Accelerating Industrial IoT Application Deployment

through Reusable AI Components

Senthamiz Selvi Arumugam∗ , Ramamurthy Badrinath† , Aitor Hernandez Herranz‡ , Jan Höller‡ ,
Carlos R. B. Azevedo‡ , Bin Xiao‡ and Valentin Tudor‡
Ericsson Research,
∗ Chennai, India, † Bangalore, India, ‡ Stockholm, Sweden,
Email: {senthamiz.selvi.a, ramamurthy.badrinath, aitor.hernandez.herranz, jan.holler,
carlos.azevedo, bin.xiao, valentin.tudor}@ericsson.com

Abstract—The Internet of Things (IoT) is penetrating almost engineering practices can be costly and time consuming if
all sectors of the global economy, addressing a wide range there is no or little reuse of software and solution practices
of opportunities by applying different Artificial Intelligence across the range of applications in different sectors.
(AI) tools to IoT data. Due to the diversity in challenges and
applications, IoT solutions are often bespoke and highly domain Our proposition and overall approach is to identify repli-
specific. With the surge of IoT applications, this approach to cability of applications by identifying recurring use case
solutions becomes very costly and time consuming if there is a patterns, and reusability of different AI-based software com-
lack of reusability and replicability across different IoT sectors. ponents that are part of an IoT AEP, advancing the work
This work presents a step towards reusability of IoT solution presented in [3]. This process involves the identification of
components applied to Industrial IoT (IIoT). We start from
the challenging position of two unique AI-driven applications a set of reusable software components with an appropriate
stemming from two separate IIoT verticals - applications which level of abstraction, with each component being application
may be realized using the same components. We identify a and domain independent, yet having well defined purposes.
set of application independent reusable AI-centric components Further on, by orchestrating the components according to
and show how they can be orchestrated into the unique IoT an appropriate blueprint, the particular application is re-
applications. Our approach shortens the time to market and
reduces costs for developing IIoT solutions, and opens a path alized. This paper focuses on the reusability aspect. We
towards reusability and replicability of IIoT components, thus demonstrate the reusability of four different and concrete AI-
accelerating the IoT market uptake. centric application-independent components by constructing
Keywords-Internet of Things (IoT); Industrial IoT (IIoT); two diverse IIoT applications from two separate industry
reusable components; application development; orchestration; domains: Smart Logistics from Intelligent Transport Systems
smart contracts; Artificial Intelligence (AI) planning; condition (ITS) and Predictive Maintenance from Manufacturing.
monitoring The rest of the paper is structured as follows: in Section II
we present the two use cases and identify the reusable com-
I. I NTRODUCTION ponents, which are described in more detail in Section III. In
IoT can be applied in virtually any sector of the global Section IV we present the related work, and in Section V we
economy or society, and in each sector there is an abundance highlight the main findings and point to the future direction.
of application examples that address different use cases. The
application diversity is very wide and problem solving easily II. I NDUSTRIAL I OT A PPLICATIONS WITH R EUSABLE
becomes niche and highly domain specific. The prevalent C OMPONENTS
practice today is to build IoT end-user applications on top The selected IIoT applications, Smart Logistics and Pre-
of an IoT Application Enablement Platform (AEP) that dictive Maintenance, are rich in terms of involved actors,
abstracts a set of common functions from the specific appli- associated data and their process flow interactions. Actors
cation logic [1]. Even though building applications on top of include suppliers, purchasers, logistics providers, transporta-
an AEP enables reuse, it is limited to either infrastructure tion vehicles, containers, pallets, manufacturing machines,
services such as connectivity and device management, or robots, workforce etc. These actors are both consumers and
basic application logic. Specifically, Industrial IoT (IIoT) producers of data employed by the reusable components.
end-user applications are about insights and automation of Next we present the two applications in more detail.
business processes using various Artificial Intelligence (AI)
and Machine Learning (ML) models and algorithms, and A. Smart Logistics
tend to be complex, thus requiring highly capable applica- A logistics application typically implies the movement
tion logic [2]. Due to the application diversity, developing of goods exchanged between participating stakeholders . It
IoT applications using traditional software development and also involves the tracking and monitoring of these goods,
including handling and environmental conditions . What IoT either due to unexpected machine failure, or due to ex-
will bring to logistics is a fully automated and adaptive pected wear of machine parts, and plan for and execute the
end-to-end process that provides optimization across all maintenance with a minimized disruption to the production
involved actors1 . Figure 1 provides an overview of the smart plan. The root cause of 80% of the problems which trigger
logistics application together with the components and actors maintenance is almost never identified [5], so applying
involved. Predictive Maintenance techniques in order to solve the
above mentioned problems is a necessity to avoid cost
inefficiency. In addition to this, pre-emptive detection of
unexpected failures can help in achieving improved operator
safety in a factory environment. Figure 3 gives an overview
of a typical Predictive Maintenance application.

