Recommendation Framework for on-Demand Smart Product

Customization

Laila Esheiba, Amal Elgammal and Mohamed E. El-Sharkawi

Faculty of Computers and Information, Cairo University, Cairo, Egypt

Keywords: Product-service Systems, PSS Customization, Recommender Systems, Big Data Analytics.

Abstract: Product-service systems (PSSs) are being revolutionized into smart, connected products, which changes the

industrial and technological landscape and unlocks unprecedented opportunities. The intelligence that smart,

connected products embed paves the way for more sophisticated data gathering and analytics capabilities

ushering in tandem a new era of smarter supply and production chains, smarter production processes, and

even end-to-end connected manufacturing ecosystems. This vision imposes a new technology stack to support

the vision of smart, connected products and services. In a previous work, we have introduced a novel

customization PSS lifecycle methodology with underpinning technological solutions that enable collaborative

on-demand PSS customization, which supports companies to evolve their product-service offerings by

transforming them into smart, connected products. This is enabled by the lifecycle through formalized

knowledge-intensive structures and associated IT tools that provide the basis for production actionable

“intelligence” and a move toward more fact-based manufacturing decisions. This paper contributes by a

recommendation framework that supports the different processes of the PSS lifecycle through analysing and

identifying the recommendation capabilities needed to support and accelerate different lifecycle processes,

while accommodating with different stakeholders’ perspectives. The paper analyses the challenges and

opportunities of the identified recommendation capabilities, drawing a road-map for R&D in this direction.

1 INTRODUCTION

Manufacturers today are seeking to fulfill orders on

demand by doing their business processes through

short-term networks where they negotiate value-

adding processes dynamically while taking into

consideration customer demands, quality, time, price,

viability, sustainability, and other dimensions

(Elgammal et al., 2017; Song, 2017; Papazoglou,

Elgammal and Krämer, 2018). In order to make

themselves unique, manufacturers are not only

offering products but they provide products

accompanied with services (Product-as-a-Service).

Product-as-a-Service starts by sensor-based products

that generate data in a continuous manner, these data

can be utilized for delivering preventive and proactive

maintenance. Product-as-a-Service often called

Product/Service Systems (Bustinza et al., 2015).

However, the current state of practice of

engineering PSSs still suffer from severe drawbacks

(Elgammal et al., 2017; Song, 2017; Papazoglou,

Elgammal and Krämer, 2018). The most noticeable

drawback is that PSS remains at conceptual level

considering a marketing or business perspective and

missing solid IT implementation. Furthermore, PSSs

do not accommodate growing user preferences or

product diversity features to enable effective

customization. They are incapable to tackle different

stakeholders’ views to automatically fit product

design to customer’s requests in real-time. PSSs are

unable to capture a full view of products and services

linking product structure with product quality,

production processes and services. More importantly,

they do not support analysis of product-related data

gathered along product lifecycles to improve data-

driven decision making.

This demands the use of novel lifecycle,

techniques, and technologies to enable manufacturers

to connect their data, processes, systems, personnel

and equipment to support customers with the aid of

product designers and engineers to co-design

customized products and services.

In a previous work, we have analysed and

conceptually designed and developed a novel PSS

customization lifecycle with supporting IT tools and

techniques taking a customer-centric approach, which

Esheiba, L., Elgammal, A. and El-Sharkawi, M.

Recommendation Framework for on-Demand Smart Product Customization.

DOI: 10.5220/0007684401770187

In Proceedings of the 21st International Conference on Enterprise Information Systems (ICEIS 2019), pages 177-187

ISBN: 978-989-758-372-8

177

assures customers’ requirements and preferences are

taken into consideration and product improvements

are attained through the process of PSS

customization. The PSS customization lifecycle is

established on the basis of the tried and tested

knowledge-intensive structures called manufacturing

blueprints (blueprints for short), which semantically

captures product-service and production-related

knowledge (Papazoglou, Van Den Heuvel and

Mascolo, 2015; Papazoglou and Elgammal, 2018).

Blueprints integrates dispersed manufacturing data

from diverse sources and locations, which includes

and combines business transactional data and

manufacturing operational data to gain full visibility

and control, and provides the basis for production

actionable “intelligence”.

The PSS lifecycle incorporates five core

processes (Papazoglou, Elgammal and Krämer,

2018), i.e., Smart product ideation, PSS

Customization, Production Planning, Production

Execution and Production Monitoring (cf. Figure 2).

The lifecycle provides a closed monitoring feedback

loop that enables continuous product and service

improvements. Big data analytics is of utmost value

to support the different lifecycle processes from the

early stages of smart product ideation and

customization all the way to smart product

monitoring and improvement (the lifecycle is

summarized in Section 6).

