Oct 14, 2015
5/24/2018 OSB Blender Optimization
1/149
DEVELOPMENT AND USE OF A DISCRETE ELEMENT MODEL FOR
SIMULATING THE BULK STRAND FLOW IN A ROTARY DRUM BLENDER
by
Graeme Dick
B.Sc., University of British Columbia, 2006
A THESIS SUBMITTTED IN PARTIAL FULFILMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
in
The Faculty of Graduate Studies
(Forestry)
UNIVERSITY OF BRITISH COLUMBIA(Vancouver)
August 2008
Graeme Dick, 2008
5/24/2018 OSB Blender Optimization
2/149
-ii-
ABSTRACT
In 2006 resin accounted for approximately 17% of the direct manufacturing costs for oriented
strand board (OSB). Because of their increased dependency on pMDI-resins, this percentage
is likely greater for oriented strand lumber (OSL) and laminated strand lumber (LSL). Thecost of PF- and pMDI-resins is expected to face upward pressure as the cost of their primary
constituents, natural gas and crude oil, continue to reach new highs. Therefore, there is
strong economic incentive to optimize the use of resin in the production of these three
products. This can be accomplished by addressing two key issues: reducing resin wastage
and optimizing resin distribution on the strands. Both issues will be overcome by focusing
on the blending process, where resin is applied to the strands.
This work focused on development and use of a discrete element model (DEM) for
simulating strand flow in a rotary drum blender using the EDEM software package. EDEM
required the input of three material and three interaction properties. Development of the
model involved creating the simulated environment (i.e. physical dimensions) and assigning
appropriate material and interaction properties given this environment and the assumptions
that were made. This was accomplished in two steps, completing baseline bench-top
experiments and a literature review to determine appropriate parameters and initial value
ranges for these properties, and then fine-tuning these values based on a validation process.
Using the validated model, an exploratory study was conducted to determine the effect of
four blender design and operating parameters (flight height, number of flights, blender
rotational speed, and blender fill level) on bulk strand flow. The results were analyzed with
regards to overall trends and by focusing on two perspectives, end users and blender
manufacturers. It was found that there was a strong relationship between these key
parameters and bulk strand flow. These results suggest that operating parameters of ablender, namely rotational speed and tilt angle, should be linked directly to the blender feed
rate to ensure an optimal blending environment is maintained. In addition, manufacturers of
blenders must take into consideration the range in final operating conditions when designing
and positioning flights.
5/24/2018 OSB Blender Optimization
3/149
-iii-
TABLE OF CONTENTS
Abstract....................................................................................................................................iiTable of Contents....................................................................................................................iiiList of Tables ........................................................................................................................... vList of Figures........................................................................................................................viiList of Abbreviations ............................................................................................................... x
Acknowledgements.................................................................................................................xi
Chapter 1: Introduction ....................................................................................................... 1
1.1 Rationale........................................................................................................................ 31.2 Objectives and structure of thesis.................................................................................. 7
Chapter 2: Literature Review.............................................................................................. 9
2.1 Rotary drum blending.................................................................................................... 92.2 Discrete element method ............................................................................................. 11
2.2.1 Applications for discrete element methods....................................................... 132.3 Friction ........................................................................................................................ 15
Chapter 3: Laboratory Determined Coefficient of Static Friction ................................ 18
3.1 Introduction ................................................................................................................. 183.2 Materials...................................................................................................................... 193.3 Procedure..................................................................................................................... 203.4 Results ......................................................................................................................... 243.5 Conclusions ................................................................................................................. 27
Chapter 4: Determination of Suitable Material and Interaction Properties for use
as Input Parameters in the RDBM ................................................................ 29
4.1 Introduction ................................................................................................................. 29
4.2 Procedure..................................................................................................................... 314.2.1 Screening design ............................................................................................... 314.2.2 Strand representation in EDEM........................................................................ 344.2.3 Characterization of resination potential ............................................................ 34
4.3 Validation process selection of appropriate input parameters.................................. 394.4 Results ......................................................................................................................... 49
4.4.1 Screening design material properties ............................................................. 494.4.2 Screening design interaction properties ......................................................... 504.4.3 Coefficient of rolling friction............................................................................ 514.4.4 Validation - coefficient of static friction........................................................... 52
4.5 Conclusions ................................................................................................................. 60
Chapter 5: Measuring the Effect of Rotary Drum Blender Design and Operating
Parameters on the Bulk Strand Flow using a Response Surface Design .... 62
5.1 Introduction ................................................................................................................. 625.2 Methodology ............................................................................................................... 625.3 Results and discussion................................................................................................. 66
5.3.1 Overall predictive trends................................................................................... 665.3.1.1 Skewness ............................................................................................. 675.3.1.2 Effect of an atomizer boom on the skewness results........................... 73
5/24/2018 OSB Blender Optimization
4/149
-iv-
5.3.1.3 Kurtosis ............................................................................................... 775.3.1.4 Effect of an atomizer boom on the kurtosis results ............................. 805.3.1.5 Average time a strand spent in the resination region .......................... 835.3.1.6 Effect of an atomizer boom on the average time a strand spent
in the resination region ........................................................................ 855.3.1.7 Discussion ........................................................................................... 85
5.3.2 Research applications........................................................................................ 865.3.2.1 Wood strand-based product manufacturers......................................... 875.3.2.2 Effect of the atomizer boom for wood strand-based product
manufacturers ...................................................................................... 895.3.2.3 Blender manufacturers ........................................................................ 915.3.2.4 Effect of the atomizer boom for blender manufacturers ..................... 945.3.2.5 Discussion ........................................................................................... 97
5.4 Conclusions ................................................................................................................. 98
Chapter 6: Summary and Future Work......................................................................... 100
6.1 Future Work .............................................................................................................. 101
Literature Cited ................................................................................................................. 103
APPENDIX A: Coefficient of Friction SAS Analysis and Results.................................... 108APPENDIX B: Mechanical Properties of UHMW and HDPE.......................................... 113APPENDIX C: Blender Drawing Provided by Coil Manufacturing.................................. 114APPENDIX D: Write-out Every Time Interval Calculation .............................................. 116APPENDIX E: VBA Macro for Sorting, Filtering and Analysing EDEM Data ............... 121APPENDIX F: Macro for Performing Analysis in Image Pro Plus................................... 130APPENDIX G: ANOVA Results for the Mechanical Properties of Aspen Strands .......... 132APPENDIX H: ANOVA Results for the Interaction Properties of Aspen Strands
and Polyethylene....................................................................................... 134
APPENDIX I: Results from the Student t-test for Shoulder and Toe Angles .................. 136APPENDIX J: Results from the Exploratory Study Without and With an
Atomizer Boom......................................................................................... 137
5/24/2018 OSB Blender Optimization
5/149
-v-
LIST OF TABLES
Table 1. OSB costs for benchmark north-central US mills from 2000 to 2006....................... 5Table 2. Pricing data for the constituents of PF- and pMDI-resin from 1999 to 2005............ 6Table 3. Strand combinations used for static friction coefficient tests. ................................. 21Table 4. Summary of test results for the coefficient of static friction between two wood
strands at 22C and 55% relative humidity ............................................................. 25
Table 5. Summary of test results for the coefficient of static friction between a woodstrand and HDPE at 22C and 55% relative humidity............................................. 26
Table 6. Regression parameters for determining the coefficient of static friction betweentwo wood strands at 22C and 55% relative humidity............................................. 27
Table 7. Regression parameters for determining the coefficient of static friction betweena wood strand and HDPE at 22C and 55% relative humidity. ............................... 27
Table 8. Required material and interaction properties for the RDBM. ................................. 29Table 9. Materials used and materials that may come in contact in the RDBM.................... 29Table 10. (Left)Aspen wood strand material properties and (right)interaction properties
simulation design.................................................................................................... 31Table 11. Factor levels for Quaking Aspen material properties............................................. 32
Table 12. Factor levels for interaction properties................................................................... 32Table 13. Fixed factor levels for the blender operation and design. ...................................... 33Table 14. Fixed factor levels for the liner and flights ............................................................ 33Table 15. Blender rotational speed and fill level combinations for laboratory video
recordings. .............................................................................................................. 41Table 16. Simulation settings. ................................................................................................ 33Table 17. Frame export settings used in Adobe Premiere Pro CS3. ...................................... 46Table 18. Factors and response variables for simulations investigating the impact of the
material properties.................................................................................................. 50Table 19. Factors and response variables for simulations investigating the impact of the
interaction properties.............................................................................................. 50
Table 20. Rolling and static friction coefficients for the simulations aimed at determininga suitable coefficient of rolling friction.................................................................. 51
Table 21. Pairs of static friction coefficients used to identify a suitable set of values........... 53Table 22. Student t-test for two sample means assuming unequal variance for the shoulder
and toe angles in Run 4 and the laboratory results................................................. 54Table 23. Student t-test for two sample means assuming unequal variance for the shoulder
and toe angles in Run 5 and the laboratory results................................................. 54Table 24. Student t-test for two sample means assuming unequal variance for the shoulder
and toe angles in Run 6 and the laboratory results................................................. 55Table 25. Summary of runs 4 to 6 and the laboratory taken shoulder and toe angle results.
