=Paper=
{{Paper
|id=Vol-2370/paper-01
|storemode=property
|title=A Big Data Perspective on Cyber-Physical Systems for Industry 4.0: Modernizing and
Scaling Complex Event Processing
|pdfUrl=https://ceur-ws.org/Vol-2370/paper-01.pdf
|volume=Vol-2370
|authors=Carina Andrade
|dblpUrl=https://dblp.org/rec/conf/caise/Andrade19
}}
==A Big Data Perspective on Cyber-Physical Systems for Industry 4.0: Modernizing and
Scaling Complex Event Processing==
https://ceur-ws.org/Vol-2370/paper-01.pdf
A Big Data Perspective on Cyber-Physical Systems for
Industry 4.0: Modernizing and Scaling Complex Event
Processing
Carina Andrade1[0000-0001-8783-9412]
1
ALGORITMI Research Centre, University of Minho, Guimarães, Portugal
carina.andrade@dsi.uminho.pt
Abstract. Nowadays, organizations have devices integrated into their business
processes and producing data that reflects the events happening in their systems.
This data relevancy is widely recognized by the community, but there is no com-
mon system architecture that categorizes how and what can be done with this
streaming data to maximize its usefulness. This document discusses the work
proposed for a doctoral thesis in this research topic, presenting: i) the main goal,
objectives and expected contributions; ii) the state-of-the-art that supports the
identified problem, describing some system architectures and relating them with
the architecture being proposed; iii) the research methodology; iv) and, the logi-
cal and technological system architecture and preliminary results of the demon-
stration case at Bosch Car Multimedia Portugal.
Keywords: Big Data, Complex Event Processing, Cyber-Physical Systems,
Rules Engine, Machine Learning.
1 Introduction
Currently, several business sectors are trying to catch up with the Big Data era, namely
dealing with multiple data sources, in different formats, and different velocities. With
the Internet of Things (IoT) proliferation, organizations can have their business pro-
cesses complemented with sensors that produce event data contributing to monitoring
their processes. Considering the existence of these hardware capabilities, some tech-
nologies to handle all the data that is constantly being produced are required, such as
Spark1, Druid2 or Storm3, which are of major relevance for processing, aggregating or
analyzing streaming data in real time. Besides the technologies needed to handle the
streaming data, some technologies for Big Data Warehousing (e.g. Hive4) can be rele-
vant to complement the streaming data analysis. Considering these technologies, the
system’s scalability regarding data processing and storage is guaranteed. However,
1
https://spark.apache.org/
2
http://druid.io/
3
http://storm.apache.org/
4
https://hive.apache.org/
2
concepts such as Complex Event Processing (CEP) and rule-based technologies (e.g.,
Drools5) are called into this context to allow the processing of different types of events,
finding patterns between them and using rules for that, reflecting the business require-
ments identified in the context of each organization. The integration of these concepts
and technologies is already mentioned in some identified works, but it is not considered
that they can be complemented with Machine Learning (ML) techniques that will allow
the system to make predictions or recommendations using pre-determined ML models
over events that arrive at the system. In addition, no other work mentions the im-
portance of a complete and innovative visualization component for monitoring this type
of systems, which is being considered in this work.
In this context, the goal of this doctoral thesis supervised by Professors Maribel Yas-
mina Santos and Carlos Costa, is the proposal of a logical and technological system
architecture that provides to organizations the capability of using all their event data in
real-time fashion, considering: i) business requirements that should be easily integrated
into the system; ii) powerful ways of doing predictions and recommendations to im-
prove daily operations; iii) system self-management and monitoring to prevent uncon-
trolled growth of the system. As can be observed in Section 2, the idealized system
architecture addressing the mentioned points was not identified in the literature review,
reason why this work discusses the proposal of a CEP system for the Big Data era.
This document is divided as follows: Section 2 presents the state-of-the-art; Section
3 mentions the expected contributions; Section 4 dissects the research methodology;
Section 5 highlights the proposed approach and current results; and, Section 6 summa-
rizes the presented and future work.
