Abaqus Analysis User's Manual

6.2.7 Low-cycle fatigue analysis using the direct cyclic approach

Products: Abaqus/Standard Abaqus/CAE

References
�Procedures: overview,� Section 6.1.1

�Static stress analysis procedures: overview,� Section 6.2.1

�Direct cyclic analysis,� Section 6.2.6

�Crack propagation analysis,� Section 11.4.3

�Damage and failure for ductile materials in low-cycle fatigue analysis,� Section
21.4

*DAMAGE EVOLUTION

*DAMAGE INITIATION

*DEBOND

*DIRECT CYCLIC

*FRACTURE CRITERION

*CONTROLS

�Configuring a direct cyclic procedure� in �Configuring general analysis

procedures,� Section 14.11.1 of the Abaqus/CAE User's Manual

Overview

A low-cycle fatigue analysis:

is characterized by states of stress high enough for inelastic deformation to

occur;

is a quasi-static analysis on a structure subjected to sub-critical cyclic loading;

can be associated with thermal as well as mechanical loading;

uses the direct cyclic approach to obtain the stabilized cyclic response of the
structure directly;

models progressive damage and failure in bulk ductile material based on a continuum
damage approach, in which case damage initiation and evolution are characterized by
the accumulated inelastic hysteresis strain energy per stabilized cycle;

models progressive delamination growth at the interfaces in laminated composites,

in which case the onset and growth of fatigue delamination at the interfaces are
characterized by the relative fracture energy release rate;

uses the damage extrapolation technique to accelerate the low-cycle fatigue

analysis; and

assumes geometrically linear behavior and fixed contact conditions within each
loading cycle.

Approaches to low-cycle fatigue analysis

The traditional approach for determining the fatigue limit for a structure is to
establish the curves (load versus number of cycles to failure) for the materials
in the structure. Such an approach is still used as a design tool in many cases to
predict fatigue resistance of engineering structures. However, this technique is
generally conservative, and it does not define a relationship between the cycle
number and the degree of damage or crack length.

One alternative approach is to predict the fatigue life by using a crack/damage

evolution law based on the inelastic strain/energy when the structure's response is
stabilized after many cycles. Because the computational cost to simulate the slow
progressive damage in a material over many load cycles is prohibitively expensive
for all but the simplest models, numerical fatigue life studies usually involve
modeling the response of the structure subjected to a small fraction of the actual
loading history. This response is then extrapolated over many load cycles using
empirical formulae such as the Coffin-Manson relationship (see Coffin, 1954, and
Manson, 1953) to predict the likelihood of crack initiation and propagation. Since
this approach is based on a constant crack/damage growth rate, it may not
realistically predict the evolution of the crack or damage.

Low-cycle fatigue analysis in Abaqus/Standard

The direct cyclic analysis capability in Abaqus/Standard provides a computationally
effective modeling technique to obtain the stabilized response of a structure
subjected to periodic loading and is ideally suited to perform low-cycle fatigue
calculations on a large structure. The capability uses a combination of Fourier
series and time integration of the nonlinear material behavior to obtain the
stabilized response of the structure directly. The theory and algorithm to obtain a
stabilized response using the direct cyclic approach are described in detail in
�Direct cyclic algorithm,� Section 2.2.3 of the Abaqus Theory Manual.

The direct cyclic low-cycle fatigue procedure models the progressive damage and
failure both in bulk materials (such as in solder joints in an electronic chip
packaging) and at material interfaces (such as in laminated composites). The
response is obtained by evaluating the behavior of the structure at discrete points
along the loading history (see Figure 6.2.7�1). The solution at each of these
points is used to predict the degradation and evolution of material properties that
will take place during the next increment, which spans a number of load cycles, .
The degraded material properties are then used to compute the solution at the next
increment in the load history. Therefore, the crack/damage growth rate is updated
continually throughout the analysis.

Figure 6.2.7�1 Elastic stiffness degradation as a function of the cycle number.

The elastic material stiffness at a material point remains constant and contact
conditions remain unchanged when the stabilized solution is computed at a given
point in the loading history. Each of the solutions along the loading history
represents the stabilized response of the structure subjected to the applied period
loads, with a level of material damage at each point in the structure computed from
the previous solution. This process is repeated up to a point in the loading
history at which a fatigue life assessment can be made.

