Metallography, Microstructure, and Analysis (2025) 14:1271–1277

[Link]

ORIGINAL RESEARCH ARTICLE

Impact of Microstructural Phase Percentages Quantified Using KNN

Algorithm on Torsional Strength and Microhardness of AISI 4140
Heat‑Treated Steel
Zainab A. Ibrahim1 · Nathera A. Saleh1 · Murtadha A. Jabbar1

Received: 17 July 2024 / Revised: 8 August 2025 / Accepted: 24 August 2025 / Published online: 15 September 2025
© ASM International 2025

Abstract
Different microstructures were generated through various heat treatments on AISI 4140 low alloy steel. The microstructures,
including ferrite-pearlite, ferrite-bainite, martensite, and ferrite-martensite, were created through a series of metallurgical
processes to build a dataset of microscopic images. The specimens were subjected to heat treatments at different tempera-
tures in the range of 800–950 °C (in increments of 25 °C), followed by quenching in various media (RO water, brine, oil,
and static air). The images were then segmented using the KNN algorithm, and the phases were quantified, to investigate
the influence of phase fractions and types on torsion strength and microhardness properties. Torsion and microhardness tests
were performed at room temperature on 30 groups of specimens. The results showed that the type and distribution of phases
were the dominant factors in determining the properties of a specific microstructure, rather than the exact phase percentages.

Keywords Microstructure · Microhardness · Torsion · KNN algorithm · Heat treatment

Introduction images, by considering the influence of the shape, distribu-

tion, and volume fractions of the microstructural phases [1].
In metallurgy, the properties of metals and alloys are closely Steel, one of the most widely used engineering materials,
linked to their internal structures, known as microstructures. exhibits a range of microstructures such as ferrite, pearl-
These microstructures evolve through processes such as ite, bainite, martensite, and austenite, depending on com-
alloying, mechanical deformation, and heat treatment, which position, processing, and heat treatment conditions [2–4].
significantly influence the resulting mechanical properties The final properties of steel are highly sensitive to the heat
of the material. Traditional metallurgy has long relied on treatment process, which controls phase transformations.
destructive and non-destructive mechanical testing to assess Despite this, many conventional studies focus predominantly
properties such as strength, hardness, impact, torsion, and on mechanical property evaluation, often overlooking the
ductility. However, recent advancements in image analy- microstructural origins of these properties.
sis have enabled researchers to derive meaningful insights Several previous studies have examined the influence of
about mechanical performance directly from microstructural heat treatment and torsional loading on metallic materials
using different approaches. Bressan et al. [5] designed and
validated a torsion test machine capable of accurately measur-
Nathera A. Saleh and Murtadha A. Jabbar authors contributed
equally to this work. ing the torsional behavior of materials, providing an experi-
mental platform for studying the relationship between twisting
* Zainab A. Ibrahim loads and material response. Silva et al. [6] investigated how
[Link]@[Link]
heat treatment affected the torsional resistance and surface
Nathera A. Saleh roughness of nickel–titanium instruments, demonstrating
[Link]@[Link]
that thermal processes can significantly alter both mechanical
Murtadha A. Jabbar performance and surface characteristics. Hjorth [7] focused
[Link]@[Link] on an aluminum alloy, revealing that deformation by torsion
1
Mechanical Engineering Department, College followed by heat treatment-induced notable microstructural
of Engineering, University of Basrah, Basrah, Iraq changes, which, in turn, influenced mechanical behavior. Cota

