Real-time Visualization of Sensor Data in Smart Manufacturing using

Lambda Architecture

Nadeem Iftikhar

, Bartosz Piotr Lachowicz

, Akos Madarasz

, Finn Ebertsen Nordbjerg

Thorkil Baattrup-Andersen

and Karsten Jeppesen

University College of Northern Denmark, Aalborg 9200, Denmark

Dolle A/S, Frøstrup 7741, Denmark

Keywords:

Industry 4.0, Real-time Visualization, Sensor Data, Smart Manufacturing, Lambda Architecture, Machine

Learning.

Abstract:

Smart manufacturing technologies (Industry 4.0) as solutions to enhance productivity and improve efﬁciency

are a priority to manufacturing industries worldwide. Such solutions have the ability to extract, integrate, ana-

lyze and visualize sensor and data from other legacy systems in order to enhance the operational performance.

This paper proposes a solution to the challenge of real-time analysis and visualization of sensor and ERP data.

Dynamic visualization is achieved using a machine learning approach. The combination of real-time visual-

ization and machine learning allows for early detection and prevention of undesirable situations or outcomes.

The prototype system has so far been tested by a smart manufacturing company with promising results.

1 INTRODUCTION

Manufacturing industry nowadays is looking to in-

crease productivity and enhance efﬁciency through

integration of new digital industrial technology (In-

dustry 4.0) within the production system. The pro-

duction system includes but is not limited to: enter-

prise resource planning (ERP), manufacturing execu-

tion system (MES), control and hardware. Hence, to

improve production performance it is critical to mon-

itor the entire production system and predict any fail-

ures in order to avoid unplanned machine stoppages.

This paper presents real-time production monitoring

techniques capable of providing a live perspective on

production performance using multiple data sources

including sensor data. Additionally and importantly,

the real-time sensor data is matched against predicted

sensor data to alert in case of anomalies. Real-time

production monitoring allows production managers

and machine operators to comprehend and master in-

cidents by taking timely actions.

To summarize, the main contributions in this pa-

per are as follow: proposing a lambda architecture to

analyze and visualize real-time and batch data, pro-

viding methodologies for real-time production moni-

toring, presenting a machine learning model to raise

alerts at real-time, and demonstration of the proposed

solution at a smart manufacturing company.

The paper is structured as follows. Section 2

describes the proposed approach to visualize sensor

data. Section 3 provides the details about real-time

production visualization methods. Section 4 intro-

duces the machine learning model. Section 5 illus-

trates the data visualization dashboards. Section 6

presents the related work. Section 7 concludes the

paper and points out the future research directions.

2 APPROACH

This section provides insight into the approach taken

to build a complete solution of relevant tools and tech-

niques to analyze and visualize real-time and histori-

cal data. In order to build this system, lambda archi-

tecture approach is chosen (Marz and Warren, 2015).

The main idea behind lambda architecture is to build

a system using a layered approach. Each layer has its

own set of properties, techniques and technologies,

and builds upon the functionality provided by the lay-

ers underneath it. The three layers of the lambda ar-

chitecture are: speed, batch and serving. The speed

layer processes recent data. The batch layer processes

all the data present in the master data storage and

the serving layer stores the real-time as well as batch

views for visualization and later analysis.

Iftikhar, N., Lachowicz, B., Madarasz, A., Nordbjerg, F., Baattrup-Andersen, T. and Jeppesen, K.

Real-time Visualization of Sensor Data in Smart Manufacturing using Lambda Architecture.

DOI: 10.5220/0009826302150222

In Proceedings of the 9th International Conference on Data Science, Technology and Applications (DATA 2020), pages 215-222

ISBN: 978-989-758-440-4

215

Figure 1: Conceptual lambda architecture.

Some of the beneﬁts of the proposed solution

are listed below: Industry 4.0 (fourth industrial rev-

olution) connectivity by connecting production ma-

chines, sensors and ERP data in order to visualize

the entire production system; enhancing the business

process by minimizing downtime; boosting real-time

visibility by using dashboards for prompt alerts; and

providing end-users with uniﬁed views.