In bulk material the cyclic loading leads to stress reversals and the accumulation
of plastic strains, which in turn cause the initiation and propagation of cracks.
The damage initiation and evolution are characterized by the stabilized accumulated
inelastic hysteresis strain energy per cycle as illustrated in Figure 6.2.7�2.

Figure 6.2.7�2 Plastic shakedown in a direct cyclic analysis.

At interfaces of laminated composites the cyclic loading leads to interface

strength degradation causing fatigue delamination growth. The onset and growth of
delamination are characterized by the relative fracture energy release rate at the
crack tip based on the Paris law (Paris, 1961).

Both the progressive damage mechanism in the bulk material and the progressive
delamination growth mechanism at interfaces can be considered simultaneously, with
the failure occurring first at the weakest link in a model.

Defining a low-cycle fatigue analysis using the direct cyclic approach is similar
to defining a direct cyclic analysis. See �Direct cyclic analysis,� Section 6.2.6,
for details on how to specify the number of Fourier terms, number of iterations,
and the increment sizes. You specify the maximum numbers of cycles, , when you
define the low-cycle fatigue analysis step.

Input File Usage:

*DIRECT CYCLIC, FATIGUE
first data line
, ,

Abaqus/CAE Usage:
Step module: Create Step: General: Direct cyclic; Fatigue: Include low-cycle
fatigue analysis, Maximum number of cycles: Value:

Determining whether to use the Fourier coefficients from the previous step
A low-cycle fatigue step using the direct cyclic approach can be the only step in
an analysis, can follow a general or linear perturbation step, or can be followed
by a general or linear perturbation step. Multiple low-cycle fatigue analysis steps
can be included in a single analysis. In such a case the Fourier series
coefficients obtained in the previous step can be used as starting values in the
current step. By default, the Fourier coefficients are reset to zero, thus allowing
application of cyclic loading conditions that are very different from those defined
in the previous low-cycle fatigue step.

As in a direct cyclic analysis, you can specify that a low-cycle fatigue step in a
restart analysis should use the Fourier coefficients from the previous step, thus
allowing continuation of an analysis to simulate more loading cycles. In a low-
cycle fatigue analysis a restart file is written at the end of the stabilized
cycle. Consequently, a restart analysis that is a continuation of a previous low-
cycle fatigue analysis will start with a new loading cycle at (see �Restarting an
analysis,� Section 9.1.1).

Input File Usage: Use the following option to specify that the current
step is a continuation of the previous low-cycle fatigue step using the direct
cyclic approach:
*DIRECT CYCLIC, FATIGUE, CONTINUE=YES
Use the following option to reset the Fourier series coefficients to zero:

*DIRECT CYCLIC, FATIGUE, CONTINUE=NO (default)

Abaqus/CAE Usage: Use the following option to specify that the current step
is a continuation of the previous low-cycle fatigue step using the direct cyclic
approach:
Step module: Create Step: General: Direct cyclic; Basic: Use displacement Fourier
coefficients from previous direct cyclic step; Fatigue: Include low-cycle fatigue
analysis

Use the following option to reset the Fourier series coefficients to zero:

Step module: Create Step: General: Direct cyclic; Fatigue: Include low-cycle
fatigue analysis

Progressive damage and damage extrapolation in bulk ductile material

Low-cycle fatigue analysis in Abaqus/Standard allows modeling of progressive damage

and failure for ductile materials in any elements whose response is defined in
terms of a continuum-based constitutive model (�Material library: overview,�
Section 18.1.1). This includes cohesive elements modeled using a continuum approach
(�Modeling of an adhesive layer of finite thickness� in �Defining the constitutive
response of cohesive elements using a continuum approach,� Section 29.5.5). The
inelastic definition in a material point must be used in conjunction with the
linear elastic material model (�Linear elastic behavior,� Section 19.2.1), the
porous elastic material model (�Elastic behavior of porous materials,� Section
19.3.1), or the hypoelastic material model (�Hypoelastic behavior,� Section
19.4.1).

After damage initiation the elastic material stiffness is degraded progressively in

each cycle (as shown in Figure 6.2.7�1) based on the accumulated stabilized
inelastic hysteresis energy. It is impractical and computationally expensive to
perform a cycle-by-cycle simulation for a low-cycle fatigue analysis; Instead, to
accelerate the low-cycle fatigue analysis, each increment extrapolates the current
damaged state in the bulk material forward over many cycles to a new damaged state
after the current loading cycle is stabilized.