Vol.:(0123456789)
1272 Metallography, Microstructure, and Analysis (2025) 14:1271–1277

et al. [8] characterized the microstructural evolution of bainitic While these studies mark significant progress, most focus
steel subjected to torsion testing and interrupted accelerated either on classification or segmentation of microstructures
cooling, highlighting the interplay between deformation, cool- with generic descriptors, and they seldom correlate micro-
ing rate, and phase transformation. Jabbar [9] applied artificial structural phase fractions directly with mechanical test
neural networks to study the effect of retained austenite on the results, such as microhardness or torsional strength.
mechanical properties of low alloy steel, demonstrating the The present study addresses this gap by integrating quan-
potential of predictive tools to link microstructure and perfor- titative microstructural analysis with mechanical testing for
mance. Funsho [10] investigated the effect of rapid cooling AISI 4140 low alloy steel. A series of heat treatments were
after heating on the torsional properties of 0.4% carbon steel applied to induce diverse microstructures, which were then
but did not quantitatively characterize the resulting micro- analyzed using optical and electron microscopy. Phase iden-
structures. Similarly, Lima [11] and Abaas [12] explored heat tification and quantification were conducted using image
treatment processes aimed at improving strength in steels, but processing techniques suitable for the specific microstruc-
the relationship between mechanical improvements and the tural features of this steel. These results were correlated
specific phase fractions present was not thoroughly addressed. with torsion and microhardness tests to establish a clearer
Although these works have provided valuable insights understanding of how phase fractions influence mechanical
into torsional behavior and other mechanical properties, heat properties. This dual approach not only deepens the under-
treatment effects, and microstructural characterization, most standing of structure–property relationships in heat-treated
have not quantitatively linked specific phase fractions to the steels but also demonstrates the potential of image-based
observed mechanical property changes. Moreover, the inte- phase quantification as a complementary tool for mechani-
gration of advanced image analysis techniques to directly cal evaluation.
correlate microstructural features with mechanical perfor-
mance remains limited. This gap highlights the need for fur-
ther research that combines mechanical testing with detailed Experimental Work
microstructural quantification to better understand the origins
of material behavior. After selecting and supplying the material, a metallographic
In recent years, researchers have begun addressing this chal- process consisting of several steps was followed to generate
lenge by turning toward computational image analysis and a collection of various microstructures for image analysis.
machine learning techniques to classify and quantify phases These steps included specimen cutting and facing, heat treat-
from microstructural images. Banerjee et al. [13] applied seg- ment, tempering, specimen preparation involving grinding
mentation techniques to dual-phase steel micrographs, ena- and polishing, etching, and imaging using optical micro-
bling quantification of ferrite and martensite. Liu [14] used scope (OM) and scanning electron microscope (SEM). This
the K-Nearest Neighbor (KNN) to classify various steel micro- methodology enabled the identification and quantification
structures, such as two-phase, fully pearlite, and mixed micro- of the present phases, which were evaluated under different
structures, and compared the results to a manual evaluation, heat treatment conditions.
while Naik et al. [15] applied supervised machine learning for Material and Heat Treatment AISI 4140 low alloy
texture-based phase identification in structural steel. Tsutsui medium carbon steel was the selected material for this study,
et al. [16] used textural features from scanning electron micro- its chemical composition is shown in Table 1. Ninety speci-
scope (SEM) images and a Random Forest (RF) classifier to mens were divided into 30 groups for testing, including the
identify phases in low-carbon steels. Paul et al. [17] expanded as-received and annealed cases, and 28 groups heated above
on this by integrating local entropy and regional contour pat- the austenitizing temperature in the range of 800–950 °C
terns as features to quantify bainite, martensite, and ferrite. (in increments of 25 °C), followed by quenching in various
Fernandes et al. [18] focused on graphite, pearlite, and ferrite media (RO water, brine, oil, and static air). The final step
in nodular cast iron using thresholding and segmentation tech- was tempering at 222 °C, to change the brittle martensite
niques, and Gola et al. [19] combined pixel- and morphology- phase into a more ductile tempered martensite phase. All
based descriptors for classifying phases such as pearlite and these procedures were carried out following the ASM heat
bainite using support vector machines. treating standard [20].

