Sei sulla pagina 1di 20

CONCEPTUAL DESIGN USING

PROBABILISTIC INTERVAL CONSTRAINT SATISFACTION

TERM PROJECT REPORT

CIVL 518

Prepared by Caroline Villiard (Student ID: 10104099)

Date of Submission: 11th of December 2009

Department of Civil Engineering

The University of British Columbia

I
Abstract
Conceptual phases of a structural design involve uncertainties. Engineers have to deal with those
as a decisive challenge to pursue the safest and most economic design. Several methods have
been developed over time to make this first concept phase much straight forward. Although the
primary design might have improved to produce safer solutions, it has not become particularly
more economic. This is due to larger safety coefficient found in codes and to lack of software
tools to carry out risk and reliability analysis.

The methodology proposed by Nathan Loewen in his PhD research seeks a solution to this
problem. The most important decisions are made during the conceptual phase, so the present
method would lead to a decision based on an assumed probability distribution of variables. The
Probability Constraint Satisfaction analysis method is the foundation of Loewen’s hybrid method
which includes the uncertainties input. The output defines a design space combining probabilities
of various data. The methodology involves subdividing the design space and estimating the
potential of respecting the constraints according to the probability distribution of the data in each
interval. A Matlab function was next developed to to show the user a most comprehensible
design space with a probability chart. Additionally, this method offers wider calculation options
provided by the use of multidimensional arrays.

The current work intends to present the essence of the thesis to the reader in a case study. It is
hoped that simplified explanations along with an example aim at better understanding for the
reader. Since multiple engineering applications are possible with the method, a large range of
examples was available to show his utilities. Seeking more knowledge in reliability as well as in
the steel structural field of study, a design example involving bolted connection was chosen.
Additional computations were added to the thesis work increasing the versatility of the Matlab
calculations. Those are related to the log-normal distribution often encountered with
engineering’s set of data. The two observed parameters for the end plate connection example
were analyzed based on the sampled set of data. A logical conclusion was subsequently made
and used for calculations. In conclusion, the Probability Interval Constraint Satisfaction method
can used produce a visual description of a particular design space as shown to the reader of this
report.

II
Table of Contents
1. INTRODUCTION ....................................................................................................................................1
2. METHODOLOGY ...................................................................................................................................1
2.1. DATA PROCESSING ......................................................................................................................................1

2.2. INTERVAL CONSTRAINT SATISFACTION PROBLEM ......................................................................................3

2.3. HYBRID METHOD ........................................................................................................................................4

2.3.1. Interval Narrowing method ..........................................................................................................5


2.3.2. Grid-Search method .....................................................................................................................5

3. EXAMPLES ..............................................................................................................................................6
3.1. EXAMPLE 1 BOLT SHEAR STRESS RESISTANCE .............................................................................................6

3.1.1. Results ..........................................................................................................................................6


3.1.2. Analysis ........................................................................................................................................7

3.2. EXAMPLE 2 BOLTED ASSEMBLY (END PLATE CONNECTION)........................................................................8

3.2.1. Problem Description ....................................................................................................................8


3.2.2. Computations ...............................................................................................................................9
3.2.3. Results ........................................................................................................................................ 11
3.2.4. Analysis ...................................................................................................................................... 13

4. CONCLUSIONS AND RECOMMENDATIONS ................................................................................ 14


5. REFERENCES ....................................................................................................................................... 14
6. APPENDICES......................................................................................................................................... 15
6.1. APPENDIX 1 – PARTIAL SET OF DATA WITH DETAILS OF THEIR ORIGIN....................................................... 15

6.2. APPENDIX 2 - SPREADSHEET CALCULATIONS FOR AN END PLATE CONNECTION......................................... 16

6.3. APPENDIX 3 – LOG-NORMAL DISTRIBUTION CALCULATIONS ..................................................................... 17

III
1. Introduction
Engineers need methods and softwares to help them dealing with uncertainties at the first stage of
structural design. A Probabilistic Interval Constraint Satisfaction (PICS) method was developed by
Nathan Loewen in his Ph.D research. This term project intends to introduce his work to the reader and to
widen it. An overview of the initial methodology is presented along with some general knowledge
necessary to understand the core of the hybrid method. As part of this project, two examples were used
to test the method and help visualizing the large extension of the method. Additional computational
modules were created to extend the versatility of his work. Multiple cases could be tested with this
method; therefore it would be interesting to extend the work to provide engineers with more tools.

