Conference PaperPDF Available

Simulation of Product Performance Based on Real Product-Usage Information: First Results of Practical Application to Domestic Refrigerators


Abstract and Figures

Today's connected products increasingly allow us to collect and analyze information on how they are actually used. An engineering activity where usage data can prove particularly useful, and be converted to actionable engineering knowledge, is simulation: user behavior is often hard to model, and collected data representing real user interactions as simulation input can increase realism of simulations. This is especially useful for (i) investigating use-related phenomena that influence the product's performance and (ii) evaluating design variations on how they succeed in coping with real users and their behaviors. In this paper we explored time-stamped usage data from connected refrigerators, investigating the influence of door openings on energy consumption and evaluating control-related design variations envisaged to mitigate negative effects of door openings. We used a fast-executing simulation setup that allowed us to simulate much faster than real time and investigate usage over a longer time. According to our first outcomes, door openings do not affect energy consumption as much as some literature suggests. Through what-if studies we could evaluate three design variations and nevertheless point out that particular solution elements resulted in better ways of dealing with door openings in terms of energy consumption.
Content may be subject to copyright.
Postprint version of paper to be presented at the ASME 2018 Computers and Information in Engineering Conference,
CIE 2018, Quebec City, Quebec, Canada
Simulation of Product Performance Based on Real Product-Usage Information: First Results
of Practical Application to Domestic Refrigerators
Wilhelm Frederik van der Vegte
; Fatih Kurt, Oğuz Kerem Şengöz
Today’s connected products increasingly allow us to collect and analyze information on how they are actually
used. An engineering activity where usage data can prove particularly useful, and be converted to actionable
engineering knowledge, is simulation: user behavior is often hard to model, and collected data representing
real user interactions as simulation input can increase realism of simulations. This is especially useful for (i)
investigating use-related phenomena that influence the product’s performance and (ii) evaluating design vari-
ations on how they succeed in coping with real users and their behaviors. In this paper we explored time-
stamped usage data from connected refrigerators, investigating the influence of door openings on energy con-
sumption and evaluating control-related design variations envisaged to mitigate negative effects of door open-
ings. We used a fast-executing simulation setup that allowed us to simulate much faster than real time and
investigate usage over a longer time. According to our first outcomes, door openings do not affect energy
consumption as much as some literature suggests. Through what-if studies we could evaluate three design
variations and nevertheless point out that particular solution elements resulted in better ways of dealing with
door openings in terms of energy consumption.
1. Introduction
Product usage information (PUI) can be considered a valuable source of knowledge for predicting usage and
behaviors of current and future products, and related services
. With ‘usage’ we mean, in this particular context,
the way users use the product and how they interact with it. It is not to be confused with ‘usage’ as consumption
of resources and supplies, which we consider here as an aspect of product performance.
Traditional ways of collecting PUI include observation of human subjects and conducting user surveys. How-
ever, now that products are increasingly becoming equipped with their own capabilities of collecting use-
related data, and a growing number of products is getting connected to the Internet, it becomes easier for
manufacturers to collect data from fielded products [1].
In the EU-funded FALCON project (see Section 3), we have investigated the opportunities of exploiting such
collected data in several ways. The main deliverable of this project was a software platform to collect and
process data generated by connected products and related social media, with the objective to extract actionable
knowledge that could be used as input for (re)design of products and related services [2]. One of the studies
conducted in this context aimed to report on, and implement, methods and tools for forecasting and simulation
based on time-stamped PUI, or TPUI i.e., each data sample holds information about the time of usage or
non-usage. Both forecasting and simulation enable predictions, i.e. descriptions of expected future processes
as potential carriers of actionable knowledge.
On the one hand, TPUI potentially holds patterns that may repeat themselves in the future or represent trends
that can be projected into the future. Applying methods and tools to that end is generally known as forecasting.
It is typically based on data analytics and data mining, with data mining typically associated with computa-
tionally executed knowledge-discovery [3], and data analytics the umbrella term that also includes human
tasks such as interpreting data visualizations [4].
Simulations, on the other hand, can be used to predict product performance under different circumstances, or
of design alternatives. Having realistic usage data available makes it possible to (i) investigate the influences
of different ways of using the product, and based on these, explore design variations tailored to these uses, and
(ii) compare design variations under the same real-life usage circumstances.
The investigation of simulation potential in FALCON is the topic of this paper, which is structured as follows.
In Section 2, we report on related work on simulations with data. Section 3 clarifies the role of simulations in
the project. Section 4 introduces the simulation reference scenario provided by consortium partner Arçelik
regarding the influence of usage and design considerations on energy consumption of a domestic refrigerator.
Subsequently, Section 5 summarizes the relevant literature reporting on energy consumption of refrigerators,
Delft University of Technology, Delft, The Netherlands
Arçelik A.Ş., Istanbul, Turkey
Services are not addressed in this paper
and how it is influenced by usage. Next, in Section 6 we discuss considerations regarding data collection and
sampling. Section 7 presents our simulation model. Selection and preprocessing of the usage data is discussed
in Section 8. In Section 9 we describe the simulation setup and the explored scenarios and Section 10 presents
our first simulation results. In Section 11 we discuss these results, and, finally, in Section 12, we discuss what
could be done next.
2. Simulation with data: Related work
Simulations are typically applied in the beginning-of-life stage of a product to evaluate design proposals based
on mathematically defined behavioral models. Shannon [5] defined simulation as conducting experiments with
an input-output model of a real system in order to predict probable future output of a system for a given input,
to understand the system behavior and/or to evaluate system operation strategies. He points out that that gath-
ering reliable input data can be time consuming and that questionable input data cannot be compensated by a
good simulation model.
In many cases, products operate based on frequently applied and well-understood physics principles (e.g.,
electric motor, heat pump), that can adequately be captured in well-validated engineering simulation models.
Yet, other processes may be involved that cannot be straightforwardly described by mathematical models, e.g.,
human behavior or the weather. This is where the dependability of the input data becomes crucial – consider
for instance makeshift models that are created to generate input signals based on assumptions, such as pulse
signals representing load patterns [e.g., 6,7]. Instead of such workarounds, we propose to use real-life TPUI,
as it is increasingly becoming available from connected products.
A concept related to TPUI-based simulation is data-driven simulation, where data from the process to be sim-
ulated, corresponding to outputs as well as inputs of the simulation model, is used to optimize the simulation
model [e.g., 8]. In some cases, data from an ongoing process is even used to continuously fine-tune a simulation
that is running ahead [e.g., 9].
In the case of TPUI-based simulation, real-life data is only used as simulation input, while the simulation
model itself is considered to be sufficiently dependable. Hence, it is assumed that the simulation results, can
be used to (i) evaluate the performance of the product in realistic circumstances, (ii) identify mismatches be-
tween assumed inputs and real inputs, (iii) support finding directions to improve the design based on (i) and/or
(ii), and (iv) evaluate (virtual simulation models of) alternative designs based on real inputs.