Figure 1. Smart Logistics overview

We identify the following components specific to the

logistics solution, which are depicted in Figure 2: a) Smart
Contract (SC) to provide an optimal selection of suppliers for
a given purchase, and to monitor and verify the fulfillment
of the contract along the process, b) AI Planner (AIP) for
computing an optimized routing of goods by the distribution
network of trucks, containers, and warehouses according to
conditions established in the contract, c) Analytics Compo-
Figure 3. Predictive Maintenance overview
nent (AC) to define how the system is expected to behave
according to the conditions of the contract and to provide the
Figure 4 depicts the components identified in the Predic-
valid models to the Condition Monitor (CM) and AIP for any
tive Maintenance solution: a) Analytics Component (AC) to
given situation and d) Condition Monitor (CM) to provide a
provide models to predict the type of service and an appro-
system that continuously monitors the products and notifies
priate time window for the service of a machine based on
the SC and AIP when the conditions are deviating and
input received from Condition Monitor (CM), b) Condition
triggers actuation when needed. More details on a specific
Monitor to provide continuous monitoring of all machines
implementation of a Smart Logistics solution are available
and robots and notify the AIP when the maintenance has
in [4].
to be planned, c) AI Planner (AIP) to optimize the process
of maintenance by reducing the production line downtime
taking into consideration the production line load schedule,
resources, spare parts availability and d) Smart Contract
(SC) in-conjunction with the respective Enterprise Resource
Planning (ERP) systems to select subcontractors required to
perform the maintenance given the cost and estimated time.