2. Methodology
In this section, it will be described all steps followed throughout this term project. The first part present
the treatment of data that enable their use. Thereafter, the basic method of the PICS analysis method will
be presented for sake of clarity, as this understanding will be needed to follow the subsequent
discussion. The main subject will be described in details with an outline of the existing probabilistic
method, along with the input needed to state the problem properly. More details of the procedure
executed to get those interesting and expected output will be described among the examples description
in the following sections.

2.1. Data processing


Data processing is explained here with the discrete example of steel proprieties data handle in this
project.

The yield and tensile strength of any material has to be considered as random variables since they are
imprecise parameters of the actual material. A set of data were available for a descriptive statistic
analysis. The details of the provenance of those steel test results are available at the Appendix 1. It can
be seen in Table 1, results for the mean and the standard deviation of both the yield strength and the
tensile strength. (Fy expected= 345 MPa, Fu expected= 450 MPa).

Table 1 Probabilistic parameters


Parameters of Fy Data set result Parameters of Fu Data set result
(MPa) (MPa)

Mean (μFy) 373 Mean (μFu) 506

Standard deviation (σFy) 13 Standard deviation (σFu) 10

Mean log(μFy) 5.938 Mean log(μFu) 6.227

Standard dev. log(σFy) 0.033 Standard dev. log(σFu) 0.019

1
From comparison of the cumulative frequency diagram, it was seen that either the normal distribution or
the log-normal distribution will fit the actual distribution. Those two plots are shown in Figure 1. Note
that the curve of the tensile strength could have more goodness-of fit, that is probably due to the error
generated by a small set of data.

The initial assumption made by N. Loewen while developing his method was to assign a Gaussian
distribution to the yield strength. However, in the literature (Alpsten 1972), the steel strength is seen as a
log-normal distributed variable. Another reliable source outlining this theory is the code calibration of
the National Building code [2]. This paper refers to some data coming from the literature although the
source is not listed. In addition, considering the fact that the strength of any material is strictly positive,
it increases the relation with a natural logarithmic. Taking those facts into consideration, the main
example presented at section 3.2 will take into account that the steel yield and tensile strength have log-
normal distribution with parameters coming from the sample of data.

Figure 1 : Comparison of cumulative distribution of Fy and Fu

A clarification has to be done to determine the type of probabilistic values that has to be chosen to
perform the method. The computations made in the thesis imply that those variables should not be
correlated to each other. One of the reasons for that is seen by the generation of an output space design
that includes every possible combination regardless of the physical meaning of the values. In other
words, it is known that if two parameters are highly correlated, their opposite extreme values are not
likely to occur at the same instant of time. Even though the yield strength and the tensile strength seem
to fall into that category, it is actually not the case. Figure 2 shows how poorly the index R2 is, showing
low correlation for the present set of data. This is explainable based on the large blend of metal
contained in steel. The all have their different mechanical proprieties hence the steel sample could stay
elastic longer but necking could occurs soon after.
2
Figure 2 : Scatter plot of the set of data

2.2. Interval Constraint Satisfaction Problem


The general formulation of a Constraint Satisfaction Problem (CSP) is established by defining variables
and their respective domains, and formulating appropriate constraints associated with those variables.
The variables must then be instantiated by assigning a specific domain of values to the subset of
variables. A partial solution for the total CSP is defined when the constraints of the problem are locally
respected. Two general techniques are used to solve those problems; the backtracking and the
consistency techniques, which are described in the Loewen’s paper [4].

Mainly, the hybrid method used for this work is based on defining variable intervals. Thereafter, we
observe either at the bounds of the intervals to find the “hull” of solutions, or simply at the mid points.
Those intervals are an easier way of representing data, especially those uncertain data so called
probabilistic data.

The constraint problems are formulated using the well known “spreadsheet” method. Engineer built
those spreadsheets frequently for each design problem that they face. The procedure is

1. Defining design variables that will iteratively be changed by the designer.


2. Entering fixed referenced variables that will not be changed.
3. Computing the formulas and constraints for the problem related to those design variables.
4. Designing an iterative work to optimize the design variables.

3
This method has some weaknesses. The engineers have to spend a lot of time to find good values, and
frequently, they would never converge to the optimal one. When we are dealing with probabilistic
variables, it is actually impossible to incorporate them in that method. The same spreadsheet used by all
companies can simply be modified to get the Matlab languages and hence, be able to optimize the
previous work done. Two snapshot views are shown below in relation with the main example of this
work. Additionally, the full spreadsheet is display at the Appendix 2.