Big data analytics are classified into descriptive,

predictive and prescriptive techniques (Donovan et

al., 2015; Nagorny et al., 2017). Predictive analytics

is an advanced branch of analytics that uses data

mining, statistics, machine learning and artificial

intelligence to make predictions about unknown

future events. Predictive maintenance, in which data

are gathered from smart, connected machines to

predict when and where failures could occur,

potentially minimizing unnecessary downtime

(Coleman et al., 2018).

Descriptive analytics uses data integration and

data mining to describe or summarize what happened

in the past. For example, reports that provide

historical insights regarding the company’s

production. Prescriptive analytics can be applied to

recommend the best course of action for a given

situation, such as, the analysis of equipment

monitoring data can alert the factory-floor operators

of a detected emergent situation that need their

attention/action, or may trigger automated corrective

action(s) to mitigate the detected disturbances, and

prevent any further damage. Prescriptive analytics

ICP4Life project: http://www.icp4life.eu/

falls under the bigger class of Recommendation

Systems (RSs), which has a potential role throughout

the different processes of the smart product lifecycle,

which has not been tackled in the literature.

RSs are software tools that are used to make

useful suggestions to users taking into consideration

their preferences/requirements (Priyanka, 2017). In

PSS customization lifecycle, recommendation

facilities can be utilized to assist various involved

stakeholders in making informed decisions and

enable the re-usability of previous successful

customization artefacts that are maintained in the

blueprints knowledge base. For example, during the

early stage of smart product ideation, the customer

may be recommended by the top smart product

variants (that’s previously customized smart product

requests/designs). The recommendation in this

example is based on the customer’s preliminary

requirements and the information stored on customers

profile such as (business type, business size, location,

companies/customers she cooperates with, etc.).

The contributions of this paper is three-fold:

 The analysis and development of a

recommendation framework that supports the

different processes of the PSS lifecycle

introduced in (Papazoglou, Elgammal and

Krämer, 2018). The framework is iteratively

built on the basis of case study conducts

(Hevner et al., 2004) and our intensive

involvement with four major industrial partners

as part of the H2020 ICP4Life

project. The

framework identifies the recommendation

needs of different stakeholders involved in

each lifecycle process, which enables the re-

usability of manufacturing knowledge, and

assists in informed decision making.

 We have differentiated between the

recommendations needs of two distinct

business models: Business to Consumer (B2C)

and Business to Business (B2B). In the later

model (B2B), the customer is actually a

business that our findings indicate that her

recommendation requirements varies from the

former model (B2C). It is worth noting that

recommendation approaches proposed in the

literature to support B2B is scarce, as opposed

to B2C, e.g., (Lu et al., 2015);

 Challenges and opportunities for each

identified recommendation feature have been

analysed for its realization from both a

theoretical and technical perspective, which

acts as a roadmap for R&D in this direction.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

178

The rest of this paper is organized as follows:

Section 2 presents the background about the different

types of recommendation techniques. Related work

is analysed in section 3. This is followed by

presenting a pilot case in section 4, which will be used

as a running example throughout this paper.

Manufacturing blueprints are presented in section 5,

followed by the proposed recommendation

framework for on-demand customization PSS

lifecycle in section 6. Finally, the paper is concluded

in section 7 by highlighting ongoing and future work

directions.

2 BACKGROUND

Recommendation technology is a growing domain of

research, and is considered a hot topic in the

information technology industry. RSs have been

applied in many research areas such as e-commerce,

fraud detection, logistics, e-learning, health,

transport, etc. RSs are being used to give advice to the

user about a decision to make or action to take (Beel

et al., 2016). These recommendations are based on the

user behavior, preferences, context and/or actions

during interaction with a website or an application.

There are several types of recommendation

techniques. The most common techniques are:

 Collaborative Filtering (CF) Techniques: these

techniques make predictions of what might

interest a person based on the taste of many

other users. CF techniques are divided into

user-based and item-based CF approaches, the

former makes suggestions by considering the

users having similar interest, while the latter

suggests items that are similar to the items that

are similar to those items that the people have

liked before (Beel et al., 2016);

 Content-Based Techniques: focus on the

features of products themselves and the

preferences of the user. They recommend items

that are similar in features to those items

enjoyed by a user in the past. These techniques

do not depend on the interaction of other users

before recommending a product (Beel et al.,

2016);

 Hybrid Techniques: are built based on joining

the best features of two or more

recommendation techniques into one hybrid

technique, to enhance the performance of the

traditional recommendation techniques;

 Knowledge-Based Recommender System

(KBRS): are presented to tackle the problems

of the above techniques. These include: new

user problem (cold start), new item problem as

well as the grey sheep problem (which occurs

when a user can be classified in more than one

group of users) (Priyanka, 2017).