Our focus has been on input data that represents human (inter)actions. To add realism to simulations if inputs
by human users have to be considered, interactive simulations with real humans in the loop [e.g., 10,11] have
been put forward. These have the drawback that they must run in real time and cannot be accelerated to inves-
tigate usage over a longer time interval [12]. Moreover, deploying real users in testing is known to be expensive
[13]. Figure 1 illustrates how TPUI from real-life usage of fielded products can fill the gap by providing
realistic human inputs and thus contribute to more realistic results [14], without the need to slow down to real-
time execution or recruit human subjects.
Our literature search revealed that in most other reports on product simulations with TPUI, also output data
was processed in a data-driven simulation setup, and the focus was on optimizing models – particularly dis-
crete-event simulation models of manufacturing systems [15–17]. A notable exception is the work reported by
Pei et al. [18], who did not only compare and optimize different simulation models of electronics-packaging
degradation based on TPUI from 100 mobile computing devices, but also used the simulation results to derive
more realistic requirements for next-generation designs. However, their research was a one-time trial, with
TPUI-collection capabilities added to product units just for this particular study. In such a case, researchers
can optimally match the collected amounts and frequencies of data to their needs. TPUI-recording products
often impose practical limits restricting the amount of
available simulation input data.
A rare example where PUI was only used as simulation
input, and not for optimizing simulation models, was re-
ported by Urban and Roth [19]. In simulations compar-
ing performance of smart thermostats, temperature set-
points based on real values collected from end users
were used. However, these simulations used fixed con-
stellations of set-points per user, i.e., non-time-stamped,
whereas our intention is to consider dynamically chang-
ing inputs to dynamic simulations.
Figure 1. Filling the gap between virtual-user in-
put and real-user real-time input (arrow depicts
increasing realism)
testing hardware with
generated input
simulations with
physical prototypes
simulation of
hardware with data
from real-life usage as
real productvirtual product
virtual userreal user
fully virtual human-
artifact simulation
simulations with
virtual prototypes; VR
simulation of virtual
artifact model with
data from real-life
usage as input
3. Simulations in FALCON
In the FALCON project we aimed to obtain actionable information from conventional engineering simulations
through exploitation of TPUI. In that context, ‘actionable’ means providing insights in how the product can be
improved in terms of performance, by a design that better anticipates actual usage. Here, ‘performance’ denotes
any output measure that determines the quality of the product’s functioning to any involved stakeholder. Ex-
amples of performance indicators are speed of operation, supplies consumption, noise production, and quality
of product outputs.
Our reasoning has been that dynamic simulation with TPUI as input can only produce actionable information
with added value if the investigated performance measure of the product (i) can actually be assessed based on
the simulation results and (ii) is influenced by the timing of changes in the TPUI.
As prerequisite (i) suggests, some performance measures cannot be assessed based on simulation results. This
is for instance the case for subjective performance measures, such as the taste of coffee produced by a coffee
Concerning (ii), for many products the available PUI is likely to represent the type, intensity and timing of
user interactions. If, for instance, we consider a washing machine, the predominant interactions are program
selection and inserting/removing the laundry. The timing of these interactions usually does not influence typi-
cal performance measures such as energy consumption and program duration. These are determined by what
happens when the program is executed, after program selection and laundry insertion and before laundry re-
moval – in other words, there is no direct interplay between user interactions and the part of product operation
that determines performance, unless the user interrupts program execution – which can be considered an ex-
ceptional case. In order to assess performance, the simulation only needs the input parameters related to user
interaction for each washing cycle (i.e., selected program and characteristics of the laundry), not their timing:
the implicit assumption that these inputs have taken place before the start of the program is enough. Except
for determining the total time span of data collection, the time stamps have no added value.
In the case of a refrigerator, on the other hand, there is direct interplay between interactions with its doors and
its contents and the part of product operation that determines performance, which is actually its continuously
ongoing. In this case we need to consider use interactions with their timing as input for dynamic simulations.
The effect of two door openings of 5s at 10s apart is likely to differ from the same openings at 60s apart.
With the simulations, we aimed to perform what-if studies to assess the influence of interactions (door open-
ings) on performance measures – energy consumption and temperature of stored food items – and to explore,
by comparison, design variations that may potentially compensate for negative effects of the interactions.
The main deliverable of the FALCON project was a virtual open platform (VOP) that enables, among other
things, the collection of TPUI and performing descriptive analytics on the collected data. The VOP supports
simulations by offering a Data Export Module that converts user-specified selections from the collected data
to a comma-separated values (CSV) file, a basic table format that can be read by most simulation packages.
The user-specified selection of the TPUI to be listed in the CSV file is handled by a VOP module called PUI
query builder, which works together with another module responsible for Knowledge Consolidation & Cross
sectoral Management (KCCM) [2].
4. Reference scenario: effect of door openings on refrigerator power consumption
To demonstrate the potential of TPUI-based simulation a simulation model was implemented to explore a
business scenario provided by Arçelik, a consumer electronics and household appliance manufacturer based
in Turkey, partner in the FALCON consortium and envisaged user of the VOP. In this scenario, a product
development team wants to exploit TPUI by performing what-if type simulations to explore improvement
options for upcoming refrigerator models, or firmware updates for the current model. For simulation modeling
and execution, we have used MATLAB/SimulinkTM, as it is widely used for engineering simulations [20], and
provides a basic refrigeration model that we could adapt and extend for use in our investigations.
The concrete case that we elaborated concerned an investigation on how the door openings by the end user
affect energy consumption, and/or the course of the inside temperature. To that end, the VOP user starts out
using the PUI Query Builder and the KCCM to select a representative refrigerator unit – for instance a ‘worst-
case’ sample of which the doors are opened very frequently. Using the Data Export Module in conjunction
with the KCCM, they create a CSV file that can be read by Simulink. In the Simulink simulation, the effect of
the selected users’ door-opening behavior on energy consumption can be studied. Based on these outcomes
different control regimes for the thermostat and/or the interior fan can be considered and the influence on these
regimes on the energy consumption can be evaluated.
5. Energy consumption of Refrigerators: related work
The influence of door openings on the performance of refrigerators is subject of ongoing debate [21]. The
frequency and duration of door openings have influence on the thermodynamic performance and the energy
consumption of a refrigerator [22]. On the one hand, there are authors who point out that other use-related
factors, such as temperature setting and room temperature have a much stronger influence [23], on the other
hand, with other factors constant, door openings are reported to increase energy consumption by 1-8% accord-
ing to several sources reviewed in [24]. Considering that the refrigerator is known to be one of the largest
electricity consumers in a household for instance, according to data from [25], refrigerators, refrigerator-
freezer combinations and freezers are accountable for 33.6% of the total electricity consumption per household
in the United Kingdom – reducing the influence of door openings can have a large impact. However, note that
in countries where HVACs are common, the relative share will likely be lower.
