METHOD FOR MECHANICAL PROPERTIES EVALUATION OF STEEL STRUCTURES

WITH CONSIDERATION OF ITS MICROSTRUCTURE

Maksym Karpash¹, Evgen Dotsenko¹, Oleg Karpash¹, Amitava Mitra²

¹Ivano-Frankivsk National Technical University of Oil and Gas, Ivano-Frankivsk, 76019, Ukraine
²National Metallurgical Laboratory, Jamshedpur, 831007, India

Metallic structures of long-term operation life in oil and gas industry are exposed to such factors as
the cyclic loads, temperature changes, pressure pulsations, aggressive medias, etc. For example, the
operation of pipelines under the influence of residual internal pressure steel pipeline withstands
tensions that are close to the normative strength characteristics of metal pipes. Such situation relates
with a complicated scheme of pipeline interaction with the soil, presence of deviations from
regulatory requirements during construction of the pipeline, presence of flaws in metal, weldments
and failuring modes of technological operations. Moreover, ‘aging’ of pipelines has been reported
from many operators based on their performance experience. In oil and gas pipeline steels most
common failures are caused by strain ageing. This is mostly due to not only cyclic nature of the
loading, but also due to some specific alloying elements. Availability of the atomic hydrogen in the
soil electrolytes, hydrogen sulfide oils, its generation during cathodic reactions leads to a significant
growth of embrittlement of tube steels in case of prolonged cyclic loading. Combination of these
processes cause change of the microstructure and physical-mechanical properties of steel, i.e.
ageing [1]. Thus aging of the pipeline steels during operation can be associated with changes at the
level of crystalline structure of metal and with redistribution of impurity atoms and dislocations.
Some investigators report about significant changes of such mechanical properties as yield/tensile
strength (changes up to 20%), impact toughness (up to 30%) etc. Accurate determination of the
actual technical state of biggest number of long-term metallic structures in the basic industries is
linked with determination of actual mechanical properties and the estimation of stress-strain state
using non-destructive methods [2]. Factors involved in ageing process negatively affect the
remaining life of the structure and can cause fatigue defects and structure damage [3]. Therefore,
determination of actual mechanical properties and evaluation of the stress-strain state of steel
structures, especially of long-term operation is an important scientific and technical problem, which
solution will eventually lead to more accurate evaluation of their technical state. We consider the
results of experimental investigations aimed on improvement of methods for determination of actual
mechanical properties. Previously new method for evaluation of mechanical properties of steel has
been proposed [4], which lays in measurement of hardness and the heat conductive parameter and
use of artificial neural networks for calculation of yield strength. Special experimental device FMH-
1 has been develop for implementation of the proposed method [5]. However, in these works due
attentioan was not paid to the investigation of dependence between mechanical properties and
electric resistivity as only in work [3] only structural sensitivity of this parameter was mentioned
without any experimental verification.

The aim of this work was to study structural sensitivity of the electrical resistivity and improve
method of mechanical properties evaluation. For the evaluation of microstructure of steel
measurement of structural-sensitive parameters such as thermal conductivity and electrical
resistivity shall be done. Given the structural sensitivity of thermal conductivity and electrical
resistivity, it is reasonable to investigate how the parameters are combined depending on the type of
steel structures. The fact of the relationship of electrical resistivity and thermal conductivity of pure
metals has been known and lay the foundation of Widemann-Franz law [6]. Firstly, based on
reference data for different structural types of steel verification of the law has been done.
Widemann and Franz based on experimental data found that for all metals ratio of thermal
conductivity λ to electrical conductivity σ is constant at the same temperature:
λ
= const , (1)
σ
 W ⋅m Sm
where λ - thermal conductivity coefficient, , σ - electric conductivity .
К m

λ
Also Lorentz showed that the ratio is directly proportional to absolute temperature T:
σ

λ
= L ⋅T, (2)
σ

where L – constant same for all metals, called the Lorentz number [6].