Figure 2. Smart Logistics flow and constituting components

B. Predictive Maintenance
Predictive Maintenance in manufacturing is used to pre- Figure 4. Predictive Maintenance flow
dict when machines in a production line need servicing,
1 https://www.forbes.com/sites/insights-inteliot/2018/06/14/ In both examples, we observe the four components solving
logistics-4-0-how-iot-is-transforming-the-supply-chain/ particular problems like planning, monitoring etc. recurring
in the applications, but that they are orchestrated differently engine is closely related to the planner. It takes a plan
depending on the application. It demonstrates the reusability and then dispatches individual steps so as to orchestrate its
aspect, and that the application differences lie in the process execution. It also monitors both the plan itself to see that
flow. In the following section we will present these compo- the agents it is dispatching to are making progress, and also
nents in more detail. it monitors the individual actions themselves to make sure
it has the desired effect. Since we aim that the component
III. D ESCRIPTION OF REUSABLE COMPONENTS would be reusable, we require a few additions to it. We
In this section we present the reusable components identi- need to have knowledge reflecting the state of the world
fied in the two Industrial IoT (IIoT) applications. Our focus and sufficient details of a formal model to enable planning.
is to: a) describe why the component presented is selected Additionally we need to have an API driven interface and
for reusability, b) describe how the reusability could be our experience suggests a REST based interface is suitable
enabled and which generalizations and specializations are for the task.
needed to provide the interfaces and c) present what are the
potential or existing implementations. C. Condition Monitor (CM)
The digital transformation starts by deploying sensors and
A. Smart Contract (SC) monitoring any relevant characteristics to the IoT application
In 1997, Nick Szabo described a smart contract as a set in question. Gathering data and orchestrating all sensors
of promises, specified in digital form, including protocols is a task on its own. The IoT ecosystem is heterogeneous
within which the parties perform on these promises [6]. The and there is a variety of devices in terms of software and
SC component helps to automate the entire smart contract hardware capabilities. In addition to this, the devices use a
life-cycle using a blockchain based substrate encompass- large number of communication protocols both in short and
ing contract negotiation, partner selection, contract final- wide range. Once the data is collected from the different
ization,contract enactment, contract monitoring and issue sources, specialized processing is required in order to be
resolution during contract enactment. This component can be able to perform the computation and retrieve insights and
modeled according to the business needs of the application knowledge.
and setup to achieve the above mentioned tasks seamlessly. Reusability can be achieved by providing a flexible
Advancements in generation of smart contracts from existing and configurable component that supports data collection,
artifacts like natural language, business process, state ma- transformation and analytics. This enables different types
chines, non smart-contract code makes it possible to reuse of applications with a simple definition of the data and
the generic object models facilitating reusability. intelligence pipeline.
In the Smart Logistics use case, the SC identifies and An implementation of the CM is described in [4] as an
recommends suitable suppliers for the purchaser and enables abstraction layer built on top of the Analytics Component,
the negotiation of pricing and contract terms with these which exposes a REST-based interface.
suppliers. Once the contract terms are accepted by both sup-
pliers and the purchaser, a purchase order is created. Features D. Analytics Component (AC)
such as quality and the delivery time are considered as the Designing reusable ACs is needed to support other com-
key obligation terms in the contract and are provided as an ponents in different solution blueprints in the path to achiev-
input to the CM module. More details on the functionality ing replicability. ACs are also necessary to scale Internet of
of this component can be found in [4]. Things (IoT) operations, especially when multiple analytics
nodes need to be distributed for coping with the large variety
B. AI Planner (AIP) of pools of heterogeneous devices required in different
This is a component consisting of two inter-related pieces: applications. Reusability can be enabled in AC by ensuring
(a) an AI planner and (b) an execution engine. The AI the component can offer supporting features for different
planner is often used in creating a required action sequence types of incoming data and analytics requests.
for a given intent. In order to provide high degree of Architecture-wise, REST-based query front-ends capable
automation, the intent needs to be specified in some formal of routing analytics requests to appropriate analytics mod-
way and similarly the available actuations also need to be els served either in Cloud or at the network Edge are
specified in a related way so that the system may derive recommended. This approach enables integrating advanced
the plan. In our work we use Planning Domain Definition data routing decision-making into analytics model server
Language (PDDL) [7] to specify the intention and the action capable of handling meta-data describing different types
semantics. A standard AI planner is then able to convert the of incoming data (e.g. numeric, symbolic, structured), of
intent to an action sequence. The advantage is that that the processing requirements (e.g. batch, stream), of data flow