To investigate the influence of door openings, simulations have also been devised [26,27], but the door-opening
patterns in these simulations were not based on data collected from real usage. Also, the papers reporting on
these simulations do not discuss the simulation speed, and therefore we can assume that investigating longer
periods of use, as in our case, has not been considered.
6. Data collection and sampling considerations
The original data produced by Arçelik’s connected refrigerators contains time-stamped values of readouts from
various sensors. The defaulted interval between successive readouts is tsample » 1h. Among these are (i) the end
time of the interval (the time stamp), (ii) the total door opening times for the top and bottom freezer
compartment, and (iii) the numbers of door openings during the elapsed interval. These and other readouts,
such as compartment temperatures, can be selected for inclusion in the CSV file using the PUI Query Manager.
Currently, exact timings of door openings are not included: To further increase realism in simulations, data
would have to be collected at shorter intervals. For now, we have approximated the occurrence of door open-
ings by taking the total opening time per hour, starting at the time of data transfer. If during the interval (ttransfer
– tsample, ttransfer] the door has been open n times for the cumulative duration
with the individual Dtop en,i not specified in the data, we have simulated that, starting at ttransfer, the door was
open for Dtopen(ttransfer). With this processing scheme, a future setup in which event-based data transfer provides
data at the end of every door opening, so that n = 1 for each transfer and Dtopen(ttransfer) is no longer cumulative,
would enable us to simulate the actual door openings.
7. Simulation model
Figure 2 shows our simulation model of the refrigerator. It is based on a refrigeration model provided with
Simulink [28] (Refrigeration cycle model […]), which was modeled using Simscape, a Simulink environment
for simulating physical systems. As our main goal was to investigate the opportunities TPUI-based simulation
offers for conducting what-if studies, we have not spent efforts in fine-tuning the simulation model so that it
gives the best possible behavioral approximation of a particular specific refrigerator design. Assuming that
door-opening behaviors in using refrigerator-freezer combinations do not depend on the particular make of the
tt t()
open transfer open i
Figure 2. Refrigerator simulation model
P E(kWh)
FanPowerT _evaporator
deviation mass_flow
appliance, our investigations in this paper can be said to apply to a hypothetical
refrigerator design and variations on it. This way, we also did not have to ex-
pose company-confidential design information.
To consider the effect of door openings we applied the following modifications
and extensions (names in italics refer to block names in Figure 2):
1. Adding a TPUI Data Import block to import the CSV file using the ‘Sig-
nal Builder’.
2. Adding a Stateflow chart Interpolation removal to remove meaningless
interpolated values that the Signal Builder adds between entries in the
CSV file. Stateflow is a Simulink environment for modelling decision
logic based on state machines and flowcharts.
3. Adding a subsystem Manual Override to allow interactive checking of the effect of door openings
(Figure 3). It consists of two manual switches, which toggle between their two input ports if the simu-
lation user double-clicks on them, even while the simulation is running. To use the TPUI data, the left-
hand switch is set as shown in Figure 3. In the other position, the simulation interactively receives its
door openings from the bottom switch. To run a reference simulation with the door closed all the time,
the manual input is permanently set to its ‘0’ port. A comparison of this reference situation with a sim-
ulation based on TPUI reveals the actual influence of door openings.
4. Modification of the Simscape model of the Compartment of the refrigerator. A starting point for this
compartment model was provided in [28] as well, but it was considerably altered to include the effect
of door openings in its heat management. Also, a compartment fan was added to ‘upgrade’ the model to
that of a frost-free refrigerator with interior fan, in accordance with Arçelik’s connected refrigerators.
The design variations that we wanted to compare are: (i) no fan, (ii) a fan that is controlled based on
compressor activity only and (iii) a controlled fan that is off as long as the door is open. More details on
model modifications are described after this listing.
5. Replacing the relay that was used to model the Thermostat by a Stateflow chart, to allow more complex
control regimes for the compressor in what-if studies.
6. Adapting values regarding dimensions, etc., to values corresponding to those of a typical household
7. Creating outputs to allow assessment of (i) energy consumption by the compressor and the fan (Energy,
numeric), and (ii) Average temperatures inside the compartment, including some food items. These are
the performance measures targeted by our investigations. In addition, graphical output of the tempera-
tures as a function of time is provided by the block Temperature graphs.
8. Adjusting the simulation duration to start and end times from the CSV file.
Figure 4 shows the adapted model of the refrigerator compartment to allow investigation of door openings. We
added a subsystem Door Influence […] that reacts on the door-opening data. We also added two blocks repre-
senting food items in the refrigerator, one of which has as its initial temperature the initial ‘cold’ starting tem-
perature of the refrigerator interior, and the other is initially at room temperature, i.e., it represents food that
has just been put into the refrigerator. The Evaporator Convection block is a varconvection block that is
discussed in the next
paragraph. Its convec-
tion coefficient is an in-
put signal, based on
which it can represent
the situation with no
fan, with a non-con-
trolled fan and with a
controlled fan.
Figure 3. Manual override.
Figure 4. Simscape model of the compartment
Figure 5 shows the
‘Door Influence & Fan
control’ subsystem. It
features custom blocks
which are modifica-
tions of the standard
blocks for heat convec-
tion and heat conduc-
tion between two
points, i.e., in our case
between the inside and
the outside of the re-
frigerator. Instead of
having a fixed value for the convection coefficient, the varconvection block ‘Door convection […]’ takes a
variable value as input from an input port. In our model, this variable is controlled by the Statechart shown on
the left. Likewise, the varconduction block ‘Insulation […]’ receives a variable for the insulator area, which
is reduced by the size of the door if the door is open. The logic of the Statechart is shown as a truth table in
Table 1.
The top compartment and the freezer compartment each have a 3W fan. Through iterative exploration we found
that optimum positive influence on the energy consumption is achieved if the fans are synchronized with the
compressor, with the switch-on timing delayed by 21s for the top compartment and 141s for the freezer, while
switching off at the same time as the compressor for the top compartment and delayed by 18s for the freezer.
These timings were implemented in the same Statechart.
8. Input data: product usage information
The data that we used originated from 43 fielded refrigerator units collected over a 432-days’ time span. The
total number of samples over all units was 67,234, out of which 7,826 turned out to be duplicates. Figure 6
shows some overall statistics, after removal of duplicates, obtained with RapidMinerTM data-mining software.
