound Scenes

and Events, pages 13–40. Springer.

Janda, K., M

alek, J., and Re

cka, L. (2017). Inﬂuence of

renewable energy sources on transmission networks in

central europe. Energy Policy, 108:524–537.

ahler, O., Prisacariu, V. A., and Murray, D. W. (2016).

Real-time large-scale dense 3d reconstruction with

loop closure. In European Conference on Computer

Vision, pages 500–516. Springer.

Klutke, G.-A., Kiessler, P. C., and Wortman, M. A. (2003).

A critical look at the bathtub curve. IEEE Transac-

tions on reliability, 52(1):125–129.

Koester, L. W., Taheri, H., Bigelow, T. A., Bond, L. J., and

Faierson, E. J. (2018). In-situ acoustic signature mon-

itoring in additive manufacturing processes. In AIP

Conference Proceedings, volume 1949, page 020006.

AIP Publishing LLC.

Lee, E., Hong, J., and Kim, D. (2018). Automatic instru-

ment status monitoring using deep neural networks.

ECCV Demo Sessions.

Morabito, R. (2017). Virtualization on internet of things

edge devices with container technologies: A perfor-

mance evaluation. IEEE Access, 5:8835–8850.

Palazzolo, E. and Stachniss, C. (2017). Change detection

in 3d models based on camera images. In 9th Work-

shop on Planning, Perception and Navigation for In-

telligent Vehicles at the IEEE/RSJ Int. Conf. on Intel-

ligent Robots and Systems (IROS).

Park, D. H., Darrell, T., and Rohrbach, A. (2019). Robust

change captioning. In Proceedings of the IEEE/CVF

International Conference on Computer Vision, pages

4624–4633.

Pliatsios, D., Sarigiannidis, P., Lagkas, T., and Sarigianni-

dis, A. G. (2020). A survey on scada systems: secure

protocols, incidents, threats and tactics. IEEE Com-

munications Surveys & Tutorials, 22(3):1942–1976.

Pudlo, G., Tenbohlen, S., Linders, M., and Krost, G. (2002).

Integration of power transformer monitoring and over-

load calculation into the power system control surface.

In IEEE/PES Transmission and Distribution Confer-

ence and Exhibition, volume 1, pages 470–474. IEEE.

Raith, J., Bonini, C., and Scala, M. (2020). Simulation of

long-term transformer operation with a dynamic ther-

mal, moisture and aging model. In 5th International

Colloquium on Transformer Research and Asset Man-

agement, pages 211–226. Springer.

Sakurada, K., Shibuya, M., and Wang, W. (2020). Weakly

supervised silhouette-based semantic scene change

detection. In 2020 IEEE International conference on

robotics and automation (ICRA), pages 6861–6867.

IEEE.

Sattler, T., Zhou, Q., Pollefeys, M., and Leal-Taixe, L.

(2019). Understanding the limitations of cnn-based

absolute camera pose regression. In Proceedings of

the IEEE/CVF conference on computer vision and

pattern recognition, pages 3302–3312.

Seitlinger, W. (2000). A thermo-hydraulic transformer

model. Proceedings of the Conference of Electrical

Power Supply Industry.

Sels, S., Verspeek, S., Ribbens, B., Bogaerts, B., Vanlan-

duit, S., Penne, R., and Steenackers, G. (2019). A cad

matching method for 3d thermography of complex ob-

jects. Infrared Physics & Technology, 99:152–157.

Shaﬁullah, G. (2016). Impacts of renewable energy integra-

tion into the high voltage (hv) networks. In 2016 4th

International Conference on the Development in the in

Renewable Energy Technology (ICDRET), pages 1–7.

IEEE.

Shi, K., Chen, Y., Yu, B., Xu, T., Yang, C., Li, L., Huang,

C., Chen, Z., Liu, R., and Wu, J. (2016). Detecting

spatiotemporal dynamics of global electric power con-

sumption using dmsp-ols nighttime stable light data.

Applied energy, 184:450–463.

Siebald, H., Hensel, O., Beneke, F., Merbach, L., Walther,

C., Kirchner, S. M., and Huster, J. (2017). Real-

time acoustic monitoring of cutting blade sharpness

in agricultural machinery. IEEE/ASME Transactions

on Mechatronics, 22(6):2411–2419.

Sinsel, S. R., Riemke, R. L., and Hoffmann, V. H. (2020).

Challenges and solution technologies for the integra-

tion of variable renewable energy sources—a review.

renewable energy, 145:2271–2285.

X-AMINOR: A Mobile Multi-Sensor Platform for Lifecycle-Monitoring of Transformers

127

Stent, S., Gherardi, R., Stenger, B., and Cipolla, R. (2015).

Detecting change for multi-view, long-term surface

inspection.

Tabernik, D.,

Sela, S., Skvar

c, J., and Sko

caj, D.

(2019). Deep-learning-based computer vision system

for surface-defect detection. In International Con-

ference on Computer Vision Systems, pages 490–500.

Springer.

Taketomi, T., Uchiyama, H., and Ikeda, S. (2017). Visual

slam algorithms: A survey from 2010 to 2016. IPSJ

Transactions on Computer Vision and Applications,

9(1):1–11.

Tuballa, M. L. and Abundo, M. L. (2016). A review of the

development of smart grid technologies. Renewable

and Sustainable Energy Reviews, 59:710–725.

Vidas, S., Moghadam, P., and Bosse, M. (2013). 3d thermal

mapping of building interiors using an rgb-d and ther-

mal camera. In 2013 IEEE international conference

on robotics and automation, pages 2311–2318. IEEE.

Wang, C., Daneshmand, M., Dohler, M., Mao, X., Hu,

R. Q., and Wang, H. (2013). Guest editorial-special is-

sue on internet of things (iot): Architecture, protocols

and services. IEEE Sensors Journal, 13(10):3505–

3510.

SMARTGREENS 2023 - 12th International Conference on Smart Cities and Green ICT Systems

128