Italicized values indicate those angles that are significantly different (= 0.05)
from the image results. ........................................................................................... 55Table 26. Summary of shoulder and toe angles obtained at 15.5 to 25.5 RPM with the
coefficients of static friction set at 0.14 and 0.07................................................... 56Table 27. Summary of material and interaction properties for use with EDEM.................... 61Table 28. Response surface design matrix. ............................................................................ 64Table 29. Response surface design factor levels .................................................................... 64Table 30. Factor levels used in the skewness and average time spent in resination region
analyses. ................................................................................................................. 67
5/24/2018 OSB Blender Optimization
6/149
-vi-
Table 31. List of effects for the skewness showing (left)the significant effects and (right)the significant effects as well as those that were included to maintain the modelhierarchy. ................................................................................................................ 68
Table 32. List of effects for the skewness when an atomizer boom is present, showing (left)the significant effects and (right)the significant effects as well as those that wereincluded to maintain the model hierarchy. ............................................................. 74
Table 33. List of effects for the kurtosis showing (left)the significant effects and (right)the significant effects as well as those that were included to maintain the modelhierarchy. ................................................................................................................ 78
Table 34. List of effects for the kurtosis when an atomizer boom is present, showing (left)the significant effects and (right)the significant effects as well as those thatwere included to maintain the model hierarchy. .................................................... 81
Table B1: Mechanical properties of UHMW ...................................................................... 113Table B2: Mechanical properties of HDPE ......................................................................... 113Table G1: ANOVA results for the impact the material properties have on the skewness
of the resulting histogram. .................................................................................. 132Table G2: ANOVA results for the impact the material properties have on the kurtosis of
the resulting histogram........................................................................................ 132Table G3: ANOVA results for the impact the material properties have on the count of
the resulting histogram........................................................................................ 132Table G4: ANOVA results for the impact the material properties have on the processing
time of the respective simulation. ....................................................................... 133Table H1: ANOVA results for the impact the interaction properties have on the skewness
of the resulting histogram. .................................................................................. 134Table H2: ANOVA results for the impact the interaction properties have on the kurtosis
of the resulting histogram. .................................................................................. 134Table H3: ANOVA results for the impact the interaction properties have on the count
of the resulting histogram. .................................................................................. 134Table H4: ANOVA results for the impact the interaction properties have on the
processing time of the respective simulation...................................................... 135Table I1: Student t-test for two sample means assuming unequal variance for the
shoulder and toe angles in Run 4 and the laboratory results ran at 15.5 RPM. .. 136Table I2: Student t-test for two sample means assuming unequal variance for the
shoulder and toe angles in Run 4 and the laboratory results ran at 25.5 RPM. .. 136Table J1: Results from exploratory study without an atomizer boom................................ 137Table J2: Results from exploratory study with an atomizer boom ..................................... 138
5/24/2018 OSB Blender Optimization
7/149
-vii-
LIST OF FIGURES
Figure 1. Canadian structural panels production from 1982 to 2005 ...................................... 1Figure 2. Size of the total North American framing lumber market and the potential
market size for engineered wood products............................................................... 4Figure 3. Flow chart showing the basic constituents that are used in the production of PF-
and pMDI-resin. The shaded constituents are included in Table 2 ......................... 6
Figure 4. Turners spinning disc blender showing a vertical spray pattern........................... 10Figure 5. Lignex spinning disc blender showing a diagonal spray pattern............................ 11Figure 6. Flow chart for operations performed in a typical DEM algorithm......................... 12Figure 7. Diagram showing the pair of forces that determines the friction torque................ 16Figure 8. Photograph of sliced aspen veneer strands, (a) primary surface and
(b) secondary surface.............................................................................................. 20Figure 9. Inclined plane jig with the various components indicated ..................................... 21Figure 10. Two sleds used for the inclined plane test, (left)sled equipped with dowels for .... larger contact pressures and (right)sled equipped with adhesive surface for
lower contact pressures. ........................................................................................ 22Figure 11. Photograph of the testing procedure for the coefficient of static friction of wood
strands using the inclined plane technique, showing a parallel parallelorientation. ............................................................................................................ 23
Figure 12. Schematic of inclined plane jig showing the measurement locations forEquation 5 ............................................................................................................. 24
Figure 13. Static coefficient of friction between two wood strands for increasing contactpressures and different strand sample orientations............................................... 25
Figure 14. Static coefficient of friction between a wood strand and HDPE for increasingcontact pressures and different strand and HDPE sample orientations. ............... 26
Figure 15. Schematic showing the representation of sticks using a series of six spheres(left)and a schematic showing the placement of a template over top of thesix spheres to aid in the visual analysis process(right). ....................................... 34
Figure 16. Blender schematic showing the resination region outlined in blue. ..................... 35Figure 17. Simulation and photographed examples of increasing skewness caused by
increasing rotational speeds from 15.5 RPM to 25.5 RPM .................................. 37Figure 18. Sample histogram with a respective skewness, kurtosis, and count of -0.3024,
-0.2794, and 22 453. ............................................................................................. 38Figure 19. Schematic showing the placement of the lights and camera/video camera
relative to the laboratory blender, with the axis indicated in blue. ....................... 40Figure 20. Photograph showing the placement of the lights and camera/video camera
relative to the laboratory blender. ......................................................................... 40Figure 21. Example of (left)a screen shot taken of an animated GIF illustrating the
simulation results and (right)a screen shot taken of the video footage taken
in the laboratory. ................................................................................................... 41Figure 22. Schematic showing the shoulder, , and toe, , angles for two points of
detachment. The 0oand 90
oreference angles are shown in blue. ......................... 42
Figure 23. Illustration showing the identification of the shoulder and toe angle from thelaboratory video footage. ...................................................................................... 44
Figure 24. Illustration showing the identification of the shoulder () and toe () anglefrom the simulation results using the streaming effect. ........................................ 45
Figure 25. Schematic of the x, y, z coordinate system relative to the blender....................... 46
5/24/2018 OSB Blender Optimization
8/149
-viii-
Figure 26. Screen shot taken in Image Pro Plus v6 showing the placement of the thick,line profile and the cooresponding grayscale values ............................................ 48
Figure 27. Example of simulation results overlaid on top of grayscale results...................... 49Figure 28. Skewness as a function of the coefficient of rolling friction ................................ 52Figure 29. Baseline grayscale results for the laboratory blender running empty................... 57Figure 30. Grayscale results for the blender running at 15.5 RPM and 1/8
thfull. ................. 58
Figure 31. Grayscale results for the blender running at 20.5 RPM and 1/8th
full. ................. 59Figure 32. Grayscale results for the blender running at 25.5 RPM and 1/8thfull. ................. 59Figure 33. Schematic of a blender fitted with an atomizer boom, shaded grey..................... 66Figure 34. Prediction profiles generated in SAS showing the relationship between the
skewness and the (top-left)number of flights, (top-right)flight height,(bottom-left)fill level, and (bottom-right)blender rotational speed..................... 68
Figure 35. Schematic showing the angle of repose, , for a pile of wood strands on ahorizontal surface.................................................................................................. 69