Damage initiation and evolution

Damage initiation refers to the beginning of degradation of the response of a
material point. In a low-cycle fatigue analysis the damage initiation criterion is
characterized by the accumulated inelastic hysteresis energy per cycle, . and
material constants are used to determine the number of the cycle in which damage is
initiated, . At the end of a stabilized loading cycle, , Abaqus/Standard checks to
see if the damage initiation criterion is satisfied in any material point;
material stiffness at a material point will not be degraded unless this criterion
is satisfied. The calculations and output associated with damage initiation are
discussed in detail in �Damage initiation for ductile materials in low-cycle
fatigue,� Section 21.4.2.

Once the damage initiation criterion is satisfied at a material point, the damage
state is calculated and updated based on the inelastic hysteresis energy for the
stabilized cycle. Abaqus/Standard assumes that the degradation of the elastic
stiffness can be modeled using the scalar damage variable, . The rate of the damage
in a material point per cycle, , is calculated based on the accumulated inelastic
hysteresis energy, the characteristic length associated with an integration point,
and material constants. For details, see �Damage evolution for ductile materials in
low-cycle fatigue,� Section 21.4.3.

Typically, a material has completely lost its load carrying capacity when . You can
remove an element from the mesh if all of the section points at all integration
locations of the element have lost their load carrying capability.

Damage extrapolation technique in the bulk material

If the damage initiation criterion is satisfied in any material point at the end of
a stabilized cycle, , Abaqus/Standard extrapolates the damage variable from the
current cycle forward to the next increment over a number of cycles, . The new
damage state, , is given by

where is the characteristic length associated with an integration point, and and
are material constants (see �Damage evolution for ductile materials in low-cycle
fatigue,� Section 21.4.3, for more information).
You specify the minimum () and maximum () number of cycles over which the damage is
extrapolated forward in any given increment. The default values are 100 and 1000,
respectively.

Input File Usage:

*DIRECT CYCLIC, FATIGUE
first data line
,

Abaqus/CAE Usage:
Step module: Create Step: General: Direct cyclic; Fatigue: Include low-cycle
fatigue analysis, Cycle increment size: Minimum: , Maximum:

Progressive delamination growth at interfaces

Low-cycle fatigue analysis in Abaqus/Standard also allows the modeling of

progressive delamination growth at the interfaces in laminated composites. The
interface along which the delamination (or crack) propagates must be indicated in
the model using a fracture criterion definition. The fracture energy release rates
at the crack tips in the interface elements are calculated based on the virtual
crack closure technique (VCCT). VCCT uses the principles of linear elastic fracture
mechanics. Therefore, VCCT is appropriate for problems in which brittle fatigue
delamination growth occurs along predefined surfaces, although nonlinear material
deformations can occur in the bulk materials. For more information about defining
fracture criteria and VCCT, see �Crack propagation analysis,� Section 11.4.3.

To accelerate the low-cycle fatigue analysis, the damage extrapolation technique is

used, which releases at least one element length at the crack tip along the
interface after each stabilized cycle. When both brittle fatigue delamination at
interfaces and ductile damage in bulk materials are considered in an analysis,
failure occurs first at the weakest link.

Onset and growth of fatigue delamination

The onset and growth of fatigue delamination at a defined crack interface are
characterized by using the Paris law, which relates the relative fracture energy
release rate, , to crack growth rates. Two criteria must be met to initiate fatigue
delamination growth: one criterion is based on material constants, , and the
current cycle number, ; the other criterion is based on the maximum fracture energy
release rate, , which corresponds to the cyclic energy release rate when the
structure is loaded up to its maximum value. Once the onset of delamination growth
criterion is satisfied at the interface, the delamination growth rate, , is a
piecewise function based on material constants and (the Paris law). The criteria
for fatigue delamination onset and growth are discussed in detail in �Low-cycle
fatigue criterion� in �Crack propagation analysis,� Section 11.4.3.

