Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

International Journal of Production Research, 2015

Vol. 53, No. 21, 6629–6636, http://dx.doi.org/10.1080/00207543.2015.1055844

Real-time dynamic simulation of continuous bulk material flow to improve the statistical
modelling of final product strength properties
Martin Rieglera*, Nicolas Andréb, Manfred Gronaltc and Timothy M. Youngb
a
Wood K plus – Competence Centre for Wood Composites and Wood Chemistry, Tulln, Austria; bCenter for Renewable Carbon,
University of Tennessee, Knoxville, TN, USA; cInstitute of Production and Logistics, University of Natural Resources and Life
Sciences, Vienna, Austria
(Received 8 January 2014; accepted 10 May 2015)

In regression analyses, correlations between independent variables (e.g. process variables) and dependent variables
(e.g. product quality) are of major interest. However, only statistically significant correlations ensure a reliable inter-
pretation of how process variables affect product qualities. In this respect, accurate time alignment of independent
variables is crucial to obtain regression models with acceptable validation that are influenced by temporal phenomena
(e.g. industrial processes) only. In this study, the commonly used static form of time alignment, where only the
distances between consecutive process parameters and the average production speed are considered, is compared to a
newly developed dynamic calculation of time lags. The dynamic calculation of time lags was achieved by modelling
the continuous bulk material flow. The two different methods of calculation were then applied on an industrial
production of particleboards to predict final board strength properties. Results of regression models showed that the
use of dynamically calculated time lags improved the predictability of the internal bond strength of boards by 67%
compared to statically calculated time lags. Consequently, final product strength properties could be predicted more
accurately, which should lead to lower costs of rejects and a higher efficiency of material inputs.
Keywords : modular flow sheet simulation; time lags; process engineering; multivariate regression analysis; data mining;
data fusion

Introduction
For the optimisation of industrial processes, statistical analyses can serve as useful tools for identifying relationships
between raw materials, processes and product properties. Statistical analyses for process optimisation and control can be
divided into descriptive and exploratory data analyses. A commonly used descriptive analysis for analysing and control-
ling the manufacturing process of wood composite panels are Shewhart control charts (Young, Winistorfer, and Wang
1999). On the other hand, modelling techniques such as regression analyses or neural networks are often used for explo-
rative data analyses, which allow the prediction of technological properties of wood composite panels and the adaptation
of their manufacturing processes (Bernardy and Scherff 1997; Cook and Chiu 1997; Maness, Kozak, and Staudhammer
2003; Hasener 2004; André et al. 2008; Clapp, Young, and Guess 2008, Young et al. 2008; Esteban, Fernández, and De
Palacios 2011; André and Young 2013; Riegler et al. 2013; Young et al. 2014).
A better understanding of the relationships between specific production processes supports manufacturers in analys-
ing faults of production and in adapting production processes. For a reliable interpretation of those relationships, it is
important to know how the components within the test specimen were processed and quantitatively merged throughout
the manufacturing process. In this respect, the elapsed time between online determined process parameters and the point
in time (t(0) in Figure 1) when a board sample is withdrawn from the production line for destructively testing its strength
properties (e.g. internal bond strength (IB), bending properties, etc.) has to be considered. This can be achieved by
tracking the flow of materials throughout the manufacturing process.
In this study, the time intervals between the point in time when the material passed a process sensor and the point
in time when the final board property was measured are called ‘time lags’. These time lags are required for proper time
alignment of process parameters with data obtained from destructive testing. This process is known as elementary
distributed data fusion (also called track-to-track fusion) where data from multiple diverse sensors is combined to make
inferences about a physical event, activity or situation (Hall 1992). Clapp, Young, and Guess (2008) used statically

*Corresponding author. Email: m.riegler@kplus-wood.at

© 2015 Taylor & Francis


6630 M. Riegler et al.

Figure 1. Layout of the material flow in the industrial manufacturing process of particleboards with physically separated core and
face material flows (single and double line); particles are stored in a bin (form) before forming the particle mat (top, bot); arrows
indicate transportation between processes and bins; time (t(0)) indicates point in time for cutting specimens for IB testing.

