Journal Pre-proofs

Consideration of random loading processes and scatter of fatigue properties

for assessing the service life of welded bus bodyworks

Miloslav Kepka, Miloslav Kepka Jr.

PII: S0142-1123(21)00184-5
DOI: https://doi.org/10.1016/j.ijfatigue.2021.106324
Reference: JIJF 106324

To appear in: International Journal of Fatigue

Received Date: 23 November 2020

Revised Date: 10 May 2021
Accepted Date: 13 May 2021

Please cite this article as: Kepka, M., Kepka, M. Jr., Consideration of random loading processes and scatter of
fatigue properties for assessing the service life of welded bus bodyworks, International Journal of Fatigue
(2021), doi: https://doi.org/10.1016/j.ijfatigue.2021.106324

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2021 Published by Elsevier Ltd.

Consideration of random loading processes and scatter of fatigue properties

for assessing the service life of welded bus bodyworks

Miloslav Kepka1), Miloslav Kepka Jr.1)

1) Regional Technological Institute, R&D centre of Faculty of Mechanical Engineering, University of West

Bohemia, Univerzitni 2732/8, 301 00 Pilsen, Czech Republic)

ABSTRACT

As in the field of passenger cars, the development of a new bus (trolleybus, electric bus) includes computational

and experimental activities: CAD - MBS - FEM - stand tests - measurements with prototypes - fatigue life

calculations. Current challenges include the rise of electromobility and battery-powered electric buses, which open

up new fields of research. Heavy batteries significantly change both the vehicle's dynamics and the stresses in the

bodywork and undercarriage frame. The main structures of bus bodyworks are welded from thin-walled profiles.

In addition to conventional structural steels and, in the case of some manufacturers, also aluminium alloys, stainless

steels and high-strength steels are beginning to be widely used. If a manufacturer decides to develop a conceptually

new vehicle, it initially does not have the exact input information necessary to assess the strength and fatigue life

of a new bodywork. Therefore, the challenge is to analyse the experience from the development and operation of

the previous generation of vehicles and to introduce new research methods. One of these trends is to estimate the

service fatigue life and future reliability of the bodywork at an early stage of its development. The paper briefly

describes the process of development of a new bus bodywork and summarizes the general assumptions for

assessing the fatigue life of welded bodywork nodes based on the concept of nominal stresses. In several specific

case studies, the variable amplitude loading and the scatter of fatigue properties of welded bodywork nodes are

considered.

KEYWORDS
bus bodywork, MBS model of vehicle, FEM model of bodywork, welded bodywork node, S-N line, stress-time

history, design stress spectrum, fatigue damage calculation, permissible maximum stress amplitude, fatigue life

distribution function

NOMENCLATURE

𝐴 ratio of the ordinary service life and the life achieved in the accelerated test

C0 parameter defining the mean line S-N relationship

𝑑 number of standard deviations 𝑠 below the mean S-N line

𝐷 fatigue damage

𝐷𝑐 limit fatigue damage

𝑓 frequency

ℎ𝑖 cumulative number of cycles with amplitude 𝜎𝑎𝑖

𝐻𝑚𝑎𝑥 number of cycles having amplitude 𝜎𝑎,𝑚𝑎𝑥

𝐻𝑡𝑜𝑡 total number of cycles inside spectrum

𝐿1 travelled distance, in which the maximum stress amplitude 𝜎𝑎𝑚𝑎𝑥 occurs once

𝐿𝐶 estimated fatigue life under city conditions

𝐿𝑑 design life of the bodywork

𝐿𝑇𝐺 estimated fatigue life under service on polygon testing grounds

𝑁𝑐 number of cycles at knee point of S-N line

𝑁𝑓 number of cycles to limit fatigue stage

𝑛𝑖 number of cycles on i-stress level with amplitude 𝜎𝑎𝑖

𝑁𝑖 number of cycles at loading 𝜎𝑎𝑖 - read-out from S-N line

𝑃 probability of occurrence of the maximum stress amplitude 𝜎𝑎𝑚𝑎𝑥

𝑠 standard deviation of log 𝑁𝑓, also shape parameter of stress spectrum

𝑣 average speed

𝑤 inverse slope of the S-N line

𝑤𝑑 inverse slope of the lower part of S-N line according to Haibach

𝜎𝑎, 𝑚𝑎𝑥 maximum stress amplitude

𝜎𝑎, 𝑚𝑎𝑥,𝑝 permissible maximum stress amplitude

𝜎𝑎 stress amplitude

𝜎𝑎𝑖 stress amplitude on i-level of loading

𝜎𝑐 so called fatigue limit

1 INTRODUCTION

The development methodologies of all vehicle manufacturers combine computer modelling and computational

simulations with a suitable experimental program at the level of the vehicle's structural system and its components

and structural nodes. The methodology which was used for designing many trolleybuses and buses has already

been presented to the public, including at the VAL2 conference [1]. This paper is an extended version of

contribution [2] presented at the VAL4 conference. The presented methodology is highly complex. The

development of a new bus structure involves both computational and experimental activities: CAD, MBS, FEM

calculations, stand test and measurement with bus prototypes. The final activity is the fatigue life assessment.

The methodology uses the advantages of virtual prototyping in combination with various tests in laboratories and

testing of vehicles on real roads and test tracks including special roads on test polygons, Fig. 1.

Virtual prototype Real prototype

Laboratory tests

Multibody simulations

Fatigue life Stand tests

evaluation

FEM simulations
Track tests
 Fig. 1. Methodology for fatigue life evaluation of bus bodyworks.

Development of buses, trolleybuses and newly also battery electric buses involves the following conditions which

shall be met in all relevant structural details of the vehicle body:

𝜎𝑎, 𝑚𝑎𝑥 ≤ 𝜎𝑎, 𝑚𝑎𝑥,𝑝 (1)

𝜎𝑎, 𝑚𝑎𝑥 is maximum amplitude of stress response determined for the structural detail when the vehicle rides over a

standardized obstacle which simulates a severe irregularity in a road surface, 𝜎𝑎, 𝑚𝑎𝑥,𝑝 is permissible maximum

stress amplitude acting on the structural detail. This condition must be met at all stages of vehicle development

(projecting and design of vehicle; investigation of vehicle function sample on test stand; measurement of vehicle

prototype).

In the past, the fatigue strength at the knee point of S-N line (so called fatigue limit [3, 4]), was considered as the

maximum permissible value. Thus, the design concept of permanent fatigue strength (infinite fatigue life), was de

facto applied. However, the weight of the bodies was high and the pressure of transport companies to increase the

occupancy of the vehicle grew. Hand in hand with this came the requirements for guaranteed fatigue life (safe life

approach). It was necessary to allow the operational load of bodywork nodes above the so called fatigue limit and

use fatigue damage calculations, estimate the fatigue life and compare it with the required design life. The process

of determining the maximum permissible value 𝜎𝑎, 𝑚𝑎𝑥,𝑝 is shown in Fig. 2.

 Fig. 2. Schematic representation of the procedure

to determine the permissible maximum stress amplitude 𝜎𝑎, 𝑚𝑎𝑥,𝑝.