Table 1  Chemical composition Elements Values type C Cr Mo Mn Si Ni

of AISI 4140 steel
Measured values Wt. % 0.38 0.94 0.222 0.88 0.291 0.058
Standard values Wt. % [21] 0.38–0.43 0.90–1.2 0.15–0.3 0.60–0.85 0.15–0.35 …
Metallography, Microstructure, and Analysis (2025) 14:1271–1277 1273

Specimen Preparation AISI 4140 low alloy steel was sup-

plied in 50-cm-long hexagonal bars. The material was cut
using a band saw and machined on a lathe to the desired
shape and dimensions of specimens, to be used later for
microscopic and mechanical tests. The hexagonal shape was
intended as the standard torsion specimens had hexagonal
shoulders to ensure a proper fit in torsion testing machine
grips. For microstructural analysis, 3 cm length specimens
were taken from the hexagonal bars. After heat treatments,
the specimens were ground with emery paper ranging from
180 to 2000 grit, then polished with 35, 25, 20, 14, 5, 1.5,
1, and finally 0.5-micron particle sizes diamond paste. The
specimen preparation process was conducted according to
ASTM Standard E3-11 [22].
Etching To reveal the microstructures, specimens were
etched with a solution containing 10 grams of picric acid,
Fig. 1  Distribution of microhardness points from surface to the center
5 drops of HCL, and 100 ml of methanol (95%) [23]. This of the specimen
etchant provided clear images with well-defined matrix and
second-phase constituents. The specimens were immersed
in the etchant for 20–30 seconds depending on the type of test parameters and results. The equipment comprises a
the present phases. horizontal torsion machine featuring an electric motor, a
Microscopy To create an image dataset for microstruc- wheel, a horizontal shaft, and a control and data acquisition
ture identification and phase quantification processes, both system. Ninety specimens were subjected to torsion testing,
optical microscope (OM) and scanning electron microscope with three specimens for each case of the 30 cases outlined
(SEM) were utilized. For this purpose, an Olympus light in Table 2. The torsion specimen dimensions are shown in
microscope and Axia Thermos Scientific scanning elec- Fig. 2. The testing machine was connected to a Versatile
tron microscope were used to capture the microstructures Data Acquisition System (DAS); which facilitated the dis-
at magnifications of X1000 and X5000, respectively. The play of torsion data.
test procedures were carried out according to ASTM E986
standard [24].
Images Segmentation and Quantification

Mechanical Tests After capturing the microstructural images, the next step was
to quantify the present phases, through computerized image
Microhardness Test segmentation and quantification processes. Image segmen-
tation is the first step in phase quantification, in which the
In the microhardness test, a diamond indenter of specific image is divided into separate, distinct, and non-overlap-
geometry is pressed into the surface of the prepared speci- ping regions [27, 28]. Various segmentation techniques are
men using a known applied force. The hardness value is designed to suit different types of images. Simple methods,
determined by measuring the dimensions of the resulting such as thresholding, are based on pixel intensities, whereas
indent on the specimen's surface. In the Vickers test, the more advanced approaches utilize complex statistical fea-
indent formed has a diagonal shape, and both diagonals are tures such as texture, pattern orientation, distance, similarity,
measured to calculate the average value for the Vickers hard- and other attributes. These advanced methods enable more
ness value (HV) [25]. Vickers hardness test was employed precise recognition and classification of objects of interest
with a 0.3 kgf load, following the ASTM E92 standard [26]. within an image [29]. Microstructure images present a sig-
Multiple measurements were taken from the surface to the nificant challenge due to their diverse microstructural char-
center of the specimen to establish a hardness value distribu- acteristics, including overlapping colors, indistinct bounda-
tion curve across this distance, as shown in Fig. 1. ries, varying textures within phase regions, and low contrast
between phases [30]. Furthermore, the images reveal phases
Torsion Test colonies (or patches) of infinite shapes and sizes distrib-
uted across the image, connected and disconnected regions,
The SM 1001 Torsion Testing Machine from TecQuipment, varying grain and colony orientation angles, and poorly
with a torque capacity of 30 N·m was used to obtain torsion defined grain boundaries, etc. In this study, the colored
1274 Metallography, Microstructure, and Analysis (2025) 14:1271–1277