The data-collection time span per unit varied from 0 to 431.7 days (Figure 6a). The majority of units collected
data for less than 200 days. The seven units that collected data for less than one day were excluded from further
analysis. Figure 6b shows the average number of data samples collected per unit per day for the remaining 36
units. Ideally, this number should be 24 (one sample per hour) but the majority of the units present a consider-
ably lower sample density. The lower numbers of data samples in this figure can partially be accounted to units
that were installed after the beginning of the data collection period, but further analysis also revealed time gaps
in the data series: in total, the data contained 202 gaps of more than a day, each unit showing one or more such
Figure 5. Simulation model of door influence and fan control
Table 1. Influence of door openings on heat transfer mode
Heat transfer
natural convection
natural convection
natural convection
forced convection
forced convection
natural convection
natural convection
natural convection
infinite conduction*
and convection**
infinite convection**
forced convection
infinite convection**
infinite conduction*
and convection**
infinite convection**
*enabled by reducing area in varconduction block to include only walls, bottom and top; ** based on the principle that if the door is
open, there should be only one heat-transport barrier (convection or conduction) between the inside air and the outside (if the door is
closed, there is convection on both sides of the door).
Figure 6. Statistics of collected refrigerator data
a) total data collection
time, days (N=43)
range: 0 - 431.7
average: 104
σ= 113
c) average daily freezer
open duration, s(N=36)
25 40
range: 0.03 – 41.2
average: 3.78
σ= 7.63
b) average # data
samples/day (N=36)
range: 0.18 – 23.0
average: 14.6
σ= 6.54
gaps in its data series. Since these gaps cannot be ascribed to lack of user interactions – which would have led
to data samples reporting zero door openings – we have assumed that they are caused by connectivity prob-
To get a first impression of how the units installed in different households compare in terms of door-openings,
we extracted the average time the freezer of each unit was open each day. We focused on the freezer compart-
ment first, because it uses the bulk of the total energy. The distribution of the average daily freezer-open
duration shown in Figure 6c. It suggests that, on average per day, the majority of users opens their freezer for
only a few seconds, but that there are a small number of users that open their freezer for more than 20s. How-
ever, the averages in this result concern the whole dataset per unit, including possible multi-day gaps, from the
first day until the last day they were online.
To obtain ‘cleaner’ data, intervals lacking long gaps had to be selected manually. We selected 9 out of the 36
units that covered a reasonable spread over the daily open durations in Figure 6c, and for each, selected the
longest possible contiguous interval of samplings that did not contain gaps of more than a day. The overview
of the selected refrigerators in Table 2 shows that the average daily open duration over the whole observed
interval is not proportional to the daily average over the simulated contiguous intervals.
9. Simulation setup and scenarios
Figure 7 shows an example of typical simulation output. Since in simulations based on TPUI interesting phe-
nomena in the graph are too far apart in time to produce an illustrative picture, it was created interactively by
manually operating the switch in Figure 3. Figure 7 shows the course of the temperature in the compartment,
as well as the temperatures of already-cold food (T_food_1) and the just inserted food at room temperature
(T_food_2). The influence of door openings is obvious (annotated as ‘door open’ and ‘door closed’, respec-
tively). The figure also gives evidence of a boot-up effect that reflects the commissioning of the refrigerator.
Since this is a one-time event that is atypical for everyday steady-state use, we have eliminated its influence
by ignoring the first 4,000 seconds of each simulation.
The actual refrigerator from which the data was collected uses one compressor for both compartments. We
simplified this set-up by running separate simulations for the top compartment and the freezer unit, each with
their own door-opening data and set temperatures (277K and 255K, respectively), and merged the results af-
terwards. Consequently, we also did not consider heat exchange between the two compartments.
To complete our base scenario, we have assumed that the refrigerator was situated in a kitchen in Turkey with
room temperature 296K (73°F / 23°C). The three design variations specified in Section 7 (number 4 in the
listing), each applied to the two compartments, provided six what-if scenarios to be simulated, and to be com-
bined to represent the refrigerator as a whole.
Apart from the performance of the refrigerator ac-
cording to the simulation model, we also meas-
ured the performance of the simulation itself. Ac-
cording to [12], where it was applied to unrelated
other simulations, simulation performance can be
defined as
where Tvirtual is the time elapsed in the virtual, sim-
ulated world and Tsim the duration of the simula-
tion computation on a given system. A value psim
> 1 indicates a performance psim times faster than
Table 2. Overview of simulated refrigerator units
Figure 7. Simulation output with annotations
all -t ime 24.5 0.4 0.9 41.2 4.2 1.2 0.5 2.2 2.2
sim ulat ed
14.6 27.5 1.9 41.2 5.3 1.9 2.0 4.0 4.0
46.2 15.1 50.8 21.2 49.4 51.3 32.8 15.8 27.9
length of simulated
interval, days
average daily
fre eze r op e n
duration, s
boot-up effect
door open
door closed
t (s)
10. Obtained results
10.1. Refrigerator performance and influence of door openings
In Table 3 we have brought together our simulation outcomes for the freezer compartment. The relative influ-
ence of door openings on the energy was calculated as a percentage, based on comparing a scenario with TPUI
input with a reference scenario in which the door was always closed, all other options being the same.
Beneath the double line, the table shows the outcomes of our three scenarios for the freezer. In the reference
scenario with closed door the energy consumption is 0.80 kWh/day, which is reduced by 0.03 kWh/day with a
fan (both variants). For three units we have also simulated the use of the top compartment. The results are
shown in Table 4, together with the consequences for the refrigerator as a whole. In the reference scenario, the
top compartment consumes 0.057 kWh/day, which is reduced by 0.002 kWh/day with a fan.
In all investigated cases, the relative influence (%) of door openings is larger for the top compartment than for
the freezer, and, on the other hand, the absolute influence (kWh) is larger for the freezer. The largest relative
influence, namely about 6% increase of energy consumption caused by door openings, could be seen in the
case of a fan controlled by the compressor only, in the top compartment. This is an unlikely design choice (and
therefore not included in Table 4), since practically every refrigerator has lighting in its top compartment,
operated by a door-controlled switch that, at the same time, can easily be deployed as a door-open sensor to
control the fan. For the freezer compartment this is different, since unlike the Arçelik refrigerators from
which we collected data many refrigerators have no door sensor and lighting in the freezer compartment.
Here, the design variant with fan that is controlled based on compressor activity only is a realistic design choice
that is worth to be evaluated and that, of all remaining options, shows the largest absolute increase of energy
consumption as a consequence of door openings.
Based on the computed daily consumption rates shown in the tables, the yearly energy consumption can be
derived in order to validate the realism of the simulation model through comparison with findings from litera-
ture. In our case, the total energy consumption would be in the range of 330-340 kWh/year. Considering that
the investigated refrigerator-freezer is a recent model, and that average energy consumption values from the
literature typically include older units [22,24] while energy savings advance with every next generation of
refrigerators [29], this appears to be consistent with the averages of 390 kWh/year that Biglia et al. [30] found
from 483 fielded refrigerator-freezers, in which, on average, the freezer was set at 2.5K colder than in our
simulation and the top compartment at 1K warmer.
Regarding the average temperatures that were computed in the simulations, it can be said in all cases that the
temperature of the air in the compartments averaged at exactly the set value, while the temperatures of the
included food items were always slightly lower in the top compartment and slightly higher in the freezer com-
partment. Influences of the various design choices and door openings were marginal and did not indicate the
existence of a consistent relationship explaining the differences.