In essence, the main components of any KBRS

are:

 Knowledge Base: the nature of the KB varies

depending on the type of KBRS; that’s, it might

be a simple database, a set of domain

ontologies, or a case base (Bouraga and Jureta,

2016). In this paper the manufacturing

blueprints, discussed in section 5 acts as our

rich KB;

 User profile: due to the fact that KBRS

provides personalized recommendations, a user

profile is a major component and must be

maintained. A user profile consists of user’s

preferences, interests, and needs. These pieces

of information can be elicited explicitly or

implicitly. Explicit elicitation implies for

example, using elicitation requirements

engineering techniques, such as interviews,

while implicit elicitation means an analysis of

the user behavior over time to gather

information about her preferences.

KBRS distinguishes itself by providing

recommendations based on the domain knowledge, it

does not take into consideration the behavior of other

users. Case-Based Reasoning (CBR) is a common

expression of KB recommendation techniques. CBR

is the process of solving new problems by reusing the

solutions of the most similar past problems based on

the assumption that similar problems will have

similar solutions. CBR working cycle consists of four

sequential steps around the knowledge of CBR

system (Aamodt and Plaza, 1994) as follows: (i)

Retrieve: involves retrieving the most similar case(s)

from the case base using a similarity measure; (ii)

Reuse: reusing the retrieved case(s)to attempt to solve

the current problem; (iii) Revise: revising the

proposed solution -if any- by taking feedback either

in the form of a correctness rating of the result or in

the form of a manually corrected revised case; (iv)

Retain: the updated solution is stored in the case base

as a part of the new case.

3 RELATED WORK

To keep the discussion focused, this section is mainly

focused on surveying prominent related work efforts

in (KBRS) in different domains, which represent the

Recommendation Framework for on-Demand Smart Product Customization

179

basic chosen technique for our recommendation

framework presented in Section 6. Recommendations

in KBRS depend only on the domain knowledge of

the considered problem and do not take into

consideration the behavior of similar users. The

nature of the knowledge in this direction may take the

form of a simple database (Ghani and Fano, 2002), or

it may exist in the form of domain ontology (Ajmani

et al., 2013) or the knowledge may amount to a case

base (Khan and Hoffmann, 2003). Most of the

KBRSs apply a case-based recommendation

approach, where recommendations are achieved by

retrieving the most similar case(s) to the user query

by following CBR working cycle as discussed in

section 2. Quantitative KB typically applies some sort

of a similarity measure (Hsu, Chang and Hwang,

2009) as the recommendation strategy, while

qualitative KB follows some sort of a matching

technique (Blanco-Fernandez et al., 2008).

Influential related work efforts in case-based

recommendations are reported in (Chattopadhyay et

al., 2012; Yuan et al., 2013) A case-based reasoning

system for medical diagnosis was developed in

(Chattopadhyay et al., 2012), where the system

focused on a particular medical diagnosis, namely,

Premenstrual syndrome. After a number of similar

cases are retrieved, human experts verify whether the

cases are satisfactory or not. If not, the search process

is refined and process continues iteratively until the

correct and acceptable diagnosis is reached.

A case-based recommendation system to the real-

estate domain was presented in (Yuan et al., 2013),

where users are required to input some information,

including for example, the desired location, price, and

housing unit property. Then the recommendation is

carried out based on the similarity between the

problem description and the cases on the case base.

Other stream of research work efforts utilizes a

conversational case-based approach to perform the

recommendations. The purpose of the conversational

part is to build users profiles, this conversation is

done through a list of questions directed to the user,

and then the recommendations are performed based

on the Knowledge Base (KB) and the induced user

profiles. Work efforts in (Lee, 2004) and (Aktas et al.,

2004) follow this direction.

Some authors have adopted a technique similar to

the content-based approach (cf. Section 2) (Carrer-

Neto et al., 2012), (Kaminskas et al., 2012). Research

efforts in this direction typically built a KB and users

profiles, and then, a similarity is measured to match

items in the KB with a specific user’s preferences. In

(Carrer-Neto et al., 2012) the authors proposed a

social knowledge based recommender system for the

movie domain. Elements in users’ profiles are

categorized according to their preferences. The

system gathers information to initiate a movie domain

ontology, and then, the recommendation is calculated

based on analyzing the user’s profile and her links to

other users. Analogously, the approach in

(Kaminskas et al., 2012) is based on a KB music

recommendation system for places of interest. The

goal of the system is to generate music corresponding

to the place of interest. Similarly, in (Ajmani et al.,

2013), a KBRS for personalized fashion

recommendation is constructed. The system

determines the visual personality of the user, and

subsequently, generates recommendation using the

ontology for fashion recommendation given the user

personality and the occasion.

To the best of our knowledge, no previous work

has considered the utilization of recommendation

capabilities to assist in the manufacturing domain. In

addition, the recommendation approaches in

literature to support Business-to- Business (B2B) are

scarce, as opposed to Business-to-Consumer (B2C).