Most of the components of the proposed solution

are successfully deployed and tested in a manufac-

turing company (Dolle). Dolle is recognised as the

market leader in Europe for manufacturing timber loft

ladders.

2.1 Lambda Architecture

The conceptual lambda architecture is presented in

Fig. 1. The proposed architecture consists of six main

modules: a sensor network; ERP data integration;

integration of other external data, for example, in-

door/outdoor weather data; master data storage layer

for persistence; speed layer to process data streams

in real time; batch layer to perform extended time

analysis; and serving layer to store the output from

the speed and batch layers. The solid arrows both

in Fig. 1 and Fig. 2, represent streaming data, while

dashed arrows represent batch data. Further, Fig. 2

presents the concrete lambda architecture. The sensor

network consists of sensor nodes deployed through-

out the machines to measure the status of the ma-

chines such as, machine on/off, pace in, pace out and

fault/error. The ERP data contain details about prod-

ucts, job executions and work calendar. In addition,

other data may include indoor temperature, humid-

ity and so on. The data transmitted by the sensors is

streamed into the master data storage (MongoDB) by

Phoenix controller

. The Phoenix controller captures

data (ranging from 1 to 10 times per second) from the

https://hexdocs.pm/phoenix/Phoenix.Controller.html

sensor network and submits the data to MongoDB.

The controller submits the data of a sensor in Mon-

goDB, only if there is a change in the state of the sen-

sor. If there is no change in the state of the sensor

the value will not be stored. Further, the external data

such as ERP data, is also submitted to the master data

storage.

Sensor data snapshot:

[{‘timestamp’: ‘2020-02-03 08:45:30’,

‘value’: ‘true’,

‘port’: ‘0103’},

{..}, ..}]

ERP data snapshot:

[{‘planned_start’:‘2020-02-03 08:00:00’,

‘planned_stop’:‘2020-02-03 11:00.00’,

‘actual_start’:‘2020-02-03 08:15:00’,

‘actual_stop’:‘2020-02-03 11:45:00’,

‘item_id’:‘550583’,

‘machine_id’:‘1404’,

‘name’:‘SW365/3section/12threads/..’, ..}]

The sensor data snapshot provides a quick

overview of the data. The sensor data (JSON format)

contains three attributes. The data has varying granu-

larity as explained above. The ﬁrst three rows in the

sensor data snapshot read as follows. Timestamp, rep-

resents the date and time of the sensor data acquisi-

tion. Value, corresponds to the state of the binary sen-

sor (either TRUE or FALSE) and port, indicates the

sensor number. For example, port = 103 represents

pace out sensor (rolling out of a ladder). Similarly,

the ERP data snapshot reads as follows. Planned start

and stop, represent the planned job execution date and

time. Actual start and stop, denote the actual job ex-

ecution data and time. Item-id, identiﬁes the prod-

uct that is being produced. Machine-id, represents

the machine and ﬁnally name, represents the type of

product being produced. A custom-built web appli-

cation programming interface (Dolle API) allows for

access to the sensor and external data from the mas-

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

216

Machine n

Machine 2

Machine 1

Phoenix Controller

API

Storm

based Script

(Real-time

Views)

Near Real-time

Data Visulatization

Historical

Data Visulatization

Sensors Data

ERP Data

Other Data

Map-Reduce

based Script

(Batch Views)

Real-time

Data Visulatization

Speed Layer

Batch Layer

Serving Layer

Master

Data Storage

SQLL DB

Figure 2: Concrete lambda architecture and the software platforms.

ter data storage by means of an HTTP GET request

to provide the data to the speed layer for processing.

The speed layer running on the Apache Storm plat-

form provides real-time views, while the batch layer

running on the Hadoop MapReduce platform delivers

precomputed batch results. The real-time views are

directly streamed to the dashboard(s) for real-time vi-

sualization using Dash, which is a Python framework

for building web applications. The real-time views

are also delivered to the serving layer with some la-

tency for near real-time visualization.

3 METHODOLOGY

3.1 Overview

This section describes the proposed real-time visual-

ization methods. Method 1 measures the productivity

in terms of the number of ladders produced during a

given period of time, whereas, method 2 calculates

the length of time elapsed since last ladder produced.