Figure 3 Snapshot view of a regular spreadsheet

Figure 4 Snapshot view of a transformed spreadsheet

2.3. Hybrid Method


The following description shows the main improvement that this method called the Hybrid Method had
brought to the CSP analysis method. The introduction of probabilistic variables as well as lack of
restraint about higher-dimensionality spaces are both achieved goal in this new method referred to as
PICS. Here is the method presentation overview, followed by the description of each implicit
methodology.

The hybrid method integrates different traditional methodologies to define the distribution of the
resulting probability of a given problem. First of all, the user gets to select his variables and specify their
type. There will be two primary classifications for variables
4
• Design variables: These have to be chosen carefully as they will be used by any user to
design their own structure. It has to be significant parameters that would yield optimal
solution to the concept. Those are usually the main concern of the designer so they could
provide the potentially valid interval to use. They are referred to the plotting variables since
they are the top-level design variables that will be locate on the axis.

• The probabilistic variables: These have uncertainties about their intensity or accuracies.
Tolerances of scientific instrumental or set of data that occurred with a certain probability
distribution are both examples for those variables. They are referred as the narrowing
variables as they have respectively their range of values where a solution yields.

This approach allows the user to define the amount of each type of variables that would be plot, and
consequently to give him free rein when choosing which variables to plot. Different axis could be
defined as well as different objective function to illustrate different relationship. For instance, the cost
for two different types of prototype could be compare on two different plots having the same axis label.

It is of the user responsibility to define the problem like it has been mentioned above with the variables.
The objective function is also part of his task and can be changed quite easily if aiming to provide the
designer with more awareness.

2.3.1. Interval Narrowing method


One could see the engineer judgment and experience as the first step of narrowing the intervals of the
variables. To reduce to the possible success intervals, the engineer has to use general knowledge about
the minimum and maximum values that would lead to a solution. Interval narrowing also provides the
elimination of intervals containing no valid solution. So far, the Matlab file used in the hybrid method
does not include that part of the methodology; this work is still in progress. Instead, the narrowing work
has to be done whether by the user within the selection criteria or has to be excluded from this method.
One of the advantages of the Narrowing method is to guarantee both a bounded solution for each entry
and a non-existence of solution outside the solution space. Obviously, in the present case, those
irregularities might occur so it is the user responsibility to choose variables and number of boxes
adequately and to keep in focus this weakness while analyzing the results.

2.3.2. Grid-Search method


Subsequently, with the narrowed intervals, the Grid-Search method plays his role. First, the user defines
the number of subdivisions to use for each variable, both the design and probabilistic one. The number
of those divisions defined the dimension of the matrix generated by the method. A CSP test is carried
out by instantiating those variables as mentioned before in the actual definition of a CSP. Matlab has the
components to calculate such a matrix, which in some cases, would have more than 3-Dimensions. The
summation depth wise allowed the user to get the total probabilities at each box. Furthermore, this
allows displaying those results in a simpler and more meaningful way for the designer.

5
3. Examples

3.1. Example 1 Bolt shear stress resistance


A simplified model is shown in this section to give the reader an overview of the method. This example
will provide information along with few details of the different steps that constitute the global procedure
used in the main design project of section 3.2. The problem is simply expressed by using the “g-
function” defined by the reliability theory. The analysis is around one variable “g” defined as if g>0, we
have success but if g<0, we have failure.

The shear resistance of a bolt subject to shear force is given by [3] {13.12.1}

V୰ = 0.6 ∗ ϕୠ ∗ n ∗ m ∗ Aୠ ∗ F୳ (1)

where

• ߶௕ = 0.8
• n = number of bolts = ሼ1,2,3,4,5ሽ
• m = number of shear surfaces = 2
஠∗ୢమ
• Aୠ = effective area of the bolt assuming threads = 0.7 ∗

• F୳ = Tensile strength of the connected material = 450 MPa
• d = bolt diameter = ሼ16,18; 32ሽmm

The “g-function” is stated as follows

g = V୰ − V୤ (2)
where

• V୰ is deϐined in equation ሺ1ሻ.


• V୤ is a probabilistic variable with μ୚౜ = 350 kN and σ୚౜ = 50 kN

The other probabilistic variable is the tensile strength of the connected material (Fu) with normal
distribution parameters defined as follows:

μ୊౫ = 450 MPa and σ୊౫ = 25 MPa


The design plot will have an x-axis of the diameters of bolts (d) used and an y-axis of the number of
bolts (n). Note that those normal distributions are used in that example for sake of clarity. Normally
distributed variables are more easy to visualize thus this is the purpose of that simplified example.

