SIMPRO – Task 2-1 – Guide Rail

Optimization by Isight
Authors: Timo Avikainen

Confidentiality: Public
 RESEARCH REPORT VTT-R-04402-15
1 (22)

Report’s title
SIMPRO – Task-2-1 – Guide Rail Optimization by Isight
Customer, contact person, address Order reference
Tekes, Matti Säynätjoki Tekes: 40204/12
Kyllikinportti 2, P.O. Box 69, FI-00101 Helsinki, FINLAND
Project name Project number/Short name
SIMPRO – Task 2.1.2 Use case of parameter variation and 78634/SIMPRO
parameter optimization using FEM simulation
Author(s) Pages
Timo Avikainen 22 pages + 1 Appendix
Keywords Report identification code
FEM, Abaqus, Isight, optimization, DOE, elevator, guide rail VTT-R-04402-15
Summary
The objectives of this Simpro subproject work was to create an optimization model by
applying an industrial real case. The KONE Oy used elevator guide rail system was used as an
example to the demonstration purposes. KONE produced the original guide rail model
including also the external frame, connection brackets and the roller wheels. The model
development in VTT was divided to the two parts. The guide rail based on the one
dimensional beam element modelling enable lighter models which gives benefit in model
modifications and computational times, Jani Wennerkoski continued the work with the beam
models. Applying the solid element mesh to the guide rails enable more detailed modelling but
on the contrary is heavier in terms of the computing times and model modifications. The solid
mesh approach is reported in this study.

In practise the guide rails were created again since the cross section was defined parametric.
The existing sketch can be modified and updated. The length of the guide rails were made
parametric when extruding them. The brackets used here are as they came from KONE and
the horizontal pipes were cut from the external frame. The locations of the each brackets were
made parametric by applying a python code which moves them. In a same way also the height
of the elevator was made parametric and the exact location of the roller wheels. The elevator
can be driven throught the existing guide rail in steps but without the dynamics and time
which can be added later. The buckling of the guide rail is also calculated in the model.

Both the Design of Experiment (DOE) and optimizations results of the solid model case are
described and discussed in this report. There exist lack of the perfect optimizations results
available. In order to achieve the trustable and comprehensive results more work is needed to
update and modify the Abaqus CAE model for the optimization.
Confidentiality Public
Espoo 28.9.2015
Written by Reviewed by Accepted by

Timo Avikainen Petteri Kokkonen Johannes Hyrynen

Research Scientist Senior Research Scientist Head of Research Area
VTT’s contact address
VTT Technical Research Centre of Finland, P.O. Box 1000, FI-02044 VTT, FINLAND
Distribution (customer and VTT)
Tekes/Matti Säynätjoki, 1 copy
VTT/archive, 1 copy

The use of the name of the VTT Technical Research Centre of Finland (VTT) in advertising or publication in part of
this report is only permissible with written authorisation from the VTT Technical Research Centre of Finland.
 RESEARCH REPORT VTT-R-04402-15
2 (22)

Contents
1. Preliminary Guide Rail Optimization Demo ....................................................................... 3
1.1 Introduction ............................................................................................................... 3
1.2 Basic model .............................................................................................................. 3
1.3 Design of Experiment by Isight ................................................................................. 7
1.3.1 Basics of DOE analysis ................................................................................. 7
1.3.2 The guide rail beam parameters ................................................................... 7
1.3.3 Move of the brackets as a parameter .................................................................. 8
1.3.4 Move of the roller wheel locations ................................................................... 11
1.4 Optimization by Isight ............................................................................................. 16
1.4.1 Basics of optimization ................................................................................. 16
1.4.2 Optimization algorithms............................................................................... 17
1.5 Buckling .................................................................................................................. 20
1.6 Discussion .............................................................................................................. 22
1.7 References ............................................................................................................. 22
 RESEARCH REPORT VTT-R-04402-15
3 (22)

1. Preliminary Guide Rail Optimization Demo

1.1 Introduction
This document is a part of the Simpro project subtask 2.1 optimization case of the KONE Oy
elevator case in which the guide rail system is analysed. The main content of the work
reported here was to develop an optimization model for a Isight optimization software and
create a demonstration example by using the simplified guide rail system. The actual
simulation is based on the Finite element method (FEM) applied in the Abaqus CAE
software.