Both methods use a sliding window approach to get

the data from the master data storage at predeﬁned in-

tervals (5 seconds in this case) and pass the data to

the speed layer. In Fig. 3, the ﬁrst window (blue) con-

tains sensor data that arrives between the zeroth and

nine-hundredth second. The second window (green)

contains sensor data that arrives between the ﬁfth and

nine-hundred and ﬁfth second. Note that data from

ﬁfth through nine-hundredth second is present in both

windows. When the second window (green) is exe-

cuted at time t = 5, data between the zeroth and ﬁfth

second is dropped from the data queue.

window length = 900 seconds

sliding interval

= 5 seconds

fresh data read from the data storage

approximately every 5 seconds

…...............................................................................................

….

0 5 900 905

time in seconds

Figure 3: Sliding window.

3.2 Production Monitoring

The purpose of the production monitoring method is

to measure the productivity with respect to number of

ladders produced and display it as a live update ev-

ery time the method is executed (approximately every

5 seconds). The main idea behind the method is to

monitor the production process at real-time and to de-

tect errors/faults at earliest possible time in order to

take counter measures. Later in Section 5, this mon-

itoring method is combined with prediction models,

enhancing its ability to raise alerts at real-time.

The following outlines the four main steps of the

productivity measurement method:

1. Read the sensor data from the master data storage

every 5 seconds with a sliding window size of 900

seconds;

• The sliding window’s start time = (current time

- interval) and end time = current time. Where,

the interval is 900 seconds.

2. The data is then fed to the speed layer;

3. The speed layer performs the following function-

ality:

• The data is ﬁltered by selecting the values,

where port equals to TRUE and ignoring the

values, where port equals to FALSE;

• The values of the port are aggregated in key-

value pairs;

• The aggregated results are displayed on dash

board;

• The aggregated data for real-time views are

stored in the serving layer every 300 seconds

for near real-time visualization;

• The key-pairs are reset to zero.

4. GOTO step 1.

The following example will help in understanding

how the sensor data is being processed by the speed

layer. For this example, port/sensor 103 (pace out)

is considered. The sliding window technique is used

to visualize sensor data at real-time. The proposed

Real-time Visualization of Sensor Data in Smart Manufacturing using Lambda Architecture

217

method divides the data in slices of constant length

of 900 seconds. Further, the algorithm writes to the

SQL database every 300 seconds for near real-time vi-

sualization. After executing the real-time processing

method at 2020-02-24 14:15:00 (current date/time), it

is of interest to see the snapshot of the data in the sen-

sor data table (Table 1). As, the predeﬁned window

size is 900 seconds for that reason the start of window

date/time is (2020-02-24 14:15:00 - 900 seconds) =

2020-02-24 14:00:00 and end of window date/time is

2020-02-24 14:15:00. In this table, the granularity of

each rows differs. This is due to the fact that the sen-

sor value for a speciﬁc port is only registered when

it changes (different from the previous value). Each

row in this table represents the state of the pace out

sensor either TRUE or FALSE. TRUE means that the

pace out sensor detects the rolling out ladder section

and FALSE means vice-versa.

Table 1: Sensor data.

Timestamp Port Value

2020-02-24 14:00:00 103 TRUE

2020-02-24 14:00:01 103 FALSE

2020-02-24 14:00:17 103 TRUE

2020-02-24 14:00:19 103 FALSE

. . .

2020-02-24 14:01:19 103 FALSE

2020-02-24 14:01:42 103 TRUE

2020-02-24 14:01:43 103 FALSE

2020-02-24 14:02:01 103 TRUE

. . .

2020-02-24 14:14:29 103 TRUE

2020-02-24 14:14:31 103 FALSE

2020-02-24 14:14:46 103 TRUE

2020-02-24 14:14:47 103 FALSE

Table 2: Key-Value pair.