2 State-of-the-Art
A CEP system can be described as a system that analyses events through different per-
spectives like pattern matching or inference. In these processes, the system can filter
and aggregate the relevant information, complementing it with external data [1]. Be-
sides the concept itself, [2] presents CEP systems as being a challenge for the Data
Stream Management Systems, considering that besides processing the data in real-time,
it is necessary to take action over it. The Rapide project [3, 4] is often considered as the
first work to explore the CEP concept [1, 5]. This project started in the 90s and provided
the capability of identifying the temporal and causal relationship among events. How-
ever, the current amount of available data requires improvements to this concept, adapt-
ing it to Big Data environments.
In this context, and analyzing the existing architectures that aim to integrate the CEP
and Big Data concepts, the work of [6] uses the recognized Lambda and Kappa archi-
tectures as the base for the proposal of BiDCEP, an architecture that integrates CEP
and Big Data Streaming concepts. A relevant point mentioned by the authors is the
relevance of the IoT concept, which should be considered as an enabler for descriptive,
predictive and prescriptive analytics, although it is not noticeable where these aspects
5
https://www.drools.org/
3
are considered in the proposed architecture. Another architecture is highlighted in the
FERARI project context [7], with a prototype for real-time CEP that process vast
amounts of event data in a distributed way. These two architectures share some princi-
ples, such as the components responsible for the connection to the data sources; the
components associated with the event processing; and, the components related to the
data consumers.
The main properties that should be considered in the development of a Big Data CEP
system are mentioned by [8]: parallelism, elasticity, multi-query and distributed re-
sources. However, the authors also mention that, although there are some works that
try to address these issues, the integration of CEP and Big Data technologies is still
something significantly unexplored.
Nevertheless, more than integrating CEP in Big Data contexts, some works [9–11]
emphasize the relevance of combining Big Data, CEP, and IoT to support the manufac-
turing industry through Cyber-Physical Systems (CPSs). The work of [9] proposes a
framework for a Manufacturing CPS that considers the physical world (e.g., manufac-
turing facilities and shop-floor resources); the cyber world (e.g., simulation and predic-
tion models); and, the interface between these two worlds (e.g., sensor networks and
structured and unstructured data). In this case, the CEP system is a component in the
cyber world, responsible for processing events and return results in (near)real-time that
could provide operational visibility and awareness for the manufacturing system. In
[10], the authors explore the use of event-based predictions for manufacturing planning
and control. In this case, sensors installed in the manufacturing plant are the data source
for the CEP system that, combining the events with historical data, will provide the
possibility of achieving the event-based prediction models for production planning and
control. The work of [11] proposes a framework that can be applied to monitor the
status of a CPS through the use of IoT data. This framework presents a Publish-
Subscribe Messaging System that receives the data for further identification of
meaningful events in a rule-based CEP System (running in a distributed way). The
processing results are published in the self-healing mechanism and predictive
maintenance component for the execution of the actions previously defined.
3 Expected Contributions
Although the main concepts and components of the analyzed architectures are being
considered in the architecture to be proposed in this doctoral thesis, current contribu-
tions: i) do not clearly and rigorously refer or detail how CEP and Big Data concepts
and technologies can act together for distributed data, rules and events processing with
(near)real-time aggregations and Key Performance Indicators (KPIs) at data ingestion
time; ii) do not combine batch data arriving from a Big Data Warehouse [12], comple-
menting the event data that arrives in a streaming way and bringing more value to the
results and actions of the system; iii) do not discuss techniques similar to the ML mod-
els lake component (presented in section 5), which can be significantly helpful for pat-
terns discovery and it is identified as a major gap in these systems [13]; and, iv) do not
4
consider the relevance of monitoring the functioning and evolution of this type of sys-
tems that can quickly become untraceable in Big Data contexts.