calculated time lags for this process of data fusion to predict the IB of medium density fiberboards (MDF). Clapp,
Young, and Guess (2008) claimed that the usage of statically calculated time lags limits the reduction in the generalised
error when predicting the IB in MDF. Thus, it can be assumed that if process parameters are inaccurately aligned,
predictions from generated models may be less robust. The accurate calculation of time lags will become even more
important if the short-term variability of process parameters increases.
As already mentioned above, one possibility to consider time lags and consequently obtain robust models with
acceptable validation is to identify accurate material flows during production. Commonly used approaches for analysing
material flows in logistic systems are well defined in queuing theory (Gross and Harris 1998). Many queuing analyses
focus on discrete events that are Poisson distributed with output data that is exponentially distributed (Arnold and Furmans
2009). In this study, general distributions would have to be used due to the continuous flow of bulk material and the rather
normally distributed process data (Arnold and Furmans 2009). Additionally, the constantly changing amount of retained
material within silos, bins and bunkers that is common in wood composite panels manufacturing would have to be taken
into account. To overcome the challenges mentioned above, a modular dynamic simulation could be applied to enable an
accurate identification of material flows due to the possibility of simulating process steps independently.
In general, process simulations have been extensively applied in the chemical process industry to track unit opera-
tion performances and component flows (Ayres and Ayres 2002). Werther et al. (2011) suggested the usage of dynamic
flow sheet simulations in combination with multi-scale models to cope with the complex process structures in solids
processing. The calculation of retention times of bulk material within silos for manufacturing wood composite panels is
similar to residence time theory. The residence time of molecules within a reactive system, e.g. piston flow reactors,
affects their probability of reacting (Nauman 2008). Reimers, Werther, and Gruhn (2008) reconciled data to adjust model
parameters for more reliable flow sheet simulations of solids processes, using the software SolidSim. Dosta, Heinrich,
and Werther (2010) included modular and simultaneous simulation strategies for simulating granulation processes. A
modular simulation of the flow of bulk material should potentially improve the accuracy and flexibility of the data
fusion process, as processes can separately be simulated using individual algorithms.
In the wood composites industry, only few studies were found that deal with the appropriate alignment of process
data to the final board properties during the process of data fusion. Bernardy and Scherff (1997) used the so-called
‘material flow tracked’ data for predicting mechanical properties of boards. However, the specifics of the material flow
tracking were not documented in the study mentioned above. Hasener (2004) ascribes lower predictabilities of regression
models to, inter alia, inaccurately calculated time lags, due to the error within the process data (i.e. independent vari-
ables). Hasener (2004) did not specify the time lag calculation in the published study. Kruse et al. (1997) considered the
accurate alignment of process variables to their corresponding board properties to be essential for obtaining reliable
regression models. However, they were not able to measure process parameters before the gluing stage of the
International Journal of Production Research 6631

manufacturing process and thus tried to withdraw information from the glued particles by detailed characterisation on
the laboratory scale.
The aim of this study was to develop a technique to dynamically simulate the continuous flow of bulk material
within the industrial process of manufacturing wood composite panels, specifically particleboards. A dynamic simulation
should improve the time alignment of process and raw material data with final product strength properties. This should
consequently improve predictabilities of validated regression models, which lead to more reliable interpretations of
correlations between process and raw material parameters and final product strength properties.