Port Value

103 42

Further, the count of TRUE’s is saved in a key-

value pair of which port is the key and the count of

TRUE’s is the value (Table 2). This table shows that

in the last 900 seconds 42 ladder sections were pro-

duced. Moreover, the data also posts live to the dash-

board. The tachometers in Fig. 4 shows two readings

at 15 minutes and hourly levels.

180

[82 %]

25 50

15 Minutes Hourly

(a) (b)

Figure 4: The pace of rolling out ladder.

The tachometer in Fig. 4(a) shows that 14 ladders

of type SW365/3section/12threads were produced in

the last 15 minutes and that the average number of

ladders of this type are produced during this time is

17, so the performance is approximately 82%. The

formula to calculate the pace out per 15 minutes is

number of ladder sections / number of sections = 42

/ 3 = 14. The number of ladder sections are divided

by three since the ﬁnished product (ladder) is a three

section ladder (Fig. 5).

section

3 –

tio

Figure 5: Three section ladder.

Similarly, the tachometer in Fig. 4(b) shows that

56 ladders are expected to be produced in an hour and

the hourly capacity for the speciﬁc type of the lad-

der is 68, so the performance is approximately 82%.

The formula to calculate the hourly production rate

(the number of items produced per hour) is (num-

ber of ladder sections * 4) / number of sections =

(42 * 4) / 3 = 168 / 3 = 56 ladders per hour. The

number of ladder sections are multiplied by four is

for the reason that there are four quarters in an hour

and it is divided by three since the ﬁnished product

(ladder) consists of three sections.

Figure 6: Hourly production rate (below average).

Further, a below average production situation is

shown in Fig. 6. It shows an hourly production rate

of 32 ladders against an expected average hourly pro-

duction rate of 68 ladders. The grey lines represent

the ladder output. In this case the rolling out ladder

pace is 47%. A fault situation is shown in Fig. 7. The

tachometer shows an hourly production rate of zero

ladders. The reason is that current time is 09:12 and

Figure 7: Hourly production rate (fault).

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

218

the last ladder was produced at 08:56. As a result

no ladders were produced within the last 15 minutes

dropping the hourly production rate of ladders to zero.

3.3 Time Elapsed

The purpose of the time elapsed method is to keep

track of when the last time the sensor triggered and

display it as a live update every 5 seconds. The idea is

to be able to follow the sensor triggering pattern and

see how many seconds have elapsed since the sen-

sor was last triggered. This value is then compared

to the expected range of the patterns. The method

works as follows. First, the Dolle API gets the sen-

sor data from the master data source in a sliding win-

dow size of 900 seconds. Following this, the acquired

data is passed to the speed layer. The initial data for-

mat is JSON, which is then transformed into inter-

pretable variables, such as integers and real numbers.

Finally, calculations on the data are performed, only

considering timestamps where the port value is TRUE

and directly followed by a value of FALSE. Each

time the method is executed it calculates the differ-

ence between the last timestamp of TRUE to FALSE

transition and the current timestamp (time elapsed =

current time - last output time). For example, if the

value of port 103 was TRUE at 2020-02-24 14:14:01

and at 2020-02-24 14:14:04 it became FALSE, then,

if the method is executed at 2020-02-24 14:14:53 in

that case it will show time elapsed since last ladder

was produced = 2020-02-24 14:14:53 - 2020-02-24

14:14:01 = 52 seconds. In addition, the method resets

the time elapsed to zero if the machine turns off. This

ensures that the time elapsed value is displayed only

if the machine is running and expected active in the

production.

The following outlines the ﬁve main steps of the

time elapsed method:

1. Read the sensor data in the sliding window size of

900 seconds from the master data storage approx-

imately every 5 seconds;

2. Filter the data where the values are TRUE and

transitioning to FALSE;

3. Get the time stamp for the last transitioning event

for the speciﬁc port;

4. Calculate the time elapsed that is by subtracting

the last timestamp from the current timestamp;

5. GOTO step 1.

Furthermore, the tachometers in Fig. 8 demon-

strate the time elapsed. In Fig. 8(b), it can be seen that

the time elapsed for port 103 (rolling out ladder) is 52

seconds, whereas, the average rolling out time of this

(a) (b)

Figure 8: Time elapsed.

ladder type is 40 seconds. An alert situation (in red) is

shown in Fig. 8(a). The tachometer shows an elapsed

time of 176 seconds. That means that there were no

ladders produced in the last 176 seconds. This indi-

cates a need for immediate action as could be caused

by a fault in the machine.