Table 2  The type and percentage of the phases for all cases as seg- not satisfying. This algorithm yielded the best results, so it
mented and calculated using KNN algorithm was chosen to obtain the phases fractions.
Heat treatment cases Pearlite Ferrite Bainite Tempered Lath
martensite mar- K‑Nearest Neighbor Algorithm (KNN)
tensite

As received 88.17 12.1 … … … The K-Nearest Neighbor (KNN) is a supervised machine

Annealed 71.16 28.845 … … … learning algorithm, that can be used for both image seg-
800 Ro 63.46 36.54 … … … mentation and classification. In this study, KNN was used
800 Brine 32.7 67.3 … … … for segmentation purpose.
800 Oil 45 55 … … … In the segmentation process, KNN groups the pixels into
800 Air 61.34 38.66 … … … different regions based on their similarity in color, texture,
825 Ro … 42.56 57.43 … … or other features according to predefined training examples.
825 Brine 78 22 … … … The training examples generate vectors in a multidimen-
825 Oil 45.93 54.075 … … … sional feature space, each associated with a class label. In
825 Air 60.58 39.425 … … … the next step, an unlabeled vector (test point) is categorized
850 Ro … 12 … 88 … by assigning the label most frequent among the k-nearest
850 Brine … 3 … 97 … training samples [31, 32]. In simpler terms, each pixel is
850 Oil 37.8 62.205 … … … considered as a point in a high-dimensional feature space,
850 Air 71.1 28.905 … … … and its neighbors are identified based on their distance to the
875 Ro … 9 … 91 … pixel. The KNN algorithm assigns the pixel to the region
875 Brine … 14 … 86 … that has the most similar neighbors. This process is repeated
875 Oil … 5 … 95 … for all pixels in the image, resulting in a segmentation map
875 Air 66.47 30.31 … … … that divides the image into different regions based on their
900 Ro … … … 100 … similarities. This method has several advantages, such as
900 Brine … 35 … 65 … intuitive design, flexible decision boundaries, simplicity,
900 Oil 47 53 … … … and ease of implementation. It does not require huge train-
900 Air 77 23 … … … ing data or complex models, making it suitable for small
925 Ro … … … 100 … datasets. Figure 3 shows example for segmentation steps and
925 Brine … 5 … … 95 results for pearlite-ferrite microstructure.
925 Oil … 2 … … 98 The new segmented images obtained after applying the
925 Air … 63 37 … … KNN algorithm was further processed into binary images
950 Ro … 3 … 97 … using MATLAB code to quantify the phase fraction accord-
950 Brine … 5 … 95 … ing to threshold values occupied by each phase. All software
950 Oil … 6 … … 94 codes utilized are available upon request.
950 Air 89.16 10.84 … … …

Results and Discussion

optical microscopic images were used for phase quantifi-
cation purpose, as they carry more features than the SEM Microhardness Test Results
gray scale images. A supervised machine learning algorithm
was selected for segmentation purpose by classifying pixels Generally, the results of hardness test coincided with those
based on their similarity to neighboring pixels. Many meth- reported in engineering standards for AISI 4140, so there
ods were experimented for this purpose, such as thresholding is no need to list the obtained values extensively. Regard-
using MATLAB and ImageJ software, but the results were ing the relationship between the microstructures and

Fig. 2  The dimensions of tor-

sion test specimen
Metallography, Microstructure, and Analysis (2025) 14:1271–1277 1275