The process of determining the maximum permissible value 𝜎𝑎, 𝑚𝑎𝑥,𝑝 for safe life approach is based on the concept

of linear accumulation of fatigue damage. The input data to the fatigue damage calculation are: S-N line of the

assessed structural node and stress spectrum in its critical cross-section. The stress spectrum represents a random

loading process over the required service life of the vehicle. The calculation of fatigue damage can be performed

with consideration of various boundary conditions, according to hypotheses proposed by various authors, see

Chapter 2.1. The parameters describing the S-N line of the assessed structural node can be specified relatively

accurately already at the design stage of the new body (most accurately on the basis of laboratory fatigue tests),

see Chapter 2.2. The stress spectra can only be determined accurately on the basis of measurements with a vehicle

prototype. In the stage of initial dimensioning, it is necessary to estimate the stresses of the load-bearing profiles

and welded bodywork joints during their service life. The parameters of the so-called design stress spectra can be

estimated on the basis of empirical experience, see Chapter 2.3. Then, the dependence of the fatigue damage on

the magnitude of the stress can be constructed. The maximum permissible value 𝜎𝑎, 𝑚𝑎𝑥,𝑝 can be deduced from this

dependence for the critical value of fatigue damage. The practical example is performed in the chapter 3.1 - Case

study 1.
Alternative design variants are compared using computational models. For a variety of reasons, only a limited

number of variants are considered. Dynamic models of the vehicle are constructed on the basis of drawings,

projected technical parameters of the vehicle and, finally, on CAD data together with specific information about

geometric and material data of all important components of the vehicle. Key assemblies of the vehicle (body, axles,

wheels and tyres, suspension and guiding elements and others) are modelled using multibody simulation (MBS)

software. The characteristics of tyres, shock absorbers and air springs are very important. The choice of these

suspension elements (and their combinations) has a strong impact on dynamic properties of the vehicle and the

levels of dynamic stresses in the body structure. This fact is demonstrated on an example of specific data by the

Case study 2 in Chapter 3.2.

Using the dynamic MBS model of an empty and fully-loaded vehicle, vehicle ride over standardized obstacles

(with left wheels, right wheels, both wheels on a single axle simultaneously) at a chosen speed is simulated.

Combinations of different variants lead to a wide range of load states. A prominent obstacle on the road surface

(kinematic excitation) is simulated using a cylinder segment 500 mm in width and 60 mm in height. The same

artificial obstacle is also used to measure the stresses of the bodywork nodes when driving on the model test track

and is shown in Fig. 16 and Fig. 17 in case study 2 in Chapter 3.2.

First, the relative movements and velocities between the body and axles in response to excitation of this kind are

examined. Then, using the known suspension characteristics, force-time histories in individual suspension

elements are derived.

The computational model of the body structure is constructed using a finite element method (FEM). It should

provide a sufficiently accurate description of the stress state in the structure, and facilitate dynamic calculations

within reasonable computing times. The vehicle body FEM model is then subjected to variable forces acting on

the suspension elements, as determined by the multibody simulations. Then, stress responses for all important

structural details of the body are calculated for the investigated load conditions, therefore for empty and fully-

loaded vehicle [5]. The maximum stress amplitudes are identified and compared with the maximum acceptable

values 𝜎𝑎, 𝑚𝑎𝑥,𝑝.

The case studies presented later in this article concern city buses (including battery buses, trolleybuses). These

vehicles move around the city at low speed and similarly start, brake and turn. The influence of the longitudinal

and transverse forces induced by these driving manoeuvres is considered only in the dimensioning and evaluation

of the axles and associated components. Vertical excitation is crucial for body stress. Fatigue damage of the welded
bodywork joints causes variable stress when driving the vehicle on uneven road surfaces, which are often of

varying quality. The payload of the vehicle (passenger load) is also very variable. From the point of view of the

bodywork, it has proven to be sufficiently effective to concentrate only two calculated load cases: direct driving

on uneven roads with an empty and fully loaded vehicle. At the request of customers, vehicle overload, which

occurs in real traffic, was sometimes assessed. For example, a 20% exceedance of the occupancy of the vehicle

interior was considered. From many computer simulations, verified by measurements, the following practical

knowledge emerged. While the undercarriage components are logically more stressed with a more loaded (more

occupied) vehicle, this is not the case for the bodywork. The bodywork often has the largest stress amplitudes

(stress ranges) when driving an empty vehicle.

Measurements in real city traffic or while the vehicle is driving on a mix of different roads on the test site confirmed

that the stress spectra of bodywork nodes can be reasonably approximated, see Chapter 2.3, case study 3. This

finding is preferably used to generate (estimate) so-called design stress spectra in stage of initial dimensioning of

bodywork profiles and their structural nodes (computational solution). Of course, this is always an estimate, but it

approximates well the future random service loading. This is refined in the later stages of vehicle development,

based on systematic strain gauge measurements.

Bus bodyworks are welded from thin-walled (usually closed) steel or aluminium profiles. Their thickness is usually

2, 3, 4, exceptionally 5 mm. On the contrary, various standards and regulations for welded structures contain

recommendations for weldments made of plates with a thickness of 5 mm and more. Nevertheless, most of these

recommendations can be accepted. Generally, the fatigue strength of welded structural details can be assessed

using three basic approaches: nominal, structural/hot-spot and local ones.

An effective tool for an analysis of a complex structure, such as a bus bodywork, is a computational mesh for

nominal stress analysis. The FEM body model is made of beam elements. Dynamic calculations are performed in

the usual way. First, the natural frequencies and the corresponding body vibration shapes are calculated.

Subsequently, calculations of forced vibrations of the structure are performed. The excitation forces are taken from

the dynamic MBS model. By suitable postprocessing, time series of stress components from the acting axial forces

and moments can then be calculated for each element of the FEM model, i.e. for its nodes. From them, the resulting

time histories of nominal stresses at specified points of the bodywork profiles are compiled, see the example in

Fig. 3.
 Fig. 3. Scheme of evaluation of nominal stresses in cross-section of structural node.

Then, of course, the fatigue resistance S-N lines of the considered structural details must be based on the nominal

stress, disregarding the stress concentrations due to the welded joint. Therefore, the measured nominal stress must

exclude the stress or strain concentration due to the corresponding discontinuity in the structural component. Thus,

strain gauges must be placed outside the stress concentration field of the welded joint [4].

In practice, it may therefore be necessary to first evaluate the stress gradient and determine the field with the stress

concentration due to the welded joint. This is done by means of strain gauge measurements during laboratory

fatigue tests of structural nodes, supported by detailed FEM calculations, see examples in Figures 4 and 5. For

measurements on the vehicle, a simple application of a strain gauge outside this field is sufficient.
 Fig. 4. Example of detailed FEM calculation of stresses around the weld of bodywork profiles

(simulation of loading applied at laboratory fatigue test).

Fig. 5. Example of installation of strain gauges around the weld of bodywork profiles at laboratory fatigue test.
Sometimes a more accurate possibility of measuring the structural stress in a hot spot is considered, but

methodically correct installation of strain gauges on thin-walled profiles of a fully finished bus is usually not

possible due to external plating, inner covers, glued glass, various welded or glued auxiliary mounts, etc. Thus,

the concept of nominal stress is usually used, for which strain gauges with a base of 6 or even 10 mm can be

applied without any problems.

Tests of the functional sample of the vehicle on a test stand provide effective means of validation of computational

models, validation of alterations to the body structure, and means for choosing the best variant in terms of service

strength and fatigue life. The vehicle is placed onto cylinders of a computer-controlled electrohydraulic test stand.

The vertical dynamic behaviour of the vehicle is of major interest. Dynamic loads are simulated (using controlled

vertical movement of the cylinder piston rod), which represent the vehicle driving over a standardized irregularity

in the road surface [6, 7]. Stress responses are recorded using strain gauges attached to the functional sample of

the vehicle in up to hundreds of locations. The authors of this paper have addressed the issue of multiaxial fatigue

in the past [8], however in the examined critical points in the case studies solved below the first principle stress

was dominant. The second and third main stresses were close or equal to zero. For this reason, linear strain gauges

and the first principal stress are used. The largest recorded stress amplitudes are compared against pre-determined

acceptable values 𝜎𝑎, 𝑚𝑎𝑥,𝑝. Testing in an electrohydraulic loading stand can be performed in a highly flexible and

versatile manner, as the equipment can replicate relevant load states at any time and to full extent. The “vehicle”

under test need not be roadworthy. The weight of its special units and components, which may be still under

development, is applied by dummy weights, as is the payload. In the course of functional sample testing on the

test stand, alterations to the design of the structure, application of loading to only selected parts of articulated

vehicles, and replacement of suspension elements can be made relatively easily.