4 MACHINE LEARNING

This section introduces the real-time pattern detec-

tion machine learning model. To detect abnormal pat-

terns is an important feature of smart manufacturing.

For example, if the performance of the machine is di-

verging from the standard pattern then it could cause

an unwanted stop to the whole production system.

Hence, to achieve production efﬁciency it is vital to

detect errors/faults or abnormal patterns in real-time

in order to take early and appropriate actions to keep

the production running without unwanted stops. The

incoming real-time sensor data is compared to stan-

dard patterns by using machine leaning models and

an alert could be triggered when some abnormal event

is predicted to happen such as, the machine is going

to stop. Further, Fig. 9 describes the pattern detec-

tion machine learning algorithm in detail. As men-

tioned in Fig. 2, the speed layer and batch layer based

on Python Storm and MapReduce, respectively. The

serving layer consists of a SQL Server. The proposed

algorithm uses batch layer (MapReduce) to train the

model to detect abnormal patterns. The MapReduce

job runs at predeﬁned intervals and updates the coef-

ﬁcients of the pattern detection models in the serving

layer (SQL Server). Further, the speed layer reads

the incoming streaming data from the MongoDB and

detects abnormal patterns by applying the pattern de-

tection algorithm. The pattern detection algorithm dy-

namically uses the latest calculated model coefﬁcients

obtained from the serving layer. In Fig. 9, the solid

arrows represent streaming data, while dashed arrows

represent batch data.

Several forecasting models are available, in this

paper Auto-regressive model (AR) is chosen. An AR

model is a widely used linear model that works on

stationary time series. AR models use the dependent

relationship between a value y and some number of

Real-time Visualization of Sensor Data in Smart Manufacturing using Lambda Architecture

219

ERP Data

Other Data

Machine Learning (ML)

Models

Speed Layer

Batch Layer

Pattern Detection

Algorithm

Sensor Data

Serving Layer

Figure 9: Pattern detection machine learning method.

lagged values y (Black, 2011). The regression coefﬁ-

cients in the standard AR models do not vary, while

in the proposed model the regression coefﬁcients may

vary with the ladder type. For this reason, regression

coefﬁcients are calculated for each ladder type, sepa-

rately. In other words, it means that for each ladder

type a separate model will be trained. The value of t

indicates the ladder type. As, there are approximately

30 different main ladder types, the time index param-

eter will be i

, where t = 1, .., n and n is the number

of ladder types (n = 30). The proposed AR model of

order P can be written as follows, Equation 1.

ˆy

= c

∑

λ=1

t,λ

−λ

+ ε

(1)

where c

is the intercept with the y − axis (a con-

stant), α

t,λ

is the regression coefﬁcient, ε

is the value

of the white noise.

The following outlines the six main steps of the

pattern detection algorithm:

1. Compute the AR model for the speciﬁc ladder

type at the batch layer;

2. Update the coefﬁcients of the model to the serving

layer every 15 minutes;

3. Read the sensor data with a window size of 900

seconds from the master data storage every 5 sec-

onds and predicts the reading for each sensor us-

ing the AR model;

4. Calculate the distance between the predicted and

the actual sensor reading;

5. If the distance is above a certain threshold then

raise an alert;

6. GOTO step 1.

5 DATA VISUALIZATION

DASHBOARDS

The section describes the use case to deploy the pro-

posed prototype system at Dolle (a smart manufactur-

ing company in Frøstrup, Denmark). Dolle has de-

ployed 14 sensors on two machines to collect pro-

duction related data. The prototype system is demon-

strated from the perspective of production managers

and machine operators. The system uses real-time

data collected from the deployed sensors from both

the machines as well as ERP data. The sensor data

consists of machine status, alarm, pace of side rails

and wood rungs, pace of output products (ladders),

faulty side rail, screwing machine error and so on.