Beyond the current results presented in Section 5, this work reveals to be of signifi-
cant relevance to several contexts considering that the proposed system architecture
should be generic to be applied in different areas, such as industry, smart cities, agri-
culture, among others, where several data sources are involved (as it is intended to be
shown during this thesis, using different demonstration cases). Consequently, this sys-
tem can be considerably helpful for the monitoring of business processes and events,
also using complementary data, to prevent possible problems through the capabilities
of its Predictors and Recommenders component, which is directly linked to the actions
that could be triggered, components not seen in other identified works. In industrial
contexts, the role of this system can be easily identified in the manufacturing shop floor
where machines and other interconnected devices are increasing, or even for the anal-
ysis of customers’ reviews in social media, for example. In this context, this system
can, for example, monitor the production data and, if some business rule is activated
(e.g., for a defective product), predict if the next products will also be defective products
and, if true, stop the production machine. Also, the system can predict and prevent a
brand crisis related to an organization's publication or a defective product, which is
generating negative comments in social media. In smart cities and agricultural contexts,
the goal is fundamentally the same, i.e., use the data that is being produced by several
sources and process it in (near)real-time, sending, for example, an accident warning to
a street screen or changing the amount of water for irrigation due to a temperature
change.
4 Research Methodology
This work follows a design science [14] research approach with the goal of extension
of the boundaries of human or organizational capabilities through the creation of new
and innovative artifacts, in this case, proposing a solution for organizations pursuing
(near)real-time decisions and automated actions based on Big Data streams and CPSs.
In this context, [14] presents a set of guidelines to conduct design science research in
Information Systems: i) propose an artifact to address an organizational problem; ii)
understand the relevance of the problem that is being solved; iii) evaluate the utility,
quality and efficacy of the proposed artifact; iv) provide clear and verifiable research
contributions; v) apply rigorous methods on the artifact development and evaluation;
vi) consider the design of the artifact as a search process utilizing the available means
to fulfil the goals and satisfy the constraints; and, vii) present the research to technology
practitioners and researchers, as well as management-oriented audiences. Moreover,
[14] considers that the Information Technology (IT) artifacts, resulting from a design
science research process, can be: constructs (vocabulary and symbols); models (ab-
stractions and representations); methods (algorithms and practices); and, instantiations
(implementations and prototype systems).
Therefore, the widely recognized Design Science Research Methodology for Infor-
mation Systems [15] is used in this doctoral thesis, considering an objective-centred
5
approach for the designing of a logical and technological system architecture that must
meet the following objectives:
1. Handle Big Data produced by several sources inside and outside the organizations
(e.g., data from production lines, cars, citizens, smartphones, among others);
a. Consider several possible data sources and their interfaces’ differences, in order
to design a system that ensures that new data sources can be easily added;
b. Consider the volume, variety and velocity of the data that could arrive at the sys-
tem, in order to define its scalability, multi-query, parallelism and distributed re-
sources characteristics.
2. Consider the business requirements and indicators defined by organizations, besides
the data itself;
3. Process the data within the time frame needed for several decision makers, with or-
ganizations providing inputs regarding the latency for the different system use cases,
as this can be used to evaluate the timeliness of the automated actions triggered by
the system;
4. Provide predictions and recommendations for the organization’s daily activities;
a. Design and evaluate an adequate system architecture to efficiently integrate pre-
dictive and prescriptive ML models into the streams of data/event processed by
the system, providing high throughput and low latency predictions/prescriptions.
5. Autonomously execute adequate actions to avoid considerable problems for the or-
ganization (e.g., stop a production machine);
a. Design and evaluate the most adequate way to communicate with external sys-
tems, executing automated actions directly related to the business requirements
(rules) and indicators defined by the organization.
6. Consider the relevance of self-management and monitoring, preventing the uncon-
trolled growth of the system with the constant monitoring and visualization of what
happened in the system (e.g., Which producers are introducing more data into the
system? What are the most triggered actions?).
a. Design and evaluate the monitoring system and the visualization platform that
should consider strategic endpoints in which data is collected and provide user-
friendly analysis of the system status (e.g., immersive and drill-down visualiza-
tion, virtual or augmented reality).