Measurement on the functional sample or prototype involves generation of dynamic stresses by driving on

simulated test tracks. The physically modelled test track includes standardized artificial obstacles placed on a

smooth asphalt road surface, see case study 2 in Chapter 3.2. Usually the stress response of an empty and fully-

loaded vehicle to driving is measured using strain gauges. Thus, the same load cases are simulated in previous

computational simulations. The selection of measurement locations can be specified on the basis of previous FEM

calculations and knowledge from bench tests. Attention is focused on bodywork nodes that can be considered

potentially as “unsafe”. The principle of evaluation is the same as in the previous stages. The condition specified

by Eq. (1) must again be fulfilled for all locations on the prototype body being measured.
The final assessment of the fatigue life of the body is based on a strain gauge measurement, which is carried out

with a prototype vehicle in real operation, when driving on the unevenness of real roads. As a rule, it is measured

with an empty and fully loaded vehicle on predetermined, sufficiently long sections of roads, which by their quality

(unevenness) represent the future typical operation of the newly developed vehicle. The measured stress-time

histories are analysed by the rain-flow method. For such experimentally determined service stress spectra and

known parameters of the S-N lines of the assessed structural nodes, their in-service fatigue life is estimated. The

service life determined by this procedure must be higher than the required design life of the vehicle. If this

condition is met for all assessed nodes, the development cycle is completed.

The procedure described is not fundamentally different from that used for passenger cars. However, passenger

cars have been produced in large series and the mass occurrence of fatigue failures would pose significant problems

for their manufacturers. Therefore, extensive service life tests are part of the development of passenger cars. For

economic and time reasons, it is appropriate to accelerate these tests. Fatigue life testing of a vehicle can be

accelerated, essentially in two ways. Under laboratory conditions, a pre-defined (severe) loading cycle can be

imposed in electrohydraulic test rigs [9]. However, accelerated fatigue testing can also take the form of a test ride

along special tracks at a testing ground. The existence of a special testing ground in the Czech Republic has

prompted domestic manufacturers of public transport vehicles to consider the possibility of accelerated testing of

prototypes on a predefined test circuit. The results of the authors' cooperation with a bus manufacturer are

presented in case study 4 in Chapter 3.4.

With regard to random loading processes and scatter of fatigue properties of welded bodywork nodes, the authors

consider a suitable interpretation of the calculated service fatigue life in the form of the so-called fatigue life

distribution function [10]. This method is gaining in importance with long-term and ideally online monitoring of

the operational load of structures [11, 12]. A practical example of the engineering application of this probabilistic

approach is the Case study 5 in Chapter 3.5.

2. THEORETICAL BACKGROUND

2.1 Fatigue damage calculation

The linear damage rule was applied in the presented case studies (see chapter 3). According to this rule, the limit

state of fatigue damage is achieved just when the following condition is fulfilled:
𝑛𝑖
𝐷 = ∑𝑖𝑁𝑖 = 𝐷𝑐 (2)
A concept of the linear damage rule was suggested by Palmgren [13] suggested for the first time. Later, Miner [14]

proposed the first mathematical expression in the form (2).

The rule of linear damage is still used most often, even though the limit value of damage 𝐷𝑐 is often not equal to

value 1. The causes may be more, there is documented at least one that results from the type of loading process in

this paper. Schütz [15] made an analysis of ∑𝑛 𝑁-values obtained in various test series reported in the literature.

He considered two groups of load sequences: sequences with large variations of the mean stress in addition of to

amplitude changes (57 tests series, mainly non-randomized), and sequences with a constant mean stress and a

random variation of the load amplitude (29 test series), see Fig. 6.

Fig. 6. Comparison between test results and estimations on fatigue life in variable amplitude fatigue tests, data

collected by Schütz and copied from [15].

The logarithmic horizontal scale gives the experimental live divided by the Miner-estimated life. This ratio is the

experimental ∑𝑛 𝑁-value. Along the vertical axis the probability scale of the normal distribution function is used.

It illustrates the scatter of ∑𝑛 𝑁-values obtained in 57 different test programs. The average value is ∑𝑛 𝑁 = 0.6,

but ∑𝑛 𝑁 varies approximately from 0.15 to 2.0. In 29 random load tests, the average value is ∑𝑛 𝑁 = 1.05, while

individual values range approximately from 0.3 to 3.0.Schijve [16] comments the question why such large

deviations of ∑𝑛 𝑁 = 1 can occur. Low values of ∑𝑛 𝑁 are possible if many small cycles are present in the load

history, and if a zero mean stress is applicable. High ∑𝑛 𝑁-values can be obtained if the load history has a positive
mean stress, which promotes the possible occurrence of favourable residual stress at notches (i.e. compressive

residual stress). The favourable compressive residual stress effect is absent for unnotched and axially loaded

specimens. Residual stresses may also be present on smooth specimens under bending loads due to the stress

gradient across the specimen cross section. Also tensile residual stresses may lead to a deviation of ∑𝑛 𝑁 from

value 1. Because of this qualitative understanding, it should not be a surprise that significant deviations from the

Miner rule are observed.

Schütz [15] considered the unconservative results of the Miner rule, and he introduced the idea that systematic

unconservative results could be accounted by replacing ∑𝑛 𝑁 = 1 by “a relative Miner rule”, ∑𝑛 𝑁 = 𝑞, with

𝑞 < 1. The value of 𝑞 had to be selected by experience of variable amplitude tests with similar load-time histories

relevant to the problem under the consideration. The relative Miner rule can also be interpreted as using the Miner

rule with a safety factor to account for possible unconservative life estimations. Although such a safety approach

appears reasonable, Schütz also pointed out that realistic calculations require test results obtained under realistic

load sequences to be applied to the structure or component itself.

An essential shortcoming of the Miner rule is due to ignoring damage contributions of load cycles with stress

ranges below the so called fatigue limit. It then appears to be reasonable to extrapolate fatigue life lines to lower

stress ranges in order to assign some damage increments to cycles with smaller stress ranges.

Haibach [17] proposed the extending of the fatigue life lines below the so called fatigue limit and his approach

was used in the formula (2), too.

Various rules apply various boundary conditions to fatigue damage calculation. A schematic representation of this

boundary conditions is shown in Fig. 7. Damage caused by cycles with small amplitudes (𝜎𝑎𝑖 < 𝜎𝑐), which occur

very often, is taken into account. Only damage caused by cycles with very small amplitudes that are less than the

considered threshold 𝜎ath is neglected. It has been proven to consider 𝜎ath = 0.5 ∙ 𝜎𝑐, this value corresponds to the

proposal given by Liu and Zenner in their hypothesis [18].

A limit value is set for the fatigue damage. According to Miner, 𝐷𝑐 = 1, but it has been shown experimentally that

this value may decline towards hazardous levels 𝐷𝑐 < 1.

 Fig. 7. Boundary conditions for calculating cumulative fatigue damage:

a - Palmgren-Miner original, b - Palmgren-Miner elementar, c – Haibach.