10.2. Simulation
To conduct the simulations, we re-
lied on hardware with a level of
processing power that is easily ac-
cessible within typical engineering
. Overall, simulation
performance was between 450 and
1050, with values around 1000 for
2017 Apple MacBook Pro with 3.1 GHz Intel Core i5 processor and 16 GB of RAM, which was also used for other tasks in parallel.
Table 3. Overview of simulation outcomes (freezer compartment)
Table 4. Simulation outcomes including top compartment
unit A B C D E F G H J average
14.6 27.5 1.9 41.1 5.3 1.9 2.0 4.0 4.0 11.4
5.5 9.4 1.0 10.0 3.3 1.5 1.0 3.4 2.1 4.1
46.2 15.1 50.8 21.2 49.4 51.3 32.8 15.8 27.9 34.5
no fan 0.14% 0.17% 0.02% 0.24% 0.03% 0.02% 0.08% 0.08% 0.22% 0.11%
fan controlled by compressor only 1.15% 1.20% 0.33% 1.59% 0.47% 0.30% 0.57% 0.57% 0.37% 0.73%
fully controlled fan 0.34% 0.34% 0.25% 0.37% 0.26% 0.23% 0.33% 0.33% 0.23% 0.30%
average dai ly freezer open frequency (simul ated in terval)
average dai ly freezer open durati on, s (si mul ated in terv al)
influence of door
opening s on energ y
length of simulated interval, days
Maxima in bold, m inima in bold italics
no fans 0.50% 0.16% 3.82% 0.48% 0.27% 0.05%
freezer fan controlled by
compressor only,
top fan fully controlled
1.11% 1.69% 0.32%
fully controlled fans 0.35% 0.55% 0.25%
average dai ly top comp artment open dur ation , s
average dai ly top comp . open frequ ency
Maxim a in bold, minima in bold italics
length of simulated interval, days
influence of
opening s on
the top compartment and around 550 for the freezer. We could not find any evidence indicating that having to
import and process TPUI would slow down the simulations.
10.3. Regression analysis
To investigate whether our simulations could be replaced by a single generalized relation to predict the relative
influence of door openings d based on daily opening duration t and frequency f, we applied regression analysis.
As input we used the values from the scenario “fan controlled by compressor only” applied to the freezer
compartment, i.e., the scenario with the largest absolute impact. To that end we used the Regression Learner
app in MATLAB, which offers semi-automated analysis using 23 different regression models. The best fit
resulted from a simple linear regression model d = α + β·t + γ·f, with α = 3.05E-03, β = 2.08·10-4 and γ =
4.48·10-4, found after 6-folds cross validation, with root mean square error RMSE = 1.74·10-3. This error is
rather large considering that the response values are in the range 0 to 2% (0 to 2·10-3). Table 5 demonstrates
that the practical value of predictions based on the regression model is indeed limited.
We also explored whether a model could be found that provides an (almost) exact fit if no validation is applied.
The existence of such a model
might indicate that, after vali-
dation with more training
data, the simulation can in-
deed be replaced by one rela-
tion that provides the simula-
tion end results much faster.
The best fit was provided by
an exponential Gaussian pro-
cess model with RMSE =
6·10-4, which we consider still
too far from an exact fit to
merit further investigation, es-
pecially since it is a more
complex model and therefore
more prone to overfitting [cf.
11. Discussion and conclusions
Using TPUI as input for dynamic simulation models only makes sense if performance measures are investi-
gated that are actually influenced by the timing of changes in the TPUI and if these measures form an assess-
able part of the simulation outputs. A domestic refrigerator is a typical product that lends itself for such simu-
lations: an important, quantitative performance measure is its energy consumption, which is dynamically in-
fluenced by detectable user interactions while the refrigerator is operating and consuming energy.
In this paper we presented first results of using simulations to assess the effect of user interactions (door open-
ings) on performance (energy consumption) of a product (refrigerator), and to review, in various scenarios,
how possible design variations can influence these effects. To allow using TPUI as input during simulations
we created custom simulation-modeling elements that accept input signals in order to vary values that are
normally assumed to be constant. Door openings and closings could effectively be modeled by varying areas
and heat transmission properties of refrigerator construction components during runtime. Although our ap-
proach and the available hardware allowed us to perform simulations at a speed of 450-1050 times real-time
and investigate use over a longer time interval, an average simulation run still took 1-2 hours, which, so far,
limited us to investigating only three units with both refrigerator compartments and different control-regime
variations as well as accompanying reference scenarios (with door always closed), and six additional units
where this was done for the freezer only – which has the largest impact. Based on our findings, we could at
least assess some of the statements regarding influence of door openings that we cited in Section 5. It turns out
that even when the doors are being opened rather frequently, the lower end of the 1-8% range mentioned in
[24] was reached, but only if the freezer compartment has a fan that does not react on door openings. In the
corresponding design variation in our simulations, door openings had a 0.57 up to 1.66% impact on energy
consumption. In such refrigerators it might be worthwhile to consider adding a door switch to control the fan,
or not to have a fan in the freezer compartment. Obviously, the latter is not an option as the fans also have a
role in defrosting, which we did not consider in our model. At any rate, considering the fact that a refrigerator
Table 5. Predictions based on linear regression
A14.62 5.52 1.15% 0.86% 0.293% 0.293% 25.5% 25.5%
B27.52 9.39 1.20% 1.30% -0.099% 0.099% -8.2% 8.2%
C1.89 1.04 0.33% 0.39% -0.061% 0.061% -18.6% 18.6%
D41.15 9.96 1.59% 1.61% -0.018% 0.018% -1.1% 1.1%
E5.34 3.32 0.47% 0.57% -0.095% 0.095% -20.2% 20.2%
F1.91 1.48 0.30% 0.41% -0.112% 0.112% -37.2% 37.2%
G2.01 1.01 0.57% 0.39% 0.178% 0.178% 31.1% 31.1%
H3.99 3.42 0.57% 0.54% 0.028% 0.028% 4.9% 4.9%
J4.02 2.12 0.37% 0.48% -0.114% 0.114% -30.7% 30.7%
avg. 11.38 4.14 0.73% 0.73% 0.000% 0.11% -6.1% 19.74%
avg. daily
open freq.
avg. daily
durati on
prediction deviation
substantially contributes to a household’s overall electricity consumption, even attempts to deal with the small
influence of door openings that we found in our investigations could make sense.
What we did not include in our simulations was the putting in and taking out of food items that goes together
with door openings. Items that are put in typically have a higher temperature than the compartment, which
might partly explain the differences with findings from literature. With the sensing technologies currently
implemented in refrigerators, it does not seem likely that such data can be added to the TPUI already collected.
Another factor of influence could be the inclusion of older, less efficient refrigerators in the field studies dis-
cussed in [24].
The fact that we could not identify an equivalent regression model to replace the simulations seems to confirm
that TPUI as input for simulations offers added value compared to synthetic data such as cyclic load patterns.