Fig. 3  Example for segmen-

tation steps and results for
pearlite-ferrite microstructure

hardness, the annealed specimens, with ferritic-pearlitic Torsion Test Results

microstructure and about 71% pearlite, showed the lowest
HV values (209 HV) among all the 30 cases. This result The results of the torsion test are consistent with those
was anticipated as the annealing process is known to sof- reported in engineering standards, so there is no need to
ten the materials. It is well-documented that ferrite in an list the obtained torsion parameters extensively. But gener-
annealed state typically has low hardness, ranging from 60 ally, the results showed an increase in the maximum shear
to 100 HV, which can increase up to 250 HV through sig- stress for all the heat-treated cases compared to the as-
nificant grain refining [33]. However, the results revealed received specimens (except for four failed cases), along
many cases with ferritic-pearlitic microstructures, but with an increase in the maximum angle of twist endured
with different ferrite/pearlite fraction and different micro- for most cases. This observation highlights the signifi-
structural and spatial characteristics. The type and proper- cance of heat treatments in enhancing torsion resistance
ties of ferrite and pearlite formed during annealing have properties. Another noteworthy finding is the inverse
lower hardness and higher ductility than that formed in the relationship between material ductility and the maximum
quenched cases due to the very slow cooling rate. Thus, angle of twist experienced by the specimens, as well as
the other microstructures even with lower pearlite percent- the reverse relationship between ductility and maximum
ages exhibited higher hardness values. shear strength. Specifically, more ductile specimens exhib-
The three highest HV values were observed in the cases ited a greater angle of twist before failure and lower shear
of 850 brine, 875 RO water, and 900 RO water, with an strength, whereas less ductile ones showed the opposite.
increase of 184%, 150%, and 150%, respectively, com- Therefore, the annealed and other pearlite/ferrite micro-
pared to the as-received case, with almost martensitic structures demonstrated the highest angle of twist.
microstructures. There were other cases with close or even Four cases out of the 30 cases were failed quickly in
more martensite percentages at the temperatures above the torsion test. These cases were the 925 RO water, 925
900 °C, but they recorded decreased hardness due to the brine, 950 RO, and 950 brine with almost full martensitic
carbide coarsening, grain growth. The three lowest HV microstructure. They already had unseen quenching cracks
values were recorded in the annealed case, 800 RO water, extending from the surface to the center and along the
and 825 air cases, with ferrite-pearlite microstructures. length of the specimens. These quenching cracks likely
Where the results of these cases indicate a degradation arose due to the high-temperature difference between the
in hardness, as these values were even lower than the as- specimens and the severe quenching mediums (RO water
received case. There were also many microstructures with and brine).
lower and higher pearlite/ferrite fractions and different
hardness values, due to the different characteristics of the
pearlite and ferrite formed at different conditions.
1276 Metallography, Microstructure, and Analysis (2025) 14:1271–1277

Microstructure and Segmentation Results present phases. For instance, both the as-received and
annealed cases exhibited a pearlite-ferrite microstruc-
Figure 4 shows a sample of the microstructures as observed ture with varying phase proportions. Despite having
in the OM and SEM, while Table 2 shows the type and the approximately 71% pearlite, the annealed specimen
calculated phase percentages for all cases, where the per- showed lower hardness than the as-received speci-
centages listed in the table are the average percentages cal- men having approximately 88% pearlite. Other cases
culated for three images for each case. The microstructures revealed instances of ferritic-pearlitic microstructures
observed with the SEM images were consistent with those with a lower pearlite fraction than the annealed sample
identified and labeled in the OM images, but with more but exhibited higher hardness. This contradictory obser-
detailed characteristics due to the higher magnification of vation leads to the conclusion that the type, shape, and
SEM images. distribution of pearlite in the microstructures are more
The results indicate that half of the cases revealed ferri- significant and influential than its percentages. This con-
tic-pearlitic microstructures with varying percentages and clusion applies with the bainitic microstructures as well.
characteristics, two cases displayed ferritic-bainitic micro-
structures. While the remaining 13 cases showed martensitic Accordingly, the following sub-conclusions drawn from
microstructures, including tempered martensite, lath mar- the microhardness results support this main prior conclusion:
tensite, and ferritic-martensitic microstructures with differ-
ent percentages. Martensite formation began above 850 °C. • Agglomerated pearlite with compact colonies and fine
lamella spacing exhibited hardness below 300 HV.
• Agglomerated pearlite with medium lamella spacing
Conclusions showed hardness exceeding 300 HV.
• Homogeneous pearlite distribution with coarse lamella
The following conclusions can be drawn from this study: spacing resulted in hardness exceeding 400 HV.
• Uniformly distributed pearlite without distinct colonies
1. The results indicate that the hardness and torsion resulted in hardness exceeding 500 HV.
strength are more influenced by the type, distribution, • The fine and agglomerated in small distant colonies bai-
and heat treatment conditions of the resulting micro- nitic microstructures, contributed to increased hardness
structure rather than the precise percentages of the ranging between 500 and 600 HV.