2.2 S-N line

In order to calculate fatigue damage, one needs a representative stress spectrum and an S-N line for the structural

detail in question. The following equation can be used for description of S-N relation:

𝜎𝑐 𝑤
𝑁𝑖 = 𝑁𝑐 ∙ ( )
𝜎𝑎𝑖 (3)

The following procedures can be used for determination of S-N line:

a) The most reliable method of determination of S-N line parameters is based on statistical evaluation of a

sufficiently large set of laboratory fatigue tests of identical test specimens. The conditions of fatigue test are

described in international standards, e.g. [19].

b) For some typical joints, S-N lines can be considered according to various design standards, industry

regulations and recommendations, e.g. [4], [20]. The word "typical" in this case means (similarity) in the

geometry, material and technological design.

c) The S-N line can also be derived using open access publications, catalogs and test reports summarizing the

results of fatigue tests of typical structural nodes or components. There are also commercially available

databases [21].

d) The S-N line of the construction node can also be derived from known fatigue properties of the material [22].

It is most often the derivation of the so called fatigue limit and slope of S-N line of the critical cross-section

of the real component from the known material fatigue line or, even from the static strength characteristics
 of the material (tensile test diagram). It should be noted that all of the approaches described above are more

accurate than this one. In addition, this approach is unsuitable for welded joints, because their fatigue strength

is only very slightly dependent on the strength of the base material.

Test laboratories are approached by vehicle manufacturers to perform tests, usually as contract research jobs.

Therefore, for all presented tests only general results and conclusions can be summarized. Specific information on

the slope, the number of cycles at the knee point and the endurable stresses at the knee point of S-N lines are given

only for case studies, where their publication was allowed to us by the test sponsor (bus manufacturer).

Fig. 8 shows sketches of some investigated structural details; and Fig. 9 includes results of individual fatigue tests

with mean S-N lines.

Tests were performed with symmetrical alternating loading and the stress amplitudes in S-N lines are valid for

such cross-sections of welded profiles in which a technical fatigue crack was initiated on the test specimens. The

stresses were monitored by strain gauge technics during the tests. The linear hydraulic motor acted perpendicularly

on the end of the arm and this caused its loading by an alternating bending moment. The failure criterion was the

identification of a macroscopic crack 1 to 2 mm long. Cracks mostly initiated in the transition zone of the fillet

weld. An example of a test set-up is shown in the photograph which is part of Fig. 23, Case study 1.

Fig. 8. Welded bodywork nodes – shown schematically.

Fig. 9. Results of individual fatigue tests with mean S-N lines.
This figure also schematically shows the critical cross-sections of the test specimens for which the nominal stress

could be easily calculated. The stresses were also monitored during the tests. Linear strain gauges were positioned

to measure nominal (normal) stresses. During the laboratory fatigue tests, other possible strain gauges were glued

in such places near the critical cross-sections, so that they can later be similarly installed on a real vehicle. The

theoretical ideal placement of sensors on the finished vehicle is not always possible for various reasons. The values

measured by such strain gauges could be referred to as "comparative".

The presented S-N lines indicate differences between the results, depending on whether the structural detail has

been reinforced or not. The authors are allowed to disclose that metallographic examination of the fatigue fractures

revealed a major role played by the quality of welds and less the strength of base material.

For the application of the probabilistic approach in the assessment of fatigue life, it is necessary to take into account

the scatter of fatigue characteristics of the assessed welded joint. It is appropriate to describe the fatigue line, for

example, as given in the British Standard [23]:

log (𝑁𝑓) = 𝑙𝑜𝑔𝐶0 ― 𝑠 ∙ 𝑑 ― 𝑤 ∙ log (𝜎𝑎), (4)

Standard deviation 𝑠 of log 𝑁𝑓 can be calculated on basis of experimental data, the value of standard deviation 𝑠

can be also found in the literature e.g. in the mentioned standard [23].

Fatigue lines for a various certainty of survival can be described by selecting standard deviation. For instance, at

𝑑 = 0, equation (4) describes a mean fatigue line (failure probability of 50 %). Fatigue lines shifted by two standard

deviations (𝑑 = 2) below the mean line, provided that log-normal distribution applies, represent a failure

probability of 2.3% (this means a probability of survival of 97.7%). In the case studies below, the standard

deviations evaluated by analysing the results of laboratory fatigue tests were used.

2.3 Design stress spectra

Stress spectra can often be approximated and described analytically, in relative coordinates 𝜎𝑎𝑖 𝜎𝑎,𝑚𝑎𝑥 (see Fig. 2)

according to the following equation [24]:

𝑠
(𝜎𝑎𝑖 𝜎𝑎𝑚𝑎𝑥)
ℎ𝑖 = 𝐻𝑡𝑜𝑡 ∙ ( )
𝐻𝑚𝑎𝑥
𝐻𝑡𝑜𝑡 (5)
ℎ𝑖 cumulative frequency of cycles with amplitude 𝜎𝑎𝑖;

𝜎𝑎𝑖 stress amplitude on i-level of loading;

𝜎𝑎, 𝑚𝑎𝑥 maximum stress amplitude;

𝐻𝑚𝑎𝑥 number of cycles having amplitude 𝜎𝑎,𝑚𝑎𝑥;

𝐻𝑡𝑜𝑡 total number of cycles inside spectrum;

𝑠 shape parameter of stress spectrum.

With this interpretation of various oscillation processes, the cumulative frequency distribution of cycles ℎ𝑖 with

amplitude 𝜎𝑎𝑖 is plotted in semi-log graphs using various shape parameters 𝑠 (Fig. 10):

𝑠 = ∞ rectangular distribution for a constant-amplitude harmonic process;

𝑠 = 2 normal distribution for a steady-state random Gaussian process defined by its standard deviation;

𝑠 = 1 linear distribution for a process consisting of a number of steady-state random Gaussian sub-

processes with various standard deviations.

Fig. 10. Characteristic shapes of one-parameter design stress spectra.

Measurements of service loads in road vehicles were interpreted several times in literature (for example [25]) with

the following observations: long-term monitoring of rides on an irregular road surface yields linear-distribution

stress spectra (𝑠 = 1) and manoeuvres (curve riding, braking, etc.) lead to stress spectra with normal distributions

(𝑠 = 2).

An estimate of the total number of cycles 𝐻𝑡𝑜𝑡 in the design stress spectrum can be obtained from the equation:

𝐿𝑑
𝐻𝑡𝑜𝑡 = 𝑣 ∙ 3600 ∙ 𝑓 (6)
The number of cycles 𝐻𝑚𝑎𝑥 having maximum service stress amplitudes 𝜎𝑎,𝑚𝑎𝑥 during the design life of the body

𝐿𝑑 depends on operating conditions of the vehicle given by the user. Operating condition indicators P and 𝐿1, were

established to indirectly express the severity of the conditions [26]:

𝐻𝑚𝑎𝑥
𝑃= 𝐻𝑡𝑜𝑡 (7)

𝐿𝑑
𝐿1 = 𝐻𝑚𝑎𝑥 (8)

Between both indicators the following relationship is valid:

𝐿𝑑
𝑃 = 𝐿1 ∙ 𝐻𝑡𝑜𝑡 . (9)

The single-parametric histogram of cycles (𝜎𝑎𝑖;𝑛𝑖) can be derived from the cumulative stress spectrum (5). At a

constant width of the classification interval 𝑑𝑎, one can derive the discrete stress spectrum 𝜎𝑎𝑖 ― 𝑛𝑖:

- absolute class frequency is calculated: 𝑛𝑖 = ℎ𝑖 – ℎ𝑖 + 1,

- this frequency is assigned to the mid-point of the class (or, safely, to its upper limit) 𝜎𝑎𝑖.