Apparently, the more complex and/or irregular usage patterns captured in the TPUI lead to results that cannot
be predicted based on average door-opening times and frequencies alone. However, if we would extract more
features from the data, such as for instance average times between door openings, it might be possible to find
a suitable equivalent model after all. In that case, only a limited series of simulations based on TPUI might be
enough to derive a sufficiently reliable model based on regression or some other machine-learning approach.
Once such a model would be available, fast predictions of the influence of use patterns based on a limited set
of features extracted from the data would be conceivable. However, such models do not lend themselves to
incorporation of design modifications, and can therefore not be used to evaluate design alternatives.
The possibility to exploit TPUI by performing simulations is likely to have impact on the way future products
will be designed. Firstly, in designing each first generation of a product range to collect and transfer data,
anticipative consideration must be paid as to what data collection capacities will be included in the design. For
example, in the case of refrigerators, changes in ambient temperature are known to affect the performance.
Since the investigated refrigerators were not equipped with external temperature sensors, we could not inves-
tigate this effect. Moreover, if the refrigerator would be able to keep track of its own energy consumption,
simulations would no longer be useful in the case where only effects on the current design would be studied
but they would still add value if design alternatives are to be explored.
Secondly, once product units are out on the market, TPUI-based simulations can be used to study how real-life
usage affects performance. If certain manifestations of usage emerging from the data raise suspicion of nega-
tively affecting performance (as in our case the door openings), comparison with reference data that lack these
manifestations (in our case fictitious input with the door always closed) can reveal the severity of the problem.
If serious enough, designers can ideate possible solutions to mitigate the negative effects, implement these in
the simulation model and run simulations with the real-life data to compare the effectiveness of the proposed
After selecting an effective solution, it can be implemented in a next-generation redesign, or if it can be realized
in software, as an update for fielded products. TPUI-based simulations will mostly facilitate redesign or de-
signing variations on existing designs. After all, the usage-related input signals to the original simulation model
must also be meaningful in a modified model. If the hinged door of a refrigerator is replaced by a sliding door
or a lid, the collected door data are likely no longer meaningful.
12. Future work
Up till now, we have applied several simplifications and shortcuts in our simulations, which we applied to a
limited set of units. We could think of several options to further improve the realism and the usefulness of
TPUI-based simulations. Among other things, it seems worthwhile to consider and investigate:
influence of usage phenomena such as environment temperature, quantity and temperature of items put
in and taken out.
inclusion of physics effects currently ignored in the model, such as heat exchange between compart-
ments, energy consumption by the light, interior geometry, etc.
fine-tuning model parameters by comparison with a physical specimen of the refrigerator – which is up
to the company, and which might not lead to publishable results due to confidentiality issues
spreading multiple openings during an hour evenly or randomly over that hour
more fielded units, and to apply machine learning to simulation results with more features from the data
to create faster-computing models for investigating the influence of door openings.
Finally, in the context of generalization, it would be interesting to investigate how TPUI-based simulation can
be applied to other products and how these may benefit from it. Perhaps our approach of customizing simula-
tion elements to allow variations of values that are normally considered to be fixed will turn out to be generally
applicable solution for introducing human manipulations into engineering simulations.
Part of this research has been funded under the EC Horizon 2020 Programme, in the context of the FALCON
project, “Feedback mechanisms Across the Lifecycle for Customer-driven Optimization of iNnovative prod-
uct-service design” ( The authors wish to acknowledge the Commission and all
the FALCON project partners for fruitful collaboration.
[1] Porter, M. E., and Heppelmann, J. E., 2014, “How Smart, Connected Products Are Transforming
Competition,” Harv. Bus. Rev., 92, pp. 11–64.
[2] Hribernik, K., Franke, M., Klein, P., Thoben, K.-D., and Coscia, E., 2017, “Towards a Platform for
Integrating Product Usage Information into Innovative Product-Service Design,Proceedings of
ICE/ITMC, IEEE, pp. 1407–1413.
[3] Kotu, V., and Deshpande, B., 2014, Predictive Analytics and Data Mining: Concepts and Practice with
Rapidminer, Morgan Kaufmann.
[4] Cooper, A., 2012, “What Is Analytics? Definition and Essential Characteristics,” CETIS Anal. Ser., 1(5),
pp. 1–10.
[5] Shannon, R. E., 1998, “Introduction to the Art and Science of Simulation,” Winter Simulation
Conference, IEEE Computer Society Press, pp. 7–14.
[6] Quesnel, G., Duboz, R., and Ramat, É., 2009, “The Virtual Laboratory Environment–An Operational
Framework for Multi-Modelling, Simulation and Analysis of Complex Dynamical Systems,” Simul.
Model. Pract. Theory, 17(4), pp. 641–653.
[7] Geppert, J., 2011, “Modelling of Domestic Refrigerators’ Energy Consumption under Real Life
Conditions in Europe,” Friedrich Wilhelms University, Bonn.
[8] Huang, Y., and Verbraeck, A., 2009, “A Dynamic Data-Driven Approach for Rail Transport System
Simulation,” Winter Simulation Conference, pp. 2553–2562.
[9] Hu, X., 2011, “Dynamic Data Driven Simulation,” SCS MS Mag., 5, pp. 16–22.
[10] Cummings, M. L., 2004, “The Need for Command and Control Instant Message Adaptive Interfaces:
Lessons Learned from Tactical Tomahawk Human-in-the-Loop Simulations,” Cyberpsychol. Behav.,
7(6), pp. 653–661.
[11] Prevot, T., Lee, P., Callantine, T., Mercer, J., Homola, J., Smith, N., and Palmer, E., 2010, “Human-in-
the-Loop Evaluation of NextGen Concepts in the Airspace Operations Laboratory,” AIAA Modeling &
Simulation Technologies Conference, p. 7609.
[12] van der Vegte, W. F., and Horváth, I., 2012, “Bulk Simulation of Using Information-Intensive Products
and Product-Service Systems: Formal Underpinnings,” Tools and Methods of Competitive Engineering,
Karlsruhe, pp. 999–1012.
[13] Jung, S., Lee, C., Kim, K., Jeong, M., and Lee, G. G., 2009, “Data-Driven User Simulation for
Automated Evaluation of Spoken Dialog Systems,” Comput. Speech Lang., 23(4), pp. 479–509.
[14] Holler, M., Uebernickel, F., and Brenner, W., 2016, “Understanding the Business Value of Intelligent
Products for Product Development in Manufacturing Industries,” Proceedings ICIME, ACM, pp. 18–
[15] Paul, R. J., 1991, “Recent Developments in Simulation Modelling,” J. Oper. Res. Soc., pp. 217–226.
[16] Qiao, G, S., Riddick, P., and McLean, D., 2003, “Data Driven Design And Simulation System Based
On XML,” Winter Simulation Conference, New Orleans, pp. 1143–1148.