Materials and methods


Calculation of time lags
In this study, the continuous material flow of wood particles (species mixture with Pinus sp or ‘pine’ as highest
proportion) used for industrially manufacturing particleboards (thickness 20 mm and density 670 kg/m³ on average) was
modularly simulated to align relevant process parameters in time with final mechanical board properties. In general,
industrially manufactured particleboards consist of disintegrated wood particles that are usually bonded by synthetic
resin that is cured under pressure and high temperature (Dunky and Niemz 2002).
The first key initiative of the study was to identify relevant processing steps, metering devices, conveyance and
intermediate storages of the process. These consecutive steps were flow charted (Figure 1). In particular, wood particles
are disintegrated (‘refining’) and dried before synthetic resin is added to the industrial manufacturing process at the glu-
ing stage. After that, the glued wood particles are formed to a mat (top, bot) using special rolls before the mat is cured
within the continuous hot press under heat and pressure (Figure 1).
The flow of wood particles through the manufacturing process was divided into a transportation time tt (conveyors),
a retention time rt (bunkers, silos) and a processing time pt (actual process steps). Due to a merging of these time
sections, the material flow was consistently traced back by
tl ¼ t ð0Þ  tt  rt  pt; (1)
resulting in the so-called time lags tl by subtracting the time sections from the point in time t , when the final board
(0)

sample was cut from the production line at the diagonal crosscut circular saw directly after the final stage of pressing.
The transportation time tti was determined by measuring the length of conveyors (using a measuring tape) and
considering the constant conveyor speeds by
li
tti ¼ ; i ¼ 1; . . .; p; (2)
vi
whereas li is the length and vi the speed of each of the p conveyors. The retention time rti of the continuous flow of
material within bins (bunkers or silos) was determined by
 !
X k 
 intv  intv
rti ¼ min k :  dti h   ati h   ; i ¼ 1; . . .; s; (3)
k¼1;...;qi  h¼1 3600  60

with dti h being the discharge rate of material in kg per hour (transformed into kg per time interval intv, here set to
10 s) at time interval ti − h, ati h being the actual amount of material within bins at time interval ti − h and k being the
time interval where the lowest absolute deviation occurred between the cumulated discharge and the amount of material
within a bin. ti indicates the point in time when the retraced material was discharged from the bins, which was the start-
ing point for calculating the retention time. In this study, the continuous material flow of particles was discretised using
time–discrete intervals of 10 s. Table 1 shows the maximum number of time intervals qi for the s bins within the

Table 1. Maximum number of intervals q for calculating retention times within s bins.

Bins Max. intervals Equates to time (min)

Top, bot 30 5
Form 500 83
Surge bins 900 150
Production silos 7500 1250
6632 M. Riegler et al.

manufacturing process. These values represent the maximum retention time of material within the bins and depend on
the size of the bins and the average discharge rates. In this study, possible mixture phenomena within silos were
neglected. Instead a first-in first-out principle was assumed.
The processing time pt was determined by timing the flow of particles within manufacturing process steps from the
previous to the subsequent metering device (i.e. process parameter).
For a precise backtracking, the material flow had to be divided into a core and a face flow regarding the core and
face material used for particleboard production (Maloney 1993), and a top and bottom flow regarding the top and bot-
tom layers at the forming stage. The core and face flows were physically separated (Figure 1). Within each flow, the top
and bottom flows were imaginarily separated, only dependent on the different time when the material passed through
the production process. Consequently, process parameters prior to the forming stage appeared twice in the subsequent
multivariate data analysis, as top and bottom flows. This approach is similar to the so-called ‘shift operators’ that are
commonly used in time-series analyses, particularly in autoregression moving average models (Thome 2005). In this
study, the shift of time intervals was not equidistant, but depended on the calculated time lags tl. The resulting X matrix
for the multivariate data analysis was formed by
h      i
ð0Þ ð0Þ ðtopÞ ð0Þ ðbotÞ
Xm: ¼ xj tm  tlmj ; xi tm  tlmi ; xi tm  tlmi ; j ¼ 1; . . .; f  1; i ¼ f ; . . .; n; (4)