Damage extrapolation technique at the interface elements

If the onset of delamination growth criterion is satisfied at any crack tip in the
interface at the end of a stabilized cycle, , Abaqus/Standard extends the crack
length, , from the current cycle forward over a number of cycles, , to by releasing
at least one element at the interface. Given the material constants and (as
defined in �Low-cycle fatigue criterion� in �Crack propagation analysis,� Section
11.4.3), combined with the known node spacing at the interface elements at the
crack tips, the number of cycles necessary to fail each interface element at the
crack tip can be calculated as , where j represents the node at the jth crack tip.
The analysis is set up to release at least one interface element per increment
after the loading cycle is stabilized. The element with the fewest cycles is
identified to be released, and its is represented as the number of cycles to grow
the crack equal to its element length, . The most critical element is completely
released with a zero constraining and a zero stiffness at the end of the stabilized
cycle. As the interface element is released, the load is redistributed, and a new
relative fracture energy release rate must be calculated for the interface elements
at the crack tips for the next cycle. This capability allows at least one interface
element at the crack tips to be released after each stabilized cycle and precisely
accounts for the number of cycles needed to cause fatigue crack growth over that
length.

Controlling the solution accuracy

Low-cycle fatigue analysis utilizes the direct cyclic approach to obtain the
stabilized cyclic solution iteratively by combining a Fourier series approximation
with time integration of the nonlinear material behavior using a modified Newton
method. The accuracy of the algorithm depends on the number of Fourier terms used,
the number of iterations taken to obtain the stabilized solution, and the number of
time points within the load period at which the material response and residual
vector are evaluated. Some methods for controlling the solution accuracy in a
direct cyclic analysis are described in detail in �Direct cyclic analysis,� Section
6.2.6. They all remain valid in a low-cycle fatigue analysis using the direct
cyclic approach. In addition, the accuracy of a low-cycle fatigue analysis depends
on the number of cycles over which the damage is extrapolated forward, as described
below.

Controlling the accuracy of damage extrapolation in the bulk material

To accelerate the low-cycle fatigue analysis, the damage extrapolation technique is
used at the end of a stabilized cycle. In addition to specifying the minimum and
maximum number of cycles over which the damage is extrapolated (see �Damage
extrapolation technique in the bulk material� above), you can specify the damage
extrapolation tolerance, , to control the accuracy of damage extrapolation in the
bulk material. The default is .

Input File Usage: Use the following option to specify the damage
extrapolation tolerance:
*DIRECT CYCLIC, FATIGUE
first data line
, , ,

Abaqus/CAE Usage:
Step module: Create Step: General: Direct cyclic; Fatigue: Include low-cycle
fatigue analysis, Damage extrapolation tolerance:

Determining the increment over which damage is extrapolated forward

Abaqus/Standard uses an adaptive algorithm to determine the number of cycles over
which the damage is extrapolated forward in each increment. By default,
Abaqus/Standard starts with 500 cycles (half of the default value of maximum
increment in number of cycles) and determines the maximum damage increment at any
material points based on
If the maximum damage increment, , is greater than the damage extrapolation
tolerance that you specify, the number of cycles over which the damage is
extrapolated forward is reduced accordingly to ensure the maximum damage increment
is less than the damage extrapolation tolerance. On the other hand, if the maximum
damage increment at all material points is less than half of the damage
extrapolation tolerance that you specify, the number of cycles is increased
accordingly to ensure the maximum damage increment is equal to the damage
extrapolation tolerance.
Initial conditions

Initial values of stresses, temperatures, field variables, solution-dependent state

variables, etc. can be specified (see �Initial conditions in Abaqus/Standard and
Abaqus/Explicit,� Section 30.2.1).

Boundary conditions

Boundary conditions can be applied to any of the displacement or rotation degrees

of freedom. During the analysis, prescribed boundary conditions must have an
amplitude definition that is cyclic over the step: the start value must be equal to
the end value (see �Amplitude curves,� Section 30.1.2). If the analysis consists of
several steps, the usual rules apply (see �Boundary conditions in Abaqus/Standard
and Abaqus/Explicit,� Section 30.3.1). At each new step the boundary condition can
either be modified or completely defined. All boundary conditions defined in
previous steps remain unchanged unless they are redefined.

Loads

The following loads can be prescribed in a low-cycle fatigue analysis using the
direct cyclic approach:

Concentrated nodal forces can be applied to the displacement degrees of freedom

(1�6); see �Concentrated loads,� Section 30.4.2.