[17] Barlas, P., and Heavey, C., 2016, “Automation of Input Data to Discrete Event Simulation for
Manufacturing: A Review,” Int. J. Model. Simul. Sci. Comput., 7(01), p. 1630001.
[18] Pei, M., Han, R., Kwon, D., Lucero, A., Vasudevan, V., Kwasnick, R., and Polasam, P. S., 2013, “Define
Electrical Packing Temperature Cycling Requirement with Field Measured User Behavior Data,” IEEE
Electronic Components & Technology Conference, pp. 159–165.
[19] Urban, B., and Roth, K., 2014, A Data-Driven Framework For Comparing Residential Thermostat
Energy Performance, Fraunhofer USA, Boston.
[20] Kulakowski, B. T., Gardner, J. F., and Shearer, J. L., 2007, Dynamic Modeling and Control of
Engineering Systems, Cambridge University Press.
[21] Fiore, E., 2018, “Systemic Design for the Innovation of Home Appliances,” Politecnico di Torino.
[22] James, C., Onarinde, B. A., and James, S. J., 2017, “The Use and Performance of Household
Refrigerators: A Review,” Compr. Rev. Food Sci. Food Saf., 16(1), pp. 160–179.
[23] Geppert, J., and Stamminger, R., 2010, “Do Consumers Act in a Sustainable Way Using Their
Refrigerator? The Influence of Consumer Real Life Behaviour on the Energy Consumption of Cooling
Appliances,” Int. J. Consum. Stud., 34(2), pp. 219–227.
[24] Haines, V., Lomas, K., Thomson, M., Richardson, I., Bhamra, T., Giulietti, M., Tang, T., Lawton, C.,
Guo, L., and Allinson, D., 2010, “How Trends in Appliances Affect Domestic CO2 Emissions: A
Review of Home and Garden Appliances,” Dep. Energy Clim. Change DECC Lond. UK.
[25] Yao, R., and Steemers, K., 2005, “A Method of Formulating Energy Load Profile for Domestic Buildings
in the UK,” Energy Build., 37(6), pp. 663–671.
[26] Borges, B. N., Melo, C., and Hermes, C. J., 2015, “Transient Simulation of a Two-Door Frost-Free
Refrigerator Subjected to Periodic Door Opening and Evaporator Frosting,” Appl. Energy, 147, pp. 386–
[27] Mastrullo, R., Mauro, A., Menna, L., Palma, A., and Vanoli, G., 2014, “Transient Model of a Vertical
Freezer with Door Openings and Defrost Effects,” Appl. Energy, 121, pp. 38–50.
[28] Ng, L., 2014, “Refrigeration Cycle in Simscape” [Online]. Available:;
[Accessed: 07-Feb-2018].
[29] Bakker, C., Wang, F., Huisman, J., and den Hollander, M., 2014, “Products That Go Round: Exploring
Product Life Extension through Design,” J. Clean. Prod., 69, pp. 10–16.
[30] Biglia, A., Gemmell, A. J., Foster, H. J., and Evans, J. A., 2017, “Temperature and Energy Performance
of Domestic Cold Appliances in Households in England,” Int. J. Refrig., pp. 172–184.
[31] Wang, Y., Miller, D., and Clarke, R., 2008, “Approaches to Working in High-Dimensional Data Spaces:
Gene Expression Microarrays,” Br. J. Cancer, 98(6), p. 1023.
... Regarding the usage and analysis methodology of PUI, the reviewed application cases often use data science methods and tools (e.g., machine learning, text and data mining, and statistic analysis) to derive useful knowledge or models for development tasks. In some cases, researchers use a set of integrated PUI directly for, for example, simulation analysis [28]. ...
Full-text available
Product design is crucial for product success. Many approaches can improve product design quality, such as concurrent engineering and design for X. This study focuses on applying product usage information (PUI) during product development. As emerging technologies become widespread, an enormous amount of product-related information is available in the middle of a product’s life, such as customer reviews, condition monitoring, and maintenance data. In recent years, the literature describes the application of data analytics technologies such as machine learning to promote the integration of PUI during product development. However, as of today, PUI is not efficiently exploited in product development. One of the critical issues to achieve this is identifying and integrating task-relevant PUI fit for purposes of different product development tasks. Nevertheless, preparing task-relevant PUI that fits different product development tasks is often ignored. This study addresses this research gap in preparing task-relevant PUI and rectifies the related shortcomings and challenges. By considering the context in which PUI is utilized, this paper presents a systematic procedure to help identify and specify developers’ information needs and propose relevant PUI fitting the actual information needs of their current product development task. We capitalize on an application scenario to demonstrate the applicability of the proposed approach.
A critical factor that makes a product successful is its acceptance in its market. To achieve this goal, producers oftentimes collect and analyze feedback information from the market. This information allows them to get a deeper understanding about the product behaviors, customers, their usage patterns, future needs and expectations. The academic literature describes a variety of use cases that outline how development-related tasks can benefit from Product Usage Information (PUI). They differ, for instance, in the investigated product, task, communication channel, information source, and the type of the result. This diversity and the lack of a generally agreed vocabulary in this research domain facilitates a fragmentation of PUI-related research. This paper provides a first systematic overview of a selection of PUI application cases in product design improvement. The study sample consists of 17 research papers from the last 20 years. We characterize and classify the papers mainly through three dimensions: product type, product development phase, and information sources and channels. The results indicate that PUI can support different tasks during product improvement, both in the task clarification phase and in the product conceptual, embodiment and detailed design phases. Our findings suggest that organizations need to know more about PUI-related information sources and channels.
Full-text available
The real-life use of a product is often hard to foresee during its development. Fortunately, today’s connective products offer the opportunity to collect information about user actions, which enables companies to investigate the actual use for the benefit of next-generation products. A promising application opportunity is to input the information to engineering simulations and increase their realism to (i) reveal how use-related phenomena influence product performance and (ii) to evaluate design variations on how they succeed in coping with real users and their behaviors. In this article we explore time-stamped usage data from connected fridge-freezers by investigating energy losses caused by door openings and by evaluating control-related design variations aimed at mitigating these effects. By using a fast-executing simulation setup we could simulate much faster than real time and investigate usage over a longer time. We showed that a simple, single-cycle load pattern based on aggregated input data can be simulated even faster but only produce rough estimates of the outcomes. Our model was devised to explore application potential rather than producing the most accurate predictions. Subject to this reservation, our outcomes indicate that door openings do not affect energy consumption as much as some literature suggests. Through what-if studies we could evaluate three design variations and nevertheless point out that particular solution elements resulted in more energy-efficient ways of dealing with door openings. Based on our findings, we discuss possible impacts on product design practice for companies seeking to collect and exploit usage data from connected products in combination with simulations.