Correlation of electric conductivity and thermal conductivity is explained by the fact that both these
properties of metals caused, basically, by movement of electrons. Widemann-Franz law is valid
only for pure metals. Given the fact that steel is a complex compound or mechanical mixture of
iron, carbon and alloying elements assumption that ratio (1) is not constant for different steel grades
depending on its microstructure can be made. In the course implementation of the State Scientific-
Technical Programme “Resurs” in Ivano-Frankivsk National Technical University of Oil and Gas a
specialized problem-oriented database that contains information about the physical and mechanical
properties of steels that are used in oil and gas industry has been created [5]. In order to check the
above mentions theoretical idea, we used data from the developed database for selected properties
of 142 foreign steel grades, which were selected according to the different types of microstructure:
1) austenitic steels (88 grades);
2) ferritic steels (12 grades);
3) duplex steels (26 grades);
4) martensitic steels (16 grades).
Here are the ranges of the target properties of the selected steel grades:
- Tensile strength: 400-2200 MPa;
- Yield strength: 145-1800 MPa;
- Hardness: 140-332 HB;
- Thermal conductivity: 9.1-32.3 W/m·K;
- Electrical resistivity - 500-1450 nOhm·m.
Electrical resistivity and electrical conductivity are linked by inversely proportional dependence:
1
σ=
ρ (3)
Substituting expression (3) into (1) and given the assumption that the ratio (1) is not constant for
steel within different microstructural groups index Ki was calculated for every steel grade:
K i = ρ i ⋅ λi , (4)
as well as the mean value Kc for each microstructure:
n

∑K i
Kc = i =1
,
n (5)
where: де ρ i - electrical resistivity of i-index steel grade, Ohm ⋅ m ; λi - thermal conductivity of і-
W
index steel grade, m ⋅ К ; n – number of grade in microstructural group. Results of Kc calculation
are given in Table 1.
 Table 1 - Results of the calculation of the Kc coefficient
No Microstructure Value of Kc 10-6 Range of values К 10-6
1 Austenitic 12.05 9.9-13.6
2 Duplex 12.55 11.8-13.6
3 Martensitic 15.45 13.7-18.0
4 Ferritic 15.72 14.0-17.7

As can be seen from Table 1, coefficient Kc is different for different types of steel microstructure.
But clear identification of the of the steel grade belonging to any of the microstructure is possible
only for two groups of steels. The first group includes austenitic and duplex steels, second one
includes ferritic and martensitic steels. Obviously that after some additional investigation with
possible use of nonlinear classification methods this approach can lead to determination of the steel
grade.

Considering complicated nature of the relationship between input parameters and mechanical
properties [3] neural networks were used for non-linear approximation of the mechanical properties
as function of hardness, electrical resistivity and thermal conductive parameter [7]. As neural
networks routine require training and testing procedures the structural sets of steels were formed
into training and testing sets:

o 1st set includes austenitic and duplex steel grades (90 grades were used for training
and 5 grades – for testing);
nd
o 2 set includes ferritic and martensitic steel grades (37 grades were used for testing
and 3 for testing):
rd
o 3 set includes all steel grades (127 steel grades were used for testing and 8 grades
for testing.

Electrical resistivity and thermal conductive parameter were used as input parameters for the neural
network while yield strength was used as output. It shall be noted that according to the established
practice [7], testing the neural networks need to be performed on data which were not used for
training (unknown for the neural network) in order to ensure the ability of the network to
approximate the desired function and meet the NDT requirements.

Levenberg-Marquardt algorithm has been used as training algorithm for all data sets as this
algorithm is recommended for small networks training. Neural network architecture selection, data
preparation, selection of testing sets and training algorithms usually are based on the investigators
experience [7] and depends on the nature and complexity of the task to be solved. After
comparative studies of the various neural network architectures scheme which can be described as
fully connected neural network with 2 hidden layers (20 and 10 neurons respectively) and single
output neuron.

