CIEM5810

Engineering Risk, Reliability and Decision

Instructor: Prof. Anthony Leung

Topic 5: Reliability-based design

Introduction

Traditionally, civil engineering design is based on factor of safety (FOS).

Limitations:
1. FOS cannot explicitly consider the uncertainties in a design;
2. The same FOS does not imply the same level of reliability in different conditions

Transforming to reliability-based design

Method described herein is based on Taylor’s series of expansion. The key is to

derive equations for converting mean and nominal FOS to reliability index, 𝜷
Reliability-based design

Factor of safety method

Illustrating example

A FOS is used to consider all uncertainties in a design.

For instance, in the design of pile foundations, the limit state function is:

𝑅 = 𝑄𝐷 + 𝑄𝐿
In the FOS method, the required resistance is:
𝑅𝑛
= 𝑄𝐷𝑛 + 𝑄𝐿𝑛 , where 𝑛 represents the nominal value
𝐹𝑠𝑛
Analysis of uncertainties

In practice, neither the loads nor the resistance can be predicted accurately.

Suppose the mean of 𝑄𝐷 , 𝑄𝐿 and 𝑅 are 𝜇𝑄𝐷 , 𝜇𝑄𝐿 and 𝜇𝑅

the standard deviation are 𝜎𝑄𝐷 , 𝜎𝑄𝐿 and 𝜎𝑅

What is the reliability of the design for a nominal design FOS, 𝐹𝑠𝑛 ?
Uncertainties in loads

Define 𝑄 as the total load. It can be computed as

𝑄 = 𝑄𝐿 + 𝑄𝐷

Assume 𝑄𝐿 and 𝑄𝐿 are statistically independent, the mean and standard deviation of 𝑄
are calculated as follows:

𝜇𝑄 = 𝜇𝑄𝐿 + 𝜇𝑄𝐷 𝜎𝑄 = 𝜎𝑄2𝐿 + 𝜎𝑄2𝐷

The COV of 𝑄 can then be calculated as follows:

2 2 2
𝜎𝑄 𝜎𝑄 𝜇𝑄
2 +𝜎 2 𝐷 + 𝜇 𝐿 𝐿
𝜎𝑄 𝑄 𝜇𝑄 𝜇𝑄 𝐶𝑂𝑉𝑄2 +𝜌2 𝐶𝑂𝑉𝑄2 𝜇𝑄𝐿
𝜎𝑄 𝐿 𝐷 𝑄𝐿 𝐷 𝐿
, where 𝜌 =
𝐷 𝐷
𝐶𝑂𝑉𝑄 = = = 𝜇𝑄 =
𝜇𝑄 𝜇𝑄𝐿 +𝜇𝑄𝐷 1+𝜇 𝐿 1+𝜌 𝜇𝑄𝐷
Live load ratio
𝑄𝐷
Uncertainties in FOS

The FOS can be computed based on the total load, 𝑄

𝑅
𝑄 = 𝑄𝐿 + 𝑄𝐷 𝐹𝑠𝑛 =
𝑄

By Taylor’s series of expansion, the mean and standard deviation of 𝐹𝑠

2 2
𝜇𝑅 𝜇𝑅 1 𝜇
𝜇𝐹𝑠𝑛 = = 𝑉𝑎𝑟 𝐹𝑠𝑛 ≈ 𝜎𝑅2 + 𝑅4 𝜎𝑄2
𝜇𝑄 𝜇𝑄𝐿 +𝜇𝑄𝐷 𝜇𝑄 𝜇𝑄

Hence,
2
2 𝜇𝑅 2
𝜎𝑅 + 4 𝜎𝑄 2 2
𝑉𝑎𝑟 𝐹𝑠𝑛 𝜇𝑄 𝜎𝑅 𝜎𝑄
𝐶𝑂𝑉𝐹2𝑠𝑛 = 2 = 2 = 2 + 2 = 𝐶𝑂𝑉𝑅2 + 𝐶𝑂𝑉𝑄2 , Thus, 𝐶𝑂𝑉𝐹𝑠𝑛 = 𝐶𝑂𝑉𝑅2 + 𝐶𝑂𝑉𝑄2
𝜇𝐹 𝑠𝑛
𝜇𝑅 𝜇𝑅 𝜇𝑄
𝜇𝑄
𝜷 − 𝝁𝑭𝒔𝒏 relationship