Figure 36. Simulation images showing the charge level per flight and the discharge patternwhen a relatively small number of flights are employed. The simulated blenderhas 4-6 inch flights and is rotating at 23.39 RPM and is 1/8thfull ....................... 70
Figure 37. Simulation images showing the charge level per flight and the discharge patternwhen a relatively large number of flights are employed. The simulated blenderhas 16-6 inch flights and is rotating at 23.39 RPM and is 1/8
thfull ..................... 71
Figure 38. Simulation image showing strands rolling in the corner of the drum, where thereare 8-4 inch flights and the blender is rotating at 18.71 RPM and is 1/4 full....... 71
Figure 39. (Left) Simulation image showing the dispersion of strands across relatively fewflights when the blender is rotating at 18.71 RPM and (right) across many flightswhen the blender is rotating at 28.07 RPM. In both cases the blender has 16-4inch flights and is 1/8
thfull. .................................................................................. 72
Figure 40. Prediction profiles generated in SAS showing the relationship between theskewness and the (top-left)number of flights, (top-right)flight height,(bottom-left)fill level, and (bottom-right)blender rotational speed when an
atomizer boom is included in the simulation ........................................................ 75Figure 41. Simulation images showing the dispersion of strands across the blender diameter
when there is (top-left)no atomizer boom and there are 2 inch flights,(top-right)no atomizer boom and there are 6 inch flights, (bottom-left)anatomizer and there are 2 inch flights, and (bottom-right)an atomizer boomand there are 6 inch flights. In all cases there were 16 flights and the blenderrotated at 23.39 RPM. ........................................................................................... 76
Figure 42. Simulation image showing strands as they become wedged between theatomizer boom and blender wall when operating at elevated fill levels,indicated by the dashed oval. In this case the blender is full and is equippedwith 8, 4 inch flights and is rotating at 28.07 RPM.............................................. 77
Figure 43. Prediction profiles generated in SAS showing the impact of the (left) numberof flights on the relationship between (right) the kurtosis and the flight height... 79
Figure 44. Prediction profiles generated in SAS showing the impact of the (right) flightheight on the relationship between (left) the kurtosis and the number of flights.. 79
Figure 45. Prediction profiles generated in SAS showing the impact of the (left) numberof flights on the relationship between (right) the kurtosis and the flight heightwhen an atomizer boom is present........................................................................ 82
5/24/2018 OSB Blender Optimization
9/149
-ix-
Figure 46. Prediction profiles generated in SAS showing the impact of the (right) flightheight on the relationship between (left) the kurtosis and the number of flightswhen an atomizer boom is present........................................................................ 82
Figure 47. Prediction profiles generated in SAS showing the relationship between theaverage time spent in the resination region and the (left)flight height and(right)blender rotational speed............................................................................. 83
Figure 48. Simulation images showing (top-left)the clustering of strands at relativelylow speeds with 2-inch flights, (top-right)the dispersion of strands at relativelyhigh speeds with 2-inch flights, (bottom-left)the clustering of strands atrelatively low speeds with 6-inch flights, (bottom-right)the dispersion ofstrands at relatively high speeds with 6-inch flights. .......................................... 84
Figure 49. Contour graphs for the skewness based on the fill level and blender rotationalspeed using 3, 4, 5, and 6-inch flights. The number of flights has been fixedat 14....................................................................................................................... 88
Figure 50. Contour graphs for the skewness based on the fill level and blender rotationalspeed using 3, 4, 5, and 6-inch flights when an atomizer boom is present. Thenumber of flights has been fixed at 14.................................................................. 90
Figure 51. Contour graphs based on number of flights and flight height. The rotationalspeed ranged from 23 to 28 RPM and the fill level was fixed at 25%. ................ 93
Figure 52. Contour graphs for the inclusion of an atomizer boom based on number offlights and flight height. The rotational speed ranged from 23 to 28 RPM andthe fill level was fixed at 25%............................................................................... 95
Figure 53. Simulation images showing the streaming of strands off of the atomizer boomat (a)18.71 RPM, (b)23.39 RPM, and (c)28.07 RPM. The simulated blenderswere each equipped with 8-4 inch flights and filled 1/4 full. The angles that thestrands stream off of the boom are approximately 14, 12, and 7 from verticalrespectively. .......................................................................................................... 96
Figure C1: Schematic of blender layout and atomizer spray patter. ................................... 114Figure D1: Schematic of an example where the write-out time interval is set too large..... 117
Figure D2: Schematic of the extreme case scenario where an object falls from the top ofthe blender, A, through the resination region, B to C, and collides with thebottom of the blender, D.................................................................................... 118
Figure D3: Location of the write-out every time intervals within the resinating region,relative to the top of the blender, when the first interval is located marginallyless than one full twfrom the top of the region. .............................................. 119
Figure D4: Location of the write-out every time intervals within the resinating region,relative to the top of the blender, when the last interval is located marginallyless than one full twfrom the bottom of the region......................................... 119
Figure D5: Location of the write-out every time intervals within the resinating region,relative to the top of the blender, when the first interval is located marginally
below the top of the region................................................................................ 120Figure D6: Location of the write-out every time intervals within the resinating region,
relative to the top of the blender, when the last interval is located marginallyabove the bottom of the region.......................................................................... 120
5/24/2018 OSB Blender Optimization
10/149
-x-
LIST OF ABBREVIATIONS
ANOVA Analysis of variance
BBF Billion board feet
BSF Billion square feet 3/8 inch basis
COV Coefficient of variation
DEM Discrete element modeling
EWP Engineered wood products
GIF Graphics interchange format
HDPE High density polyethylene
IPP Image pro plus
LSL Laminated strand lumber
OSB Oriented strand board
OSL Oriented strand lumber
PE Polyethylene
PF Phenol formaldehyde
pMDI Polymeric diphenyl methane diisocyanate
RPM Revolutions per minute
RSM Response surface methodology
RDBM Rotary drum blending modelSF Square feet
UHMW Ultra high molecular weight polyethylene
VBA Visual basic for applications
5/24/2018 OSB Blender Optimization
11/149
-xi-
ACKNOWLEDGEMENTS
I would like to extend my sincere gratitude to those who have helped ensure that this project
was a success. To my committee members: Drs. Gregory Smith, Paul McFarlane, and Erik
Eberhardt, your guidance throughout this process has certainly been appreciated.
To my colleagues: Jo Chau, Emmanuel Sackey, Solace Sam-Brew, Dr. Kate Semple, and
Chao Zhang, your assistance and support have made this experience enjoyable.
To my family and friends, your ongoing support and devotion have helped me reach this
point. And to my wife, Sara, you have kept me grounded throughout this experience by
simply listening to my challenges and always being there.
A special thank you goes to Weyerhaeuser Canada for their financial support and guidance
throughout this project, and to the Natural Sciences and Engineering Research Council of
Canada for their financial support.
5/24/2018 OSB Blender Optimization
12/149
-1-
CHAPTER 1
INTRODUCTION
Beginning in the mid-1970s, structural products composed of reconstituted wood strands
have increasingly become a major component of Canadas forest products industry. This
transition began with the advent of waferboard and quickly progressed to oriented strand
board (OSB), a direct substitute for plywood in the construction market (Figure 1). In 2006,
oriented strand board accounted for approximately 63% of all structural panel production in
Canada (Louisiana-Pacific Corporation 2008; Spelter et al. 2006). Laminated strand lumber
(LSL) and oriented strand lumber (OSL) were subsequently developed to compete with solid
sawn lumber in the same market. In todays North American residential construction market,
these wood strand-based products can be found everywhere from the sheathing on exterior
walls, to the headers used to span garage door openings, and to the specialty studs used
behind kitchen cabinets.
Plywood
OSB
0
2,000
4,000
6,000
8,000
10,000
12,000
14,000
16,000
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Year
Millionsf-3/8"Basis
Figure 1. Canadian structural panels production (million SF - 3/8" basis) from 1982 to 2005(data from: International Wood Markets Group 2006).
5/24/2018 OSB Blender Optimization
13/149
-2-
The North American production capacity for OSB is expected to continue increasing into the
future. In fact, if all of the OSB projects tabled in 2005 were to go ahead as planned,
capacity would balloon from 23.7 million m3(27.5 BSF) in 2005 to 31.9 million m3(37 BSF)
by 2010 (International Wood Markets Group 2006). However, because of recent slowdowns
in the US housing market, largely driven by the sub-prime mortgage crisis, and an
accompanying over-supply of OSB, these previous projections have been stifled. According
to a report by Dixon (2008), North American OSB capacity had already reached 26.2 million
m3by the end of 2007; however, any additional capacity that was planned to come on stream
by 2010 had been postponed indefinitely.
