Scribd
Upload
32 views

2.5.3 Structural Uncertainty

Structural uncertainty arises from a lack of knowledge about the true physical system. Mathematical models are used to describe real systems, but are not fully representative and can only predict results approximately. Other sources of uncertainty include the future behavior of loads on structures, material properties, and the understanding and modeling abilities of engineers. Monte Carlo simulation and Latin hypercube sampling are statistical techniques used to simulate risks and generate models of predictions containing uncertainties. Bayes' theorem provides a framework for updating probabilities as new data is incorporated using prior probabilities, likelihoods, and posteriors. Generative models show how observed data arises from underlying causes and processes.

Uploaded by

ChrisYap
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
32 views

2.5.3 Structural Uncertainty

Structural uncertainty arises from a lack of knowledge about the true physical system. Mathematical models are used to describe real systems, but are not fully representative and can only predict results approximately. Other sources of uncertainty include the future behavior of loads on structures, material properties, and the understanding and modeling abilities of engineers. Monte Carlo simulation and Latin hypercube sampling are statistical techniques used to simulate risks and generate models of predictions containing uncertainties. Bayes' theorem provides a framework for updating probabilities as new data is incorporated using prior probabilities, likelihoods, and posteriors. Generative models show how observed data arises from underlying causes and processes.

Uploaded by

ChrisYap
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 7

2.5.

3 Structural Uncertainty

Warren (2006) said that the structural uncertainty arise due to the lack of the
knowledge about to the true physical. For real life situation, the mathematical model is
used to describe the true system. It is obvious that the high degree of accuracy of the
mathematical model will show the low degree of uncertainty. Although the
mathematical model possess high accuracy, but the mathematical model is not fully
represent the reality. It is only almost always near to the reality system. According to
Bulliet, time is one of the elements in structural uncertainty. This element focuses on
the uncertainty in predicting the future. For example, the load exert by the mountainous
waves toward the marine structure. Also, the uncertainty of the old structure included
strength, porosity, moisture content and the health of the old concrete structure. All of
these elements are only able to predict and obtain an approximately result.

2.5.4 Model Uncertainty

Based on the statement made by Cont (2006), the model uncertainty of structure
model may arise due to the model simplified the analysis and many aspect of the
structure. This is due to the structure made by a lot of material and design should
consider of many aspect such as environmental condition, quality material and etc.
There are two types of model uncertainties:
1) Uncertainty related to the prediction equation models and the data.
2) Uncertainty related to how effective a structure model predicts the behaviour of
the structure.
According to Jalayer (2010), the two types of the model uncertainties are similar. But
the uncertainty related to the prediction equation models and the data may incorporate
with the resistance factor. Notice that the prediction equation will multiply by a model
error factor during the structural reliability analysis. The model error factor is
represented in the form of random variable. If the experiment data is sufficient about the
structure, then the reliability analysis may corporate with the distribution model error.
The second model uncertainty is related to how effective a structure model predicts the
behaviour of the structure. In this aspect, the effective of the structure model is
influence by the structure engineer. The reason that turns this uncertainty out is the

understanding of the engineer on the modelling technique being used. The more effort
the engineer affords the high degree of accuracy of the structure model.

2.5.5 Heuristics Uncertainty


Koen (2003) said that a heuristic as anything that provides a plausible aid or
direction in the solution of a problem but is in the final analysis unjustified, incapable of
justification, and potentially fallible. Heuristic in structural can refer as a guideline or a
technique used by engineer to solve the problem in order to perform the design of
building structure. For example, the dynamic of the wind load may be ignored in most
building design especially the low rise building. Although, during design of the low rise
building, only static wind load will be used to design the structure while the dynamic of
the wind load was ignored. This is a form of heuristics uncertainty as the engineer not
able to assure the static wind load is enough to resist in the wind load in reality.
Heuristic is very important in structure design, there are no problem if the engineer able
to recognize the limit. In the fact, most of the engineers are using the heuristics without
thinking about them.

2.5.6 Human Error

The collapsed of a building structure may refer to the reasons of model


uncertainty and heuristic uncertainty. But, there is another factor should not be forgot
i.e. human error. In the reality, the combination of model error, improper use of
heuristics and human error will cause the high probability of failure. The human error
dominates among the uncertainties. The reason is the model and heuristics was used is
selected by human, and the human cause the failure of structure due to the wrong
selection of heuristic and model.