Assuming 𝐹𝑠𝑛 follows lognormal distribution, the reliability index, 𝛽, is:

𝜇𝐹
ln 𝑠𝑛
1+𝐶𝑂𝑉2𝐹𝑠𝑛
𝛽= which gives the relationship between 𝝁𝑭𝒔𝒏 and 𝜷
ln 1+𝐶𝑂𝑉𝐹2𝑠𝑛

Note that:
1 2 2 𝜎2
𝜆 = ln 𝜇 − 𝜁 𝜁 = ln 1 + = ln 1 + 𝛿 2
2 𝜇2
Example 1

Determine the reliability index, by considering the following parameters

𝜌=0.2 𝐶𝑂𝑉𝑄𝐷 =0.07 𝐶𝑂𝑉𝑄𝐿 =0.29 𝐶𝑂𝑉𝑅 =0.2

𝐶𝑂𝑉𝑄2 +𝜌2 𝐶𝑂𝑉𝑄2

𝐷 𝐿
𝐶𝑂𝑉𝐹𝑠 = 𝐶𝑂𝑉𝑅2 + 𝐶𝑂𝑉𝑄2 = 𝐶𝑂𝑉𝑅2 + =0.340
1+𝜌

2.5
ln
1+0.3402
For a design value 𝜇𝐹𝑠𝑛 = 2.5, the reliability index: = 2.60
ln 1+0.3402
Implication

The same 𝜇𝐹𝑠𝑛 does

not imply the same
level of reliability

FOS is not a
consistent measure
of safety!
Target reliability index, 𝜷𝑻

• Superstructures, 𝜷𝑻 ≈3.5 (ASSHTO, 1994)

• Offshore piles, 𝜷𝑻 =1.4 ~ 3.0 (Tang et al. 1989, Wu et al. 1989)

• ASSHTO and Canadian National Codes recommendation for spread footings:

𝜷𝑻 =3.0~3.5

• ASSHTO1994 recommendation for single driven piles: 𝜷𝑻 = 2.0~2.5

• ASSHTO1994 recommendation for single bored piles: 𝜷𝑻 = 2.5~3.0

𝝁𝑭𝒔𝒏 − 𝑭𝒔𝒏 relationship

Based on the relationship between mean and nominal values, 𝑭𝒔𝑛 can be
calculated based on 𝝁𝑭𝒔 as follows:
𝜇𝑅 𝜇𝑅
𝑅𝑛 𝜆𝑅 𝜆𝑅
𝑭𝒔𝒏 = =𝜇 𝜇 =
𝑄𝐷𝑛 + 𝑄𝐿𝑛 𝑄 𝐷
+ 𝐿
𝑄 𝜇𝑄𝐷 𝜇𝑄𝐿
𝜆 𝑄 𝐷 𝜆 𝑄𝐿 + 𝜇 𝑄𝐷 + 𝜇 𝑄𝐿
𝜆 𝑄𝐷 𝜇 𝑄 𝐷 + 𝜇 𝑄𝐿 𝜆 𝑄𝐿 𝜇 𝑄𝐷 + 𝜇 𝑄𝐿
𝜇𝐹𝑠𝑛 1 + 𝜌 𝝁𝑭𝒔𝒏
= =
𝜇𝑄𝐷 𝜇𝑄𝐿 1 𝜌
𝜆𝑅 + 𝜆𝑅 +
𝜆 𝑄𝐷 𝜇 𝑄𝐷 + 𝜇 𝑄𝐿 𝜆 𝑄𝐿 𝜇 𝑄𝐷 + 𝜇 𝑄𝐿 𝜆 𝑄𝐷 𝜆 𝑄𝐷

where 𝝀𝑹 , 𝝀𝑸𝑫 and 𝝀𝑸𝑳 are the bias factor of the respective variables.
𝜷 − 𝑭𝒔𝒏 relationship

Hence, solving:
𝜇 𝐹𝑠
ln
1+𝐶𝑂𝑉2 1+𝜌 𝜇𝐹𝑠𝑛
𝐹𝑠 1+𝜌 𝐹𝑠𝑛
𝛽= (1) 𝐹𝑠𝑛 = 1 𝜌  𝜇𝐹𝑠𝑛 = 1 𝜌 (2)
ln 1+𝐶𝑂𝑉𝐹2𝑠 𝜆𝑅 𝜆 +𝜆 𝜆𝑅 𝜆 +𝜆
𝑄𝐷 𝑄𝐷 𝑄𝐷 𝑄𝐷

Substituting Eq. (2) into (1) gives the 𝜷 − 𝑭𝒔𝑛 relationship.