4 PILOT STUDY

To improve understanding, we present a

comprehensive industrial-strength pilot that was

conducted in the context of the EU H2020 ICP4Life

project. The pilot was provided by PRIMA Industries

(http://www.primaindustries.it/en/) a leading

manufacturer of laser and sheet metal machinery. The

different requirements of the pilot case are tagged as

“Req#x”, where

∈



,,…



, which will be cited in

the framework in Section 6 to exemplify the different

components of the recommendation framework.

In this pilot study we assume that a turbine engine

manufacturer (customer) is interested in a multi-axis

laser processing system and specifies its requirements

and preferences, and co-designs the product with the

help of stakeholders from an OEM, such as product

designers using the novel Product-oriented

Configuration Language (PoCL) (Papazoglou and

Elgammal, 2018), which is a user-friendly domain-

specific language aims at easing the collaborative

product design task using the same jargon familiar to

customers and other stakeholders, in an abstract and

intuitive manner. For example, the customer may

specify that the laser processing system features

should include a CO2 laser generator, its power is

4000W and its speed is 5 m/min, positioning

capability combined with a high-accuracy rotary table

motion to enable new manufacturing processes while

improving existing ones (Req#A). The work area

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

180

should be X 600 mm – Y 600 mm – Z 600 mm

(Req#B). The customer wishes to extend the laser

welding nozzle of a multi-axis laser processing

product with a cross-jet element that provides a high

velocity gas barrier to prevent molten metal spatter

and weld zone fumes from contaminating the

protective lens cover slide (Req#C). The aerospace

engine manufacturer may also demand to include

sensors that meter multiple parameters providing

services that measure the actal laser output, motion,

temperature, humidity, process gases, and process

control in both workstations (Req#D). In addition, the

customer may enhance the multi-axis laser processing

system functionality by specifying safe impact

protection services by means of including a capacitive

sensor for automatically maintaining the pre-set

stand-off from the sheet metal (Req#E).

5 MANUFACTURING

BLUEPRINT ENVIRONMENT

Manufacturing data and knowledge come from a wide

range of sources such as: shop-floor equipment,

control systems, quality tracking systems, PLM

systems, monitoring systems, CAD/CAM systems

and maintenance systems. These data and knowledge

are not completely captured nor gathered in a digital,

searchable form. These massive amounts of

knowledge and data will be useless unless they are

transformed into actionable insights. To overcome

this problem, manufacturing data must be captured,

stored, structured and inter-related through a formal

knowledge model. To achieve this objective, in a

previous work (Papazoglou, Van Den Heuvel and

Mascolo, 2015; Papazoglou and Elgammal, 2018) we

have developed a knowledge-driven manufacturing

framework.

This framework depends on the novel concept of

Manufacturing Blueprints. Manufacturing blueprints

rely on model-based design techniques to manage and

inter-link product data and information (both its

content and context), product portfolios and product

families, manufacturing assets (personnel, plant

machinery and facilities, production line equipment),

and in general, help meet the requirements

(functional, performance, quality, cost, time, etc.) of

an entire manufacturing network. This information

can be collated and put within a broader operational

context, providing the basis for manufacturing

actionable “intelligence” and a move toward more

fact-based decisions.

Figure 1: Manufacturing blueprint models.

As shown in Figure 1, the suppliers, product and

production knowledge are encapsulated in the five

interconnected extendable abstract knowledge as

described below, which called blueprint images:

 Supplier Blueprint: describes business and

technical details of a partner firm such as

production capabilities, production capacity and

stakeholder roles.

 Product Blueprint: defines the details of base or

configured product, product parts, and

materials. Such information is coupled with

other relevant data such as machine parameters,

personal skills, machine and tool data and all

entities that is necessary to represent a full

product. It also includes definitions of product

families and connects them to products, product

parts and materials.

 Production Process Blueprint: this blueprint

captures the standard assembly and production

solutions in addition to suitable production

execution plan, embedding end-to-end

processes into workflows and linking the events

of discrete activities associated with all aspects

of actual production on the factory-floor.

 Quality Assurance: Blueprint: it defines

process performance and product quality

metrics (KPIs) to monitor production operations

and solve operation problems across supply

production-chains. The objective of this

blueprint is to increase process efficiency and

asset utilization, equipment health and

consumption levels.

Recommendation Framework for on-Demand Smart Product Customization

181

 Service Blueprint: according to PSS, smart

connected products require a number of services

across the full lifecycle of the product. These

services range from how the product is

operated, maintained and upgraded. Instances

of this blue print define the characteristics of all

services that are coupled with the physical

product. These include services types, sensors,

service metrics, scope of plans, service

schedules, and work orders created from service

plans, compliance standards, service history and

cost estimates.

6 THE RECOMMENDATION

FRAMEWORK

This section presents the recommendation framework

that supports the novel smart product PSS lifecycle

we introduced in (Papazoglou, Elgammal and

Krämer, 2018) (cf. Figure 2). The results presented in

this Section have been iteratively identified, refined

and validated by ICP4Life Industrial partners, which

ascertains the applicability, efficacy and utility of the

work presented in this article (Hevner et al., 2004).