1.2 Basic model

The preliminary simulations based on the FE model provided by KONE to VTT on February
2013. As a first step the solid frame pipes were replaced by the Abaqus shell elements by the
6.3mm thickness in the pipe frame walls. The main cross section dimensions of the guide rail
were made parametric by the Abaqus CAE sketching tool. The constraints of the sketch must
be considered carefully since these can easily make errors to model in the optimization phase.
The length of the guide rails are created and made parametric by extruding the cross sections
sketches. The other details of the preliminary optimization model were the same with the
original model (one frame, four guide rails, five brackets, fishplates between the guide rails
etc.). The main dimensions (6 dimensions) of the cross section were made parametric. This
sketch can be updated to correspond even better the reality in the future.

Figure 1. The cross section of the guide rail made parametric.

 RESEARCH REPORT VTT-R-04402-15
4 (22)

In order to make a computationally “lighter” model for the heavy optimization purposes, a
simpler Abaqus beam model including also the supporting frame was created. An important
point for this was also possibility to control the distance between the brackets and make it
parametric for the optimization purposes when the supporting frame is present. This is not
possible in practise in Abaqus CAE when the frame is modelled by the shell or solid elements
without the geometry behind. In this frame as the simple beam model the frame is sketched
which enable the control of the distance between the guide rail brackets. The cross sections of
the simple frame pipes are easy to be defined by the Abaqus Edit Profile function.

Figure 2. Preliminary simulations when the frame was also present.

The use of the beam element seems not to enable very detailed guide rail profile due to the
limitations in the Abaqus Edit Profile option. Existing alternatives are limited to the I or T –
profile or just applying the generalized profile in which the basic anonym cross section
surface area and the needed moment of Inertia values in two directions are made parametric.

Figure 3. Alternative profiles for the guide rail in the simple beam model.
 RESEARCH REPORT VTT-R-04402-15
5 (22)

The simplified solid FE model is cut from the original from the both ends of the transversal
pipes that have a close connection to the brackets. The guide rail and the perpendicular short
pipes were remeshed again but the mesh of the bracket was remained as it came from KONE.
In fact, the mesh of the brackets should be regenerated as well because according to the
Abaqus rules at least four element layers are needed in the thickness direction of the plate
when the element type C3D8R is applied. The applied Python script must include the meshing
and assembly update commands if e.g the geometry is changed, in this study this is not
needed since the question is about the movement of the components.

The gravity of the guide rail is added as a first loading step of the solution. Symmetry
boundary conditions are applied in the both ends of the guide rail system. This gives a little
bit wrong results since the ends of the simulated system are not symmetric in practise
(different lengths of guide rails in the ends). The boundary condition in the ends of the
perpendicular short pipes is all fixed except the y direction (elevator moving direction) left
free which enable the parametric movement of the brackets. The static loads are applied in the
elevator rollers.

Figure 4. The element mesh of the simulated guide rail system.

The static roller wheel forces were defined close to the values during the elevator car running.
The vertical forces applied in the both roller wheels were 180N and in the horizontal direction
100N. The roller wheel positions applied in the static forces case were in the middle between
the brackets. These locations were just estimated as the “worst case” conditions for the
preliminary analysis.
 RESEARCH REPORT VTT-R-04402-15
6 (22)

Figure 5. The used static forces in two directions, the 5m guide rail is located in the left side.

The Isight component chain is quite simple in this guide rail example since most of the
important issues can be included in preliminary CAE model itself (e.g different loadings in
own steps) and the applied userscript (userscript_cae_pre.py). The calculator component was
used here to couple the roller wheel force locations together in a way that the height of the
elevator (= distance between the rollers in the guide rail) and the position of one of the rollers
were made parametric.