3.1.1. Results
The main design plot is shown in figure 5 below. Since there are not many possibilities for the diameters
and the number of bolts, the boxes are large. In addition, those variables are discrete numbers, so it is
6
shown by using separations that referred to one discrete choice. For instance, if d=26mm and n=3, there
is a 0,65 probability of having success.

5.5

5 0.9

4.5 0.8

4 0.7
number of bolt (n)

3.5 0.6

3 0.5

2.5 0.4

2 0.3

1.5 0.2

1 0.1

0.5 0
16 18 20 22 24 26 28 30 32
bolt diameter (d) [mm]

Figure 5 Probability plot for bolt design

3.1.2. Analysis
Assuming the constraints reasonable, the plot presented shows us the possible combinations related to
different probability of success. The two normal distributed variable PDF plots are roughly represented
in the 2D plot by the curve. This curve shows a 3D relation to explain why those colors change
gradually. The greater the standard deviation would be for the variables, the largest the bright color band
would be.

Those possible outputs give guidelines to the designer. The numerical example used in the previous
section tells us that such combination will be satisfying if we accept 65% chance of success. Those
thresholds are fixed by whether the designer, the client or the society. The importance of the connection
regarding the effect of failure will guide the design process. In the previous example shown, if the
probability expected was 1 (which means 100% of chance of success), the upper part of the plot would
have to be used (dark red color). Whether 5 bolts of 25 mm or 3 bolts of 30 mm could be used,
thereafter; the cost and availability will restraint the choice. With the next example, we will see how the
cost could be included in this method.

7
3.2. Example 2 Bolted assembly (End plate connection)
This section will help visualize how helpful the hybrid method could be. The example is based on a
structural design problem often encountered by engineers; an end plate connection. Some assumptions
have been made to simplify the structural example and focus on the probabilistic issues. Here they are:

1. It is assumed that the weld is checked out of this conception. Thus, the resistance will not
involve the welding assembly. The cost will include a full length fillet weld.
2. The bolt hole pitches and edge distances are assumed to be constant as they are chosen for
the extreme cases encountered throughout this example.
3. The end plate thickness is implicitly chosen as 8mm.

Figure 6 End plate connection sketches

3.2.1. Problem Description


Like it has been said before, the first step is to define various variables and transformed the spreadsheet
used for calculations. It is the responsibility of the user to ensure that all constraints and formulas are
well documented and referred. The spreadsheet used is shown at the appendix 2 and are included, all
comments about previous assumptions.

The design variables are as follows

nThe number of bolt rows in the assembly (i.e rows of 2, see Figure 6)
dThe diameter of the bolts,

and the probabilistic variables are

Fy The yield strength of the steel (considering that the steel of the whole assembly are
the same)
FuThe tensile strength of the steel (same consideration as Fy)
VfThe factored shear stress applied to the assembly.

The range value used and the parameters for the probabilistic variables (along with their distribution
type) are display in Table 2.
8
Table 2 : Parameters for design and probabilistic variables
Design Probabilistic Fy Fu Vf
n d [mm]
variables variables (log-normal) (log-normal) (Normal)

Intervals [1,6] [16,32] Mean µ 5.938 6.227 850 kN

St. dev. σ 0.033 0.019 100 kN

The strength parameters have been discussed in section 2.1 of this report. The load Vf is assumed
normally distributed since at the preliminary step of a design, one usually know approximately which
load will be applied to the specific assembly. It is assumed that this specific load might be lower or
higher depending on the assembly solicitation. This design work is done at an early stage of the design,
and like we all doubt, people’s mind are always changing. The assembly has to be in some ways
prepared for those uncertainties. The choice of a standard deviation equal to 100kN which lead to a
coefficient of variation of more than 10% intend to represent the large uncertainties of the future
loading.

Those design variables are chosen by the user depending on how the optimization wants to be done. The
present example focuses on the bolt selection assuming their type is A490 with a Fub=1035 MPa. Their
size and quantity will be the display on each axis of the outcome.

The design is made to respect the limits establish by the CAN/CSA S16-01 [3]. The article indexes have
been removed from the spreadsheet to clear it up. The computations will lead to a probabilistic design
respecting all of the constraints imposed previously. Also, the most interesting output produced from
those computations is the design space showing the total cost of the design for all combinations of
variables. In relation with this plot, another constraint was added and state that the total cost calculated
will fall under a target cost. For instance, this constraint could be developed according to the budget and
the importance of the assembly for the structure stability,.