Considering these objectives, some metrics were identified as relevant for the system
evaluation: scalability and complexity when integrating new data sources, producers
and consumers in the system; number of events produced/consumed per second; num-
ber of rules, actions and ML models verified, triggered and applied per second, respec-
tively; average response time of the application of ML models; and, usefulness of the
analyses made available in the system’s monitoring platform. These metrics can be
evaluated considering the organizations’ requirements and/or using baselines and
guidelines identified in the literature review (e.g., throughput, latency and scalability
benchmarks).
The results of a first iteration (after finishing the second of the four years of the
doctoral program) on the Design and Development phase of the research methodology
are presented in Section 5, already fulfilling the 1st, 2nd, 3rd and 5th objectives discussed
6
above through a real-world prototype implementation at Bosch Car Multimedia Portu-
gal, although it still needs a rigorous evaluation.
5 Proposed Approach and Current Results
This section presents the first iteration of the Design and Development phase of the
research methodology explained in Section 4, using a proof of concept based on the
Active Lot Release application from Bosch Car Multimedia Portugal as a demonstra-
tion case. Taking into consideration the organization’s needs and the gap found in the
literature review, a system architecture is being proposed to fill these needs. This sys-
tem, named Intelligent Event Broker, aims to represent a Big Data-oriented CEP system
that combines a collection of software components and data engineering decisions, in-
tegrated to ensure their usefulness, efficiency and harmonious functioning, in order to
process the events that arrive in the system.
The proposed architecture for the Intelligent Event Broker (Fig. 1) considers a vast
number of components for dealing with the volume, variety and velocity of the data:
1. Source Systems: the system architecture should be prepared to receive data from sev-
eral sources: relational, NoSQL or NewSQL databases, IoT Gateways or (Web)Serv-
ers, and even components of the Hadoop ecosystem, such as Hive Tables or HDFS
files;
2. Producers: to ensure that all the Source Systems identified in 1. can be integrated
into the system, regardless of their communication interfaces, the Producers com-
ponent is proposed to standardize the collection of events entering the system.
Kafka6 (a distributed streaming platform) is proposed for the deployment of this
component;
3. Broker Beans: the events collected in the Source Systems by the Kafka Producers
are serialized into the form of classes that define the several business entities existing
in the system – Broker Beans;
4. Brokers: events serialized into Broker Beans, can be published by the Producers into
a Kafka topic that is stored in a cluster of Kafka Brokers;
5. Event Processor: events are subscribed by the Event Processor (Kafka Consumers
that are embedded into Spark Applications) that are always waiting for processing
the events arriving at the system, regardless of their frequency;
6. Complementary Data: in addition to the events published in the topics, the Event
Processor can use Complementary Data from the Source Systems, if useful for the
event processing, providing additional and relevant information;
7. Rules Engine: includes the defined rules that represent the Business Requirements
(with Strategical, Tactical or Operational rules). This work is usually associated to a
Data Engineer that creates the rules that represent the business needs. Here Drools
is used to store the rules that will be then translated by the Event Processor, using a
custom-made integration of Spark and Drools, based on previously explored paths
by the technical community [16].
6
https://kafka.apache.org/
7
8. Triggers: connectors to the Destination Systems, execute the actions previously
defined for the rules when the condition is evaluated as being true.
Fig. 1. System Architecture
9. Destination Systems: the results of processing the events can be sent to, for
example, IoT Gateways that can activate an actuator, Text or E-mails Messages, or
even Transactional or Analytical Applications;
10. Predictors and Recommenders: the concept of a Lake of ML Models, which are
trained beforehand, is proposed in this system architecture as being of major
relevance. Allow the application of those ML models to the data that is being
processed, providing the capability to predict occurrences or recommend actions
based on the events that are arriving at the system;
11. Event Aggregator: stores the raw event data (events that arrived at the system) or
processed event data (the events processing result, such as results from the
Predictors and Recommenders component) used to calculate the KPIs relevant to
the business. This component is supported by Druid, a columnar storage system
useful for aggregating event data at ingestion time [17, 18];
12. Mapping and Drill-down System: allows the constant monitoring of the Intelligent
Event Broker and includes:
a. Graph Database: stores the relevant metadata of the Intelligent Event Broker,
allowing the exploration of the flows of the events in the Intelligent Event Broker;
b. Web Visualization Platform: provides an interactive and immersive visualization
regarding the Intelligent Event Broker metadata, stored in the Graph Database,
taking into account the various implementation contexts of the system.