The increased production of OSB is also being experienced beyond North American borders.
While most of the additional capacity that was expected to come on stream in the near future
is located in North America, Europes OSB industry has also been expanding, albeit at a
considerably lower volume (International Wood Markets Group 2006). This is largely
because of the relatively slow adoption of OSB into European building codes (World Forest
Institute 2007). This progress has been further hindered by the effect of North American
OSB producers dumping excess supply in Europe, discouraging the development of new
domestic facilities (Higgs 2008). The total European capacity reached 3.9 million m3by the
end of 2007. By 2009 an additional 1.4 million m3is expected to come on stream.
As mentioned, there are three products that fall beneath the umbrella of wood strand-based
products: OSB, LSL, and most recently OSL. During the manufacture of these products a
mat consisting of a large number of strands is consolidated under heat and pressure to form a
single entity, or billet. In order for the consolidation to be effective, resin must be employed
to hold the final product together. The application of resin onto the strands is perhaps the
least studied and understood aspect of the manufacturing process; however, it has one of the
most significant effects on the strength and durability of the final product and in 2006
accounted for nearly 17% of the direct manufacturing costs (Spelter et al. 2006).
The application of resin onto the strands begins when the strands are fed from dry strand bins
and deposited in the blender. Blenders are between 8 and 11 feet in diameter and extend 20
to 35 feet in length. As the drum rotates at typically between 8 and 22 RPM the strands
tumble along its length and become resinated (Smith 2005).
5/24/2018 OSB Blender Optimization
14/149
-3-
The strands are resinated using either a series of spinning disc atomizers mounted along the
drums axis of rotation for liquid resin, or using a conveyor metering system for powdered
resin. Spinning disc atomizers are the predominant system employed in blenders
commissioned since the early-1990s. The objective of blending is to achieve a uniform resin
distribution on both sides of all strands. The ability of the process to effectively resinate the
strands is dependent on the overall blender design, such as the drum diameter, atomizer
locations, and flight design; as well as on the operational environment, such as the rotational
speed, fill level, and tilt angle of the blender (Coil 2007b; Coil 2008; Maloney and Huffaker
1984; Smith 2005; Smith 2006).
1.1 Rationale
The respective market share of structural products composed of reconstituted wood strands is
forecast to continue increasing. This is particularly true for OSB where its share of the
structural panel demand in North America has increased from approximately 35% in 1995 to
63% in 2008. OSB market share is forecast to further increase to approximately 72% by
2012 (Louisiana-Pacific Corporation 2008).
In addition to OSBs market share growth however, there is great potential for LSL and OSL
to increase their share of the framing lumber market. LSL and OSL are both members of thefamily of products referred to as engineered wood products (EWP). Currently, EWP only
capture approximately 30% of the potential 12 BBF North American framing lumber market,
with the sub-sector of wood strand-based products only accounting for 5% (Figure 2)
(Louisiana-Pacific Corporation 2008).
5/24/2018 OSB Blender Optimization
15/149
-4-
Figure 2. Size of the total North American framing lumber market and the potential marketsize for engineered wood products (adapted from: Louisiana-Pacific Corporation 2008).
Although the five year period between 2001 and 2006 saw the growth of the wood strand-
based product sector outpace the growth of the overall construction market (by 11%
compared to 2% (Louisiana-Pacific Corporation 2008), continued growth will be largely
dependant on the relative manufacturing costs. As an example, in 2004 the average total
manufacturing cost for structural lumber, OSL, and LSL was similar. Lumber was 188
US$/m3,while OSL and LSL were approximately 180 US$/m3(International Wood Markets
Group 2006; Spelter et al. 2006). These figures assume that OSL and LSL have a similar
cost structure as OSB. In reality the cost of OSL and LSL will be marginally greater than
OSB because of the type of resin employed, product density, and wood utilization.
Subsequently, the costs are likely nearer, or even past those of structural lumber.
In recent years the total manufacturing cost of wood strand-based products has faced
increased pressure. In 2006 the average cost reached 201 US$/m3of OSB (Table 1) (Spelter
et al. 2006). Much of this increase was caused by escalating wood costs, affecting solid
lumber and EWP alike. In addition to wood costs however, wood strand-based products have
experienced escalating resin costs, increasing by 61% between 2000 and 2006 (Table 1)
(Spelter et al. 2006). Increasing resin costs have created a need to optimize and subsequently
reduce the amount of resin employed in the manufacture of these products in order to remain
LVL & I-Joists25%
Applicationssuitable for EWP
substitution
50%
Lumber
70%
OSL & LSL5%
Total North AmericaFraming Lumber
(~24 BBF)
Current & PotentialEWP Market(~12 BBF)
Potential MarketGrowth for EWP
5/24/2018 OSB Blender Optimization
16/149
-5-
cost competitive in the structural panel and framing markets. Optimization is only possible
however if there exists a thorough understanding of the process in which resin is applied to
the strands in a rotary drum blender.
Table 1. OSB costs for benchmark north-central US mills from 2000 to 2006 (data from:
Spelter et al. 2006).Cost (US$/m
3)
2000 2001 2002 2003 2004 2005 2006
Direct costs
Wood 56 54 53 60 67 85 82Labor 20 20 20 21 21 22 22Resin 18 19 19 26 27 32 29Wax 6 6 6 7 7 8 7Energy 11 13 12 15 17 19 19Supplies 14 14 14 15 15 15 15
Total direct 125 125 124 144 154 181 175
Fixed costsGeneral 6 6 6 6 6 6 6Depreciation 23 21 20 21 20 20 20
Total fixed 29 27 26 27 26 26 26
Total costs 154 153 150 171 180 207 201
There are only a few studies on the operation of a rotary drum blender in the literature with
these dating from the mid-1980s (Beattie 1984; Coil and Kasper 1984; Lin 1984). More
recently, Smith (2005) examined the modes of tumbling in a full-sized rotary drum blender.
All of these studies were focused on OSB. To date there have not been any published studies
on the blending of OSL and LSL. The blending of these products differs from OSB in
several important aspects. First, OSL and LSL strands exceed 6 inches in length, while OSB
strands rarely exceed 5 inches. Second, LSL operations only employ polymeric diphenyl
methane diisocyanate (pMDI) resin, compared with OSB operations that typically use a
combination of phenol formaldehyde (PF) and pMDI resins.
During blending approximately 3.5% resin based on the oven dry weight of furnish is added.The precise amount depends on the operation, resin type, and product grade (Spelter et al.
2006). As indicated, resin costs are a major materials cost in the production of wood strand-
based products. Because PF- and pMDI-resins are derived from crude-oil and natural gas
(Table 2 and Figure 3), it is very likely that resin costs will remain high or even increase over
the next five years. Significant resin savings may be possible through blending optimization.
5/24/2018 OSB Blender Optimization
17/149
-6-
Due to the large volume of resin used in the manufacture of strand technology products, even
small decreases in resin consumption will lead to significant savings.
Table 2. Pricing data for the constituents of PF- and pMDI-resin from 1999 to 2005 (datafrom: Winchester 2005).
MethanolUS$/USG
UreaUS$/ton
PhenolUS$/lb
Crude OilUS$/Barrel
Natural GasUS$/MMBTU
1999 0.26 0.40 100 140 0.25 0.36 17 3.752001 0.35 0.80 140 290 0.30 0.40 24 5.242003 0.75 1.00 160 230 0.40 0.45 26 5.812005 0.90 0.95 270 - 300 0.50 0.70 60+ 9.80
Figure 3. Flow chart showing the basic constituents that are used in the production of PF-
and pMDI-resin. The shaded constituents are included in Table 2 (adapted from: Winchester2005).