If there is a sufficiently representative record of the stress time history, it is more accurate to evaluate the stress

spectra from these data. The random processes are converted with the "rain flow" method into one-parameter or

two-parameter histograms of stress cycles depending on the magnitude of their amplitudes and mean values.

2.4 Fatigue life distribution function

Another suitable option is to take into account the scatter of fatigue properties of the assessed welded joint and its

variable amplitude loading and to interpret the life calculations using a probabilistic approach. A possible

engineering procedure for calculating the fatigue life distribution function (FLDF) depending on the scatter of the

fatigue properties (scatter of log N) and the random nature of the operational load was published by Kliman et al

[9]. The principle of a practical FLDF derivation has already been presented in the literature [27, 28], where the
FLDF indicates the probability of failure depending on the mileage. The Kliman´s method was transformed into

an engineering approach, which can be practiced with the support of commercial fatigue solvers. The schematic

diagram of the FLDF calculation is shown in Fig. 11. The advantage of this approach is that the FLDF shows at

first glance how the probability of fatigue damage changes depending on the number of kilometers (miles)

travelled.

Fig. 11. Calculation of FLDF (fatigue life distribution function).

3. CASE STUDIES

3.1 Case study 1

Based on the desired service life 𝐿𝑑, S-N line parameters, and a design stress spectrum, the engineer determines

maximum allowable service stress levels (𝜎𝑎, 𝑚𝑎𝑥,𝑝) for a particular cross-section of components. This procedure

was schematically outlined in Fig. 2.

Therefore, the design stress spectrum parameters (𝑠, 𝐻𝑡𝑜𝑡, 𝜎𝑎𝑚𝑎𝑥 and 𝐻𝑚𝑎𝑥) for designing and sizing the beam

joints of the bodywork were selected on the basis of the following considerations. Very simply, the consequence

of driving the vehicle on various road irregularities is the oscillation of the bodywork. The bodywork's own shapes
and the bodywork's own frequencies are projected with varying intensity into the resulting body oscillation. This

oscillation causes stress responses in the bodywork nodes that meet the criteria of random processes. Theoretically,

this is a situation which is schematically shown in Fig. 10. And also based on experience with processing many

measured random stress processes, the shape parameter of the design stress spectrum was chosen as 𝑠 = 1. The

design life of the trolleybus bodywork was set as 𝐿𝑑 = 500,000 km. The most significant bodywork vibration

frequencies were considered to be around 𝑓 = 10 Hz. At the design stage of the vehicle, of course, this is only an

assumption, but it is based on experience and frequency analyses of many bus bodies. The average travel speed

was set as 𝑣 = 50 kph. In this way it is possible to estimate the total number of cycles during the operation of the

vehicle at the initial stage of its design. After substituting the values into Eq. (6), the estimation of total number of

cycles in the specified bodywork life was found as 𝐻𝑡𝑜𝑡 = 3.6 ∙ 108 cycles. The occurrence of a cycle with

maximum stress amplitude 𝜎𝑎𝑚𝑎𝑥 was considered once per 100 km, that is, 𝐻𝑚𝑎𝑥 = 5 ∙ 103 cycles. According to

experience from many measurements in cities, this estimate corresponds to the usual facts. Of course, it would not

be a problem to adapt this estimate of 𝐻𝑚𝑎𝑥 to some unusual operating conditions (in a city with poor roads, the

incidence of extreme load values would be higher).

The structural nodes with and without reinforcement were tested and two materials (low carbon steel and stainless

steel) were considered in this study, see schematic pictures of nodes A, B and C in Fig. 12. In order to determine

the fatigue strength of the evaluated structural detail, laboratory fatigue testing was carried out. The critical cross-

section of tested welded nodes was subjected to reverse bending load (the cycle stress ratio was 𝑅 = ―1). The

limit state was defined by the instant at which a macroscopic fatigue crack forms (1 to 2 mm). In all cases, fatigue

cracks initiated in the transition zone of the fillet weld. It has not been investigated in detail where the fatigue crack

was initiated, but with respect to the load mode it is possible to assume growth from the surface to the depth of the

profile.

Statistical evaluation of the fatigue test data yielded the parameters of the S-N line in the form Eq. (4), see Fig. 12.

The scatter of fatigue properties of assessed construction nodes was considered by British Standard BS 7608.
 Fig. 12. Bodywork structural nodes and S-N lines.

In the present case, the Haibach-modified version of the Palmgren-Miner rule was chosen for calculating fatigue

damage. The limit number of cycles Ni was determined as follows:

𝜎𝑐 𝑤
- 𝜎𝑎𝑖 ≥ 𝜎𝑐: 𝑁𝑖 = 𝑁𝑐 ∙ ( )
𝜎𝑎𝑖 (10)

and

𝜎𝑐 𝑤𝑑
- 𝜎ath < 𝜎𝑎𝑖 < 𝜎𝑐: 𝑁𝑖 = 𝑁𝑐 ∙ ( )
𝜎𝑎𝑖 , (11)

Inverse slopes of S-N line 𝑤 for investigated nodes are written in Fig. 12, Haibach recommends the exponent for

the lower part of the S-N line to be set as 𝑤𝑑 = 2 ∙ 𝑤–1. The threshold stress amplitude for taking the fatigue

damaging into account was given as 𝜎ath = 0.5 ∙ 𝜎𝑐 in the present case. The limit value of fatigue damage was

taken as 𝐷𝑐 = 0.5. This value is widely recommended for such cases where the structure is loaded by variable

amplitude loading. For example, the recommendation of International Institute of Welding [20]. The final results

of this case study are shown in Fig. 13, 14 and 15.

Fig. 13. Maximum calculated stress amplitude (black) and permissible values – node A.

Fig. 14. Maximum calculated stress amplitude (black) and permissible values – node B.

Fig. 15. Maximum calculated stress amplitude (black) and permissible values – node C.
Based on the case study, it is possible to formulate a clear conclusion. The structural node A does not absolutely

meet the required service life. Even the structural node B does not achieve the required service fatigue life with

sufficient reliability. Only the structural node C could be recommended because it meets the required service

fatigue life with the usual design reliability (survival probability higher than 97.7%).

3.2 Case study 2

It has been confirmed that the main factors which play a role in the fatigue life of a bodywork are vertical dynamic

characteristics of the vehicle and the stress response in critical structural details of the bodywork while the vehicle

is driven across an uneven road surface. The effects of severe irregularities of a road surface can be simulated

effectively by driving the vehicle across an exactly defined bump. The operational tests on an uneven test track

(so called bump tests) were carried out in a standard way according to the road vehicles testing methodology which

scheme is given in Fig. 16 and described in [29]. An artificial test track is created on a common bitumen road with

a set of four portable standard bumps (obstacles) spaced out on the smooth road surface 20 meters, the shape of

which is defined in the ČSN 30 0560 Czech Standard, Fig. 17. Vertical coordinates 𝑧(𝑥) of the standardized

artificial obstacle are given by the formula

𝑙 2
(
𝑧(𝑥) = 𝑅2 ― 𝑥 ― 2 ) ― (𝑅 ― ℎ) (12)

where 𝑥 is obstacle coordinate in the vehicle driving direction (in millimetres), obstacle radius 𝑅 = 551 mm,

obstacle height ℎ = 60 mm and obstacle length 𝑙 = 500 mm.

The first obstacle is run over only with right wheels, the second one with both and the third one only with left

wheels at vehicle speed about 40 km/h.

Fig. 16. Scheme of the track according to the road vehicles testing methodology.
 Fig. 17. The standardized artificial obstacle.