Figure 6. The simulation flow of the demonstration case and overall in Isight.
 RESEARCH REPORT VTT-R-04402-15
7 (22)

1.3 Design of Experiment by Isight

1.3.1 Basics of DOE analysis

Design of Experiment (DOE) as the sensitivity analysis describes and makes possible to
compare the importance of the different input design parameters. The objective parameters
develop to the target value or the most possible extreme value according to these design
variables. The following examples hopefully clarify this. The default DOE technique in Isight
is Latin Hypercube. It allows many more points and more combinations for each factor.

1.3.2 The guide rail beam parameters

The following DOE results were calculated by using the static roller force locations according
to the figure 5 and the bracket locations as in the original model. The influence of the flange
width of the guide rail is more dominating on the maximum displacement and Von Mises
stress than the two other dimensions. These two other dimensions were the height and width
of the top part of the web in the guide rail cross section. The maximum displacement covers
the simplified solid guide rail model including also the brackets and the vertical pipes.

Figure 7. The influence of the guide rail profile parameters by the static force locations.

The results look different in the case of the moving roller force locations. The elevator is
moved 9.5m through the existing guide rail from the end to end, distance between the rollers
is constant. The importance of the width of the flange is still dominating in the terms of
maximum displacement and stress but not e.g single detection point of the guide rail.

Figure 8. The influence of the guide rail profile to the max. overall and a single point
displacement by the moving roller force locations.
 RESEARCH REPORT VTT-R-04402-15
8 (22)

Figure 9. The influence of the guide rail profile parameters on the max. Von Mises stress by
the moving roller forces.

The influence of the 3.5 m guide rail on the maximum displacement is described. The
importance of the three 3.5m guide span on the displacement is clearly higher compared to the
single 5m guide rail alone according to the Pareto plot (linear results). Pareto plot rank factors
by how much they contribute to the variance of the response. Pareto plots in Isight software
use normalized coefficients of fitted 2nd order polynomial, which represent the % total effect
on the response. The larger the magnitude the greater the influence on the response. Higher
2nd order effect indicates a single parameter’s effect on a response is not linear which doesn’t
seem to be a case in the following figure.

Figure 10. The influence of the guide rail length on the displacement. by static forces.

1.3.3 Move of the brackets as a parameter

The potential move of the each five bracket instances were also made parametric. The
brackets were firstly named according the following figure. The “low” status means the
starting from -0.4m and continuing up to the “high” +0.4m around the original bracket
locations. The coordinate axes are also described in the figure. The maximum volume of the
each four guide rails were minimized as the objective function.
 RESEARCH REPORT VTT-R-04402-15
9 (22)

Figure 11. The guide rail brackets added to the model, static roller wheel loads used.

The next graph describes what is the influence of moving the brackets to the volume of the
3.5m guide rail. The location of the bracket 1 and 3 seem to have the most clear influence, the
move of the bracket 1 to the minus direction would give benefit as the bracket 3 has the best
position in the original position.

Figure 12. The influence of the moving the brackets to the first 3.5 m guide rail volume.
 RESEARCH REPORT VTT-R-04402-15
10 (22)

Naturally the brackets s4 and s5 have the most clear influence to the volume of the 5m guide
rail since they are in a physical contact to that.

Figure 13. The influence of the moving the brackets on the 5.0 m guide rail volume.

The pareto plot describes the influence of the different combinations of the brackets on the 5m
guide rail volume as well as the most important single bracket movements (linear s2 and s5).
The higher the magnitude the greater the influence on the response. The higher interaction
effects like e.g s3-s5 in the plot indicate that changing two parameters together will greatly
influence the response. Higher 2nd order effect indicates a single parameter’s like s3^2 effect
on a response is nonlinear. The blue colour represents positive main effect when as factor
increases then also the response increases. On the contrary the red colour represents the
negative main effect which means as factor increases then responses decreases.

Figure 14. The Effect of the moving the brackets in terms of the 5m guide rail mass
minimization.
 RESEARCH REPORT VTT-R-04402-15
11 (22)

The influence of the moving the brackets are visualized in the following figure. Left side
describes the displacement in one single point of the guide rail between the brackets s2 and s3
(in the joint of two guide lines) which explains the reason why the corresponding curves
separate clearly from others. According to the picture in right side the variations in the bracket
coordination along the guide rail do not make big change to the maximum overall
displacement which take place in the brackets itself (not in the guide rail).