with x containing values of the n independent process variables at time t(0) minus the time lags tl, considering top (top)
and bottom (bot) flows for process variables after the forming stage f. The above-described dynamic calculation of time
lags was carried out every time the IB of boards was measured offline (m), i.e. in approximately 4 h intervals. The IB
of boards was determined according to ASTM D1037 (2006), whereas the tensile strength perpendicular to the plane of
the board is analysed. The available process data spanned 31 days of production, which corresponds to 267,840
ten-second intervals.
To evaluate the presented approach, dynamically calculated time lags were compared to statically calculated time
lags, where only the distances between process parameters and the average production speed were used. Statically calcu-
lated time lags were used as reference in this study because they are commonly used in industrial manufacturing pro-
cesses of wood composite panels. In addition, the persistent determination of continuous bulk material flows was not
possible and would have been beyond the scope and budget of this study. Time-lag calculations were carried out in
MATLAB® using a newly developed algorithm.

Partial least squares regression modelling


To evaluate the calculation of time lags, time-lag-corrected data were analysed using common multivariate data analysis
(Figure 2) known as partial least squares regression (PLSR) (Wold, Sjöström, and Eriksson 2001). Initially, the time-lag-
corrected data set (290 observations, 313 variables) was divided into a train set (80% of observations used for model

Figure 2. Scheme of evaluating time-lag calculations using PLSR analysis.


International Journal of Production Research 6633

calibration) and a test set (20% of observations used for model validation). For the model calibration, the data were
pretreated by imputing missing values, where a PCA model was iteratively fitted to the data, at which missing values
were initially replaced by the arithmetic mean of their corresponding variables. Afterwards, the initially replaced values
were iteratively imputed (maximum 100 times) until the change in the replaced values dropped below the threshold of
1E-6, using PLS Toolbox (Wise et al. 2012). Additionally, variables with a coefficient of variation below 0.2% and vari-
ables with values that were out of a plausible predefined range (as specified by the manufacturer) were excluded. In the
course of PLSR analysis, significant variables were selected out of 313 possible variables, using interval partial least
squares regression (IPLS) (Norgaard et al. 2000). At this step of variable selection, only those variables were iteratively
added to a new PLSR model that improved the predictability. To validate the so obtained calibrated PLSR model, it was
applied on the test set and the resulting error of predictions was interpreted. The detailed approach of data pretreatment
(data quality), model generation and model validation can be found in Riegler et al. (2013).

Results and discussion


Dynamic calculation of time lags
Time lags were dynamically calculated by considering the flow of bulk material (Figure 1) every time a specimen was
withdrawn from the process for testing the IB. In particular, transportation times (e.g. production speed), retention times
(e.g. silo levels and outputs) and processing times were recalculated. These times were merged to compare dynamic and
static forms of calculation (Figure 3). When using static calculations, the mean time lags of 27.1 min were considerably
shorter than those obtained from dynamic calculations (137.2 min). As the dynamic calculation is able to consider the
actual material flow in real time, this technique is considered to represent more reliable time lags. Consequently, the
usage of statically calculated time lags leads to an overestimation of the velocity of material flows through production,
resulting in 110.1 min shorter time lags on average.
The maximum dynamic time lag of 1464.3 min was more than five times higher than the maximum static time lag
of 281.5 min. Additionally, short-term variations of the material flow could be taken into account by using the dynamic
calculation of time lags. In particular, the 290 sample means of recalculated dynamic time lags in Figure 3 showed a
coefficient of variation of 17%.
The difference between static and dynamic calculations was mainly caused by the fact that static calculations were
not able to consider retentive material within bins. Thus, especially bins with long retention times induced differences
between the two calculation methods. The longest retention time within bins was observed at the production silo for the
face material where dynamically calculated retention times of continuously fed material ranged from 268.7 to
1167.3 min in the analysed period of 31 days (Figure 4). The maximum retention time (1167.3 min) occurred at the
241th sampling of specimens for IB testing. Additionally, the accuracy of dynamically calculating retention times using
the approach presented in this study should be increased by simulating the blending behaviour within silos that usually
occurs during production. This simulation will be included in future research.