Statistics of design methods for piles
Reliability-based design

Method of reliability-based design

Three levels of design

Level 1: considering the joint distribution of random variables

(Monte Carlo Simulation)

Level 2: considering the first two moments of the random variables

(design point method)

Level 3: design based on partial factors

(limit state design; load and resistance factor design (LRFD))
Factored strength – European approach
Factored strength – European approach
Factored resistance – North American approach

• Comply with structural design

• Adopt in the latest Eurocode
• Consider different sources of uncertainty
Load and resistance
factor design (LRFD)
Load and resistance factor design (LRFD)

Geotechnical resistance: ∅𝑅𝑛 ≥ 𝜂 σ 𝛾𝑖 𝑄𝑖

where 𝑅𝑛 is nominal resistance; ∅ is resistance factor;
𝑄𝑖 is load effect; 𝛾𝑖 is load factor, defined in structural design
𝜂 is a factor to account for effects of ductility, redundancy and
operational importance, > 0.95

Service limit state: 𝛿𝑖 ≤ 𝛿𝑛

where 𝛿𝑖 is estimate displacement under the nominal load effect;
𝛿𝑛 is tolerant displacement;
Structural resistance factors for pile foundations (PDCA 1998)
Resistance factors for bearing capacity of shallow foundations (AASHTO 1996)
Resistance factors for axially loaded driven piles (AASHTO 1996)
Resistance factors for axially loaded bored piles (AASHTO 1996)
LRFD for pile foundation

A load or resistance factor is applied to each random variable for more accurate
evaluation of safety:
𝛾𝑅 𝑅𝑛 = 𝛾𝐷 𝑄𝐷𝑛 + 𝛾𝐿 𝑄𝐿𝑛 International
𝑅𝑛
= 𝛾𝐷 𝑄𝐷𝑛 + 𝛾𝐿 𝑄𝐿𝑛 China
𝛾𝑅
where 𝛾𝑅 is partial factor for 𝑅, or resistance factor
𝛾𝐷 is partial factor for 𝑄𝐷 , or load factor
𝛾𝐿 is partial factor for 𝑄𝐿 , or load factor
Key: how to determine the resistance factor (note: load factors are already
specified by structural codes)?
𝜸𝑹 − 𝝁𝑭𝒔 relationship

Given: 𝛾𝑅 𝑅𝑛 = 𝛾𝑄𝐷 𝑄𝐷𝑛 + 𝛾𝑄𝐿 𝑄𝐿𝑛

𝜇𝑅 𝜇𝑄𝐷 𝜇𝑄𝐿
𝛾𝑅 = 𝛾𝑄𝐷 + 𝛾𝑄𝐿
𝜆𝑅 𝜆𝑄𝐷 𝜆𝑄𝐿
𝜇𝑅 𝜇 𝑄𝐷 𝜇𝑄𝐿
𝛾𝑅 = 𝛾𝑄𝐷 + 𝛾𝑄𝐷
𝜇𝑄𝐷 + 𝜇𝑄𝐿 𝜆𝑅 𝜇𝑄𝐷 + 𝜇𝑄𝐿 𝜆𝑄𝐷 𝜇𝑄𝐷 + 𝜇𝑄𝐿 𝜆𝑄𝐿
𝜇𝑅 𝜇𝑄 𝐿
Since 𝝁𝑭𝒔 = , 𝜌=
𝜇𝑄𝐷 +𝜇𝑄𝐿 𝜇𝑄 𝐷