8
Currently, a demonstration case was already implemented using data from the Bosch
Car Multimedia Portugal plant [19]. This data comes from its ALR System that supports
the quality control used in the manufacturing and packaging processes. This system is
based on rules that are applied to the products contained into lots before they are
shipped to customers. The ALR system provides a stream of events that contain
information about the quality control process, being considered, at this point, the lot
identification, its packaging date, the production line, and the status (“Valid” or “Inva-
lid” lot).
Therefore, for this demonstration case, one Operational Rule and two Tactical Rules
were defined, as well as their own Triggers (store data into Cassandra for further anal-
ysis and send an E-mail Message to a stakeholder) that are activated if the result for the
rule condition is true. Considering these two types of rules, two types of dashboards
were created on the Analytical Application, one oriented for operational analysis and
the other one oriented for a more tactical point of view. These dashboards, and more
details about the proposed aproach and current results, can be seen in [19].
6 Conclusions and Future Work
Considering the current evolution of the industrial world, this document presents the
status and main research goals of a doctoral thesis that is dedicated to exploring the Big
Data and CEP concepts in the Industry 4.0 movement. The context for the emergence
of this topic of interest, as well as the research agenda, were explained in this work, and
the state-of-the-art was detailed to highlight the contributions that distinguish this work
from the already existing contributions. From the methodological point of view, the
research methodology was presented, and the current status of the proposed contribu-
tion was also highlighted.
At this moment, a first version of the system was already implemented with the
Bosch Car Multimedia Portugal demonstration case, where the 1st, 2nd, 3rd and 5th ob-
jectives discussed in Section 4 were fulfilled. With this prototype, a CEP system in Big
Data contexts that reveals its adequacy to the problem and contains several components
already developed (e.g., Data Producers and Consumers, Rules and Triggers with the
business requirements, an Event Aggregator and an Analytical Application as Destina-
tion System) is presented to practitioners and researchers.
As future work, the components of the architecture should be quantitatively evalu-
ated (e.g., benchmarking), as mentioned in the methodology section. Moreover, three
other relevant components will be developed and evaluated: Predictors and Recom-
menders based on ML; Graph Database for metadata management; and, Web Visuali-
zation Platform for the system’s monitoring.
Acknowledgements. This work has been supported by FCT – Fundação para a
Ciência e Tecnologia, Projects Scope UID/CEC/00319/2019 and PDE/00040/2013, and
the Doctoral scholarship PD/BDE/135101/2017. This paper uses icons made by
Freepik, from www.flaticon.com.
9
References
1. Leavitt, N.: Complex-Event Processing Poised for Growth. Computer. 42, 17–20 (2009).
doi:10.1109/MC.2009.109
2. Chakravarthy, S., Qingchun, J.: Stream Data Processing: A Quality of Service Perspective:
Modeling, Scheduling, Load Shedding, and Complex Event Processing. Springer US (2009)
3. Luckham, D.C.: Rapide: A Language and Toolset for Simulation of Distributed Systems by
Partial Orderings of Events. Stanford University, Stanford, CA, USA (1996)
4. Luckham, D.C., Vera, J.: An event-based architecture definition language. IEEE Transac-
tions on Software Engineering. 21, 717–734 (1995). doi:10.1109/32.464548
5. Cugola, G., Margara, A.: Processing Flows of Information: From Data Stream to Complex
Event Processing. ACM Comput. Surv. 44, 15:1–15:62 (2012).
