Content uploaded by Georgi I. Nalbantov
Author content
All content in this area was uploaded by Georgi I. Nalbantov on May 21, 2021
Content may be subject to copyright.
Predictive Maintenance Decision Support System
March 2, 2021
Predictive Maintenance Tool for Non-Intrusive Inspection
Systems
Georgi Nalbantov georgi.nalbantov@danlex.bg
Dimitar Todorov dimitar.todorov@danlex.bg
Nikolay Zografov nikolay.zografov@danlex.bg
Stefan Georgiev stefan.georgiev@danlex.bg
Nadia Bojilova nadia.bojilova@danlex.bg
Danlex EOOD, Bulgaria
Editor:
Abstract
Cross-border security is of topmost priority for societies. Economies lose billions each year
due to counterfeiters and other threats. Security checkpoints equipped with X-ray Security
Systems (NIIS-Non-Intrusive Inspection Systems) like airports, ports, border control and
customs authorities tackle the myriad of threats by using NIIS to inspect bags, air, land,
sea and rail cargo, and vehicles. The reliance on the X-ray scanning systems necessitates
their continuous 24/7 functioning being provided for. Hence the need for their working
condition being closely monitored and preemptive actions being taken to reduce the overall
X-ray systems downtime. In this paper, we present a predictive maintenance decision
support system, abbreviated as PMT4NIIS (Predictive Maintenance Tool for Non-Intrusive
Inspection Systems), which is a kind of augmented analytics platforms that provides real-
time AI-generated warnings for upcoming risk of system malfunctioning leading to possible
downtime. The industrial platform is the basis of a 24/7 Service Desk and Monitoring
center for the working condition of various X-ray Security Systems.
Keywords: Predictive maintenance, Early-warning system, Condition monitoring, Ma-
chine Learning, Asset Health Monitoring, Remote Maintenance, Early Warning Alerts
1. Introduction
In the past decades, the world economies have faced unprecedented threats to their security
and financial interests. It is estimated that up to USD 870 billion, or 1.5% of global GDP
(OECD (2016)) are lost annually due to counterfeit, illegal trafficking of goods, and other
threats. The main means for detecting various threats is invariably X-ray Security Systems,
which are used for scanning bags, air, land, sea and rail cargo, and vehicles. Other, mostly
concomitant, choices for detection are goods profiling, using meta-data for risk calculation,
and others.
The predominant reliance on the X-ray scanning systems necessitates their continuous
24/7 functioning being provided for. Hence the need for their working condition being
closely monitored and preemptive actions being taken to reduce the overall X-ray systems
downtime.
1
arXiv:2103.01044v1 [cs.LG] 1 Mar 2021
Nalbantov, Zografov, Georgiev, Todorov, and Bojilova
In this paper, we present a predictive maintenance decision support system, abbreviated
as PMT4NIIS (Predictive Maintenance Tool for Non-Intrusive Inspection Systems), which
is a kind of augmented analytics platforms that provides real-time AI-generated warnings
for upcoming risk of system malfunctioning leading to possible downtime. The industrial
platform is the basis of a 24/7 Service Desk and Monitoring center for the working con-
dition of various X-ray systems. To the best of our knowledge, this is the first predictive
maintenance solution for non-intrusive inspection systems for non-medical purposes.
The paper is organized as follows. Section 2 presents a description of the predictive
maintenance platform, including IT infrastructure, security considerations, and AI pre-
dictive models for systems’ health status and risk of forthcoming malfunctions. Section 3
gives examples of real-time AI-model predictions, both short-term and long-term, which are
brought to the attention of X-ray system service specialists. Section 4 discusses implications
for ends users such as airports, national customs agencies, and X-ray system maintenance
service providers. Section 5 concludes.
2. Description of the predictive maintenance platform for X-ray systems
Figure 1: PMT4NIIS Decision Support System design.
The structure of the predictive maintenance tool for non-intrusive (X-ray based) inspec-
tion systems (PMT4NIIS) consists of the following elements: an IoT device for extracting
relevant technical data from X-ray systems, a Smart Data Infrastructure for secure data
transfer and data collection, an AI Engine for generating alerts in case of short-term and/or
2
Predictive Maintenance Decision Support System
long-term predictions for high risk of X-ray systems malfunction, and a front-end BI dash-
board for visualization of current system parameters as well as AI-generated alerts and
warnings. The users of the BI dashboard are mainly service specialists, who receive the
output of the decision support system and take an informed decision on whether to in-
tervene and perform system maintenance or do not take any action. Other users of the
BI dashboard are end users (airports, customs agencies, port administrations, etc.), who
are interested in monitoring the health state and availability of their (X-ray) assets. The
elements of PMT4NIIS who they are organized in one overall system are depicted in Figure
1.
The assets, which state is being monitored, are various types of industrial non-medical
X-ray systems. For the airports, these are predominantly low-energy X-ray imaging systems
for carry-on luggage, hand-baggage, as well as air-cargo goods. For border customs, these
are high-energy systems that generate X-ray images for trucks and trains content check-up.