Figure 15. The influence of moving brackets to the max. displacements a) guide rail, b) total.

As an example by applying the parameters (s1=0.22, s2=0.041, s3=-0.318, s4=-0.188,

s5=0.318) according to the DOE simulation gives very minimal reduction in the guide rail
mass (less than 1%), but increase lightly the displacements in the model. It could indicate that
there is not a lot of available benefit to make changes in the existing locations of the brackets.

1.3.4 Move of the roller wheel locations

The influence of the roller wheel locations were firstly calculated by the fixed distance 5m
between the roller wheels. Then the both existing roller force locations were included inside
one single parameter. This common parameter can get values starting from one end of the
guide rail up to the opposite end minus the named 5m distance between the rollers. Here in
this simple DOE run the movement of the elevator was the only design parameter and
objectives were defined to be minimizations of the parameters max Von Mises stress, max
overall displacement, displacement in a single point (figure 16).

Figure 16. The responses according to the elevator location (height=5m).

 RESEARCH REPORT VTT-R-04402-15
12 (22)

The optimum roller force locations would look like following with the 5m height of the
elevator car.

Figure 17. Optimum roller wheel locations by the constant 5m elevator height.

In the next step the height of the elevator car in other words the distance between the rollers is
also included as a design variable.

Figure 18. The roller wheel locations added to the sensitivity analysis of the guide rail.
 RESEARCH REPORT VTT-R-04402-15
13 (22)

The overall maximum Von Mises stress has a minimum point at the elevator rolling location
9.1m from the end of the guide rail with the height of the elevator car 5.18m. The height of
the elevator (distance between the rollers) has a minor influence on the stress compared to the
elevator location along the guide rail. By increasing the height of the elevator (distance
between the rollers) increase lightly the overall maximum stress.

Figure 19. The influence of the elevator location (m1) and height (h1) on the overall max.
Von Mises stress.

The situation is a little bit different with the overall maximum displacement. Both the location
and height of the elevator seem to have roughly equal influence on the displacement.

Figure 20. The influence of the elevator location (m1) and height (h1) on the overall max.
displacement.
 RESEARCH REPORT VTT-R-04402-15
14 (22)

The location of the elevator has the maximum in the low values since the detecting
displacement point is located there. The height of the elevator has a small influence.

Figure 21. The influence of the elevator location (m1) and height (h1) on the displacement in
a single point of the guide rail.

The influence of the all acute design variables to the guide rail volume are visualized in the
next picure. The results looks different when comparing them to the maximum Von Mises
stress results.

Figure 22. The influence of the guide rail design variables on the volume of the 3.5m rail.
 RESEARCH REPORT VTT-R-04402-15
15 (22)

Figure 23. The influence of the guide rail design variables on the max. Von Mises stress.

The width of the flange is the most dominating parameter in the next pareto plot according to
the linear results. The parameter “Web width top” has also the magnitude as strong but it’s
effect on the mass of the guide rail is quite nonlinear. The mass of the guide rail decreases by
increasing the value of the elevator location as the red colour describes the negative main
effect but the relationship includes a lot of nonlinearity.The height of the elevator car is less
dominating according to the magnitude procent.

Figure 24. Pareto plot describes the importance of the design variables to the 3.5m guide rail.
 RESEARCH REPORT VTT-R-04402-15
16 (22)

1.4 Optimization by Isight

1.4.1 Basics of optimization

The optimization idea can simply be described by the following figure. The objective target is
to find the highest point in the hill which means the most optimum point in the design. In the
elevator guide rail case typically this is the total mass which is wanted to be minimized. The
longitude and latitude curves in the figures describe the design variables, parameters that can
be looped in the given frames. The small dog must stay inside the fences which create the
constraints to the system (the optimization solution can be anyway started outside the fences).
Typically constraints are maximum allowed displacements, stresses, certain natural
frequencies, parameters where the variations are limited to the given exact values. In each
optimization step target is to find a search direction that will improve the objective while
staying inside the fences. Search takes place in certain direction until no more improvement is
made.