The development of an articulated low-floor trolleybus included testing the effect the axle suspension

characteristics have on the stress response in critical spots of the bodywork. It has been studying the impact of the

use of tyres from two different producers (identified as T1 and T2 in Table 1) inflated to two slightly different

levels (tyre pressures P1 and P2) and the use of air springs (AS1 and AS2) and shock absorbers (SA1 and SA2)

from different manufacturers. It was found that a certain combination of suspension parameters may lead to a

dramatic reduction of the stress response in critical locations of the bodywork. All results are collected in Table 1

and illustrated in Fig. 18 for a critical location on the right-hand side panel of the front vehicle. Step by step

changes in the configuration of the suspension elements are highlighted in Table 1.

Table 1

Suspension arrangements
suspension I. axle II.axle III.axle critical
arrangement strain
gauge

tire tire air shock tire tire air shock tire tire air shock stress
type pressure spring absorber type pressure spring absorber type pressure spring absorber amplitude
(MPa)
1 T1 P1 AS1 SA1 T1 P1 AS1 SA1 T1 P1 AS1 SA1 88
2 T1 P2 AS1 SA1 T1 P2 AS1 SA1 T1 P2 AS1 SA1 83
3 T2 P1 AS1 SA1 T2 P1 AS1 SA1 T2 P1 AS1 SA1 75
4 T2 P2 AS1 SA1 T2 P2 AS1 SA1 T2 P2 AS1 SA1 79
5 T1 P2 AS1 SA2 T1 P2 AS1 SA2 T1 P2 AS1 SA2 77
6 T1 P2 AS1 SA2 T1 P2 AS2 SA2 T1 P2 AS1 SA2 68
7 T1 P2 AS1 SA2 T1 P2 AS2 SA2 T1 P2 AS2 SA2 56
 Fig. 18. Stress amplitudes for different axle suspensions of the vehicle under test

(structural detail in the right-hand side panel of the bodywork of an articulated low-floor trolleybus).

The conclusion of this case study can be formulated as follows. Understandably, a less rigid suspension is more

gentle in terms of bodywork stresses but too soft a suspension may lead to vehicle instability during manoeuvring,

such as overtaking etc. This finding highlights the potential of adaptive elements which can be integrated into

vehicle suspension. These findings contributed to a decision to develop an adaptive shock absorber for a bus [30].

3.3 Case study 3

The stress-time histories were measured for a city bus riding on an irregular surface along an urban route whose

total length was 𝐿𝑚 ≈ 39.8 km. The signals from strain gauges were analysed in the frequency domain, and

frequencies higher than 20 Hz were filtered off before using the rain-flow technique. Higher frequencies were

reflected in the vibration of the bodywork with very little power. As part of this rain-flow step, small cycles with

amplitudes below 2 MPa were omitted. In high-cycle fatigue of welded structures, the mean stress of the cycle

does not play a major role. Therefore, only the one-parameter stress spectra were used for subsequent calculations.

Fig. 19 shows the procedure of extrapolation of measured stress-time histories into cumulative stress spectra (5).

The extrapolation is performed only in terms of the number of cycles. It is obvious that the length of the measured

track may not be sufficiently representative in relation to the required service life, especially with regard to the

occurrence of the maximum stress amplitude. The authors are aware of this fact and therefore also deal with

methods of extrapolation of load spectra. In publication [31], some extrapolation methods were tested on real data,

an overview of which is given in literature [32]. Extreme-Value Extrapolation Method [33] proved to be the most
suitable for the given case. However, the current goal is to ensure long-term online monitoring of service load.

Hardware and software for this purpose are now being developed within the project of the Ministry of Industry

and Trade, Czech Republic [34].

Fig. 19. Extrapolation of measured stress-time history into stress spectrum (city road, empty vehicle).

Section km
slope circuit 3.80
speed circuit 2.80
arrival to special roads 0.07
panel road 0.45
exit/arrival 0.13
sinus resonance road 0.40
exit/arrival 0.15
paved road 0.40
exit/arrival 0.15
paved road 0.40
exit/arrival 0.15
paved road 0.40
exit/arrival 0.16
Belgian paving 0.40
exit from special roads 0.08
speed circuit 1.90
Total 11.84

Fig. 20. Special tracks on proving ground and composition of test route.
In a similar manner to representative urban traffic, the stress spectra on the test proving ground were evaluated.

The test proving ground offers testing grounds with various longitudinal road profile and different surface quality.

The table at Fig. 20 specifies the composition of the proposed test route. The measurement was repeated three

times with an empty vehicle and three times with a fully loaded vehicle. The total length of the measured route

was 𝐿𝑚 ≈ 35.5 km.

Theoretically generated design stress spectra and measured stress spectra are plotted and compared in Fig. 21. The

difference in the aggressiveness of the individual spectra can be seen at first glance. The shape of all measured

stress spectra corresponds well to the assumed linear distribution. Due to the load, the stress increases only in the

undercarriage parts of the vehicle. Driving on an empty bus can cause higher load amplitudes in some parts of the

body than driving on a fully loaded bus. Therefore, it is necessary to test at both loading conditions. In the presented

case, it can be seen that the maximum stress values have not changed in both cases. In the Table 2 there are the

values of the shape parameter 𝑠 from equation of design spectra (5). The values were calculated from experimental

data by regression analysis. The shape parameter in all cases approaches to 𝑠 ≈ 1.

Fig. 21. Stress spectra – empty and fully loaded vehicle.

Table 2

Shape parameter of all measured stress spectra.

𝑠 – shape parameter City roads Test proving ground

Empty vehicle 1.011 1.092

Loaded vehicle 0.939 1.027

This case study confirmed the previously formulated recommendations for the selection of the shape parameter 𝑠

of the design stress spectrum.

3.4 Case study 4

Essentially, fatigue life testing of a vehicle can be accelerated in two ways. Under laboratory conditions, a pre-

defined (severe) loading cycle can be imposed in electrohydraulic test rigs. Accelerated fatigue testing can also

take the form of a test ride along special testing grounds at a test proving ground, see previous text. The test

acceleration factor is the ratio of the ordinary service life and the life achieved in the accelerated test. In our case,

the ratio is as follows:

𝐴50 50 = 𝐿𝐶,50 50 𝐿𝑇𝐺,50 50 (13)

𝐿𝐶 is estimated fatigue life under city conditions and 𝐿𝑇𝐺 is estimated fatigue life under operation on polygon

testing grounds. To obtain approximate results for actual service loads, the fatigue life 𝐿,50 50 was estimated for

operation in fully-loaded condition for 50% of time and as empty vehicle for the other 50%.

Analysis of results is most meaningful when it concerns the structural details under the most severe loads or those

with the largest fatigue damage. For this reason, the case study was focused on those structural details whose

fatigue life estimation in the city environment indicated that the required design life 𝐿𝑑 = 1,000,000 km might not

be met (in at least one of the loading states under assessment). The results are shown in Table 3.
Table 3

Estimation of acceleration factor.

Structural 𝐿𝐶,𝑒𝑚𝑝𝑡𝑦 𝐿𝐶,𝑓𝑢𝑙𝑙 𝐿𝐶,50 50 𝐿𝑇𝐺,𝑒𝑚𝑝𝑡𝑦 𝐿𝑇𝐺,𝑓𝑢𝑙𝑙 𝐿𝑇𝐺,50 50 𝐴50 50

node (km) (km) (km) (km) (km) (km)

T31 121,494 39,177 59,249 2,930 10,654 4,596 12.89

T49 417,800 162,717 234,216 22,446 32,788 27,829 8.42

T48 454,373 162,886 239,805 24,137 36,398 27,768 8.64

T3 >1,000,000 174,171 348,024 15,219 408,709 29,345 11.86

T10 >1,000,000 656,103 984,388 71,060 94,817 81,237 12.12

T25 >1,000,000 792,792 1,581,539 65,967 > 1,000,000 126,274 12.52

Based on this case study, several partial conclusions can be formulated.