Fig. 4  Sample of the obtained microstructures as appeared in the OM and SEM

Metallography, Microstructure, and Analysis (2025) 14:1271–1277 1277

• And finally, the martensitic microstructures alone or with 17. A. Paul, A. Gangopadhyay, A.R. Chintha, D.P. Mukherjee, P. Das,
little ferrite gave hardness in the range of 600–800 HV. S. Kundu, Calculation of phase fraction in steel microstructure
images using random forest classifier. IET Image Process. 12(Is. 8),
1370–1377 (2018)
18. D. de Oliveira Fernandes, C.T.M. Anflor, J.N.V. Goulart, B.
2. The K-Nearest Neighbor machine learning algorithm Baranoğlu, Nodular cast iron GGG40, 60, 70 mechanical charac-
(KNN) can be effectively applied for the task of micro- terization from bars and blocks obtained from Brazilian foundry.
Metals. 12, 1115 (2022)
structural image segmentation, as well as classification 19. J. Gola, J. Webel, D. Britz, A. Guitar, T. Staudt, M. Winter, F.
tasks. Mücklich, Objective microstructure classification by support vector
machine (SVM) using a combination of morphological parameters
and textural features for low carbon steels. Comput. Mater. Sci. 160,
186–196 (2019)
References 20. ASM International Handbook Committee, American Society for
Metals, Heat Treating, Vol.4, ASM International, Ohio, USA, 2002.
1. H. Bhadeshia, R. Honeycombe, Steels: Microstructure and Proper- 21. ASM International Handbook Committee, Properties and selec-
ties (Butterworth-Heinemann, 2017) tion: irons, steels, and high-performance alloys (ASM International,
2. C.C. Tasan, M. Diehl, D. Yan, M. Bechtold, F. Roters, L. Schem- USA, 1995)
mann, D. Raabe, An overview of dual-phase steels: advances in 22. ASTM (the American Society for Testing and Materials), Standard
microstructure-oriented processing and micromechanically guided Guide for Preparation of Metallographic Specimens, ASTM E3-11,
design. Annu. Rev. Mater. Res. 45, 391–431 (2015) ASTM International: West Conshohocken, PA, USA ,2017
3. P. Khedkar, R. Motagi, P. Mahajan, G. Makwana, A review on 23. ASTM (the American Society for Testing and Materials),” Standard
advance high strength steels. Int. J. Curr. Eng. Technol. Special Practice for Microetching Metals and Alloys1”, book. ASTM stand-
Is(6), 240–243 (2016) ard E407 − 07, ASTM International, (2011)
4. G.S. Krauss, Steels: Processing, Structure, and Performance (ASM 24. ASTM " Standard practice for scanning electron microscope beam
International, 2015) size characterization" Annual Book of ASTM standard E986 – 04,
5. J.D. Bressan, R.K. Unfer, Construction and validation tests of a ASTM International 2010.
torsion test machine. J. Mater. Process. Technol. 179(1–3), 23–29 25. ASM Handbook, Mechanical testing and evaluation, vol 8 (Ameri-
(2006) can Society for Metals, Metals Park, Ohio, USA, 2003)
6. E.J.N.L. Silva, J.F.N. Giraldes, C.O. De Lima, V.T.L. Vieira, C.N. 26. ASTM " Standard test method for Vickers hardness of metallic
Elias, H.S. Antunes, Influence of heat treatment on torsional resist- materials” Annual Book of ASTM standard E92– 82, ASTM Inter-
ance and surface roughness of nickel-titanium instruments. Int. national 2003.
Endod. J. 52(11), 1645–1651 (2019). [Link] 27. S. Abdulateef, M.D. Salman, A comprehensive review of image seg-
13164 mentation techniques. Iraqi J. Electr. Electron. Eng. (2021). [Link]