Figure 25. The simple illustration of the optimization.

The DOE (Design of Experiment) analysis is typically solved before the optimization. That
describes the importance of the different parameters. In the optimization procedure the main
task is to find the ultimate point meaning typically minimum, maximum or target value for the
objective function. In our example the total volume of the guide rails are defined as the
objective function since the volume means the same as the mass (density given). The design
variables are listed in the following table with the original and the resulting optimum value.
The applied constraint here was the maximum displacement in the total and guide rail model.
The maximum Von Mises stress is also listed but it was not used for the optimization. The
resulting design variables differ from the table in the case when the Von Mises stress is
applied as the objective function.

Table 1. Conclusion of the optimization example.

Original Optimum
Flange width Design Variable 0.14 m 0.126 m
Upper web height Design Variable 0.053 m 0.05025 m
Upper web width Design Variable 0.02 m 0.0170 m
Rail-3.5m-length Design Variable 3.5 m 3.348 m
Volume-rail-3.5m Objective Function 0.0123m3 / 96.6kg 0.0105m3 / 82.5kg
Volume-rail-5.0m Objective Function 0.0176m3 / 138.1kg 0.0150m3 / 117.9 kg
Max.displacement Constraint 1.32e-4m bracket 1.33e-4m bracket
8.80e-5m rail 9.40e-5m rail
Max.Von Mises stress Constraint 7.622e+6 bracket 8.611e+06 bracket
2.185e+6 rail 2.468e+06 rail
 RESEARCH REPORT VTT-R-04402-15
17 (22)

The maximum displacement and Von Mises stress distributions are described in the following
figures.

Figure 26. Maximum displacement and Von Mises stress distributions in the optimized case.

Isight uses Sequential Quadratic Programming (NLPQL) as a default optimization algorithm

technique. It suits well to the long running simulation and to the highly nonlinear design
spaces but not well to the discontinuous design spaces. It exploits the local area around initial
design point and rapidly finds a local optimum design. It handles inequality and equality
constraints directly.

1.4.2 Optimization algorithms

Different Isight optimization algorithms calculated results are compared in table 1. The
operation range of the design variables as the dimensions in the following table was defined
from -10% to +10% except in the distance between the roller wheels and location coordinate
parameter m1. Constraints were defined according to the displacements calculated from the
original model, stress was not defined as the constraint in the next table. The objective
function was to minimize the volume of the guide rails. There are more optimization
algorithms available in Isight but some of them need stronger cluster resources in computing
or they are not directly suitable to this case. In case of errors Abaqus model would require
more updating steps to be applied. There is a list of the optimization algorithms presented in
the appendix 1 and information how they behaved with this guide rail example. Quite typical
message appearing in the Isight optimizations refers to the missing of the final solution when
the software gives a warning: “No feasible design found to satisfy all constraints”. Then it is
better to try smaller runs (e.g in minimum starting from one simple loop of the acute
component itself), make tests by a different optimization parameters or modifying the cae
model itself.
 RESEARCH REPORT VTT-R-04402-15
18 (22)

Table 2. Conclusion of the few Isight used optimization algorithms compared to the DOE run.

Original DOE NLPQL Downhill Evol MISQP Multi-

Analysis (Default) Simplex Objective
Particle
Swarm
Objective
Guide rail total 11 % 10 % 7.5 % 11.5% 9% 1%
volume reduction reduction reduction reduction reduction reduction