According to the calculations of fatigue lives using Miner's rule and the required bus mileage, the bus manufacturer

was alerted to several critical locations in the vehicle structure and to the need for their re-design (reinforcement

or substitution of material).

It was confirmed that driving a bus at a testing ground can accelerate its road testing for fatigue life assessment by

an order of magnitude. In a rough approximation, this means that the design mileage 𝐿𝑑 = 1,000,000 km can be

demonstrated by travelling approximately 100,000 km on a test track without failure. Half of this mileage should

be travelled with an empty vehicle and the other half with a fully-loaded vehicle. Alternating the payload regularly

(e.g. after each 10,000 km) is recommended. The payload of buses in operating conditions varies due to the number

of passengers and varies for different lines or cities. If the bus manufacturer has field experience from the

customers, this can be considered when designing the test programme. If such information does not exist, which

is our case, the most frequent alternation of driving with an empty and loaded bus can be recommended. However,

it should be noted that for each alternation, it is necessary to load / unload about 10 tons of artificial load in the

case of a two-axle bus. Therefore, changing the payload after 10,000 km, i.e. 10 times during the accelerated test,

seems practical.

In theory, an even more aggressive composition of the test track could be designed. It would involve a larger

proportion of those sections of the testing ground which produce the most severe damage. However, suspension

elements would have to be protected from degradation. The sequence of the test track sections should enable them
to “relax”; particularly the shock absorbers would need to cool down, because they might overheat during riding

on some types of road surface.

When accelerating any fatigue tests, it is not possible to allow the load peaks to cause local plastic deformations

in the tested structure. This would lead to residual stresses and affect damage behaviour.

3.5 Case study 5

The fatigue properties of the structural details, however, show scatter and their service load is random. For this

reason, the probability interpretation of fatigue life calculations is more correct than the deterministic

interpretation. An acceptable form of such a result is the fatigue life distribution function. The principle of its

derivation has already been presented in chapter 2.4.

The detail of interest was a severely stressed beam joint in the top corner of the door opening in the bus body

shown in schematic Fig. 22. The desired (design) fatigue service life of the vehicle in question was defined 𝐿𝑑 =

1,000,000 km-run.

Fig. 22. Investigated structural bodywork node – schematically.

This study is a virtual investigation because the specific values used in commercial contracts for various

manufacturers are confidential. However, the values and data employed in this study are realistic and can be

encountered in real life of a bus or a similar vehicle (trolleybus, battery bus or another vehicle).

In order to determine the fatigue strength of the evaluated structural detail, laboratory fatigue testing was carried

out. Test specimens were equivalent to the considered beam joint in terms of basic material, shape and

manufacturing technology (welding). Two types of test specimens were made from thin-walled welded closed

profiles (mild steel S235, former St37), see Fig. 23. The limit state was defined by initiation of a macroscopic

fatigue crack (1 to 2 mm). In all cases, fatigue cracks initiated in the transition zone of the joint welds. The

experimental points and the evaluated mean S-N lines are shown in Fig. 24.
 Fig. 23. Investigated welded nodes: A – without reinforcement, B – with reinforcement.

Fig. 24. Results of laboratory fatigue tests and mean S-N lines.

Statistical evaluation of the fatigue test data yielded the parameters of the S-N lines for the investigated bodywork

nodes in the form (4):

A: log (𝑁𝑓) = 15.45 ― 0.35 ∙ 𝑑 ― 6.10 ∙ log (𝜎𝑎), c = 40 MPa (14)

B: log (𝑁𝑓) = 14.54 ― 0.18 ∙ 𝑑 ― 4.53 ∙ log (𝜎𝑎), c = 60 MPa (15)

The real service stresses are measured by means of strain gauges. In this case, the service stress-time histories were

measured for a city bus riding on an irregular surface along a city route whose total length was 𝐿𝑚 = 78.4 km.
The signals from the strain gauges were analysed in the frequency domain, and frequencies higher than 20 Hz

were eliminated before using the rain-flow technique. As part of this rain-flow step, small cycles with amplitudes

below 2 MPa were omitted.

In high-cycle fatigue of welded structures, the mean stress of the cycle does not play a major role. Therefore, only

the one-parameter stress spectra 𝜎𝑎𝑖 ― 𝑛𝑖 were used for subsequent calculations.

In the present case, again the Haibach-modified version of the Palmgren-Miner rule was chosen for calculating

fatigue damage. Number of cycles Ni were calculated using the same relationships (10, 11) as in the case study 1.

The exponent for the lower part of the S-N line was set as 𝑤𝑑 = 2 ∙ 𝑤–1. The threshold stress amplitude for taking

the fatigue damaging into account was given as 𝜎𝑎𝑡ℎ = 0.5 ∙ 𝜎𝑐 in the present case. The limit value of fatigue

damage was taken as 𝐷𝑐 = 0.5. The computational estimate of service fatigue life (in kilometre run) is obtained

from equation:

𝐷
𝐿 = 𝐷𝑐 . 𝐿𝑚 (16)

Table 4 provides estimated fatigue life for the stress spectra measured with empty vehicle and fully loaded vehicle

during the real city operation. The calculated values are valid for mean fatigue S-N lines (for a 50% probability of

survival).

Table 4

Calculated fatigue life for 50% survival probability

Real city operation - Fatigue life 𝐿 (km)

Investigated node Empty vehicle Fully loaded vehicle

A 29,184 39,377

B 2,620,352 3,703,640

The fatigue properties of the investigated construction node, however, show scattering and their service load is

random. For this reason, the probability interpretation of fatigue life calculations is more correct than the
deterministic interpretation. An acceptable form of such a result is the fatigue life distribution function (FLDF,

Fig. 11).

The measured stress spectra were considered sufficiently representative for the assessment of operational

reliability (service fatigue life). When considering the scatter of fatigue properties, it was possible to calculate

FLDFs for both measured loading states (empty and fully loaded vehicle). The calculated FLDFs are shown in

Fig. 25.

Fig. 25. FLDFs for real city operation.

Based on this case study, it can be stated that the fatigue tests and calculations have shown that the welded

connection of bodywork profiles without a reinforcement (variant A) is not a suitable solution. The reinforced

variant B of this structural node meets the required operational fatigue life of the vehicle. The scatter of material

properties must be taken into account in fatigue life calculations and results of the calculations can be interpreted

by the fatigue life distribution function.

CONCLUSIONS

Partial conclusions have already been formulated for individual case studies. In summary, it can be stated that the

methodology of designing and assessing fatigue life of bus and trolleybus bodyworks continues to be developed.

Probabilistic approaches to estimating service fatigue life are used at the design stage, using design stress spectra.

Insight was gained into the possibilities of accelerated fatigue testing in a proving ground. To develop

understanding for this matter, various parametric studies are conducted and results of the theoretical calculations

are compared with data from the real vehicles operation.

ACKNOWLEDGMENTS

The contribution has been prepared under project Nr. FW01010386 Research and development of articulated

electric bus, with financial support of the TREND programme of the Technology Agency of the Czech Republic.

REFERENCES

1. Kepka M. Durability and fatigue life investigation of bus and trolleybus structures: review of SKODA

VYZKUM methodology. In: Proceedings of the Second International Conference on Material and Component

Performance under Variable Amplitude Loading (VAL2) scheduled from 23.March to 26.March 2009 in

Darmstadt/Germany, Publisher: DVM, Berlin/Germany, Deutscher Verband für Materialforschung und-

prüfung e.V. pp. 1231-1240. ISBN: 978-3-00-027049-9.