Figure 3. Comparison of statically and dynamically calculated time lags.


6634 M. Riegler et al.

Figure 4. Variation of retention time, level and discharge rate from production silo for face material.

As stated in Equation (3), the retention time of continuously fed material within silos depends on the cumulated dis-
charge rate and the amount of material within the silo at a given time. Thus, silos with relatively high capacities in com-
bination with low discharge rates usually induce retention times that are longer and show higher variations. This can
especially be seen in the variations of the level and discharge rates within the production silo for the face material, hav-
ing a capacity of 112,500 kg (Figure 4). Such variations point out the importance of dynamically calculating time lags.
The maximum difference between time lags calculated from the top and bottom material flows was 309.0 min. This
difference corresponds to 1854 ten-second time intervals and is considered to legitimate the calculation of separate top
and bottom streams. The maximum difference between time lags calculated from face and core material flows was
532.3 min. In particular, processing the face material (mean time lags = 475.2 min) took about two and a half times
longer (on average) than processing the core material (mean time lags = 194.2 min) (Figure 5).

PLSR modelling
To evaluate the influence of statically as well as dynamically calculated time lags on the modelling of product properties,
two PLSR models were calculated predicting the IB of boards. The actual IB of all tested specimens was 0.52 N/mm² on
average with a standard deviation of 0.12 N/mm². For predicting the IB of boards, a PLSR model using statically
calculated time lags revealed a mean normalised root mean squared error of prediction (MNRMSEP) (Riegler et al. 2013)

Figure 5. Comparison between time lags of face and core material flows.
International Journal of Production Research 6635

Table 2. Summary of PLSR models using static and dynamic time lags.

Time-lag calculation # Selected variables # Latent variables # Obs. MNRMSEP (%)

Static 13 3 290 138


Dynamic 15 3 290 45

of 138%. This model was built with three latent variables and 13 selected independent variables (Table 2). The most
important selected variable was the dummy variable of product type 8 which showed a positive regression coefficient of
0.31 (scaled units).
The second PLSR model, in which dynamically calculated time lags were used to predict the IB of boards, revealed
a MNRMSEP of 45%, which is a decrease of about 67% compared to the PLSR model using statically calculated time
lags. This second PLSR model was also built with three latent variables and contained 15 selected independent variables
(Table 2). The most important variable was the press temperature in the last zone of the continuous hot press showing a
negative regression coefficient of −0.29 (scaled units).
Consequently, it can be stated that the accurate alignment of data records to the final test specimens, which is
achieved by a dynamic calculation of time lags, leads to an improvement of the predictability of PLSR models, com-
pared to the state of the art static calculation of time lags. This means that the internal bond strength of boards can be
predicted more accurately, which should lead to lower costs of rejects and a higher efficiency of material inputs. The
relatively high MNRMSEP values of both models could be explained by the fact that validation was done on a separate
test data set. The usage of separate test data sets for validation tends to result in higher MNRMSEP values than the
usage of e.g. cross-validation. However, separate test data sets should increase the reliability of statistical models when
applied to real industrial data.
Due to the new data structure (Equation (4)) from aligning time lags dynamically, selected variables differed from
those selected for the model with static time lags, because IPLS selects variables that minimise the MNRMSEP. The
two so obtained best possible models can be used for comparing dynamically and statically calculated time lags. Addi-
tionally, the separation of material flow into imagined top and bottom streams enables the analysis of board properties
dependent on the layer composition of boards.
The presented approach of dynamically calculating time lags within an industrial manufacturing process is not lim-
ited to the wood-based panels industry. As long as manufacturing processes contain similar elements of transporting,
processing and intermediately storing bulk material, the presented approach should be easily transferred to other applica-
tions or industries. This is mainly achieved by the applied modular simulation of the continuous flow of bulk material.