planner itself is domain independent, however the inputs to characteristics (e.g. time-stamped vs. partially-ordered data),
the planner require some degree of modeling. The execution and of analytics parameters (e.g. prediction horizon, scalar
vs. interval-valued, deterministic vs. probabilistic output). composition and orchestration. Further investigations will be
Hiding such complexity in query front-end allows for better carried out to define models for matching stakeholder needs
isolating an AC so it can be reused in different domains. with solution blueprints and reusable components.
One example of AC was introduced within the context
R EFERENCES
of [4]. An AC was realized using a distributed system
based on an actor-based model which provided data and [1] V. Tsiatsis, S. Karnouskos, J. Höller, D. Boyle, and
intelligence composition across the whole IoT landscape, C. Mulligan, Internet of Things - Technologies and
Applications for a New Age of Intelligence. Elsevier,
from edge gateways to cloud datacenters. 2018. [Online]. Available: https://www.amazon.com/
Internet-Things-Technologies-Applications-Intelligence/dp/
IV. R ELATED WORK B07KQ4BVSS
Mineraud et al. [8] make an extensive study of the IoT
platforms identifying their capabilities and limitations and [2] N. Anderson, W. W. Diab, T. French, K. E. Harper, S.-W.
Lin, D. Nair, and W. Sobel, “Industrial Internet of Things
also the availability of the IoT marketplace. One of their Analytics Framework,” techreport, 2017. [Online]. Available:
important findings is that reusable components are seldom http://www.iiconsortium.org/industrial-analytics.htm
supported in IoT platforms, and this is an obstacle in treating
the IoT domain as a single ecosystem. [3] J. Höller, V. Tsiatsis, and C. Mulligan, “Toward a Machine
Intelligence Layer for Diverse Industrial IoT Use Cases,”
Current research of reusable IoT components targets the
IEEE Intelligent Systems, vol. 32, no. 4, pp. 64–71, 2017.
lower layers of the stack, where connectivity and device
management are the main concerns. To the best of our [4] S. S. Arumugam, V. Umashankar, N. C. Narendra, R. Badri-
knowledge, reusability at the higher layers of IoT is still nath, A. P. Mujumdar, J. Höller, and A. Hernandez, “IoT
an unexplored space. Bandyopadhyay et al. [9] address Enabled Smart Logistics Using Smart Contracts,” in 2018
8th International Conference on Logistics, Informatics and
reusability in IoT from a middleware perspective, where Service Sciences (LISS), Aug 2018, pp. 1–6.
heterogeneous domains are integrated. We argue however
that existing IoT middlewares do not tame IoT application [5] Industrial Internet Consortium, “Smart Factory Machine
development complexity but rather enable plug-and-play Learning for Predictive Maintenance,” 2018, Online: accessed
capabilities for on-boarding devices and data. on 15 January 2019. [Online]. Available: https://www.
iiconsortium.org/smart-factory-machine-learning.htm
Christidis and Devetsikiotis [10] perform a study on the
feasibility of employing smart contracts in IoT applications. [6] N. Szabo, “Formalizing and securing relationships on public
They identify issues in current implementations such as networks,” First Monday, vol. 2, no. 9, 1997.
lower transaction processing throughput compared with tra-
[7] D. McDermott, M. Ghallab, A. Howe, C. Knoblock, A. Ram,
ditional databases, but also highlight the benefits such as M. Veloso, D. Weld, and D. Wilkins, “PDDL - The Planning
cryptographically verifiable automation of workflows. In a Domain Definition Language,” CVC TR-98-003/DCS TR-
recent exercise, AI planning is used in a reusable fashion in 1165, Yale Center for Computational Vision and Control,
creating system integrated within larger applications [11]. Tech. Rep., 1998.
Zaki et al. [12] provide an early example of how predictive
[8] J. Mineraud, O. Mazhelis, X. Su, and S. Tarkoma, “A gap
analytics can be used to anticipate plan execution failures in analysis of Internet-of-Things platforms,” Computer Commu-
order to building monitors that trigger alarms for re-planing nications, vol. 89-90, pp. 5–16, Sep. 2016.
and mitigation actions.
[9] S. Bandyopadhyay, M. Sengupta, S. Maiti, and S. Dutta, “A
V. C ONCLUSION AND FUTURE WORK survey of middleware for internet of things,” in Recent Trends
in Wireless and Mobile Networks, A. Özcan, J. Zizka, and
This paper presents an approach to create IIoT solutions D. Nagamalai, Eds. Berlin, Heidelberg: Springer Berlin
in an AI-centric manner based on reusable components Heidelberg, 2011, pp. 288–296.
to reduce the time and costs for being applied in diverse
sectors. Taking the logistics and manufacturing scenarios as [10] K. Christidis and M. Devetsikiotis, “Blockchains and Smart
Contracts for the Internet of Things,” IEEE Access, vol. 4,
examples, a set of decoupled, flexible, and configurable AI-
pp. 2292–2303, 2016.
centric IIoT components have been proposed, namely the
Smart Contracts, the AI Planner, the Condition Monitor, [11] A. V. Feljan, S. K. Mohalik, M. B. Jayaraman, and R. Badri-
and the Analytics Components. Instead of providing a full- nath, “Soa-pe: A service-oriented architecture for planning
stack static solution, these components can be adapted and and execution in cyber-physical systems,” in 2015 Interna-
tional Conference on Smart Sensors and Systems (IC-SSS),
reintegrated for application solutions in various sectors. Dec 2015, pp. 1–6.
The future work will continue with investigating the
application replicability aspects via identification of recur- [12] M. J. Zaki, N. Lesh, and M. Ogihara, “PlanMine: Predicting
ring patterns and arriving at domain independent blueprints, Plan Failures Using Sequence Mining,” Artificial Intelligence
including the use of process engineering technologies for Review, vol. 14, no. 6, pp. 421–446, Dec 2000.