𝛾𝑄𝐷 𝜆𝑅 𝛾𝑄𝐷 𝜆𝑅 𝜌 𝜆𝑅 𝛾𝑄𝐷 𝛾𝑄𝐿

𝜇𝐹𝑠 = + = + 𝜌
𝛾𝑅 𝜆𝑄𝐷 1 + 𝜌 𝛾𝑅 𝜆𝑄𝐿 1 + 𝜌 𝛾𝑅 1 + 𝜌 𝜆𝑄 𝐷 𝜆𝑄𝐿
𝜸𝑹 − 𝜷 relationship
𝜇 𝐹𝑠
ln
1+𝐶𝑂𝑉2
𝐹𝑠
Since 𝛽 =  𝜇𝐹𝑠 = 1 + 𝐶𝑂𝑉𝐹2𝑠 exp 𝛽 ln 1 + 𝐶𝑂𝑉𝐹2𝑠
ln 1+𝐶𝑂𝑉𝐹2𝑠

𝜆𝑅 𝛾𝑄𝐷 𝛾𝑄𝐿
𝜇𝐹𝑠 = + 𝜌
𝛾𝑅 1 + 𝜌 𝜆 𝑄𝐷 𝜆 𝑄𝐿

These two equations of 𝝁𝑭𝒔 give the 𝜸𝑹 − 𝜷 relationship

𝜆𝑅 𝛾𝑄𝐷 𝜌𝛾𝑄𝐿 1
𝜸𝑹 = +
1+𝜌 𝜆 𝑄𝐷 𝜆 𝑄𝐿
1 + 𝐶𝑂𝑉𝐹2𝑠 exp 𝜷 ln 1 + 𝐶𝑂𝑉𝐹2𝑠
Example 2

Considering the following parameters:

𝜆𝑅 = 𝜆𝑄𝐷 = 𝜆𝑄𝐿 = 1.0; 𝛾𝑄𝐷 = 𝛾𝑄𝐿 = 1.0

𝜌 = 1, 𝐶𝑂𝑉𝑅 = 0.2, 𝐶𝑂𝑉𝑄𝐷 = 0.07, 𝐶𝑂𝑉𝑄𝐿 = 0.29

What should be the value of 𝜸𝑹 to achieve 𝜷 = 3.0

Based on the above data, 𝐶𝑂𝑉𝐹𝑠𝑛 = 0.214. Therefore the required resistance
factor can be computed as follows:
𝜆𝑅 𝛾𝑄𝐷 𝜌𝛾𝑄𝐿 1
𝜸𝑹 = + = 0.518
1+𝜌 𝜆𝑄𝐷 𝜆𝑄𝐿
1+𝐶𝑂𝑉𝐹2𝑠 exp 𝜷 ln 1+𝐶𝑂𝑉𝐹2𝑠
Reliability-based design

Determination of partial factors, or load and resistance factors

Methods

AFORM can be used to determine partial factors or load and resistance factors

Key idea: failure probability of a system is controlled by the design point (or
most probable point, MPP). Hence, a design should be safe if the reliability
index corresponding to the MPP meets the target

Steps:

1. Adjust the design to achieve the target reliability index

2. Conduct the design based on the resultant design point
Example 3: Determination of partial factors

Deep foundation example:

Considering the following limit state function:

𝑔 𝒙 = 𝑅𝑠 + 𝑅𝑇 − 𝑄𝐿 − 𝑄𝐷 = 0,

where 𝒙 = 𝑅𝑠 , 𝑅𝑇 , 𝑄𝐿 , 𝑄𝐷

Its design equation can be written as:

𝛾𝑅𝑠 𝑅𝑛𝑆 + 𝛾𝑅𝑇 𝑅𝑛𝑇 = 𝛾𝑄𝐿 𝑄𝑛𝐿 + 𝛾𝑄𝐷 𝑄𝑛𝐷

How to determine the load and resistance factors?