Conclusion
The dynamic calculation of time lags that typically occur within industrial processes enabled a more accurate alignment
of process parameters with corresponding test specimens for determining the IB of boards relative to the static calcula-
tion of time lags. Additionally, the variation of process parameters could be considered in real time by using the devel-
oped technique of dynamically calculating time lags introduced in this study. In particular, the consideration of actual
retention times within silos revealed that dynamically calculated time lags (137.2 min) were on average about five times
longer than statically calculated time lags (27.1 min). The dynamic calculation of retention times within silos will be
even more accurate if the occurring blending behaviour is considered. The incorporation of this kind of simulation will
be subject to future research.
By using dynamically calculated time lags, the predictability of PLSR models improved by 67% compared to PLSR
models using statically calculated time lags. Thus, final product strength properties can be predicted more accurately,
which improves the understanding of correlations within the manufacturing process. By using more reliable models,
costs of rejects could be lowered and material inputs could be used more efficiently. The approach presented in this
study is not limited to the wood-based panels industry. It is especially useful when dealing with highly complex manu-
facturing processes or processes with increased short-term variability of individual raw material or process parameters.

Acknowledgements
The authors gratefully acknowledge the financial support of the Competence Centre for Wood Composites and Wood Chemistry,
Wood K plus, Austria, and The University of Tennessee, Center for Renewable Carbon, Knoxville, TN. Special thanks go to Mrs.
Cornelia Riegler for proofreading.
6636 M. Riegler et al.

Disclosure statement
No potential conflict of interest was reported by the authors.