2.5.7 Clinical Uncertainty

Clinical uncertainty mention that the lack of the clinical evidence from the past
(Bojke, et al., 2006). The insufficient clinical evidence data produced a model that not
able to give the good prediction, probability of failure and expert opinion in order to do

the decision making. Without the sufficient clinical evidence data, the prior model
would not be updated well as the new data incorporated. These mention that the prior
model is very unclear and insensitive. It is obvious that the model produced by
sufficient clinical evidence data able to shown a good prediction and expert opinion
compare to the model that produced by insufficient clinical evidence data. Thus, the
insufficient clinical evidence data strongly influence the decision making.

2.5.8 Monte Carlo Simulation

Monte Carlo simulation is a risk analysis sampling method. By using Monte


Carlo simulation, the random samples will be generating. Monte Carlo simulation is
very useful in stimulates the risk in order to generate a diagram or modelling as stated
by Blanchard (2006). According to Lapeyre (2007), Monte Carlo simulation is a
technique used to understand the risk may arise when making a prediction in
management, financial and cost. When making a forecasting model, the actual value is
not possible to obtain. The reason is the predictions contain of uncertainties that human
not possible to predict. However, the prediction may make within a range. The range
exists based on the historical data, experience and expertise in the field. The prediction
may make in the range in order to reduce the error and make the prediction more
accurately.
The range of estimation that understand by human enable they understand the
risk and uncertainty in the case that need to be predicted. With the presence of the range
of estimation, the Monte Carlo simulation may able to contribute in the prediction.
Monte Carlo simulation will inform that how likely of a outcome will result in the range
of estimate that understood by human.
In Monte Carlo simulation, the range of estimation for each case is represented in
random variable value. Based on the random variable value, the model is calculated and
the result will be recorded. The process will be repeated and each time will use different
selected value.

Figure 2.8: Hit and miss data

Source: University of Southern California 2007

In figure 2.8, it is describes the hit and misses data during the simulation process. The
random variable value will make a boundary to separate the hit region (inside boundary)
and miss region (outside of boundary). During the simulation, if the selected value is
located inside the boundary, the data will be selected. However, the simulation will stop
and start again if the miss data obtained.

2.5.9 Latin Hypercube Sampling Method

Sampling is a process that drawn a value based on the random variable value.
The sampling in simulation will done repetitively for every iteration from the
probability density function which generated by the random variable value. In the figure
2.9, it is describe the Latin Hypercube sampling method.
Based on the figure 2.9, the curve has been divided to 5 intervals. The sample will be
drawn from each interval during the sampling. This sampling method reflects the
sample from each interval to make it more equally and accurately. The number of
stratifications is equal to the number of iteration has been set at the beginning of the
simulation. If the iteration is 100000 times, there will be 100000 stratifications in the
curve.

Figure 2.9: Latin Hypercube Sampling Method

Source: Universitetet I Oslo 2006


2.5.10 Bayes Theorem

Olshausen (2004) said that the Bayesian probability is a mathematical


framework to describe the probability with reasoning. The Bayes Theorem was been
used by many scientist such as Bernuolli since 200 years ago. The Bayes theorem is
often used to judge the hypothesis and uncertainties data. Also, it is used to adjust the
parameter of a model.
In Bayes rule, two events should occur and denoted as event A and B. This can
be expressed in:
P(AB) = P(AB).P(B)

(2.10)

= P(BA).P(A)

(2.11)

In Bayesian probability, event A is hypothesis and denoted as H. On the other hand,


event B is data from experimental and denoted as D. The relation may express as
follow:

P(HD) =

) ( )
( )

(2.12)

P(DH) is known as likelihood function, it is used to judge the probability arise


from the hypothesis. The likelihood is make from the experimental and it will modified
the probability arise from hypothesis to a true probability. On the other hand, P(H) is
call prior and it is describe the hypothesis system without consider any data. P(D) is
used to integrate the function P(DH) and P(H). Normally P(D) act as an normalizing
constant and ignorable. P(HD) is known as posterior which reflect the probability after
consider the data. Bayesian probability is very useful in transform the prior, P(H) to
posterior P(HD) by incorporating new data.
Generative model is one of the important aspect should understand in Bayes
theorem in order to make the inference toward a single model form a large number of
data.

Figure 2.10: Generative model


Source: Olshausen 2004
The observed data D is a set of data obtained due to the causes, . In this case,
the data D may represent the corrosion at certain depth and data at the right may be the
corrosion process. The link is representing how the corrosion process gives rise of the
corrosion at certain depth of concrete cover and how the process interact with each
other.

You might also like