Determination of partial factors

Recall AFORM (c.f. topic 3, slide 10 – 26)

𝑝𝑓 ≈ 1 − 𝛷 𝛽 𝛽 = min 𝑛𝜌𝑥−1 𝑛
𝑔 𝑥 =0

𝒙∗𝒊
Let 𝒙* denote the design point, determine partial factors by 𝜸𝒊 =
𝒙𝒏𝒊

Given: 𝑄𝐿 ~𝐿𝑁 100, 20 , 𝑄𝐷 ~𝐿𝑁 300, 30 , 𝑅𝑆 ~𝐿𝑁 600, 180 , 𝑅𝑇 ~𝐿𝑁 200, 60

𝒙 = 𝑅𝑠 , 𝑅𝑇 , 𝑆𝐿 , 𝑆𝐷 , 𝑔 𝒙 = 𝑅𝑠 + 𝑅𝑇 − 𝑄𝐿 − 𝑄𝐷 < 0

𝜆𝑅𝑠 = 𝜆𝑅𝑇 = 1.0, 𝜆𝑄𝐿 = 1.15, 𝜆𝑄𝐷 = 1.08

Excel implementation
Determination of partial factors

(c.f. Topic 3, slide 17)

𝝁𝒙
𝒙𝒏 =
𝝀𝒙
𝒙∗
𝜸𝒊 =
𝒙𝒏𝒊
Observations

1. Effects of target reliability index

𝜷 𝑪𝑶𝑽𝑹𝒔 𝑪𝑶𝑽𝑹𝑻 𝝁𝑸𝑫 𝜸 𝑹𝒔 𝜸 𝑹𝑻 𝜸 𝑸𝑫 𝜸 𝑸𝑳

3.0 0.2 0.3 455.66 0.61 0.66 1.15 1.42
2.0 0.2 0.3 455.66 0.71 0.75 1.13 1.29

2. Effect of uncertainty of resistance

𝜷 𝑪𝑶𝑽𝑹𝒔 𝑪𝑶𝑽𝑹𝑻 𝝁𝑸𝑫 𝜸 𝑹𝒔 𝜸𝑹𝑻 𝜸𝑸𝑫 𝜸𝑸𝑳

3.0 0.2 0.3 455.66 0.61 0.66 1.15 1.42
3.0 0.3 0.3 316.1 0.45 0.67 1.13 1.33
Determination of partial factors

Generalisation

For a general geotechnical problem with the following limit state function:
𝑔 𝑥 = 𝑔 𝑥1 , 𝑥2 , … , 𝑥𝑛 = 0

Its design equation can be written as: 𝑔 𝛾𝑥1 𝑥𝑛1 , 𝛾𝑥2 𝑥𝑛2 , … , 𝛾𝑥𝑛 𝑥𝑛𝑛 = 0

where the resistance factor (𝛾𝑥𝑖 ) is connected to the nominal value (𝑥𝑛𝑖 ) and
design point (𝑥𝑖 *) as follows:
𝑥𝑖 ∗ 𝜇𝑥𝑖
𝛾𝑥𝑖 = recall the nominal value is connected to the mean value: 𝑥𝑛𝑖 =
𝑥𝑛𝑖 𝜆𝑥𝑖
Challenges of implementing LRFD in geotechnical engineering

Unlike man-made materials such as concrete and steel, soil is a naturally

occurring material. Its variability varies from one site to another. It also
depends on the efforts of site investigation.

Resistance factors are a function of the variability of the soil parameters. The
resistance factors will also vary from one site to another.

It is very difficult to recommend a unique set of geotechnical resistance factors

for all construction sites in a design code
Reliability-based design

Code calibration example

National Building Code of Canada (BNCC) (Becker 1996)
Calibration with working stress design

In LRFD format

∅𝑅𝑛 ≥ ෍ 𝛾𝑖 𝑄𝑖

In working stress design (WSD)

𝑅𝑛
𝐹𝑆 =
σ 𝑄𝑖

Hence,
σ 𝛾𝑖 𝑄𝑖
∅≥
𝐹𝑆 σ 𝑄𝑖
Calibration with working stress design

For a given FS, within a practical range of load ratio, ∅ varies within a narrow range
Calibration with the reliability theory

For log-normal distributed 𝑅 and 𝑆

2
𝜇𝑅 1+𝐶𝑂𝑉𝑄
ln 𝜇𝑅
𝜇𝑄 1+𝐶𝑂𝑉2 ln
𝑅 𝜇𝑄
𝛽= ≈ , for 𝐶𝑂𝑉𝑅 and 𝐶𝑂𝑉𝑄 ≤ 3
ln 1+𝐶𝑂𝑉𝑅2 1+𝐶𝑂𝑉𝑄2 𝐶𝑂𝑉𝑅2 +𝐶𝑂𝑉𝑄2