Design
variables
Rail 3.5 m 3.5m 3.325 m 3.437 m 3.470 3,400 3.500 3.675
± 10%
Flange web 0.14 0.128 0.126 0.127 0.127 0.126 0.134
width ± 10%
Web top height 0.053 0.048 0.058 0.0526 0.050 0.047 0.0541
± 10%
Web top width 0.019 0.0196 0.0209 0.0188 0.018 0.021 0.0170
± 10%
Elevator height 4.5 – 7m 5.18 m 6.80 m 7.18 m 6.66 m 5.62 m 5.98 m
Elevator 4.7 – m1 m1: m1 m1 m1 m1
location 11m 9.1m 9.2m 10.8m 11.9m 9.1m 12.8
Bracket 1 ± 0.4m -0.262 0.228 0.060 0.228 0.080 -0.310
Bracket 2 ± 0.4m 0.097 0.4 0.223 0.148 0.206 -0.331
Bracket 3 ± 0.4m -0.014 -0.263 0.219 -0.204 -0.4 0.063
Bracket 4 ± 0.4m -0.015 0.114 0.236 -0.028 0.027 -0.180
Bracket 5 ± 0.4m 0.124 0.083 0.134 0.244 0.071 -0.002

Constraint
Max total 1.27e-4 1.261e-4 1.271e-4 1.290e-4 1.30e-4 1.272e-4 1.28e-4
displacement
U3_20 3.21e-5 4.76e-5 3.974e-5 4.184e-5 2.78e-5 3.61e-5 2.56e-5
U3_5822 1.078e-5 3.89e-6 1.207e-5 1.168e-5 1.097e-5 1.22e-5 1.15e-5
U3_8363 -9.12e-6 2.81e-6 -1.06e-5 -9.03e-6 -9.41e-6 -9.7e-6 -9.9e-6

Max Von 7842380 8740570 8181730 8001120 8346550 7970170 8165810

Mises stress

There seems to be surprisingly large differences in guide rail results appearing between the
different Isight optimization algorithms as described in table 1. The values of the parameter in
meters as the elevator location (roller wheel point m1) seem to be pretty close to each other
when the values of the height of the elavator (distance between the top and bottom roller
wheels) seem to differ clearly. Anyway typically 10% mass reduction in guide rail could be
possible according the existing Isight optimization results. The optimum locations of the
roller wheels could be for example as in the next picture in which they are calculated by the
default NLPLQ algorithm.
 RESEARCH REPORT VTT-R-04402-15
19 (22)

Figure 27. Optimized rolling load locations.

The most disadvantageous roller wheel locations were optimized by asking the Isight to
maximize the displacements in the guide rail. The results look like following.

Figure 28. The most bad loading conditions by the optimization.

 RESEARCH REPORT VTT-R-04402-15
20 (22)

1.5 Buckling
The buckling analysis was tested in Abaqus CAE included to the Isight loop. The buckling is
a linear axial buckling eigensolution with the first mode read as the lowest buckling load
result. The buckling is defined as an own load step with the changed boundary conditions
before coupling to the Isight. The unit force as the buckling load was applied to the same
nodesets in which the earlier roller related forces were deleted, as described an example of the
buckling forces in the next figure. The direction of the buckling loads are in the axial direction
of the guide rails. Any changes were not made to the buckling FE – mesh in terms of the
idealization.

Figure 29. Buckling bending in the y-z plane, two forces, lowest buckling load 1.5MN.

As the second example the buckling load was applied 0.3m offset from the guide rail top
surface (against the top roller). The offset was created by the MPC beam element. The lowest
buckling load (left) is 1.98 MN and the second one 2.06 MN which are clearly above the 150
KN based on the Mathcad calculations.

Figure 30. The buckling model coupled to the Isight, buckling load roughly 2MN by the
described modes.
 RESEARCH REPORT VTT-R-04402-15
21 (22)

As the third example the buckling force was applied to the end of the guide rail as well the
longitudinal displacement is free. The Abaqus encastre boundary condition was used in the
opposite end of the guide rail to prevent all the displacements. The resulting buckling loads
are clearly above the 150 kN. In the case of leaving the force end boundary condition free
without any fixing, the buckling load crushes down to the 91 kN. The buckling seems not to
be a problem in the studied one guide rail module, but the situation can be different in the
larger systems. Anyway the used force and the boundary conditions are in a key position.

Figure 31. Force in the end, lowest buckling modes by the load 0.7 kN (left), 0.8 kN (right).