doi:10.1145/2187671.2187677
6. Hadar, E.: BIDCEP: A vision of big data complex event processing for near real time data
streaming position paper - A practitioner view. In: CEUR Workshop Proceedings (2016)
7. Flouris, I., Manikaki, V., Giatrakos, N., Deligiannakis, A., Garofalakis, M., Mock, M., Bo-
the, S., Skarbovsky, I., Fournier, F., Stajcer, M., Krizan, T., Yom-Tov, J., Curin, T.:
FERARI: A Prototype for Complex Event Processing over Streaming Multi-cloud Plat-
forms. In: Proceedings of the 2016 International Conference on Management of Data. pp.
2093–2096. ACM, New York, NY, USA (2016)
8. Flouris, I., Giatrakos, N., Deligiannakis, A., Garofalakis, M., Kamp, M., Mock, M.: Issues
in complex event processing: Status and prospects in the Big Data era. Journal of Systems
and Software. 127, 217–236 (2017). doi:10.1016/j.jss.2016.06.011
9. Babiceanu, R.F., Seker, R.: Manufacturing Cyber-Physical Systems Enabled by Complex
Event Processing and Big Data Environments: A Framework for Development. In: Service
Orientation in Holonic and Multi-agent Manufacturing, Studies in Computational Intelli-
gence. pp. 165–173. Springer International Publishing Switzerland 2015 (2015)
10. Krumeich, J., Jacobi, S., Werth, D., Loos, P.: Big Data Analytics for Predictive Manufac-
turing Control - A Case Study from Process Industry. In: 2014 IEEE International Congress
on Big Data. pp. 530–537. , Anchorage, AK, USA (2014)
11. Dundar, B., Astekin, M., Aktas, M.S.: A Big Data Processing Framework for Self-Healing
Internet of Things Applications. In: 12th International Conference on Semantics, Knowledge
and Grids (SKG). pp. 62–68. , Beijing, China (2016)
12. Costa, C., Santos, M.Y.: Evaluating Several Design Patterns and Trends in Big Data Ware-
housing Systems. In: Krogstie, J. and Reijers, H.A. (eds.) Advanced Information Systems
Engineering. pp. 459–473. Springer International Publishing (2018)
13. Tawsif, K., Hossen, J., Emerson Raja, J., Jesmeen, M.Z.H., Arif, E.M.H.: A Review on
Complex Event Processing Systems for Big Data. In: 2018 Fourth International Conference
on Information Retrieval and Knowledge Management (CAMP). , Kota Kinabalu, Malaysia
(2018)
14. Hevner, A.R., March, S.T., Park, J., Ram, S.: Design Science in Information Systems Re-
search. MIS Q. 28, 75–105 (2004)
15. Peffers, K., Tuunanen, T., Rothenberger, M.A., Chatterjee, S.: A Design Science Research
Methodology for Information Systems Research. Journal of Management Information Sys-
tems. 24, 45–77 (2007). doi:10.2753/MIS0742-1222240302
16. Ganta, M.: How-to: Build a Complex Event Processing App on Apache Spark and Drools,
https://blog.cloudera.com/blog/2015/11/how-to-build-a-complex-event-processing-app-on-
apache-spark-and-drools/
10
17. Yang, F., Tschetter, E., Léauté, X., Ray, N., Merlino, G., Ganguli, D.: Druid: A real-time
analytical data store. In: Proceedings of the 2014 ACM SIGMOD International Conference
on Management of Data. pp. 157–168. ACM, Utah, USA (2014)
18. Correia, J., Santos, M.Y., Costa, C., Andrade, C.: Fast Online Analytical Processing for Big
Data Warehousing. Presented at the International Conference on Intelligent Systems, Ma-
deira Island, Portugal September (2018)
19. Andrade, C., Correia, J., Costa, C., Santos, M.Y.: Intelligent Event Broker: A Complex
Event Processing System in Big Data Contexts. In: AMCIS 2019 Proceedings. Cancun
(2019). Manuscript accepted for publication.