2
1+ 𝐶𝑂𝑉𝑅 /𝐶𝑂𝑉𝑄
Approximate 𝐶𝑂𝑉𝑅2 + 𝐶𝑂𝑉𝑄2 = 𝜃𝐶𝑂𝑉𝑅 + 𝜃𝐶𝑂𝑉𝑄 , where 𝜃 =
1+𝐶𝑂𝑉𝑅 /𝐶𝑂𝑉𝑄

𝜇𝑅
Hence, ln ≥ 𝛽 𝜃𝐶𝑂𝑉𝑅 + 𝜃𝐶𝑂𝑉𝑄 𝜇𝑅 exp −𝛽𝜃𝐶𝑂𝑉𝑅 ≥ 𝜇𝑄 exp 𝛽𝜃𝐶𝑂𝑉𝑄
𝜇𝑄

Compared with the LRFD format ∅ = 𝜆𝑅 exp −𝛽𝜃𝐶𝑂𝑉𝑅

Calibration with the reliability theory

1 𝜆
∅ = 𝜆𝑅 exp −𝛽𝜃𝐶𝑂𝑉𝑅 , then 𝛽 = ln 𝑅
𝜃𝐶𝑂𝑉𝑅 ∅
Note:
𝑉𝑅 = 𝐶𝑂𝑉𝑅 , 𝑘𝑅 = 𝜆𝑅

𝛽 is sensitive to
𝐶𝑂𝑉𝑅 and 𝐹𝑆
Calibration with the reliability theory

σ 𝛾𝑖 𝑄𝑖
∅≥ 1 𝜆𝑅 𝐹𝑆 σ 𝑄𝑖
𝐹𝑆 σ 𝑄𝑖 ቑ 𝛽= ln
𝜃𝐶𝑂𝑉𝑅 σ 𝛾𝑖 𝑄𝑖
∅ = 𝜆𝑅 exp −𝛽𝜃𝐶𝑂𝑉𝑅

1. The bias factor of resistance, 𝜆𝑅 , is 1.0~1.25 in general, and can be taken as

1.1 for average

2. COVR depends on soil properties and design methods, typical range 0.1~0.3
3. θ = 0.7~0.85 for COVR /COVQ = 0.2~5, can be taken as 0.75 for average
Relationships among 𝛽, 𝐹𝑆 and 𝛷

1 𝜆𝑅 𝐹𝑆 σ 𝑄𝑖
𝛽= ln
𝜃𝐶𝑂𝑉𝑅 σ 𝛾𝑖 𝑄𝑖

σ 𝛾𝑖 𝑄𝑖
∅≥
𝐹𝑆 σ 𝑄𝑖
Reliability-based design

Example: Reliability-based code revision for pile design in Shanghai

Design question

𝑅𝑛
= 𝛾𝑄𝐷 𝑄𝐷𝑛 + 𝛾𝑄𝐿 𝑄𝐿𝑛 (note the code difference in China; slide 25)
𝛾𝑅

National design code (1989): 𝛾𝑄𝐷 =1.2; 𝛾𝑄𝐿 =1.4

National design code (2002): 𝛾𝑄𝐷 =1.0; 𝛾𝑄𝐿 =1.0

Shangai design code (2000): 𝛾𝑄𝐷 =1.2; 𝛾𝑄𝐿 =1.4

𝛾𝑅 is calibrated based on local data: 1.6

Mismatch between local and national design codes

Comparison with other design codes

Implication of the Shanghai design code:

𝑅𝑛 𝛾𝑄𝐷 𝑄𝐷𝑛 + 𝛾𝑄𝐿 𝑄𝐿𝑛 𝛾𝑅 𝛾𝑄𝐷 + 𝛾𝑄𝐿 𝜌 𝛾𝑅
𝐹𝑠𝑛 = = =
𝑄𝐷𝑛 + 𝑄𝐿𝑛 𝑄𝐷𝑛 + 𝑄𝐿𝑛 1+𝜌
1.2 + 1.4 × 0.2 1.6
= ≈ 𝟐. 𝟎
1 + 0.2
Shanghai: FOS = 2.0
Hong Kong: FOS = 3.0
South Korea: FOS = 3.0

Shanghai design code too risky?