3.2.2. Computations
Matlab was chosen as the software platform for the development of the theory included in the referred
thesis. It is reasonable given that Matlab has all features capable of formulating loop to build plots of
higher dimensions. It has no constraint when dealing with multi-dimensional array as well as
representing in an easy visualized way the 2, 3 and 4D plots. This section will show some specific
coding used to produce the expected outputs.

3.2.2.1. Probabilistic function


Two special function files to calculate the CDF (Cumulative probability function) have been developed.
Originally, in the Loewen’s thesis, the normal distribution was used for the steel strength. However, the
log-normal is defined in this project as a better suitable choice. The cdf_lognorm.m is written as follows

9
function c = cdflog(x,xmean,xsd)

a=-8*(pi()-3)/(3*pi()*(pi()-4)); % parameter a

xsn=-1*(log(x)-xmean)/(xsd*sqrt(2)); % parameter for error


function

erfc=1-(sqrt(1-exp(-xsn^2*(4/pi()+a*xsn^2)/(1+a*xsn^2)))); %
error function

if log(x)>=xmean;

c=1-(1/2*erfc); % CDF

else

c=1/2*erfc; % CDF

end

As the program goes through the range of variables, arrays have to be used to define the minimum value,
step number and the maximum value. In a simple case of design variables, the formulation looks like
this

dv1=[0.5,6.5,6]; % number of bolt rows

dv1a=dv1(1)+(dv1(2)-dv1(1))*(0:dv1(3))'/dv1(3);

However, in the most complicated case of a log-normal variable, the range has to be well defined to get
all possible values on both side of the mean. It has been calculated with our specific range of values
from the set of data that more than enough values fall into the interval bounded by the mean ± three
standard deviations. The excel sheet used for that purpose is shown in Appendix 3. Here is shown the
associated Matlab syntax

beta=3;

mu1=5.938; % mu of ln(p1)

sd1=0.033; % stdev of ln(p1)

Ep1=exp(mu1+0.5*(sd1^2)); % mu of ln(p1)

Stp1=exp(mu1+0.5*sd1^2)*sqrt(exp(sd1^2)-1); % stdev of ln(p1)

pv1=[Ep1-beta*Stp1,Ep1+beta*Stp1,20,mu1,sd1]; % LOGNORM Yield strength

The length of the arrays defines the number of loop for the function calculating each probability of every
combination.

10
3.2.2.2. Solution module
The solution module will solve the constraints by the Grid-Search solver module developed in Matlab by
N.Loewen. All calculations are carried out, and a verification of the existences of the solution is made.
The average probabilities for each interval are estimated and added if they are consistent with the
constraints. It is done by assuming a target cost and assigning the value of either 0 or 1 to the truth
variable. In addition to the cost verification, throughout the loop, sub loop was implemented to verify
extra constraints related to the design itself. Those sub loops result to a total cost of 0 if the design
constraint is not met.

The outcome space expected has an easy readable shape. In this case, we are looking for a 2D plot. The
total probability of each probabilistic variables form a multi-dimensional array due to their specific
range of value. A sumdepth.m file function was developed by N.Loewen, and goes as follows

function matrix_out = sumdepth(matrix_in,ntot,nplot)

for i=ntot:-1:nplot+1;

matrix_out = sum(matrix_in,i);

matrix_in = matrix_out;

end

The matrix dimensionality is the number of the design variables added to the number of probabilistic
variables. The sumdepth function use the total probability of each boxes throughout the probabilistic
variables to display the total for each pixel related to the design variables. In accordance, it is needed to
use the sumdepth.m function as many times as the number of the probabilistic variables used for
computations.

3.2.3. Results
The main output awaited is a 2D plot scaled with the probability of meeting all constraints. Three
different plots are shown below to show the versatility of the method. Diversely, small divisions are
used for the two first plots in this example. The purpose will be explained in the following section. In
Figure 7, it is possible to see the probability of having a design that would resist to the structural
constraints. Any limits for the cost are disabled in the Matlab formulation.

11
Probability of a resistant design

6 0.9

0.8
5
0.7
Number of rows (n)
4 0.6

0.5

3 0.4

0.3
2
0.2

1 0.1

16 18 20 22 24 26 28 30 32
bolt diameter (d) [mm]

Figure 7 Design plot without cost target


Figure 8 shows the probability of an assumed 500$ cost target. The upper bound of feasible space that
represents the cost limit is a sharp line. Above this line, probabilistic variables are no longer applied to
the costing values.