The design and operation of rotary drum blenders has remained virtually unchanged since the
mid-1980s when the long-retention time, Mainland Manufacturing blenders emerged as the
blender of choice amongst the wood strand-based product industry. Although Mainland
Manufacturing has subsequently been bought out by Coil Manufacturing, very little has been
changed with regards to the operation and design of the blenders. In general, as the capacity
demands have increased over the years with new, sophisticated operations, the blender
dimensions have been scaled up. In the 1950s and 60s, rotary drum blenders were 4 to 5 feet
in diameter and 20 feet long (Coil 2002; Watkins 1981). More recently, blenders can be up
to 11 feet in diameter and 45 feet long (Coil 2007b; Coil and Kasper 1984; Smith 2005).
While this process of scaling the blenders has been widely accepted throughout the industry,
as previously mentioned resin costs have placed increased pressure on the end users of rotary
Phenol
PF
Formaldeh de
Methanol
Natural Gas
MDI
Phenol
Formaldeh de
Methanol
Crude Oil
Benzene
5/24/2018 OSB Blender Optimization
18/149
-7-
drum blenders to optimize the process. The primary challenge hindering progress is the lack
of research into the inner dynamics of the blending process and the inability to quantify and
model the bulk strand flow.
In the past, blender design and operating parameters have been selected based on visual
inspection and past experience. Although these empirical approaches have resulted in a
blender design that fits with the spectrum of end-users needs and a reasonable understanding
of the impact design and operating parameters have on the strand flow, a quantitative,
systematic approach is necessary to further enhance the process.
1.2 Objectives and structure of thesis
This study seeks to understand the blending process by developing and evaluating a
quantitative discrete element model (DEM) of the blending process.
Chapter 2: Literature review on discrete element modeling and friction. A review of static,
kinetic, and rolling friction will facilitate the selection of initial coefficients as
input parameters in the DEM. Additionally, this review will help to understand
how these coefficients should change to achieve better correlation between the
simulated and laboratory results.
Chapter 3: Determination of static friction coefficients between Aspen wood strands andbetween an Aspen wood strand and high density polyethylene. The flights and
the inside liner of the blender are constructed of either high density polyethylene
or ultra high molecular weight polyethylene. For the purpose of this project, and
because the specifications of these two materials do not differ considerably, it
will be assumed that the liner and flights are both constructed of high density
polyethylene. The friction values will be used as a starting point within the
EDEM software package and may be adjusted accordingly during a subsequent
study.
Chapter 4: Calibration of the rotary drum blending model (RDBM) with experiments
conducted in the 6 foot laboratory blender. This process will be completed by
adjusting various material and interaction properties in the model.
Chapter 5: Completion of an exploratory study aimed at determining the impact several
blender design and operating parameters have on the bulk strand flow within a
5/24/2018 OSB Blender Optimization
19/149
-8-
simulated 6 foot blender. This study will provide a profound understanding of
the impact changes made to the blending environment will have on the strand
flow through a blender.
Chapter 6: Evaluation of potential future work. As the software and computing technology
advances many of the limitations that were present during this project will
become less significant. The most logical progression will be the scaling of the
simulated blender up to a full size industrial blender. This will enable research
to be completed that focuses on the impact blender tilt angle has on the residence
time of strands.
5/24/2018 OSB Blender Optimization
20/149
-9-
CHAPTER 2
LITERATURE REVIEW
2.1 Rotary drum blending
Rotary drum blending has been the method used for coating wood strands with resin and wax
for the manufacture of oriented strand board since this products industrial emergence in the
late-1970s (Moeltner 1980). In fact, this blending technique was also employed for OSBs
predecessor, waferboard, since its emergence in 1955 (Gunn 1972). Although the resin was
applied exclusively in a powdered form until the mid- to late-1970s, the fundamental process
was similar to todays blenders that tend to employ liquid resins.
During the blending process strands enter from one end of the blender and are then lifted by a
series of flights, which extend from the inside of the drums circumference. The strands
eventually fall from the flights and migrate along the length of the blender. This process
repeats until the strands are discharged from the opposite end of the blender. The rate of
migration along the blender length is a function of the blender tilt angle. As reported by
Smith (Smith 2005), the strands move forward the most while they are in freefall. As the
strands move along blender length they are coated with wax and resin.
The manner in which powdered and liquid resin adhere to and coat the strands is
considerably different. While powdered resin adheres to the strands via the wax droplets that
are applied at the onset of the blending process, liquid resin adheres directly to the strands in
droplet form (Coil 2002). The transition towards liquid resin was largely driven by the
relative cost advantage of liquid resin, the health concerns caused by the dust from powdered
resin, and the relatively high wax content required to improve the affinity of the powdered
resin to the wood surface (Chiu and Scott 1981; Maloney and Huffaker 1984).
Early rotary drum blenders were 4 to 5 feet in diameter and 20 to 25 feet long (Coil 2002;
Watkins 1981). By the early 1980s it was widely accepted that blender diameters of 8 feet
and greater were required to ensure adequate resin coverage on the strands and/or wafers and
to meet capacity requirements (Beattie 1981). At around this same time liquid resin
5/24/2018 OSB Blender Optimization
21/149
-10-
increased in popularity and a system was developed and adopted across the industry for
metering liquid resin into the blenders and distributing it onto the strands. This system
involved using a series of spinning disc atomizers (Beattie 1984; Coil and Kasper 1984; Lin
1984). After several variations in the positioning of these atomizers within the blenders
(Figures 4 and 5), a design was eventually accepted whereby the atomizers were mounted
along a stationary shaft running down the length of the blender. The atomizers were
positioned to disperse resin in a horizontal plane. Except for the plane of the spray pattern
and the stationary shaft or boom, this design was similar to Turners design (Figure 4).
Figure 4. Turners spinning disc blender showing a vertical spray pattern (adapted from:Beattie 1984).
Resin spray
region
Spinningdisc Rotating
shaft
5/24/2018 OSB Blender Optimization
22/149
-11-
Figure 5. Lignex spinning disc blender showing a diagonal spray pattern (adapted from:Beattie 1984).
Modern blenders are provided nearly exclusively by Coil Manufacturing Limited of Surrey,
British Columbia. Their blenders range in size from 8 feet in diameter and 20 feet long up to
11 feet in diameter and 45 feet long; however, the 11 foot diameter blenders are most
common in newer operations requiring relatively high capacity (Coil 2007b; Smith 2005).
These blenders operate with strand volumes ranging from 25% to 50% of the blender volume
and with a tilt angle of approximately 3. Blenders revolve at between 8 and 22 RPM
depending on the blender diameter, number of flights, flight height, and resin type (Coil
2008). As a general rule, liquid resins require higher speeds than powdered resins and as the
diameter increases the speed decreases (Coil 2008).
2.2 Discrete element method
Discrete element methods (DEM), or distinct element methods as they are also known
(Cundall 1989), are a family of numerical techniques suitable for modeling the movementand interaction of rigid or deformable bodies, particles, or arbitrary shapes that have been
subjected to external stresses or forces (Bicanic 2004). As reported by Mustoe and Miyata
(2001), most of these methods are based on cylindrical- or spherical-shaped particles because
of the inherent ease in detecting contacts between particles. In recent years there has been an
increased number of DEMs based on noncircular-shaped bodies, such as polygonal bodies,
for specific applications. The vast majority of the commercially available software packages
Resin sprayregion Spinning
disc
Resin
feed
5/24/2018 OSB Blender Optimization
23/149
-12-
however, still rely on cylindrical- or spherical-shaped particles for 2D and 3D modeling
respectively. These particles may be clustered and/or overlapped rigidly or elastically to
form different shaped bodies (Collop et al. 2004; Mustoe 2001).
DEMs are based on Newtons Second Law of Motion (Bertrand et al. 2005; Serway 2000):
itotalii Fam ,= [1]
or, itotali
i Fdt
xdm ,2
2
= [2]
where:miis the mass of particle i,aiis the acceleration of particle i,xiis the position of particle i, and
Fiis the total force acting on particle i.
This equation is used to calculate the total force that acts on a particle due to a collision and
is subsequently integrated to find the respective particles new velocity and distance of travel
(Bertrand et al. 2005). During a simulation, the location of all particles is tracked at a
specified time interval. When a collision between particles is detected Newtons Second
Law of Motion is applied to determine each particles resulting position and velocity. Figure
6 shows the steps of a typical DEM algorithm.
Figure 6. Flow chart for operations performed in a typical DEM algorithm (Schafer et al.2001).