7. A.S. Hjorth, The Effect of Deformation by Torsion and Subsequent doi.​org/​10.​3791/​ijeee.​17.2.​18
Heat Treatment on the Microstructure of AA6082. MS thesis. 28. Y. Yu, C. Wang, Q. Fu, R. Kou, F. Huang, B. Yang, T. Yang, M.
NTNU (2018). Gao, Techniques and challenges of image segmentation: a review.
8. A.B. Cota, D.B. Santos, Microstructural characterization of bainitic Electronics. 12, 1199 (2023). [Link]
steel submitted to torsion testing and interrupted accelerated cooling. s1205​1199
Mater Charact. 44(3), 291–299 (2000) 29. W.N. Jasim, R.J. Mohammed, A survey on segmentation techniques
9. M.A. Jabbar, Effect of Retained Austenite on Mechanical Properties for image processing. Iraqi J. Electr. Electron. Eng. (2021). [Link]
of Low Alloy Steel Using Artificial Neural Networks, Ph.D. thesis, doi.​org/​10.​37917/​ijeee.​17.2.​10
Mechanical Engineering Department College of Engineering Uni- 30. Z.A. Ibrahim, N.A. Saleh, M.A. Jabbar, Microstructural images seg-
versity of Basra (2010). mentation techniques: a review. Basrah J. Eng. Sci. 24(1), 48–56
10. Funsho and Mayowa, "The Effect of Heat Treatment on the Tor- (2024). [Link]
sional Properties of TR1021 Carbon", Conference Paper, 2018. 31. X. Liu, "Microstructural Characterization of Pearlitic and Complex
11. R. Lima, F. Tolomelli, A. Clarke, K. Clarke, J. Spadotto, F. Assun- Phase Steels Using Image Analysis Methods", Ph.D. thesis, Metal-
ção, Microstructural characterization of a 1100 MPa complex-phase lurgy and Materials Science, School of Engineering, University of
steel. J. Mater. Res. Technol. Area. 17, 184–191 (2022) Birmingham, 2014.
12. A.M. Abaas, A.A. Ramadhan, F.H. Hasan, Study the effect of cool- 32. Z.A. Ibrahim, M.A. Jabbar, N.A. Saleh, Automated segmentation
ing medium on the torsion resistance and hardness of medium car- and quantification of retained austenite in AISI 4140 using image
bon steel, in IOP Conf. Ser.: Materials Science and Engineering, pp processing techniques. Metall. Res. Technol. 121(404), 1–10 (2024).
405 (2018). [Link]
13. S. Banerjee, S. Ghosh, S. Datta, S. Saha, Segmentation of dual 33. Marco Boniardi and Andrea Casaroli, Steel metallurgy vol. I,
phase steel micrograph: an automated approach. Measurement. 46, Gruppo Lucefin Research & Development, 2017.
2435–2440 (2013)
14. X. Liu, Microstructural Characterization of Pearlitic and Complex Publisher's Note Springer Nature remains neutral with regard to
Phase Steels Using Image Analysis Methods (University of Birming- jurisdictional claims in published maps and institutional affiliations.
ham, 2014)
15. D. Naik, H. Sajid, R. Kiran, Texture-based metallurgical phase Springer Nature or its licensor (e.g. a society or other partner) holds
identification in structural steels: a supervised machine learning exclusive rights to this article under a publishing agreement with the
approach. Metals. 9, 546 (2019). [Link] author(s) or other rightsholder(s); author self-archiving of the accepted
50546 manuscript version of this article is solely governed by the terms of such
16. K. Tsutsui, H. Terasaki, K. Uto, T. Maemura, S. Hiramatsu, K. publishing agreement and applicable law.
Hayashi, K. Moriguchi, S. Morito, A methodology of steel micro-
structure recognition using SEM images by machine learning based
on textural analysis. Mater. Today Commun. 25, 101514 (2020)