Based on the findings of the literature review

presented in Section 3, we have selected the KBR

technique as the main technique supporting the

proposed recommendation framework, due to the

advantages cited in Section 2. The next subsections

discuss the recommendation capabilities at each PSS

lifecycle process by accommodating with relevant

stakeholders’ views/requirements involved in each

process.

6.1 User Engagement and Smart

Product Ideation

The lifecycle starts by the Smart Product ideation

process (the left hand-side of Figure 2) such that a

customer wishing to configure and customize a base

product or a previously customized PSS variant that

s/he can retrieve from the PSS library to meet her

unique requirements (step-1 in Figure 2).

During this phase, the customer collaboratively

with the designer/engineer elicit and validate

requirements of extending base products with

pluggable parts and services, quality attributes, etc.,

to enable product-service differentiation. During the

user engagement process, customers may specify

product requirements, parts and preferences and co-

design the digital product with the help of OEM

product engineers using the novel PoCL.

As shown in Figure 2 the main identified and

validated recommendation capability for this process

concerns itself with “Recommending previously

customized product variants or base products”. This

acts as a starting point of the customization process,

and enables the re-usability of product and service

knowledge, maintained in the blueprints knowledge

rep. (cf. Figure 1). For example, reverting back to the

pilot case in section 4, Req#A describes the main

requirements of the customer, which we regarded by

the KBR technique as a new case. In addition to these

requirements, the information stored in the user

profile such as (business type, business size, location,

with whom he co-operates (companies or customers),

etc.) may be taken into consideration for doing

suggestions/recommendations. Assume that there are

three previously customized product variants V

, the content of these variants is represented in

terms of their parts’ attributes and their associated

values as shown in Table 1.

Table 1: Examples of previously customized products.

Variants

Attributes

Laser

generator

Laser

power

Laser

speed

Workpiece

laser 3000w

m/min

Rotary table

YAG laser 4000w

m/min

Rotary table

YAG laser 3000w

m/min

X-Y table

Given the initial requirements of the customer

described in the pilot case as Req#A, this

recommendation capability may find the most similar

case(s) (product variant(s)) from the variants stated

before (cf. Table 1), by using a possible similarity

measure such the Nearest-Neighbor function in (1),

that computes the similarity between the stored cases

(previously customized products) and the new input

case (Customer requirement) based on weighted

features.

similarity(CaseI,CaseR) =



×









,











∑







(1)

Where is the importance weight of a feature,

 is the similarity function and 





and 





are the

values for feature  in the input case and the retrieved

cases respectively.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

182

Figure 2: Recommendation framework for PSS customization.

If the value of feature  belongs to numerical class

then the similarity function will be defined as the

absolute difference between 





and 





as in (2)



(







,





)

=1



















(2)

If the value of the feature  belongs to categorical

class, then the similarity function will be defined as

in (3)

(





,





)=1

(







=





)

(3)

Implying that the features having the same value

get a similarity score of 1 and 0 otherwise. Now

assume that the weights of the features as follows:

laser generator (0.5), laser power (0.1), laser speed

(0.1) and workpiece (0.3).

By using the above function, the similarity values

between the customer requirements (input case) and

the stored cases (variants) will be as follows:

Similarity (case I, V

) =0.97, Similarity (Case I, V

) =

0.47 and Similarity (case I, V

) = 0.175.This

recommendation facility will rank the stored

previously customized variants and will display in a

user-friendly and intuitive manner the ranked

recommendations that the customer can scrutinize.

Assume that customer chooses to base her new

smart product customization effort on recommended

. The customer can then tweak the new customized

product in many aspects.

If there are no previous cases that match the

customer’s requirements, we identify two scenarios:

 Scenario 1: the customer may need to update

her requirements and the recommendation

process described above restarts. This process

repeats iteratively until a recommended smart

product variant is satisfactory enough to the

customer to start with.

 Scenario 2: the system will recommend a base

product (base laser machine) where the

Recommendation Framework for on-Demand Smart Product Customization

183

customer starts customization efforts from

scratch.

Any of these three identified scenarios will

eventually result in a new smart product variant that

is also stored in the blueprints KB for further

reusability.

This recommendation facility opens these

opportunities such as: (i) Offering varying levels of

product differentiation to accommodate with diverse

customers’ requirements which will increase the

customer retention and satisfaction, (ii) Reducing the

time and the cost of doing customization from

scratch.

We envision the R&D challenges to support this

recommendation facility as follow: (i) Visualizing

/presenting recommendations in a user friendly

manner, which may be incorporated by utilizing

domain-specific languages, 3D visualization, and

Augmented and Virtual Reality (White et al., 2016);

(ii) Explaining why each of the recommended

artefacts are recommended, which would assist the

customer to make a more informed decision. This

belongs to the stream of descriptive analytics

described in section 1. Prominent visualization

techniques in this area are tables, text- highlighting,

images, diagrams, rating and animation

(Richthammer, Sänger and Pernul, 2017). These may

be combined with advanced visualization capabilities

described above.