And for ports, these are high-energy imaging systems for containers. All X-ray systems
are subject to wear-and-tear, and therefore require parts to be replaced, as they break
and/or malfunction. The traditional maintenance service models are based on preventive
(performed at predetermined intervals) and corrective (run-until-it-breaks) maintenance
approaches. Both of these services are not efficient (Milojevic and Nassah (2018)) and in
many industries they are currently being replaced by Condition Based Maintenance (CBM)
and Predictive Maintenance (PdM), which ensure the reduction in breakdowns, reduction
in maintenance cost and increase of the production (Roland Berger Strategy Consultants
(2014)). What we propose is a shift from the traditional models into Service 4.0 models,
which involve real-time asset monitoring and AI-based asset-health alert system, which
help anticipate upcoming malfunction problems and consequently prepare for their repair
in advance. These actions both save operating costs and increase X-ray systems availability,
which are two crucial KPIs.
The data-extraction element of PMT4NIIS involves extraction of machine log files gen-
erated by the X-ray systems, which provide relevant data about current technical state. In
case additional technical data is needed for better monitoring, which is not readily avail-
able in the log files, we have introduced a multi-purpose IoT device, called DanlexBox,
for automatic collection and transfer of additional technical and environmental data. The
DanlexBox is interoperable across interfaces and system boundaries with autonomous AI-
driven functions for processing and cleaning data. All technical and environmental data
is collected and transfered in (near) real time using a private APN channel via the Dan-
lexBox IoT device. An alternative approach is to store the data on the IT infrastructure
of the owner of X-ray scanners, and then transfer the data from there. All technical data
is transferred to a centralized location (outside the IT infrastructure of the owner of X-ray
scanners) with a Smart Data Infrastructure, where it is stored and delivered to a predictive
analytics environment for post-processing. DanlexBox is a cyber resilient device supporting
the Open Architecture for Airport Security Systems initiative (Kramer et al. (2020)).
The predictive analytics environment takes as input raw technical data from X-ray
systems. Based on them, it generates and provides predictions for the current health state
of the systems. In essence, an AI engine, equipped with AI prediction models, provides
short-term (to up a month) and long-term (survival) estimates for the risk of X-ray system
stoppage/malfunction. In such a way, both short-term reparation actions as well as long-
3
Nalbantov, Zografov, Georgiev, Todorov, and Bojilova
term inventory management activities can be planned in advance. The AI models are
augmented with rule-based physical models, which generate alarms based on predefined
thresholds of certain censor and other technical readings (including error messages, electrical
parameters, etc.). All predictions are stored in an SQL database.
The relevant technical X-ray-system data, available either via the machine log or col-
lected via the DanlexBox IoT device, together with the asset health-state predictions, are
brought to the attention to service specialists in a 24/7 monitoring center. The service
specialists monitoring center are responsible for acting upon generated alarms, where the
possible actions are: refute the alarm, accept the alarm, take relevant action based on
accepted alarm, and escalating alarm to a higher level. The output of the monitoring cen-
ter is: 24/7 surveillance of the health-state of X-ray systems, report-generation in case of
maintenance action is recommended/advised by the service specialists, and generation of
knowledge-base for solved cases. The service specialists can be alerted not only via the
available monitoring web dashboards of the PMT4NIIS web site but also via email and
mobile phone notification.
Figure 2: An example use case of the PMT4NIIS decision support system. AI-based gener-
ated alerts (colored) as well as associated system-data events are being displayed
in a web front end for a chosen period of time (1 dec 2019 - 17 dec 2019). These
alerts are being reviewed by maintenance service specialists and physicists and
further action is taken where necessary.
The overall output of the PMT4NIIS decision support system to the X-ray owners
and X-ray systems maintenance service providers are (i) web dashboards to check over-
all system availability and health state, (ii) recommended actions in case of high risk of
system malfunction, and (iii) access to ever-increasing knowledge base for solving malfunc-
tion problems. The performance of the decision support system is measured in terms of
4
Predictive Maintenance Decision Support System
improved operating-efficiency and reduced X-ray systems downtime, resulting in improved
financial/economic KPIs.
3. Use case
In this section we present an example use case of the PMT4NIIS decision support system
(see Figure 2). The period in question is 1 dec 2019 - 17 dec 2019. The X-ray systems
being monitored are high-energy systems located at cross-border checkpoints 1. In the case
at hand, four technical service reports, which are based on on-site visits, are being pro-
vided and used by service specialists to augment the knowledge base for problem resolution
management. Various sorts of alerts are being generated through time and brought to the
attention of the service specialists at the 24/7 monitoring center. The associated (log)
events on which the alerts are being based are displayed as well. In many cases the AI
predictive engine generates alerts based on a collection of events, each one participating
with its relative weight computed by an AI model. At other times, one event is sufficient to
generate an alert, for example when the alert is rule-based (referred to as condition-based
alert).
For each alert, the service specialists at the monitoring center decide whether to accept
the alert or label it as a false alarm. In this case the alert is labeled as a false alarm, it is
provided as feedback to the AI engine, which takes it into account in a re-training process.
In such a way, the predictive models increase their performance with time. If an alert is
accepted, it is either tackled by the service specialists at hand or escalated to a higher level.