Testing results of the trained networks for 3 structural data sets are given in the Table 2.

Based on the results given in the Table 2 the following conclusion can be made:

1) Evaluation of the mechanical properties of steels within the groups of same or similar
microstructure allow to increase for 2-3 times the accuracy of the given procedure. This conclusion
is consistent with the prior idea of the relationship between thermal conductivity and electrical
resistivity of steel grades with its microstructure.
 Table 4 – Neural network testing results

Testing steel grade 1 2 3 4 5 6 7 8 Testing error

MPa %
Dataset Real values of 275 310 485 280 450 275 350 560
yield strength
1st Neural network 271 313 492 283 430 - - - 7.4 3.52
outputs
2nd Neural network - - - - - 273 384 562 12.6 4.42
outputs
3rd Neural network 276 316 483 273 379 328 345 478 28.3 9.92
outputs

2) Use of artificial neural networks allow establishment of the relationships between

informative parameters and target mechanical properties (i.e. yield strength).

3) Achieved results confirm choice of the NDT informative parameters (thermal

conductivity and electrical resistivity).

Additionally the achieved results were partly confirmed in the frame of the Indo-Ukrainian
projected “Microstructural evaluation of steel used in petrochemical industry by NDE technique”,
which is being performed by Ivano-Frankivsk National Technical University of Oil and Gas
(Ukraine) and National Metallurgical Laboratory (Jamshedput, India). Series of samples which
were put into operation in Ukrainian gas transport system were examined using non-destructive
means (including thermal conductive parameter measurement) and microscopic methods.

Figures 1 and 2 represent microstructural patterns of the two different steel samples. At Figure 1
there is ferrite-pearlite structure which is evenly distributed through out the sample. At Figure 2
grain sizes are very irregular, large patches of very fine pearlitic colony is unevenly distributed.

Figure 1 – Ferrite-pearlite structure of the steel used in gas piplines

 Figure 2 – Another steel sample
used in gas pipeline

Structural difference of the samples, shown at Figures 1 and 2 was confirmed by the measurements
of the thermal conductive parameter as their values differ for more than 20%.

Obviously, the obtained theoretical results have to be confirmed by electrical resistivity

measurements and tensile tests.

Thus, as a result of research a new approach to the evaluation of the mechanical properties of
metallic structures was developed, which lays in:

• calculation of the product of thermal conductivity and electrical resistivity (according to the
Widemann-Franz law) for determination of the microstructure group of the steel;
• use of neural network within the defined microstructural group for prediction of the
mechanical properties with thermal conductive parameter and electrical resistivity as input
parameters.

References

1. V.V.Kurochkin, N.A.Malushin, O.A.Stepanov, A.A.Moroz.,2003. Operational reliability of

oil pipelines. Nedra-Biznestsentr, Moscow, 231 pp. (in Russian)
2. The Cabinet of Ministers of Ukraine, on October 8, 2004, No.1331 "On approval of the
State Scientific and Technical Programme "Resurs".
3. Karpash M.O.,2004. Justification of the novel approach to the determination of physical and
mechanical properties of metallic structures. Methods and instruments of quality control.
Vol.12, 30-33. (in Ukrainian with English abstract)
4. Karpash O., Karpash M. New neural-based method for evaluation of mechanical properties
of steels. Proceedings of 9th International Conference of Non-destructive testing, Berlin, 25-
29 September 2006, P114.
5. Report No NC/409/2007 "Development of new methods and technical means of
determination of physical and mechanical characteristics of long-term operation
technological objects aimed to create data banks on the physical and mechanical properties
of constructional materials", Reg No. 0107U08068.
6. Ashcroft N., N. Mermin, 1979. Solid State Physics in 2 volumes. Vol.2. Mir, Moscow, 399
pp.
7. Haykin, S., 1999. Neural networks. A comprehensive foundation, Second edition.Prentice
Hall, New Jersey. 842.