Calculate force incrementcaused by each contact
between particles
Calculate velocity andposition increments caused
by forces
Find which particles havecome into contact
t = t + t
5/24/2018 OSB Blender Optimization
24/149
-13-
Although the basis of the DEM approach is relatively straight forward, the computational
requirements quickly become overwhelming when more than a few particles are present.
Because DEMs are routinely used for simulating a large number of particles, the
effectiveness of the method is largely dependent on the ability of the model algorithms to
detect particle contacts quickly and efficiently (Bicanic 2004).
There are several DEMs that have been developed for describing how the particles behave
when they come into contact with each other. A typical DEM has the following features
(Cundall 1989; Bertrand et al. 2005; Mustoe 2001):
1. They allow finite displacements and rotations of discrete bodies, including complete
detachment, and
2. They recognize new contacts automatically as the calculation progresses.
Two of the more commonly applied models include the linear spring-dashpot model and the
Hertz Mindlin model. As Bertrand (2005) described, the principal difference between these
two models is that the linear spring-dashpot model considers any particle contact to lead to
inelastic deformation, while models based on Hertz theory considers this contact to lead to
elastic deformation. There is no consensus on what model is best; however, DEM solutions
(2008) report that the linear spring model is simpler because it requires less computationaloverhead. For EDEM, the selected software package for this research project, the Hertz
Mindlin model is the default model because of its accurate and efficient force calculation
(DEM Solutions 2008). This model was also used for the duration of this project.
Ultimately, the choice of model will depend on the environment being simulated and the
ability to validate the results. For additional information concerning the choice of models,
Bertrand (2005) provides a reasonable explanation of several of the more commonly
employed models. Additionally, Cundall and Strack (1979) and DEM Solutions (2008)
provide information on the model algorithms.
2.2.1Applications for discrete element methods
The DEM was first pioneered by Cundall (1971) for problems involving rock mechanics.
Since the early-1970s this method has branched out and adapted for use in a wide range of
engineering applications. Mining has perhaps benefited most from DEMs where they have
5/24/2018 OSB Blender Optimization
25/149
-14-
been shown to be particularly effective at analyzing granular material flow, power draw, and
liner wear in semi-autogenous grinding mills (Cleary 1998; Cleary 2006; Djordjevic et al.
2004; McIvor 1983; Mishra and Rajamani 1992; Powell 1991). In addition to mining, other
industries that have benefited include: pharmaceutical, chemical, agricultural, advanced
materials, and food (Bertrand et al. 2005).
An area that has gained recent attention is the modeling of granular material mixing. This
topic covers a variety of industries, but its significance is seen most prominently in the
pharmaceutical manufacturing arena. As Bertrand (2005) reported, even slight changes to
ingredient properties or process operating conditions can have significant implications on the
quality of a drug and/or resulting health effects. Consequently, pharmaceutical companies
are reluctant to make process changes based on DEM results alone and still rely heavily on
process monitoring to ensure quality (Bertrand et al. 2005).
Despite the widespread use of DEMs in various engineering applications, it has never been
used specifically for modeling the rotary drum blending of wood strand-based particles;
however, its successful use in semi-autogenous grinding (SAG) and other rotating drum type
processes suggests that it is possible (Kaneko et al. 2000; Moakher et al. 2000; Stewart et al.
2001). In this process wood strands are deposited inside a rotating drum at the front end.
The strands are then lifted by a series of flights and cascade and tumble along the drumlength. The dynamics of the process are similar to those encountered in a SAG mill;
however, the process objectives more closely resemble those of pill coating in the
pharmaceutical industry (Thibault 2008).
In a rotary drum blender as the strands migrate along the drum length resin is applied in
either liquid or powdered form. The objective is to maximize the resin deposition on the
strands and the distribution of resin amongst the strands, while minimizing strand breakage.
In a SAG mill the aim is to breakdown and grind the rocks. As mentioned, the objectives of
resination are therefore more closely related to those of pill coating. In pill coating the pills
tumble in a drum while a coating is sprayed onto them. Although the primary objective is to
coat the pills, the pills must also remain intact in order to avoid contamination (Thibault
2008). Because DEMs, and in particular the EDEM software package, have been used for
5/24/2018 OSB Blender Optimization
26/149
-15-
modeling both processes it is believed that this is a suitable method for modeling the rotary
drum blending process as well.
For this project a DEM will be used for simulating the trajectory and distribution of strands
as the blender revolves. The primary challenges associated with its use are the relatively
high slenderness ratio and thinness of the wood strands and the large quantity of strands in
the process. Simplifications and assumptions will be required to assemble a model that can
simulate the process with reasonable accuracy and within a reasonable time span.
2.3 Friction
Friction forces are a critical phenomenon when performing discrete element modeling. For
objects that slide relative to each other the key friction properties are static and kinetic
friction (Serway 2000). The only difference between the equations used for determining
these two forms of friction is the relevant coefficient of friction, . This equation is known
as Amontons Law (Equation 3).
iNf FF = [3]
where:iis either static kinetic friction, and
FNis the normal force.
If the shape of an object permits rolling to occur, such as a sphere, then the resistance to
rolling manifests as a torque that opposes the direction of rolling, Tf(Equation 4). Rolling
friction is caused by the deformation of either the rolling sphere/cylinder or the plane (Figure
7).
RNf FT = [4]
where:FNis the normal force, andRis the rolling coefficient.
5/24/2018 OSB Blender Optimization
27/149
-16-
Figure 7. Diagram showing the pair of forces that determines the friction torque, whereFNisthe reaction force acting on the object by the plane andFgis the normal component of theobjects weight. The coefficient of rolling friction is the arm of the pair of forces (Domenechet al. 1987).
According to Amontons Law, knowledge of the coefficient of friction is vital when
examining the interaction between objects. Because this coefficient depends on the
interaction between surfaces of different objects, it is most accurately considered to be a
system property rather than a material property. This is particularly relevant for this research
as true coefficients of frictions were not known. Instead coefficients were chosen based on
the resemblance of the model system to the actual observed systems.
For modeling the rotary drum blending process for wood strands there are broadly three
systems of objects that must be considered: strands and flights, strands and drum liner, and
the interaction between strands themselves. In light of the limited published information
pertaining to the static coefficient of friction values for the aforementioned systems, a series
of tests aimed at determining the respective values for the particular materials used in the
laboratory was conducted. Because it was anticipated that during the modeling stage the
drum liner and flights would be grouped as one material type, the strand and drum liner
interaction was dropped and instead replaced with the strand and flight interaction properties.
Classic theoretical research related to static and kinetic friction has shown that frictional
coefficients are independent of surface area and contact pressure. However, more recently
Bejo et al (2000) found that these generalizations are not necessarily true if at least one of
the [objects] in the system is wood or a wood-based composite. As a result, the initial study
Lower rolling friction Higher rolling friction
Fg F
F FN
5/24/2018 OSB Blender Optimization
28/149
-17-
of this project focused predominately on determining the relationship between the range of
contact pressures that may be encountered during blending and the friction coefficient.
5/24/2018 OSB Blender Optimization
29/149
-18-
CHAPTER 3
LABORATORY DETERMINED COEFFICIENT OF STATIC FRICTION
3.1 Introduction
Friction coefficients are perhaps the most important material interaction property to consider
when developing a rotary drum blending model based on the discrete element method
because of the significant impact they have on particle dynamics (DEM Solutions 2008). As
described in section 2.3, assigning the respective friction coefficients is complicated by the
fact that the friction coefficients within systems involving at least one wood substrate are
dependent on the contact pressure. The selected software package for this research, EDEM,
assumes constant coefficients regardless of the contact pressure, as is widely accepted for
most systems of materials. In addition, most of the published friction coefficient values for
systems involving wood are based on clear wood blocks, rather than strands. Wood strands
tend to be less smooth and relatively flexible, generally resulting in higher coefficients. As a
result, a series of experiments were conducted in the laboratory to determine the impact of
contact pressure and wood grain orientation on the coefficient of friction for wood on wood
and wood on polyethylene (PE) systems of materials. Collectively, these experiments will
investigate all of the material interactions that will occur during the simulations: strand strand, strand flight, and strand blender liner.