References

André, N., and T. M. Young. 2013. “Real-time Process Modeling of Particleboard Manufacture Using Variable Selection and
Regression Methods Ensemble.” European Journal of Wood and Wood Products 71 (3): 361–370.
André, N., H. W. Cho, S. H. Baek, M. K. Jeong, and T. M. Young. 2008. “Prediction of Internal Bond Strength in a Medium Density
Fiberboard Process Using Multivariate Statistical Methods and Variable Selection.” Wood Science and Technology 42 (7): 521–534.
Arnold, D., and K. Furmans. 2009. Materialfluss in logistiksystemen [Material Flow in Logistic Systems]. Berlin: Springer.
Astm D1037. 2006. Standard Test Methods for Evaluating Properties of Wood-base Fiber and Particle Panel Materials. West
Conshohocken: ASTM International.
Ayres, R. U., and L. W. Ayres. 2002. A Handbook of Industrial Ecology. Cheltenham: Elgar.
Bernardy, G., and B. Scherff. 1997. “Process Modelling Provides Online Quality Control and Process Optimization in Particleboard
and Fiberboard Production.” Holz als Roh- und Werkstoff 55 (2–4): 133–140.
Clapp Jr, N. E., T. M. Young, and F. M. Guess. 2008. “Predictive Modeling the Internal Bond of Medium Density Fiberboard Using
a Modified Principal Component Analysis.” Forest Products Journal 58 (4): 49–55.
Cook, D. F., and C. C. Chiu. 1997. “Predicting the Internal Bond Strength of Particleboard, Utilizing a Radial Basis Function Neural
Network.” Engineering Applications of Artificial Intelligence 10 (2): 171–177.
Dosta, M., S. Heinrich, and J. Werther. 2010. “Fluidized Bed Spray Granulation: Analysis of the System Behaviour by Means of
Dynamic Flowsheet Simulation.” Powder Technology 204 (1): 71–82.
Dunky, M., and P. Niemz. 2002. Holzwerkstoffe und Leime [Wood Composite Panels and Adhesives]. Berlin [u.a.]: Springer.
Esteban, L. G., F. G. Fernández, and P. De Palacios. 2011. “Prediction of Plywood Bonding Quality Using an Artificial Neural
Network.” Holzforschung 65 (2): 209–214.
Gross, D., and C. M. Harris. 1998. Fundamentals of Queueing Theory. New York: Wiley.
Hall, D. L. 1992. Mathematical Techniques in Multisensor Data Fusion. Boston, MA: Artech House.
Hasener, J. 2004. Statistische methoden der industriellen prozessmodellierung zur echtzeitqualitätskontrolle am beispiel einer kon-
tinuierlichen produktion von faserplatten [Statistical Methods of Industrial Process Modeling for Real-Time Quality Control
using the Example of a Continuous Production of Fiberboards]. Univ. Hamburg.
Kruse, K., H. Thömen, H. Maurer, A. Steffen, and R. León-Méndez. 1997. “Optimisation of Particleboard Production by Means of
Process Modelling.” Holz als Roh- und Werkstoff 55 (1): 17–24.
Law, A. M., and W. D. Kelton. 1991. Simulation Modeling and Analysis. New York: McGraw-Hill.
Maloney, T. M. 1993. Modern Particleboard & Dry-process Fiberboard Manufacturing. Madison, WI: Forest Products Society.
Maness, T. C., R. A. Kozak, and C. Staudhammer. 2003. “Applying Real-time Statistical Process Control to Manufacturing Processes
Exhibiting between and Within Part Size Variability in the Wood Products Industry.” Quality Engineering 16 (1): 113–125.
Nauman, E. B. 2008. Chemical Reactor Design, Optimization, and Scaleup. 2nd ed. Hoboken, NJ: Wiley.
Norgaard, L., A. Saudland, J. Wagner, J. P. Nielsen, L. Munck, and S. B. Engelsen. 2000. “Interval Partial Least-squares Regression
(iPLS): A Comparative Chemometric Study with an Example from Near-infrared Spectroscopy.” Applied Spectroscopy 54 (3):
413–419.
Reimers, C., J. Werther, and G. Gruhn. 2008. “Flowsheet Simulation of Solids Processes: Data Reconciliation and Adjustment of
Model Parameters.” Chemical Engineering and Processing: Process Intensification 47 (1): 138–158.
Riegler, M., B. Spangl, M. Weigl, R. Wimmer, and U. Müller. 2013. “Simulation of a Real-time Process Adaptation in the
Manufacture of High-density Fibreboards using Multivariate Regression Analysis and Feedforward Control.” Wood Science and
Technology 47: 1243–1259.
Thome, H. 2005. Zeitreihenanalyse: eine einführung für sozialwissenschaftler und historiker [Time Series Analysis: an Introduction
for Social Scientists and Historians]. München: Oldenbourg.
Werther, J., S. Heinrich, M. Dosta, and E. U. Hartge. 2011. “The Ultimate Goal of Modeling – Simulation of System and Plant
Performance.” Particuology 9 (4): 320–329.
Wise, B. M., N. B. Gallagher, R. BrO, J. M. Shaver, W. Windig, and R. S. Koch. 2012. Eigenvector Documentation Wiki [online].
Eigenvector Research, Inc. http://wiki.eigenvector.com
Wold, S., M. Sjöström, and L. Eriksson. 2001. “PLS-regression: A Basic Tool of Chemometrics.” Chemometrics and Intelligent
Laboratory Systems 58 (2): 109–130.
Young, T. M., P. M. Winistorfer, and S. Wang. 1999. “Multivariate Control Charts of mdf and osb Vertical Density Profile Attributes.”
Forest Products Journal 49 (5): 79–86.
Young, T. M., L. B. Shaffer, F. M. Guess, H. Bensmail, and R. V. Leon. 2008. “A Comparison of Multiple Linear Regression and
Quantile Regression for Modeling the Internal Bond of Medium Density Fiberboard.” Forest Products Journal 58 (4): 39–48.
Young, T. M., N. E. Clapp Jr, F. M. Guess, and C.-H. Chen. 2014. “Predicting Key Reliability Response with Limited Response
Data.” Quality Engineering 26 (2): 223–232.
Copyright of International Journal of Production Research is the property of Taylor & Francis
Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.

You might also like