Probability and statistics
Prediction of behavior from the past data of well-understood
processes is a critical step in many tasks of the design process
Probability study is the study of a known population of data to
predict the characteristics of a single observation from a
population or sets of observations (samples) taken from the
population in a meaningful way
A population is the total set of objects or observations about which
information is sought
A sample is a subset of the population created by randomly selecting
members from the set
A random sample means each member of the population has an equally
likely chance of being included in the sample
Probability and statistics
Engineers use the principles of probability to create simulation
models of events that occur with a known probability distribution
Statistics is the study of a sample of observations from a
population to make inferences about the entire population
An engineer takes samples from a batch of manufactured parts for
measurement in order to infer key characteristics of the entire batch of
parts
Descriptive Parameter Formulas in statistics
Probabilistic Approach to Design
Conventional engineering design uses a deterministic approach
It doesn’t consider the fact that material properties, the dimensions of the
parts, and the externally applied loads vary statistically
In conventional design these uncertainties are handled by applying a factor of
safety
In critical design situations such as aircraft, space, and nuclear
applications, it is often necessary to use a probabilistic approach
Quantifies the uncertainties better, and thereby increases reliability
There are three approaches for incorporating probabilistic effects in
design. In the order of increasing sophistication, they are:
The use of a factor of safety
The use of the absolute worst case design
The use of probability in design
Basic Probability Using the Normal Distribution
A probability distribution is a table or an equation that links
each outcome of a statistical experiment with its probability of
occurrence
Many physical measurements follow the symmetrical, bell-shaped
curve of the normal, or Gaussian frequency distribution
The distributions of yield strength, tensile strength, and reduction of
area from the tension test follow the normal curve to a suitable degree
of approximation
Basic Probability Using the Normal Distribution
The equation of the normal curve is :
f (x) is the height of the frequency
curve corresponding to an assigned
value, x
μ is the mean of the population, and
σ is the standard deviation of the
population
Normal distribution extends from
x=-∞ to x=+∞ and is symmetrical
about the population mean, μ
The existence of long “tails” makes the
normal distribution a poor model for
certain engineering problems
Standardized normal frequency distribution
Normal distributions are standardized by expressing the normal curve
in terms of the standard normal variable,
The equation of the standard normal curve becomes,
; ( f (z) or P(z) is the probability density function)
Standardized normal frequency distribution
For the standardized normal
curve, μ = 0 and σ = 1
Total area under the curve is unity
Probability of a value of z falling
between z = -∞ and a specified
value of z is given by the area
under the curve
Probability is the numerical
measure of likelihood of an event
P = 0 for an impossible event, and
P = 1 for a certain event
Standardized normal frequency distribution
The area under the curve from
z = -∞ to z = -1 is 0.1587
So the probability of a value falling
into that interval is P = 0.l587, or
15.87 percent.
Since the curve is symmetric, the
probability of a value falling into
the interval z = ‒1 to z = +1 or
μ ± σ is 1.0000 ‒ 2(0.1587) = 0.6826
In a similar way it can be shown
that μ ± 3σ encompasses 99.73
percent of all values.
Areas Under Standardized Normal Frequency Curve
Some Examples
If z = ‒3.0, the probability of a
value being less than z is
0.0013 or 0.13 percent
The percentage of values
greater than this z is
100 ‒ 0.13 = 98.87 percent
If we wanted to exclude the
lowest 5 percent of the
population values we would set
z at ‒1.645
Area under the Cumulative Distribution Function for z
Area under the Cumulative Distribution Function for z
Area under the Cumulative Distribution Function for z
Area under the Cumulative Distribution Function for z
Example-1
Example-1 (Contd)
Frequency Distributions Combining Applied Stress
and Material Strength
Consider a structural member subjected to a static load that develops a stress σ
The variation in load or sectional area results in the distribution of applied stress
shown in the figure below
Here, the mean is denoted by and the standard deviation by s
Yield strength of the material Sy has a distribution of values given by
(mean) and sy (standard deviation)
The two frequency distributions overlap, and it is possible that , which
is the condition for failure
Frequency Distributions of yield strength Sy and stress σ
Frequency Distributions Combining Applied Stress
and Material Strength (Contd)
Probability of failure is, and Reliability R is,
If the stress distribution is subtracted from the strength distribution, the
distribution obtained is,
The distribution Q can be constructed by performing algebraic operations on
the two independent random variables Sy and σ
Mean and standard deviation of the distribution (subtracted) can also be determined
Here, and sQ (for this example)
Frequency Distributions… (Contd)
The part of the distribution to the left of Q = 0 represents the area for which
is negative, i.e., , and failure occurs.
Transforming to the standard normal variable, ,
at Q = 0,
From a table of cumulative normal distribution, it is seen that 0.16 of the area
falls between ‒ ∞ and z = −1
Thus, the probability of failure is, Pf = 0.16
And, the reliability is, R = 1 − 0.16 = 0.84
This is not a satisfactory situation .
If a stronger material with and z = ‒2 is selected, the
probability of failure ≈ 0.02
Variability in Material Properties
Mechanical properties of engineering materials exhibit variability
Fracture and fatigue properties show greater variability than the static tensile
properties of yield strength and tensile strength
Most published mechanical property data do not give mean values and
standard deviations
Published mechanical property data without statistical attribution is
usually taken to represent a mean value
If a range of values is given, the lower value is often taken to represent
a conservative value for design
Not all mechanical properties are normally distributed
A normal distribution is a good first approximation that usually results in a
conservative design
Variability in Material Properties (Contd)
When statistical data are not available the standard deviation can be
estimated by assuming that the upper xU and lower xL values of a
sample are ± three times the standard deviation from the mean
When the range of property values is not given, it is still possible to
approximate the standard deviation, using the coefficient of
variation,δ , which is a measure of the uncertainty of the value of the
mean
Coefficient of variation is different for each mechanical property, but it
tends to be relatively constant over a range of mean values