Full-text available
The domestic refrigerator is now a common household device with very few households in the developed world not possessing one, or more, for the storage of chilled foods. Domestic storage is the last, and in many respects the most important, link in the food chill chain. Inadequate domestic refrigeration or cooling is frequently cited as a factor in incidents of food poisoning. The authors reviewed the temperature performance of refrigerators in 2008. This new review builds on that review, covering studies that have been published since (and those that were unfortunately missed in the first review), and also seeks to put this important stage of the food cold chain in its context. It is clear from the published data that many refrigerators throughout the world are running at higher than recommended temperatures. It is also clear that, despite improvements in energy use, the temperature performance and use of refrigerators has not changed significantly in the last 40 or so years. Many consumers still remain unaware of the recommended refrigeration temperature range, how to ensure that the correct refrigeration temperature range is achieved, the importance of monitoring that the recommended temperature range is being maintained, and the potential hazards of temperature abuse.
Full-text available
Discrete event simulation (DES) is a well-established decision support tool in modeling work flows in manufacturing industry. But, there are an amount of practical and financial obstacles that deter the employment of this technology in industry. One of the main weaknesses of operating DES is the costs spent on collecting and mapping input data from different enterprise data resources into a DES model. Another issue is the cost of integrating simulation applications with other manufacturing applications. These barriers hinder the automated input of data into DES models and as a result deter use of real-time DES in manufacturing. This review presents the existing research studies in the literature that address the above issues, demonstrating in parallel the already implemented concepts. The scope of this review is to provide an overview of the input data phase, focusing on its automation and motivating researchers to re-examine this phase by highlighting future research directions.
Conference Paper
A product's value is increasingly determined by the services supporting it. Complex Product-Service Systems (PSS) – combinations of services and products – are in demand. Product usage information (PUI), that is, information about how an individual product is used, is rapidly becoming a valuable asset to industry to help inform service offers throughout the product lifecycle. Conventionally, information about how a product is used is gathered in customer relations and MRO (maintenance, repair and overhaul) processes via channels such as repair logs, call centers or helpdesks. Today, new channels of PUI are emerging from the digitalization of products, the Internet of Things, Cyber-Physical Systems, Internet sources and social media. To benefit from the user experience and knowledge about product-service usage contained in PUI, current product and service development systems (e.g. PDM/PLM, CAx and simulation tools) and the related processes and methodologies need to be extended to incorporate PUI feedback and feedforward loops. This means a suitable IT infrastructure is required to efficiently and seamlessly integrate and apply PUI to PSS (re-)design processes. This paper proposes such an IT infrastructure in the form of an architecture for a Virtual Open Platform for incorporating feedback loops into PSS (re-)design processes and the corresponding IT systems.
This paper reports the results of a large-scale survey in which 998 cold appliances were monitored in 766 properties in England. No surveys published to date analyse such a large dataset, which includes data on ambient temperature, cold appliance temperature (refrigerator and/or freezer) and electricity consumption of the cold appliance. Simultaneous measurements of the temperature inside and outside of the cold appliances and the electricity consumption were taken over a period of seven days during a nine-month period in 2015. An interview was also conducted with the householders to collect further information about the cold appliances and their usage patterns. The cold appliances monitored in the work included fridge-freezers (52%), refrigerators with ice-box (6%), larder fridges (14%), chest freezers (9%) and upright freezers (19%). It was found that for all monitored cold appliances with valid data that: the mean ambient temperature was 18.5°C; the mean refrigerator temperature was 5.3°C; the mean freezer temperature was -20.3°C; and the mean electricity consumption was 354 kWh per year. Significant differences between the electricity consumption of different types of cold appliance were determined from statistical analysis.
Conference Paper
Intelligent products are transforming manufacturing industries and offer new opportunities for business innovation. As little is known about the business value of intelligent products for product development, this exploratory paper aims to provide guidance. Empirical data from case studies and focus groups and a structured literature review suggest that a positive impact on every product development stage exists. More specifically, findings propose that intelligent products enable (1) transparency generation on market opportunities, (2) specification of requirements, (3) customer- and user-centric product portfolio planning, (4) design and process planning for usage, (5) shortening and replacing of physical prototyping and field testing, and (6) superordinate benefits. With the inherent limitations of a case study approach, we supply scholars and practitioners with first exploratory insights into this emerging research field.
Put Predictive Analytics into Action Learn the basics of Predictive Analysis and Data Mining through an easy to understand conceptual framework and immediately practice the concepts learned using the open source RapidMiner tool. Whether you are brand new to Data Mining or working on your tenth project, this book will show you how to analyze data, uncover hidden patterns and relationships to aid important decisions and predictions. Data Mining has become an essential tool for any enterprise that collects, stores and processes data as part of its operations. This book is ideal for business users, data analysts, business analysts, business intelligence and data warehousing professionals and for anyone who wants to learn Data Mining. You'll be able to: 1. Gain the necessary knowledge of different data mining techniques, so that you can select the right technique for a given data problem and create a general purpose analytics process. 2. Get up and running fast with more than two dozen commonly used powerful algorithms for predictive analytics using practical use cases. 3. Implement a simple step-by-step process for predicting an outcome or discovering hidden relationships from the data using RapidMiner, an open source GUI based data mining tool. Predictive analytics and Data Mining techniques covered: Exploratory Data Analysis, Visualization, Decision trees, Rule induction, k-Nearest Neighbors, Naïve Bayesian, Artificial Neural Networks, Support Vector machines, Ensemble models, Bagging, Boosting, Random Forests, Linear regression, Logistic regression, Association analysis using Apriori and FP Growth, K-Means clustering, Density based clustering, Self Organizing Maps, Text Mining, Time series forecasting, Anomaly detection and Feature selection. Implementation files can be downloaded from the book companion site at • Demystifies data mining concepts with easy to understand language • Shows how to get up and running fast with 20 commonly used powerful techniques for predictive analysis • Explains the process of using open source RapidMiner tools • Discusses a simple 5 step process for implementing algorithms that can be used for performing predictive analytics • Includes practical use cases and examples.
This paper describes a quasi-steady-state simulation model for predicting the transient behavior of a two-door household refrigerator subjected to periodic door opening and evaporator frosting. A semi-empirical steady-state model was developed for the refrigeration loop, whereas a transient model was devised to predict the energy and mass transfer into and within the refrigerated compartments, and also the frost build-up on the evaporator. The key empirical heat and mass transfer parameters required by the model were derived from a set of experiments performed in-house in a climate-controlled chamber. In general, it was found that the model predictions followed closely the experimental trends for the power consumption (deviations within ±10%) and for the compartment temperatures (deviations within ±2 K) when the doors are opened periodically and frost is allowed to accumulate over the evaporator.
This article presents dynamic data driven simulation as a new simulation paradigm where a simulation system is continually influenced by real time data for better analysis and prediction of a system under study. This is different from traditional simulations that are largely decoupled from real systems by making little usage of real time data. We present a framework of dynamic data driven simulation based on sequential Monte Carlo methods, describe the opportunities and challenges associated with dynamic data driven simulation, and show an illustrative example.