Distributed pressure forces or body forces can be applied; see �Distributed loads,�
Section 30.4.3. The distributed load types available with particular elements are
described in Part VI, �Elements.�

During the analysis each load must have an amplitude definition that is cyclic over
the step where the start value must be equal to the end value (see �Amplitude
curves,� Section 30.1.2). If the analysis consists of several steps, the usual
rules apply (see �Applying loads: overview,� Section 30.4.1). At each new step the
loading can either be modified or completely defined. All loads defined in previous
steps remain unchanged unless they are redefined.
Predefined fields

The following predefined fields can be specified in a low-cycle fatigue analysis

using the direct cyclic approach, as described in �Predefined fields,� Section
30.6.1:

Temperature is not a degree of freedom in a low-cycle fatigue analysis using the

direct cyclic approach, but nodal temperatures can be specified as a predefined
field. The temperature values specified must be cyclic over the step: the start
value must be equal to the end value (see �Amplitude curves,� Section 30.1.2). If
the temperatures are read from the results file, you should specify initial
temperature conditions equal to the temperature values at the end of the step (see
�Initial conditions in Abaqus/Standard and Abaqus/Explicit,� Section 30.2.1).
Alternatively, you can ramp the temperatures back to their initial condition
values, as described in �Predefined fields,� Section 30.6.1. Any difference between
the applied and initial temperatures will cause thermal strain if a thermal
expansion coefficient is given for the material (�Thermal expansion,� Section
23.1.2). The specified temperature also affects temperature-dependent material
properties, if any.

The values of user-defined field variables can be specified. These values affect
only field-variable-dependent material properties, if any. The field variable
values specified must be cyclic over the step.

Material options

Most ductile material models that describe mechanical behavior are available for
use in a low-cycle fatigue analysis. The inelastic definition in a material point
must be used in conjunction with the linear elastic material model (�Linear elastic
behavior,� Section 19.2.1), the porous elastic material model (�Elastic behavior of
porous materials,� Section 19.3.1), or the hypoelastic material model (�Hypoelastic
behavior,� Section 19.4.1).

The following material properties are not active during a low-cycle fatigue
analysis: acoustic properties, thermal properties (except for thermal expansion),
mass diffusion properties, electrical conductivity properties, piezoeletric
properties, and pore fluid flow properties.

Rate-dependent yield (�Rate-dependent yield,� Section 20.2.3), rate-dependent creep

(�Rate-dependent plasticity: creep and swelling,� Section 20.2.4), and two-layer
viscoplasticity (�Two-layer viscoplasticity,� Section 20.2.11) can also be used
during a low-cycle fatigue analysis.

Elements

Any of the stress/displacement elements in Abaqus/Standard can be used in a low-

cycle fatigue analysis (see �Choosing the appropriate element for an analysis
type,� Section 24.1.3). This includes cohesive elements with finite thickness
(�Modeling of an adhesive layer of finite thickness� in �Defining the constitutive
response of cohesive elements using a continuum approach,� Section 29.5.5).

Output

Different types of output are available for postprocessing and for monitoring a
low-cycle fatigue analysis using the direct cyclic approach.

Message file information

As in a direct cyclic analysis, low-cycle fatigue analysis using the direct cyclic
approach in Abaqus/Standard prints the residual force, time average force, and a
flag to indicate if equilibrium was satisfied in the message (.msg) file at
different time increments for each iteration in each loading cycle. You can control
the frequency in increments at which information is printed to the message file,
and you can suppress the output; the default is to print output every 10 increments
(see �The Abaqus/Standard message file� in �Output,� Section 4.1.1, for more
information).

Abaqus/Standard also prints the number of Fourier terms used, the maximum residual
coefficient, the maximum correction to displacement coefficients, and the maximum
displacement coefficient in the Fourier series in the message file at the end of
each iteration in each cycle. An example of the output is shown below:

CYCLE 5 STARTS
 ITERATION 26
STARTS
INC TIME STEP LARG. RESI. TIME AVG. FORCE
INC TIME FORCE FORCE EQUV.
10 0.250 2.50 1.008E+01 50.9 N
20 0.250 5.00 1.622E+01 76.8 N
30 0.250 7.50 4.622E-02 99.8 Y