Implementation of code revision

Teams of expert include:

1. Researchers in geotechnical reliability
2. Local experienced geotechnical engineers

Code is revised based on consensus of all participating parties

New features of the code:

1. Both within-site variability and cross-site variability are considered

2. The safety level is lowered to achieve cost-effectiveness supported by
reliability theory
3. Information-sensitive partial factors are developed based on resistance
components
Characterisation of uncertainties

Design methods:

1. Static load test method: most reliable, yet expensive

2. Design table method: based on soil profile; no tests are needed

3. CPT method: Chinese CPT has a cone cross-sectional area of 15 cm2

Uncertainties in pile capacity

Probabilistic model
𝑅 = 𝑁𝑅𝑐 = 𝑁1 𝑁2 𝑅𝑐
where
𝑅 : Actual pile capacity
𝑁1 : Correction factor for
within-site variability

𝑁2 : Correction factor for

cross-site variability
𝑅𝑐 : Predicted pile capacity
Calibration datasets and calibrated uncertainties

Calibration datasets:

1. Driven piles: 146 piles from 32 sites, min. of 3 piles at the same site
2. Bored piles: 37 piles from 37 sites, min. of 3 piles at the same site

Calibrated uncertainties:
Comparisons with previous studies
Reliability of existing piles in Shanghai

Reliability index determined by the first-order second-moment method:

Load test method is indeed most reliable

Reliability of existing piles in Shanghai

Comparison with other studies

Reliability of existing piles in Shanghai

Comparison with other studies – why a large difference?

Different geology:

1. USA: A large country with varying geology

2. Hong Kong: Residual materials with large variability

3. S. Korea: Residual materials

4. Shanghai: Uniform sedimentary materials

Calibration and recommended resistance factors

Relationship between 𝛾𝑅 and 𝛽𝑇

First-order second-moment approximation:

𝜆𝑄𝐷 = 1.0 (bias factor for dead load) 𝜆𝑄𝐿 = 1.0 (bias factor for live load)
𝜌=0.2 (ratio of live load to dead load)
𝛾𝐷 = 1.0 (partial factor for dead load) 𝛾𝐿 = 1.0 (partial factor for live load)
𝜆𝑸 𝑫 𝜆 𝑄𝐿 𝑪𝑶𝑽𝑄𝐿 𝑪𝑶𝑽𝑄𝐿 References
1.00 1.05 0.10 0.18 Ellingwod et al. (1980)
1.05 1.00 0.10 0.25 Ellingwood and Tekie (1999)
1.08 1.13 1.15 0.18 Nowak (1999), ASSHTO (2007)
1.03-1.05 0.08-0.10 1.1-1.2 0.18 Nowak (1994), FHWA (2001)
1.00 1.00 0.07 0.29 Based on MOC (2002)
Effects of load factor on calibrated resistance factor

Targeted reliability index = 3.7

Driven piles Bored piles

Load
factors Design Design
Load test CPT Load test
table table
MOE 𝜆𝑸𝑫 = 1.0
1.53 1.93 1.72 1.56 2.29
(2002) 𝜆𝑸𝑳 = 1.0

MOE 𝜆𝑸𝑫 = 1.2

1.24 1.61 1.40 1.26 1.77
(1989) 𝜆𝑸𝑳 = 1.4
Recommended partial factors

Driven piles Bored piles

Load
factors Design Design
Load test CPT Load test
table table
𝜆𝑸𝑫 = 1.0
Calibrated 1.53 1.93 1.72 1.56 2.29
𝜆𝑸𝑳 = 1.0
Engineering judgement
𝜆𝑸𝑫 = 1.0
Suggested 1.80 2.00 2.00 1.90 2.00
𝜆𝑸𝑳 = 1.0

𝜆𝑸𝑫 = 1.0
𝛽 5.09 3.75 4.67 5.34 3.29
𝜆𝑸𝑳 = 1.0
End of lecture note