The output of this process is a set of validated and

well-documented requirements that act as inputs to

the next process: “PSS Configuration and

Customization”.

6.2 PSS Configuration &

Customization

Once the user input from Step-1 is validated, the PSS

customization process begins. Here, products and

services are customized according to the user

requirements. At this stage the flow moves to the

“PSS Configuration” process (Step-2 in Figure 2)

where a customized product is created. This process

is interleaved with service customization (Step-3 in

Figure 2) where services for the customized products

are created in a manner that enables a seamless

product and service integration. It is worth noting that

PSS customization varies according to the business

model, whether it is a B2B or a B2C. In the B2B

model, the customer is an advanced customer who

can adaptively customize the PSS by

adding/removing/replacing components and parts.

However, in a B2C model, Customers are typically

novice, so, more guide and control should be

provided during customization.

Step-2 and Step-3 in Figure 2 are described as

follows:

6.2.1 Customization of Base Products & PSS

Variants

The customization of base products or PSS variants

can be done through two customization scenarios:

 Parameterized Product Customization: the

customer with the help of product

designer/engineer, performs parameterized

customization by adjusting the feature values

of the newly customizer PSS. This is done by

adjusting desirable values for parameters

defined in the respective blueprints models e.g.,

material types, product dimensions, etc. This

results in a new product configuration or

variant.

 Adaptive Customization: a more advanced

customization can be created by extending a

base product with additional pluggable parts, or

by replacing existing parts of a base product

with new pluggable parts that achieve better

functionality while preserving operational

consistency.

As shown in Figure 2 the main identified and

validated recommendation capabilities for this

customization activity are:

 Recommending top N parts, that meets

customer requirements;

 Recommending parts that are frequently

ordered in line with certain product’s part;

 Recommending accessories (e.g., laser glasses,

safety curtains, ESD protection).

These recommendation capabilities are provided

based on both the customer requirements and the

information stored in her profile.

For example, according to the customer

requirements described in the pilot case section 4,

Req#C, we could recommend the top N cross jet

elements to the customer. We may recommend other

parts or elements that are frequently ordered when

requesting this cross-jet element. In addition,

accessories (e.g., welding glasses or safety curtains)

are recommended to the customer.

6.2.2 Customization of Services

It involves expansion of existing products by adding

smart sensors or Internet of Things communication

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

184

devices to improve product usage. As shown in

Figure 2, the recommendation capabilities identified

for this customization activity are:

 Recommending services;

 Recommending sensors.

For example, according to the customer

requirements identified in Req#D and Req#E in the

pilot case, the recommendation facility will

recommend another services that are always

accompanied with the requested service such as

services to measure humidity or process gases. In

addition, sensors and IOT devices that are used to

realize the requested service(s) are recommended.

The same challenges and opportunities for the

recommendation capabilities identified in the user

engagement and smart product ideation process apply

here.

6.3 Production Planning

The main aim of this process is to interconnect every

step of the production process by transferring

individual product specifications into plans, working

instructions, and machine configurations, which are

to be dispersed to the respective facilities on the shop-

floor. As shown in Figure 2 the main identified and

validated recommendation capabilities for this

process targeting the production engineer are:

 Recommending suppliers: which may provide

the production engineer with the best supplier

for supplying a certain product’s part (e.g.,

Angle- torch). The recommendation strategy

maybe based on machine learning approaches,

i.e., Ranking Neural Network (RankNet)

(Zhang et al., 2016);

 Recommending previous production plans and

business processes: by re-using the production

plans and production business processes of

previously customized products that are most

similar to realize the new customized PSS

request.

The same approach used in the user engagement

and product ideation process will be used to find the

most similar previously customized product then, its

associated production plan and production business

process in the Blueprints KB are recommended to the

production engineer to reuse. In the example

explained in Section 6.1, V

is the most similar

product to the customer requirements, its associated

production plan will be recommended to the

production engineer as a consequence.

The presence of these recommendation

capabilities will open these opportunities: (i)

Reducing the time and cost of constructing

production plans and processes from scratch; (ii)

Avoiding mistakes by adapting the previous

successful plans/business processes; (iii) Automatic

selection of the best suppliers may reduce the time of

locating a supplier manually by the production

engineer.

Nevertheless, the existence of opportunities does

not mean the absence of obstacles and challenges: (i)

Adoption/adaptation of effective and efficient

adaptation techniques that will be used to make

updates on the recommended solution (e.g.,

Production plans); (ii) Visualizing/presenting these

recommendations in a user friendly manner.