ITERATION 26 SUMMARY
NUMBER OF FOURIER TERMS USED 40, TOTAL NUMBER OF INCREMENTS 120
CYCLE/STEP TIME 30.0, TOTAL TIME COMPLETED 31.0
AVERAGE FORCE 21.2 TIME AVG. FORCE 25.7

MAX. COEFFICIENT OF DISP. 0.142 AT NODE 24 DOF 2

MAX. COEFF. OF RESI. FORCE ON CONST. TERM 31.7 AT NODE 44 DOF 1
MAX. COEFF. OF RESI. FORCE ON PERI. TERMS 0.82 AT NODE 6 DOF 3
MAX. CORR. TO COEFF. OF DISP. ON CONST. TERM 0.002 AT NODE 50 DOF 3
MAX. CORR. TO COEFF. OF DISP. ON PERI. TERMS 0.015 AT NODE 50 DOF 3
Results output
Element and nodal output are written only when the stabilized cycle is reached. If
a stabilized cycle has not been reached at the end of a cycle, output is written
for the last iteration of the cycle. All standard output variables in
Abaqus/Standard (�Abaqus/Standard output variable identifiers,� Section 4.2.1) are
available. In addition, the following variables are available:

STATUS
Status of element (the status of an element is 1.0 if the element is active, 0.0 if
the element is not).

SDEG
Scalar stiffness degradation, D.

CYCLEINI
Number of cycles to initialize the damage at the material point.

Recovering additional results for a stabilized cycle

You may want to recover additional results for a stabilized cycle. You can extract
these results from the restart data (see �Recovering additional results output from
restart data in Abaqus/Standard� in �Output,� Section 4.1.1).

Input File Usage:

*POST OUTPUT, CYCLE=n

Abaqus/CAE Usage: Recovering additional results for a stabilized cycle is not

supported in Abaqus/CAE.

Limitations

A low-cycle fatigue analysis using the direct cyclic approach is subject to the
following limitations:

Contact conditions cannot change during a given cycle when direct cyclic analysis
is used iteratively to obtain a stabilized solution.

Geometric nonlinearity can be included only in any general step prior to a direct
cyclic step; however, only small displacements and strains will be considered
during the cyclic step.
Input file template

*HEADING
�
*BOUNDARY
Data lines to specify zero-valued boundary conditions
*INITIAL CONDITIONS
Data lines to specify initial conditions
*AMPLITUDE
Data lines to define amplitude variations
**
*MATERIAL
Options to define material properties
*DAMAGE INITIATION, CRITERION=HYSTERESIS ENERGY
Data lines to define material constants for bulk ductile material damage initiation
*DAMAGE EVOLUTION, TYPE=HYSTERESIS ENERGY
Data lines to define material constants for bulk ductile material damage evolution
**
*SURFACE, NAME=slave
Data lines to define slave surface at delamination interface
*SURFACE, NAME=master
Data lines to define master surface at delamination interface
*CONTACT PAIR
slave, master
**
*STEP (,INC=)
Set INC equal to the maximum number of increments in a single loading cycle
*DIRECT CYCLIC, FATIGUE
Data line to define time increment, cycle time, initial number of Fourier terms,
maximum number of Fourier terms, increment in number of Fourier terms,
and maximum number of iterations
Data line to define minimum increment in number of cycles,
maximum increment in number of cycles, total number of cycles,
and damage extrapolation tolerance
*DEBOND, SLAVE=slave, MASTER=master
*FRACTURE CRITERION, TYPE=FATIGUE
Data lines to define material constants used in Paris law and fracture criterion
**
*BOUNDARY, AMP=
Data lines to prescribe zero-valued or nonzero boundary conditions
*CLOAD and/or *DLOAD, AMP=
Data lines to specify loads
*TEMPERATURE and/or *FIELD, AMP=
Data lines to specify values of predefined fields
**
*END STEP
Additional references

Coffin, L., �A Study of the Effects of Cyclic Thermal Stresses on a Ductile

Metal,� Transactions of the American Society of Mechanical Engineering, vol. 76,
pp. 931�951, 1954.

Manson, S., �Behavior of Materials under Condition of Thermal Stress,� Heat

Transfer Symposium, University of Michigan Engineering Research Institute, Ann
Arbor, MI, pp. 9�75, 1953.

Paris, P., M. Gomaz, and W. Anderson, �A Rational Analytic Theory of Fatigue,� The
Trend in Engineering, vol. 15, 1961.