A technical report is being generated for each solved case. The alerts generated in this
case study have helped the service specialists diagnose an X-ray system remotely, and in a
timely fashion, without losing any travel time to the location of the X-ray system.
4. Discussion
We have presented the first, to our best knowledge, an online decision support system for
remote monitoring of X-ray non-intrusive inspection systems. It’s main features include a
secure transfer of machine-log data from X-ray systems to a centralized location with Smart
Data Infrastructure. The data is being provided to an AI engine, which applies continuously
short-term and long-term models for X-ray systems health state. In case of alerts being
generated, they are displayed on a dedicated web-based BI dashboard. The BI dashboard
is a web site, where all relevant system data is displayed in near-real time, together will all
AI-generated alerts. A monitoring center, which consists of service specialists, takes care of
handling the alerts. In case of detected or predicted problems, preemptive action is being
taken.
Similar kinds of online decision support systems are already successfully applied in a
large variety of other industries. Canizo et al. (2017) present a predictive maintenance AI-
based solution for real-time prediction of wind turbines failures. Fausing Olesen and Shaker
(2020) describe a data-driven predictive maintenance approach for pump systems and ther-
mal power plants based on a variety of technical inputs. Massaro et al. (2018) suggest a
predictive maintenance approach for milk production lines, based on inputs such a tempera-
1. We do not disclose the location of these systems for security reasons.
5
Nalbantov, Zografov, Georgiev, Todorov, and Bojilova
ture. By collecting valuable data generated by sensors, using IoT devices for data collection
and cloud computing for analyzing data C¸ oban et al. (2018) have proposed a predictive
maintenance architecture for medical equipment for improved health-care services.
5. Conclusion
It is important to societies to have the technology to inspect, in a non-intrusive way, goods
at locations such as airports, cross-border checkpoints, and ports. Such technology exists in
the form of X-ray non-intrusive inspection systems. These systems need to be operational
at all times, which is currently a bottleneck, which needs to be addressed. To help ensure
the smooth functioning of these systems at all times, we have developed a comprehensive
decision support system, which reduces the reaction time in case a problem is detected,
and also anticipates short-term and long-term problems by employing advanced AI models.
A 24/7 monitoring center is indispensable for quicker diagnostic and reaction times, which
ultimately results in increased X-ray system availability and improved operational efficiency.
The long term effectiveness of this approach vis-a-vis the predominant off-line, run-to-failure
and regular-time maintenance approaches should be further researched to provide insights
into way to improve industrial and financial KPIs.
References
Mikel Canizo, Enrique Onieva, Angel Conde, Santiago Charramendieta, and Salvador
Trujillo. Real-time predictive maintenance for wind turbines using big data frame-
works. In 2017 IEEE International Conference on Prognostics and Health Manage-
ment, ICPHM 2017, Dallas, TX, USA, June 19-21, 2017, pages 70–77. IEEE, 2017. doi:
10.1109/ICPHM.2017.7998308. URL https://doi.org/10.1109/ICPHM.2017.7998308.
Jonas Fausing Olesen and Hamid Reza Shaker. Predictive maintenance for pump systems
and thermal power plants: State-of-the-art review, trends and challenges. Sensors, 20
(8), 2020. ISSN 1424-8220. doi: 10.3390/s20082425. URL https://www.mdpi.com/
1424-8220/20/8/2425.
E. Kramer, R. Dempers, and J. Paulshus. Open architecture for airport secu-
rity systems, 2020. URL https://www.aci-europe.org/downloads/resources/Open%
20Architecture%20for%20Airport%20Security%20Systems_1st%20Edition.pdf.
A. Massaro, Vincenzo Maritati, A. Galiano, V. Birardi, and Leonardo Pellicani. Esb plat-
form integrating knime data mining tool oriented on industry 4.0 based on artificial
neural network predictive maintenance. International Journal of Artificial Intelligence &
Applications, 9:01–17, 2018.
Milos Milojevic and Franck Nassah. Digital industrial revolution with predictive main-
tenance, 2018. URL https://www.plm.automation.siemens.com/media/global/
en/PAC_Predictive_Maintenance_Siemens_Executive_Summary_2018-71130_
tcm27-33237.pdf.
S. C¸ oban, M. O. G¨okalp, E. G¨okalp, P. E. Eren, and A. Ko¸cyi˘git. [wip] predictive main-
tenance in healthcare services with big data technologies. In 2018 IEEE 11th Confer-
6
Predictive Maintenance Decision Support System
ence on Service-Oriented Computing and Applications (SOCA), pages 93–98, 2018. doi:
10.1109/SOCA.2018.00021.
OECD. Illicit Trade. 2016. doi: https://doi.org/https://doi.org/10.1787/
9789264251847-en. URL https://www.oecd-ilibrary.org/content/publication/
9789264251847-en.
Roland Berger Strategy Consultants. Predictive maintenance: ss the timing
right for predictive maintenance in the manufacturing sector?, 2014. URL
https://www.rolandberger.com/publications/publication_pdf/roland_berger_
predictive_maintenance_20141215.pdf.
7