The ERP data consists of machine-id, job-id, ladder

type, planned job start and stop date/time, actual job

start and stop date/time, number of ladders produced,

down time and so on. The sensor data is acquired with

an interval of every ﬁve second, while the ERP data

is acquired at the start of each new job. Wall mounted

dashboards reporting the hourly production rate and

elapsed time since the last ladder was produced are

presented in Fig. 10.

Figure 10: Real-Time production monitoring dashboards.

Further, Fig. 11 shows the hourly production rate

(the pace of number of items produced per hour) in

real-time. It is important to monitor the hourly pro-

duction rate in order to keep the production running.

Fig. 11, displays the hourly production rate or pace

that has dropped to zero (tachometer on the left) also

visible on the graph (right) that no ladders were pro-

duced in the last 15 minutes. The last ladder was

produced at 08:57. The grey lines in the graph rep-

resents the exit of an ladder from the machine. The

dashboard also displays the ERP data such as, job-

id, ladder type, machine number, planned start and

stop, actual start and predicted stop. In addition, the

dashboard displays the planned and actual number of

ladder produced, time since last error as well as er-

ror type and machine downtime in the last hour. It

is interesting to observe that the date/time for the pre-

dicted stop is automatically adjusted depending on the

remaining number of ladders to produce and the ma-

chine downtime for the current job.

Moreover, machine learning models depending on

the ladder type are applied to predict hourly produc-

tion rates. The difference between the predicted and

actual data is measured against thresholds and the

presence of signiﬁcant anomalies automatically trig-

ger alerts, such as changing the color coding of the

line in the graph (Fig. 11). The changing of color

from blue to yellow and later to red triggers warning

so that the situation can be actioned.

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

220

Figure 11: Real-Time visualization of hourly production rate with machine learning.

6 RELATED WORK

This section mainly concentrates on previous work

done in relation to data analysis, data visualization

and machine learning in smart manufacturing.

A near real-time and easy to use operational pro-

cess dashboard designed for the machine operators

to monitor the information about the products and

the performance of the manufacturing process is pre-

sented by (Gr

oger et al., 2013). A state-of-the-art re-

view by (Bordeleau et al., 2018) indicated that real-

time monitoring and analysis are the major areas of

research in the context of Industry 4.0. Similarly a

comprehensive study by (Zheng et al., 2018) exam-

ined smart manufacturing systems for Industry 4.0

and presented some scenarios for production monitor-

ing. In addition, (Tokola et al., 2016) suggested mul-

tiple dashboards for end-users in manufacturing in-

dustry based on different hierarchy levels to visualize

operational, tactical and strategical data. These pre-

vious works focus on various conceptual aspects and

recent advancements of data visualization and analy-

sis in medium and large-sized manufacturing compa-

nies. The work presented in this paper builds upon the

ideas presented in those previous works, however, the

focus of this paper is to provide practical application

of data analysis and visualization technologies in rela-

tion to data from operational and production systems

in small and medium-sized (SME) enterprises.

Furthermore, a real-time monitoring system that

utilizes IoT-based sensors, big data processing and

various prediction models is proposed (Syafrudin

et al., 2018). The main focus of this work is to visu-

alizer indoor climate data at real-time and classify the

values for fault prediction. In contrast to this work,

the work presented in this paper emphases on real-

time operational and production data analysis and vi-

sualization.

In connection with machine learning in smart

manufacturing. A state-of-the-art review of deep

learning techniques for machinery fault diagnosis,

predictive analysis and defect prognosis is presented

by (Wang et al., 2018). Predicting the bottlenecks

in a production system by using ARIMA model is

proposed by (Subramaniyan et al., 2018). Further,

(Shin et al., 2017) presented a model for predicting

energy consumption of manufacturing machinery. A

generic solution in environment monitoring that is ca-

pable of detecting anomalies and displaying the re-

cent situation is proposed by (Trilles et al., 2015). Fi-

nally, (Iftikhar et al., 2019) described the basic build-

ing blocks to construct a complete solution for data

analysis and visualization.