6.4 Production Execution

The purpose of this process is to execute production

processes and manage order execution, equipment

downtime, assets and manufacturing operation

execution. During this process, some machines may

be broken down. As shown in Figure 2 the main

identified recommendation capability for this process

concerns itself by “recommending previous solutions

for current equipment downtime” that will assist the

shop-floor operator.

The recommendation technique may be based on

the CBR approach. The knowledge base amounts to a

case base, the case represents a diagnostic situation

and contains description of the symptoms, the failure

and the cause, and description of a repair strategy. By

using the Nearest-Neighbor function in (1) the most

similar case will be retrieved, reused, refined and

stored as a new solution for the current machine

problem.

Obviously, the existence of this recommendation

capability will open new horizon of opportunities

including: (i) Reducing the time and cost for doing

maintenance/repair by adopting previous successful

solutions; (ii) Increasing customer’s trust and

retention by promptly and effectively reacting to

shop-floor disturbances.

These recommendation facilities face the same

challenges of the recommendation capabilities

identified in the production planning process.

6.5 Production Monitoring

The aim of this process is to continuously monitor

Key Performance Indicators (KPIs) and quality

attributes, to ensure quality manufacturing by

identifying early signs of problems. In order to assist

Recommendation Framework for on-Demand Smart Product Customization

185

the factory-floor operator, the main identified

recommendation capability for this process is

“recommending a set of corrective actions” based on

the nature of the predicted error, such as shutting

down the machine or having it serviced, activating

standby machines, etc.

The recommendation strategy to realize this

facility may be based on the analysis of machine

sensor data, operational data, and process data by

applying predictive analytics techniques such as

machine learning, data mining and deep learning.

Comparing the results of this analysis to historical

data stored in quality assurance blueprint, problems

are predicted and as a consequence a set of corrective

actions are recommended using prescriptive

techniques.

The opportunities identified for this

recommendation capability are: (i) Increasing

machine life time; (ii) Reducing maintenance cost;

(iii) Increasing customer trust by committing to

delivery time; (iv) Faster detection and correction of

problems.

The realization of these recommendations will

face some challenges such as: (i) Adoption/adaptation

of effective and efficient techniques for collecting

vast real-time data and integrating it with historical

data are required; (ii) Pre-processing and processing

of this large volume of data requires powerful

processing tools and techniques; (iii) The availability

of condition monitoring tools and sensors is very

costly.

7 CONCLUSIONS AND FUTURE

WORK

Big data analytics help organizations exploit their

data and use it to identify new opportunities. This

leads in turn to smarter business moves, more profits

and satisfied customers, and more efficient

operations. Prescriptive analytics falls under the

bigger class of Recommendation Systems (RSs),

which has a potential role in assisting involved

stakeholders throughout the different processes of the

PSS lifecycle for informed decision making.

In this paper we have analyzed and identified a

novel recommendation framework that supports all

processes of the PSS customization lifecycle

introduced in (Papazoglou, Elgammal and Krämer,

2018). In this framework a set of recommendation

capabilities are identified for each process. For each

recommendation capability, we have identified the

challenges and opportunities for its realization. The

framework is iteratively built on the basis of case

study conducts and the intensive involvement of four

major industrial partners as part of the H2020

ICP4Life project.

Future work efforts are ongoing into a number of

parallel and complementary directions. This includes

extending the blueprints models to meet the

realization of the identified recommendation

capabilities; in addition, tackling the challenges

identified for each recommendation facility and

building efficient and effective/theoretical conceptual

solutions by utilizing the recent advances in ICT, and

developing an integrated manufacturing

recommendation tool-suite.

REFERENCES

Aamodt, A. and Plaza, E. (1994) ‘CBR: foundational

issues, methodological variations and system

approaches’, AI Communications, 7, pp. 39–59.

Ajmani, S. et al. (2013) ‘An ontology based personalized

garment recommendation system.’, in Web Intelligence

(WI) and Intelligent Agent Technologies (IAT), 2013

IEEE/WIC/ACM International Joint Conferences on,

pp. 17–20.

Aktas, M. S. et al. (2004) ‘A web based conversational

casebased recommender system for ontology aided

metadata discovery’, in Proceedings of the 5th

IEEE/ACM international workshop on grid computing,

IEEE Computer Society, pp. 69–75.

Beel, J. et al. (2016) ‘Research-paper recommender

systems : a literature survey’, International Journal on

Digital Libraries. Springer Berlin Heidelberg, 17(4),

pp. 305–338. doi: 10.1007/s00799-015-0156-0.

Blanco-Fernandez, Y. et al. (2008) ‘A flexible semantic

inference methodology to reason about user preferences

in knowledge-based recommender systems’,

Knowledge-Based Systems, 21(4), pp. 305–320.

Bouraga, S. and Jureta, I. (2016) ‘Knowledge-based

recommendation systems : A survey Knowledge-Based

Recommendation Systems : A Survey’, (January). doi:

10.4018/ijiit.2014040101.