2. Kepka M, Kepka Jr M. Review of practical experiences from development of buses: service stresses, laboratory

fatigue tests and fatigue life calculations. In: Proceedings of the Fourth International Conference on Material

and Component Performance under Variable Amplitude Loading (VAL4) scheduled from 30.March to 1.April

2020 in Darmstadt/Germany, Publisher: DVM, Berlin/Germany, Deutscher Verband für Materialforschung

und-prüfung e.V., 2009. pp. 251-260. ISBN: 978-3-9820591-0-5.

3. Sonsino, C. M. : Course of SN-Curves Especially in the High-Cycle Fatigue Regime with Regard to

Component Design and Safety, Int. J. Fatigue 29 (2007), pp. 2246-2258 and Hobbacher, A. (Ed):

Recommendations for Fatigue Design of Welded Joints and Components. IIW Document XIII1823-07, updates

2008 to 2015.

4. Hobbacher AF. Recommendations for Fatigue Design of Welded Joints and Components (2nd ed.). Publisher:

Springer (2016) IIW225915 (ex XIII246013/XV144013), (ISBN: 978-3-319-23756-5).

5. Kepka M, Polach P, Kepka Jr, M. From Dynamic Models of Buses to Reliable Estimation of the Fatigue Life

of Their Bodyworks. In: Proceedings IRF2020: 7th International Conference Integrity-Reliability-Failure,

Funchal, Portugal, 14-18 June 2020. Editors J.F. Silva Gomes and S.A. Meguid, Publ. INEGI/FEUP (2020);

pp. 419-430. Paper ref.: 17200. ISBN: 978-989-20-8315-3.

6. Kepka M, Rehor P. Methodology of experimental research into operating strength and fatigue life of bus and

trolleybus bodywork. International Journal of Vehicle Design, vol. 13., no. 3, 1992, pp. 242 – 250.
7. Kepka M, Kepka Jr, M et al. Fatigue Tests – Important Part of Development of New Vehicles. In: 12th

International Fatigue Congress, Poitiers, France, May 27 to June 1, 2018, MATEC Web of Conferences 165,

22023 (2018), https://doi.org/10.1051/matecconf/201816522023.

8. Kepka M, Kepka Jr, M, Chvojan J, Vaclavik J. Structure service life assessment under combined loading using

probability approach. Frattura ed Integrita Strutturale, Vol. 10 (2016), No. 38, pp. 82-91. ISSN: 1971-899

9. Halfpenny A. Methods for accelerating dynamic durability tests. In: 9th International Conference on Recent

Advances in Structural Dynamics, Southhampton, 2006.

10. Kliman V, Kepka M, Václavík J. Influence of scatter of cyclic properties of material on operational endurance

of construction. Metallic Materials, 48 (2010), pp. 367–378, DOI: 10.4149/km 2010 6 367.

11. Chmelko V, Garan M. Long-term monitoring of strains in a real operation of structures. In: 14th IMEKO TC10

Workshop Technical Diagnostics New Perspectives in Measurements, Tools and Techniques for system’s

reliability, maintainability and safety, Milan, Italy, June 27-28, 2016, pp. 333–336.

12. Kepka M, Vaclavik J. On-line vehicle condition monitoring based on probability approach. In: 14th IMEKO

TC10 Workshop Technical Diagnostics New Perspectives in Measurements, Tools and Techniques for

system’s reliability, maintainability and safety, Milan, Italy, June 27-28, 2016, pp. 42-45.

13. Palmgren AZ. Die Lebensdauer von Kugellagcrn. Z. Ver. Deutsch. Ing. 68 (1924) 339.

14. Miner MA. Cumulative damage in fatigue. Transactions of the ASME. Series E. J. Appl. Mech. Vol. 12, (1945)

pp. 159–164.

15. Schütz W. The estimateion of fatigue life in the crack initiation and propagation stages. A state of the art

survey, Eng. Fract. Mech. 11 (1979) pp. 405–421.

16. Schijve J. Fatigue of Structures and Materials. 2nd ed., Publisher: Springer, 2009 with CD-Rom. (ISBN-13:

978-1-4020-6807-2).

17. Haibach E. Modified Linear Damage Accumulation Hypothesis Accounting for a Decreasing Fatigue Strength

During the Increasing Fatigue Damage. Laboratorium für Betribsfestigkeit, LBF, Darmstadt, German, 1970

TM Nr. 50.

18. Liu J, Zenner H. Berechnung der Dauerschwingfestigkeit bei mehrachsiger Beanspruchung, Mat.-wiss. u.

Werkstofftech., Vol. 24 (1993), pp. 240–249.

19. ISO/DIS 12107 Metallic materials — Fatigue testing — Statistical planning and analysis of data.

20. Eurocode 3: Design of steel structures - Part 1-9: Fatigue.

21. WIAM® METALLINFO: https://wiam.de/produkte/wiam-metallinfo/

22. FKM (Forschungskuratorium Maschinenbau), Analytical Strength Assessment of Components, VDMA

Verlag, 6th revised edition, 2012.

23. BS 7608:1993 - Fatigue design and assessment of steel structures.

24. Ruzicka M, Hanke M, Rost M. Dynamic strength and fatigue life. University textbook, Czech Technical

University, Praha, 1987 (in Czech).

25. Neugebauer J, Grubisic V, Fischer G. Procedure for Design Optimization and Durability Life Approval of

Truck Axles and Axle Assemblies. SAE Technical Paper Series 892535, 1989.

26. Kepka M, Kepka Jr M. Using design S-N lines and design stress spectra for probabilistic fatigue life assessment

of vehicle components. In: Proceedings of the 6th International Conference Integrity-Reliability-Failure

(IRF2018) scheduled from July 22 to July 28, 2018, Lisbon, Portugal, Publisher: INEGI/FEUP, pp. 373-384.

Paper ref.: 7059. ISBN: 978-989-20-8313-1

27. Kepka M, Kepka Jr M. Deterministic and probabilistic fatigue life calculations of a damaged welded joint in

the construction of the trolleybus. Engineering Failure Analysis, 93, 2018, pp. 257–267.

28. Václavík J, Marek P, Chvojan J, Kepka M. Determination of the Fatigue Strength of a Railway Vehicle Node

using the Probability Approach. Engineering Mechanics, 17 (2010), pp. 99-11, ISSN 1802-1484

29. Polach P, Vaclavik J, Hajzman M. Verification of the Multibody Models of the TriHyBus on the Basis of

Experimental Measurements. In: Proceedings of the 6th Asian Conference on Multibody Dynamics

ACMD2012 scheduled from 26. August to 30. August 2012 in Shanghai/China, USB flash drive only.

30. Polach P, Hajzman M.: Design of Characteristics of Air-Pressure-Controlled Hydraulic Shock Absorbers in an

Intercity Bus. Multibody System Dynamics, Vol. 19, No. 1-2, pp. 73-90.

31. Dewa RT, Kepka Jr M, Kepka M. Statistical approaches on the design of fatigue stress spectra for bus

structures. SN applied sciences, Vol. 1 (2019), No. 1, p.1360.

32. Wang J, Chen H, Li Y, Wu Y, Zhang Y. A review of the extrapolation method in load spectrum compiling.

Strojniski Vestnik - Journal of Mechanical Engineering, Vol. 62 (2016), No.1, pp. 60-75

33. Johannesson P. Extrapolation of load histories and spectra. Fatigue & Fracture of Engineering Materials &

Structures, Volume 29 (2006), No. 3, pp.179-280.

34. FV40260: On-line measurement and analysis of the operational loading of structures with adaptive virtual

models. Ministry of Industry and Trade, Czech Republic (2019-2022).

Corresponding author: kepkam@rti.zcu.cz