It was hypothesized that the coefficient of friction would increase with decreasing contact
pressure and that the coefficient of friction would increase from parallel parallel to parallel
perpendicular and to perpendicular perpendicular grain orientation. These results will
be used as a starting point for the initial development of the RDBM. Subsequently, blending
experiments and simulations will be compared to adjust these values until there is close
correspondence in the strand dynamics.
Objectives:
1. To determine the relationship between contact pressure and coefficient of friction for
wood wood and wood PE systems of materials,
5/24/2018 OSB Blender Optimization
30/149
-19-
2. To determine the relationship between grain orientation and coefficient of friction for
wood wood and wood PE systems of materials, and
3. To determine the ratio between the coefficients of friction for wood wood and wood
PE systems of materials.
3.2 Materials
A total of 40 sliced veneer, aspen wood strands were randomly selected from a 10.1 kg bag
of strands. The strands had been previously cut to approximately 12 inches long by 1
inches wide and 0.030 inches thick. Aspen strands were used in this case because it
represents the predominant species used in the manufacturing of wood strand-based products
in Canada (Industry Canada 2007). There were two requirements for the selected strands.
First, the strands had to have an area that was at least 9 inches by 1 inches void of any
splits, and second the strands could not exhibit excessive warp. Either of these flaws could
impact the experiment.
The selected strands were divided into two sets of twenty, one to be used as the primary
surface and one to be used as the secondary surface. The primary surface strands were
trimmed to 9 inches by 1 inches using a guillotine paper cutter, removing any splits or
defects. The edges of the strands were then lightly sanded using 220 grit sandpaper to
remove any burrs that might otherwise affect the test results. The secondary surface strandswere prepared in a similar manner; however, they were trimmed to 8 inches by 1 inch so that
they would easily lay flat atop the primary strands without their edges contacting. If the
edges were to come in contact the concern was that any remaining burrs may mechanically
interlock, resulting in a confounded reading of the static friction coefficient. These relatively
large strands were used for this experiment because it provided adequate room for weights to
be added to the surfaces, as described in Section 3.3. Ultimately the surface area does not
impact the friction coefficient so this was assumed to be a reasonable simplification for the
test procedure (Serway 2000).
The strands in each set were labeled 1 through 20 with the appropriate suffix added: a for
the primary surface strands and b for the secondary surface strands (Figure 8). Each set of
strands was then lightly clamped with a protective wood block on either face. Clamping
helped to prevent the strands from warping before testing.
5/24/2018 OSB Blender Optimization
31/149
-20-
Figure 8. Photograph of sliced aspen veneer strands, (a) primary surface and (b) secondarysurface.
In addition to the wood samples, two HDPE specimens were also prepared. The samples
were removed from an extra 5 inches T-flight for the 6 foot by 3-foot Coil laboratory
blender. The flight was trimmed into two specimens measuring 9 inches by 4inches and
8 inches by 1 inches. The edges of the samples were also sanded using 220 grit
sandpaper to remove any burrs and then washed in warm water and left to air dry.
3.3 Procedure
The experimental procedure is based on the inclined plane technique (American Standardsfor Testing and Materials 2002a; American Standards for Testing and Materials 2002b; Bejo
et al. 2000). This method was selected because of its relative simplicity and the shape of the
test specimens. A photograph of the testing apparatus is shown in Figure 9.
a b
1-inch
5/24/2018 OSB Blender Optimization
32/149
-21-
Figure 9. Inclined plane jig with the various components indicated ((1)scissor lift, (2)measuring guides, (3)platform with stop block, (4) raisedprimary surface platform, (5)bullseye level, (6)base, and (7)leveling glides).
The coefficient of static friction for the system involving two wood strands was determined
using five combinations of primary and secondary surface strands. These combinations were
generated using a random number generator in Microsoft Excel (Table 3).
Table 3. Strand combinations used for static friction coefficient tests.
Pair Primary surfacestrand
Secondary surfacestrand
1 9 182 5 173 14 184 19 95 8 11
Because of the orthotropic nature of wood, each combination was tested for all three
combinations of grain orientation. In addition, seven contact pressures were used to study
the impact contact pressure has on the coefficient of static friction. The included orientations
and target contact pressures are listed below:
1. Strand orientations (secondary on primary): parallel - parallel, perpendicular
perpendicular, and perpendicular - parallel.
2. Target contact pressures: 23 Pa, 47 Pa, 94 Pa, 188 Pa, 375 Pa, 750 Pa, and 1500 Pa.
2
1
3
6
7
7
4
2
5
5/24/2018 OSB Blender Optimization
33/149
-22-
For each orientation the primary surface strand was attached to the raised primary platform
using double-sided tape. The platform was then placed on the inclined plane in contact with
the stop block.
The secondary surface strand was attached to one of two sled designs (Figure 10) also using
double-sided tape. For the first two strand orientations the sled equipped with dowel
extensions was used for contact pressures greater than, and including, 188 Pa. Weights were
hung from the extensions to adjust the contact pressure (Figure 11). For contact pressures
less than 188 Pa, the sled equipped with an adhesive surface was used. This was necessary
as the first sled produced a contact pressure that was greater than 94 Pa without the addition
of any weights. Small weights were attached to the adhesive surface of the second sled to
adjust the contact pressure. For the third strand orientation the second sled was used
exclusively. Because the contact surface area was significantly less for the third orientation,
the weights had to be reduced accordingly to achieve the appropriate contact pressure. For
each pair of strands the same surfaces were in contact for the three orientations tested.
Figure 10. Two sleds used for the inclined plane test, (left)sled equipped with dowels forlarger contact pressures and (right)sled equipped with adhesive surface for lower contactpressures.
1-inch
5/24/2018 OSB Blender Optimization
34/149
-23-
Figure 11. Photograph of the testing procedure for the coefficient of static friction of woodstrands using the inclined plane technique, showing a parallel parallel orientation.
For the coefficient of static friction tests involving wood strands and HDPE the same test
format was followed, including three orientations and seven contact pressures. Instead of
using a variety of HDPE samples however, only two were used. The larger HDPE sample
was used for the first two orientations and the smaller sample was used for the third
orientation. In all three cases the HDPE was the primary surface and the wood strand was
the secondary surface. The HDPE sample was placed directly on the inclining plane in
contact with the stop block for the first two orientations. For the third orientation the HDPE
was attached to the raised platform, which was then placed on the inclined plane. The raised
platform added stability to the HDPE sample, preventing it from shifting during the tests.
The same secondary surface strands from the first set of tests were used for this second set of
tests. The secondary surface was prepared identically as before using the two sleds.
After the strands were mounted to the respective surfaces the testing apparatus was leveled
using the adjustable glides. The appropriate weight was then added to the sled. The angle of
inclination was slowly increased until the secondary surface began to slip along the primary
surface. The heights from the base of the apparatus to two predetermined points along the
inclined plane as well as the distance between those two points along the inclined plane, the
5/24/2018 OSB Blender Optimization
35/149
-24-
hypotenuse, were recorded when the sled began to slip. These measurement points remained
constant across all of the tests. The coefficient of static friction was then determined
according to Equation 5. This procedure was repeated for each contact pressure, strand
orientation, and system of materials. The complete set of results were then analyzed using
SAS version 9.1 to develop a model that predicted the coefficient of static friction based on
the material combination, orientation, and contact pressure.
=
l
hh 12arcsintan [5]
Figure 12. Schematic of inclined plane jig showing measurement locations for Equation 5.
3.4 Results
The test results aimed at determining the coefficients of static friction for strand to strand and
strand to flight interactions clearly showed a decreasing coefficient of static friction value for
increasing contact pressures (Table 4, Figure 13, Table 5, and Figure 14). This is particularly
apparent when increasing from 23 Pa to 94 Pa. The coefficient of static friction appears to be
constant for pressures greater than and including 94 Pa. These results are consistent with
hypotenuse, l
height 2, h2 height 1, h1
5/24/2018 OSB Blender Optimization
36/149
-25-
those obtained by Bejo et al. (2000). Bejo found that as the contact pressure decreased below
approximately 5 kPa the coefficient of static friction increased substantially. Above 5 kPa the
static friction coefficient began to stabilize. It was also found that the grain orientation had a
considerable impact
Table 4. Summary of test results for the coefficient of static friction between two woodstrands at 22C and 55% relative humidity
1.