To the best of our knowledge, this paper is one of

very few to deal with the practical aspects of produc-

tion monitoring and machine learning at real-time in

order to enhance operational efﬁciency in manufac-

turing industry.

7 CONCLUSIONS AND FUTURE

WORKS

This paper has presented the possibilities of real-time

production monitoring based on a real world case

study in the manufacturing industry. The proposed

solution is quite ﬂexible and salable to develop con-

crete prototype system for real-time, near real-time

and historical data analysis and visualization. To en-

hance the operational efﬁciency real-time visualiza-

tion dashboards were developed. Moreover, regres-

sion based machine learning technique was also used

to detect anomalies. The effectiveness of the produc-

tion monitoring dashboards and the performance of

the prediction method were tested at the premises of

a manufacturing company with promising results.

For the future work, a real-time interactive dash-

board based on the idea of predictive maintenance

will be developed. Further, it will be investigated how

Real-time Visualization of Sensor Data in Smart Manufacturing using Lambda Architecture

221

the real-time dashboards help the smart manufactur-

ing companies in general, in order to enhance their

operational efﬁciency and productivity. Finally, the

accuracy of the machine learning model will be tested

to see how well it can predict the likelihood of devel-

oping anomalies in a machine.

REFERENCES

Black, K. (2011). Business Statistics: For Contemporary

Decision Making. John Wiley & Sons, 9th edition.

Bordeleau, F. E., Mosconi, E., and Santa-Eulalia, L. A.

(2018). Business intelligence in industry 4.0: State

of the art and research opportunities. In 51st Hawaii

International Conference on System Sciences, pages

3944–3953. AIS e-Library.

oger, C., Hillmann, M., Hahn, F., Mitschang, B., and

Westk

amper, E. (2013). The operational process dash-

board for manufacturing. Procedia CIRP, 7:205–210.

Iftikhar, N., Baattrup-Andersen, T., Nordbjerg, F. E.,

Bobolea, E., and Radu, P. B. (2019). Data analytics

for smart manufacturing: A case study. In 8th In-

ternational Conference on Data Science, Technology

and Applications, pages 392–399. SCITEPRESS Dig-

ital Library.

Marz, N. and Warren, J. (2015). Big Data: Principles and

Best Practices of Scalable Real-time Data Systems.

Manning Publications Co., New York.

Shin, S. J., Woo, J., and Rachuri, S. (2017). Predictive ana-

lytics model for power consumption in manufacturing.

Procedia CIRP, 15:153–158.

Subramaniyan, M., Skoogh, A., Salomonsson, H., Banga-

lore, P., and Bokrantz, J. (2018). A data-driven algo-

rithm to predict throughput bottlenecks in a produc-

tion system based on active periods of the machines.

Computers & Industrial Engineering, 125:533–544.

Syafrudin, M., Alﬁan, G., Fitriyani, N. L., and Rhee, J.

(2018). Performance analysis of iot-based sensor, big

data processing, and machine learning model for real-

time monitoring system in automotive manufacturing.

Sensors, 18(9):2946.

Tokola, H., Gr

oger, C., J

arvenp

a, E., and Niemi, E. (2016).

Designing manufacturing dashboards on the basis of

a key performance indicator survey. Procedia CIRP,

57:619–624.

Trilles, S., Schade, S., Belmonte, O., and Huerta, J. (2015).

Real-time anomaly detection from environmental data

streams. In AGILE’15 Geographic Information Sci-

ence as an Enabler of Smarter Cities and Communi-

ties, pages 125–144. Springer, Cham.

Wang, J., Ma, Y., Zhang, L., Gao, R. X., and Wu, D. (2018).

Deep learning for smart manufacturing. Journal of

Manufacturing Systems, 48:144–156.

Zheng, P., Sang, Z., Zhong, R. Y., Liu, Y., Liu, C., Mubarok,

K., Yu, S., and Xu, X. (2018). Smart manufacturing

systems for industry 4.0: Conceptual framework, sce-

narios, and future perspectives. Frontiers of Mechani-

cal Engineering, 13(2):137–150.

DATA 2020 - 9th International Conference on Data Science, Technology and Applications

222