Bustinza, O. F. et al. (2015) ‘Servitization and Competitive

Advantage: The Importance of Organizational

Structure and Value Chain Position’, Research-

Technology Management, 58, pp. 53–60.

Carrer-Neto, W. et al. (2012) ‘Social knowledge-based

recommender system. application to the movies

domain.’, Epert Systems with Applications, 39(12), pp.

10990–11000.

Chattopadhyay, S. et al. (2012) ‘A case-based reasoning

system for complex medical diagnosis’, Expert

Systems.

Coleman, C. et al. (2018) ‘Making Maintenance Smarter:

Predictive Maintenace and the digital Supply Network’,

Deloitte.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

186

Donovan, P. O. et al. (2015) ‘Big data in manufacturing : a

systematic mapping study’, Journal of Big Data.

Journal of Big Data. doi: 10.1186/s40537-015-0028-x.

Elgammal, A. et al. (2017) ‘Design for Customization: A

New Paradigm for Product-Service System

Development’, Procedia CIRP, 64(December), pp.

345–350. doi: 10.1016/j.procir.2017.03.132.

Ghani, R. and Fano, A. (2002) ‘Building recommender

systems using a knowledge base of product semantics.’,

in Proceedings of the Workshop on Recommendation

and Personalization in ECommerce at the 2nd

International Conference on Adaptive Hypermedia and

Adaptive Web based Systems., pp. 27–29.

Hevner, A. et al. (2004) ‘Design Science in Information

Systems Research’, MIS Quarterly, 28, pp. 75–105.

Hsu, C. K., Chang, C. K. and Hwang, G. J. (2009)

‘Development of a reading material recommendation

system based on a multi-expert knowledge acquisition

approach.’, in Advanced Learning Technologies.

ICALT., pp. 273–277.

Kaminskas, M. et al. (2012) ‘Knowledge-based music

retrieval for places of interest.’, in Proceedings of the

second international ACM workshop on Music

information retrieval with user-centered and

multimodal strategies, ACM, pp. 19–24.

Khan, A. S. and Hoffmann, A. (2003) ‘Building a case-

based diet recommendation system without a

knowledge engineer’, 27, pp. 155–179.

Lee, W. P. (2004) ‘Applying domain knowledge and social

information to product analysis and recommendations:

An agent-based decision support system.’, Expert

Systems, 21(3), pp. 138–148.

Lu, J. et al. (2015) ‘Recommender System Application

Developments: A survey’, Decision Support Systems,

pp. 1–38.

Nagorny, K. et al. (2017) ‘Big Data Analysis in Smart

Manufacturing : A Review’, pp. 31–58. doi: 10.4236/

ijcns.2017.103003.

Papazoglou, M. and Elgammal, A. (2018) ‘The

manufacturing blueprint environment: Bringing

intelligence into manufacturing’, 2017 International

Conference on Engineering, Technology and

Innovation: Engineering, Technology and Innovation

Management Beyond 2020: New Challenges, New

Approaches, ICE/ITMC 2017 - Proceedings, 2018

(August), pp. 750–759. doi: 10.1109/ICE.2017.

8279960.

Papazoglou, M., Elgammal, A. and Krämer, B. (2018)

‘Collaborative on-demand Product-Service Systems

Customization lifecycle’, CIRP Journal of

Manufacturing Science and Technology.

Papazoglou, M. P., Van Den Heuvel, W. J. and Mascolo, J.

E. (2015) ‘A reference architecture and knowledge-

based structures for smart manufacturing networks’,

IEEE Software, 32(3), pp. 61–69. doi: 10.1109/

MS.2015.57.

Priyanka, P. (2017) ‘A Survey Paper on Various

Algorithm’s based Recommender System’, IOSR

Journal of Computer Engineering, 19(3), pp. 27–32.

doi: 10.9790/0661-1903042732.

Richthammer, C., Sänger, J. and Pernul, G. (2017)

‘Interactive Visualization of Recommender Systems

Data’, Proceedings of the 4th Workshop on Security in

Highly Connected IT Systems - SHCIS ’17,

(September), pp. 19–24. doi: 10.1145/3099012.

3099014.

Song, W. (2017) ‘Requirement management for product-

service systems: Status review and future trends’,

Computers in Industry, 85, pp. 11–22.

White, M. et al. (2016) ‘Tech Trends 2016’, Deloitte

University Press, p. 19. Available at: http://www2.

deloitte.com/global/en/pages/technology/articles/tech-

trends.html.

Yuan, X. et al. (2013) ‘Toward a user-oriented

recommendation system for real estate websites’,

Information Systems, 38(2), pp. 231–243.

Zhang, R. et al. (2016) ‘Learning to Select Supplier

Portfolios for Service Supply Chain’, pp. 1–19. doi:

10.1371/journal.pone.0155672.

Recommendation Framework for on-Demand Smart Product Customization

187