Approximate contact pressure (Pa)Primarystrand
orientation
Secondarystrand
orientation23 47 94 188 375 750 1500
Parallel Parallel Mean 0.63 0.55 0.53 0.49 0.41 0.44 0.41COV % 24 21 19 17 26 20 25
Perpendicular Perpendicular Mean 0.88 0.81 0.77 0.66 0.64 0.63 0.70COV % 13 12 19 12 12 18 6
Parallel Perpendicular Mean 0.89 0.60 0.50 0.51 0.41 0.36 0.35COV % 28 11 20 15 10 12 11
1Average ambient strand moisture content was 9.2%.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Contact Pressure (Pa)
StaticC
OF
PERP/PERP
PAR/PERP
PAR/PAR
Figure 13. Static coefficient of friction between two wood strands for increasing contactpressures and different strand sample orientations.
Perpendicular - Perpendicular
Parallel - Parallel
Parallel - Perpendicular
5/24/2018 OSB Blender Optimization
37/149
-26-
Table 5. Summary of test results for the coefficient of static friction between a wood strandand HDPE at 22C and 55% relative humidity
2.
Approximate contact pressure (Pa)Primarystrand
orientation
Secondarystrand
orientation23 47 94 188 375 750 1500
Parallel Parallel Mean 0.31 0.30 0.26 0.22 0.24 0.25 0.24
COV % 16 23 12 12 8 6 16Perpendicular Perpendicular Mean 0.38 0.31 0.29 0.29 0.29 0.28 0.28COV % 18 10 9 6 1 8 5
Parallel Perpendicular Mean 0.39 0.40 0.32 0.29 0.26 0.26 0.27COV % 20 20 12 20 27 12 17
2Average ambient strand moisture content was 9.2%.
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.45
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Contact Pressure (Pa)
StaticCOF
PERP/PER
PAR/PERP
PAR/PAR
Figure 14. Static coefficient of friction between a wood strand and HDPE for increasingcontact pressures and different strand and HDPE sample orientations.
The test results were analyzed using multiple regression analysis in SAS version 9.1. A
logarithm transformation was necessary to normalize the data so that a model could be fit
that accurately predicted the static friction coefficient (Equation 6), while meeting the
assumptions of multiple linear regression analysis at a = 0.05 (Kleinbaum 1988).
Perpendicular - Perpendicular
Parallel - Parallel
Parallel - Perpendicular
5/24/2018 OSB Blender Optimization
38/149
-27-
)()( 2101010 xbxLogbbs++=
[6]
where:sis the predicted static friction coefficient,xis the contact pressure in pascals, andbiis the predicted parameters for i= 0 to 2.
The appropriate coefficients for the respective systems of materials and orientation are shown
in Tables 6 and 7. In general, these results show an approximate 2:1 ratio between the static
coefficient of friction for wood to wood and wood to HDPE. The complete SAS analysis is
included in Appendix A. It is important to note that this relationship has only been verified
for the range of data presented above and at the ambient temperature and relative humidity
encountered during the test, i.e. approximately 21oC and 50% respectively.
Table 6. Regression parameters for determining the coefficient of static friction between twowood strands at 22C and 55% relative humidity.
Primary strandorientation
Secondary strandorientation
b0 b1 b2
Parallel Parallel 4.242 x 10-2 -1.748 x 10-1 8.835 x 10-5
Perpendicular Perpendicular 1.112 x 10-1 -1.290 x 10-1 8.835 x 10-5Parallel Perpendicular 2.454 x 10-1 -2.634 x 10-1 8.835 x 10-5
Table 7. Regression parameters for determining the coefficient of static friction between awood strand and HDPE at 22C and 55% relative humidity.
HDPE orientation Strand orientation b0 b1 b2
Parallel Parallel -3.530 x 10-1 -1.222 x 10-1 8.835 x 10-5
Perpendicular Perpendicular -2.842 x 10-1 -1.216 x 10-1 8.835 x 10-5Parallel Perpendicular -1.500 x 10-1 -1.179 x 10-1 8.835 x 10-5
3.5 Conclusions
The coefficient of static friction tests confirmed previous work by Bejo et al. (2000) where it
was found that, contrary to classical theoretical research pertaining to friction, the coefficient
of static friction is in fact dependent on the contact pressure between surfaces when at least
one surface is wood. In general, the contact pressure and coefficient of friction were
inversely related, as the contact pressure decreased the friction coefficient increased and
vice-versa. It was also found that the coefficient of friction was highest when the wood
grains were aligned perpendicular - perpendicular and least when they were aligned parallel -
parallel to the sloped plane. Finally, the relationship between the coefficient of friction
between two aspen strands versus that between an aspen strand and HDPE was found to be
approximately 2:1.
5/24/2018 OSB Blender Optimization
39/149
-28-
These findings suggest that knowledge of the contact pressures encountered during the
blending operation is necessary to fully describe the slipping of strands in the RDBM.
Because the currently available discrete element modeling software packages are unable to
accommodate for a variable coefficient of static friction, a value must be selected that results
in the most accurate resemblance between the RDBM simulations and the actual observed
dynamics. As an initial starting point for the model validation process, the average of the
two extreme strand orientations will be used together with the average of the higher, more
rapidly changing friction coefficients (contact pressure 188 pa) and the lower, more stable
friction coefficients (contact pressure >188 pa).
5/24/2018 OSB Blender Optimization
40/149
-29-
CHAPTER 4
DETERMINATION OF SUITABLE MATERIAL AND INTERACTION
PROPERTIES FOR USE AS INPUT PARAMETERS IN THE RDBM
4.1 Introduction
The RDBM required the input of three mechanical properties for each material used in a
simulation and three interaction properties for each pair of materials that may come in
contact during a simulation (Tables 8 and 9). While all of these properties must be included
for a simulation to be initiated, not all of them have a significant effect on the bulk strand
dynamics. Instead, some of these properties are only significant outside of the tested ranges
and/or are used for measuring incidents of little consequence to this study, such ascompressive forces acting on the particles. Because this research is focused predominantly
on measuring the bulk strand flow within a rotary drum blender, it is only necessary to select
representative values for those factors that are vital to the accurate representation of said
strand flow.
Table 8. Required material and interaction properties for the RDBM.
Material properties Interaction properties
Modulus of rigidity (G) Coefficient of restitutionPoissons ratio () Coefficient of rolling frictionDensity () Coefficient of static friction
Table 9. Materials used and materials that may come in contact in the RDBM.
Materials Interactions
Aspen wood strands Strand - StrandPolyethylene
1 Strand - Polyethylene
1Polyethylene (PE) has been used here to describe both the high density polyethyleneflights (HDPE) and the ultra high molecular weight polyethylene drum liner(UHMW). This will be discussed further in section 4.2.1.
Focusing on the behavior of the overall system introduces several inherent challenges. First,
published values for these properties are typically based on measurements taken of individual
pieces of clear wood samples. These values may or may not be accurate for characterizing
the behavior of large collections of strands. They do however provide a useful foundation for
beginning such research. Second, these properties must also incorporate other events that are
occurring in an actual rotary drum blender but are not able to be incorporated into the model
5/24/2018 OSB Blender Optimization
41/149
-30-
because of constraints, such as the representation of strands using a series of spheres or the
lack of any air flow dynamics in the model.
In addition to potentially affecting the strand dynamics, preliminary simulations and
literature (DEM Solutions 2008) have also shown that the material properties have a
significant impact on the processing time for a simulation. The processing time increases
with increasing shear modulus of rigidity and decreases with increasing density and
Poissons ratio. If it can be found that any of the above listed material properties do not
significantly impact the bulk material dynamics then values may be chosen that minimize the
time required for processing a simulation. The relationship between the density, ; modulus
of rigidity, G; Poissons ratio, ; particle radius,R; and processing time as represented by the
idealized time step, TRis (DEM Solutions 2008):
+=
G
RTR
8766.01631.0. [7]
In this case where strands are being modeled, the particle radius refers to the individual
particles, or spheres, that are joined to form a strand. As a result, the particle radius is inch
as will be discussed further in Section 4.2.2.
This study was therefore divided into two distinct phases. The first phase consisted of two 2-
level, full factorial experimental designs aimed at determining which material properties and