Figure 32. Force in the end, third/fourth buckling modes by roughly by the load 1.0 MN.
 RESEARCH REPORT VTT-R-04402-15
22 (22)

1.6 Discussion

All the results showen in this document are basically meant for demonstration purposes only.
There are several uncertain details involved in the used optimization models.

The solid model enable more detailed design to be optimized, e.g the loads, detection points
in the right locations etc. Also extra holes are possible to be included with certain dimensions
made parametric in the guide rail web. On the contrary solid model is naturally heavier
alternative compared to the one dimensional beam model in terms of both computing and
update modification.

Each parameter must be created in Abaqus CAE, some are very straightforward but some
require e.g use of the constraints, python scripting etc. and the coupling in Isight. The
multiparametric model may cause problems because these can be in contradiction to each
other e.g certain parameters require use of the constraints but may be harmful for some other.

Isight updates automatically the element mesh in Abaqus CAE when the element sizes are
exactly defined. In the case of the number of the elements per dimension Abaqus naturally
keeps this definition and basically not update the mesh. This can cause element quality
problems (distorted elements) if the dimensions of the body part are changed a lot. In this case
it showed that the optimization was possible in a quite narrow parameter range. In order to
expand these limits the optimum solution should be raked to find more accurately the right
sector and again concentrate the optimization there. Different optimization algorithms have
own prons and cons depending on the actual case. Certain algorihms may offer better
solutions for “raking” than others in the larger range of the parameter dimensions and finally
find out the optimum solution more efficiently than others. Finally the preliminary Abaqus
CAE model should be updated by cutting the acute guide rail parts to the shorter complexes.

The total number of the guide rails can be up to 76 (300m). Here was only applied the four as
a fixed number of the guide rails. The model do not include the loop to find out the optimum
number of the guide rails with the optimized lentgths. Now the model tries to optimize the
length of the three 3.5m and one 5m guide rails.

The deflection angle in the guide rails (same position with the brackets) as the basic parameter
for the ride comfort is not yet included to the solid model. That would clearly add the
complexity of the Abaqus CAE model since it requires several extra parameters to be defined
e.g the angles between the guide rails itself, locations to cut the guide rails in a same position
with the bracket, connection to the length of the each guide rail etc.

1.7 References
[1] Demirsoz R, Imrak E, Targit S., Finite Element Analysis of Elevator Guide Rails during Safety
Gear Operation., ASRAY Company, Turkey.

[2] KONE internal document.

[3] KONE internal document.

[2] Wikipedia article at: http://en.wikipedia.org/wiki/Meta-optimization

 RESEARCH REPORT VTT-R-04402-15
23 (22)

APPENDIX 1: Optimization algorithms used in Isight

AMGA – Archive based Micro Genetic Algorithm, not works, gives error immediately in the
beginning:

ASA – Adaptive simulated annealing, Error related to the overconstraint in some nodes,
algorithm requires computing in VTT cluster since the amount of the iterations is roughly 20
times larger compared to the NLPQL.

LSGRG – Large Scale Generalized Reduced Gradient, works yes but needs the cluster
resources because of the huge amount of iterations.

Hooke-Jeeves – Direct search, often used when feasible design can not be determined. Here
this didn’t worked and not fount the optimum solution. Run was ended to the errors.

MMFD – Modified Method of Feasible Directions, used best when starting from a feasible
design point, works yes but needs the cluster resources because of the huge amount of
iterations.

MISQP – Mixed Integer Sequential Quadratic Programming, works yes but needs the cluster
resources because of the huge amount of iterations.

MOST – Multifunction Optimization System Tool, works yes but needs the cluster resources
because of the huge amount of iterations.

Multi-Island GA Genetic Algorithm, finished to the errors.

Multi-Objective Particle Swarm, works (some errors involved) but needs the cluster resources
because of the huge amount of iterations.

NCGA – Heighborhood Cultivation Genetic Algorithm, run was ended to the errors,
computation cluster is needed

Stress Ration, NSGA-II – Non Dominated Sorting Genetic Algorithm, solving stops to the
errors

Pointer Automatic Optimizer, solving stops to the errors.