Probability of a total cost < 500$

6 0.9

0.8
5
0.7
number of rows (n)

4 0.6

0.5

3 0.4

0.3
2
0.2

1 0.1

16 18 20 22 24 26 28 30 32
Bolt diameter (d) [mm]

Figure 8 Design plot with 500$ cost target


12
Finally, Figure 9 simply shows the cost for all possible assembly. The larger size chosen for the pixel
helps seeing the conceivable price for a specific combination of “n” and “d”.
Total cost of the assembly ($)

6 500

450
5
400

350
number of rows (n)

4
300

250
3
200

150
2
100

1 50

0
16 18 20 22 24 26 28 30 32
Diameter of bolts (d) [mm]

Figure 9 Total cost of the assembly

3.2.4. Analysis
The previous section display various output plots enabled to be produced by the hybrid method. Sizes of
the pixel could be changed at ease by the designer, and alternative choices could better suit depending
on the case. For the first two plots, the pixels are three times more refined than the last one. It could be
justified as follows; one could have an allowable success probability ranging between 60 and 100%,
regarding his own constraint and maybe his stress level in life (!). For that reason, the two first plot
display smaller pixel to see all the possible combination related to the previous range, for instance.
Smaller and smaller pixels would lead to smoother transition of colors, but will need a larger
computations time. The last plot purpose was to show the likely cost for each specific assembly,
therefore, the same number of boxes than possible combinations was chosen.

Those outputs could be easily used for design during the first phase of a structure design project. The
Figure 7 could be interesting to see which assembly resist to the applied force regardless of his cost.
Budget managing and low cost structures is a big issues these days, so the Figure 8 would probably be
more useful since it is possible to enter any cost target in the computations. A virtual cost target of 500$
was entered for the sake of the presentation.

13
4. Conclusions and Recommendations
By utilizing the method in diverse common structural design, more efficient decision could be made by
engineers at early stages. After computing those examples and analyzing them, the method proved itself
of being easy to carry on as it was promoted by N. Loewen. The two examples made for the purpose of
this brief project aim, by using the coding, to test proof and find possible weakness. No major problem
has been found, but some add-up has been designed as a benefit of this project.

Initially, in the Loewen’s thesis, data analysis was not included in the procedure. For sake of more
consistency in the presented results, this step was added and led to different probabilistic distribution for
some variables. Therefore, it was motivating to implement new coding to broaden the probabilistic
generator tool of this method and thus, to use a more realistic distribution for those variables. In
addition, the amount of constraints involve in the end plat connection differs slightly from all previous
example conducted with the Hybrid Method. Some design constraints leading to a cost result “nil” were
added throughout the coding. Fortunately, the output design space reacted well to this new feature. As a
result, design constrained by codes where maxima values are frequently imposed, could be carry on
without problem.

Finally, a recommendation regarding the probabilistic distribution is to add more distribution coding
available to the users. The most common distribution should be implemented for the method to be
suitable for a wider range of variables. Obviously, more concrete example containing different amount
of variables having different probabilistic distribution should be conducted to test the new implemented
coding. To broaden the type of variables that the method can handle, implementation of coding taking
into account correlation between some variables would be a great improvement. After studying the data
to make sure they were not correlated, I assumed it would be likely to be controlled by that limitation. In
addition, an interesting alternative to the example presented would be to include, as a design variable,
some component thickness in the structural example made for this project. It would produce a 3D output
that could guide the user throughout more decisions. Those 3D plot give even more information so it
would be interesting to carry on with the bolted connection example and expand it in more possible
ways.

5. References
Alpsten, G.A. (1972).“Variation in mechanical and cross sectional properties of steel.” National
Research Council of Canada, Lehigh University, Bethlehem, pp.775-805.

Bartlett, F.M., Hong, H.P. and Zhou, W (2003). “Load factor calibration for the proposed 2005 edition
of the National Building Code of Canada: Statistics of loads and load effect. Canadian Journal of Civil
Engineering, 30 : p.441

CAN/CSA S16-01 Limit States Design of Steel Structures.

Loewen, N., Vancouver.(2009). “Conceptual design using probabilistic interval constraint satisfaction”
PhD thesis. UBC.

14
6. Appendices

6.1. Appendix 1 – Partial set of data with details of their origin.

15
6.2. Appendix 2 - Spreadsheet calculations for an end plate connection

16
6.3. Appendix 3 – Log-normal distribution calculations

17

Potrebbero piacerti anche