Comparison of Impact Toughness in Simulated Coarse-Grained Heat-Affected Zone of Al-Deoxidized and Ti-Deoxidized Offshore Steels

: The presence of acicular ferrite (AF) in the heat-affected zone (HAZ) of steels used offshore is generally seen as beneﬁcial for toughness. In this study, the effects of varying fractions of AF (0–49 vol.%) were assessed in the simulated, unaltered and coarse-grained heat-affected zones (CGHAZ) of three experimental steels. Two steels were deoxidized, with one using Ti and the other using Al. The characterization was carried out by using electron microscopy, energy-dispersive X-ray spectrometry, electron backscatter diffraction and X-ray diffraction. The fraction of AF varied with the heat input and cooling time applied in the Gleeble thermomechanical simulator. AF was present in one of the Ti-deoxidized steels with all the applied cooling times, and its fraction increased with increasing cooling time. However, in other materials, only a small fraction (13–22%) of AF was present and only when the longest cooling time was applied. The impact toughness of the simulated specimens was evaluated using instrumented Charpy V-notch testing. Contrary to the assumption, the highest impact toughness was obtained in the conventional Al-deoxidized steel with little or no AF in the microstructure, while the variants with the highest fraction of AF had the lowest impact toughness. It was concluded that the coarser microstructural and inclusion features of the steels with AF and also the fraction of AF may not have been great enough to improve the CGHAZ toughness of the steels investigated.


Introduction
The demand for steels that can withstand harsh environmental conditions is increasing due to the opening of new oil fields in ever colder climates. The fine-grained microstructure of the steels used offshore is typically provided by thermomechanically controlled hot rolling processes (TMCP). Both high strength and high toughness in the base plate are obtained by utilizing this approach. Furthermore, weldability is enhanced by using moderately low amounts of carbon and other alloying elements. The combination of strength, toughness and weldability provided by TMCP in these steels is also beneficial for other structural uses [1].
Standards such as, for example, EN 10225-1 require offshore steels to be weldable using submerged arc welding (SAW), gas-metal arc welding (GMAW) or flux-cored arc Table 1. Chemical compositions of the experimental laboratory steels (in wt.%, the remainder being Fe) analyzed using optical emission spectrometry and combustion analysis techniques. A Gleeble 3800 thermomechanical machine was used for simulating the thermal cycle of an unaltered single-pass coarse-grained heat-affected zone (CGHAZ). The samples were cylindrical, with a length of 36 mm and a diameter of 6 mm. The Rykalin 3D cooling model was used with the following parameters: a heating rate of 300 • C/s, a peak temperature of 1350 • C, holding time of 0.1 s and free span at 9 mm. The cooling times from 800 to 500 • C (t 8/5 ) were 5, 24 and 64 s, respectively. Once the samples cooled below 250 • C, they were removed from the machine and allowed to cool in the air to room temperature.

Mechanical Testing
The hardness profile was measured by using a Duramin-A300 (Struers) device under 100 N load (HV 10 ) on the center line of the surface of a section containing the transversal direction (TD) and the plate normal (i.e., thickness) direction (ND).
Standardized instrumented Charpy V-notch impact toughness tests (Zwick Roell PSW750 with TestXpertII software) according to SFS-EN ISO 14556 [40] were carried out at temperatures ranging from −80 • C to −40 • C by using sub-sized 5 × 10 × 55 mm 3 specimens that were oriented transverse to the rolling direction (T-L). Using the instrumented test, it is possible to determine the absorbed energy needed for the nucleation of the fracture crack as well as the energy needed for its propagation to complete the fracture.

Microstructural Characterization
General transformation microstructures in the simulated CGHAZ of Nital etched specimens were studied using a Zeiss Sigma field emission scanning electron microscope (FESEM) with an acceleration voltage of 5 kV and a working distance of 5 mm. The microstructural features were characterized using a Zeiss Sigma FESEM with an EDAX electron backscatter diffraction (EBSD) system with an acceleration voltage of 15 kV. The effective grain sizes were measured from the EBSD data (area 146 × 146 µm; step size 0.3 µm). In order to mitigate the frequency of minuscule grains frequently encountered with EBSD, only grains with an equivalent circular diameter (ECD) larger than 0.55 µm and misorientation greater than 15 • were used in the analysis. The EBSD image quality (IQ) analysis technique presented in [41,42] was used to deconvolute the IQ spectrum into quantitative microstructural features.

Prior Austenite Reconstruction
As revealing the PAGS using common metallography and etching techniques was practically impossible, MATLAB software supplemented with the MTEX toolbox [43] was Metals 2021, 11, 1783 4 of 20 utilized in order to reconstruct the prior austenite grains from the EBSD data. The reconstruction was carried out by two main steps on the basis of the previous works [44][45][46]. Firstly, the orientation relationship between the parent austenite and product ferritic phase, i.e., here mainly bainite, was determined using the Kurdjumov-Sachs (K-S) relationship [47] (i.e., {111} γ //{110} α , 110 γ //111 α ). Secondly, the grain map was divided into separate clusters, and parent austenite orientation was calculated for each cluster discretely in order to reconstruct the austenite orientation map and grain structure.

Non-Metallic Inclusion Characterization
The non-metallic inclusions on the polished cross-sections containing the rolling direction (RD) and the plate normal direction (ND) were characterized with JEOL JSM-7000F FESEM. The measured area varied approximately between 30 and 40 mm 2 . The chemical composition for each detected inclusion was obtained with energy dispersive X-ray spectroscopy (EDS) in the Oxford Inca Feature runs by using an acceleration voltage of 15 kV and a live time of 1 s. Simultaneously, morphological data of each inclusion were recorded; here, the maximum length was used to depict their size. The minimum size for the inclusions to be included in the results was set to 1 µm.
The inclusions were classified in the same manner as in our previous work [37]. According to the EDS analyses, the phase composition of each inclusion was estimated, resulting in the calculated fractions of Al 2 O 3 , MnO, MnS, TiN and TiO 2 components. In practice, inclusions can contain various titanium oxide phases, such as TiO 2 or Ti 2 O 3 , but for the sake of simplicity, TiO x was used to denote all variants. The inclusion classes were constructed based on a 10 wt.% threshold for the components, resulting in 31 combinations that were further combined into nine appropriate inclusion classes: Al 2 O 3 -containing oxides, Al 2 O 3 -containing complexes, MnO-TiO x -MnS (+TiN), MnO-TiO x (+TiN), MnO (+MnS-TiN), TiO x (+MnS-TiN), MnS, MnS-TiN and TiN.

Precipitate Characterization
Qualitative and quantitative analyses of the precipitates within the CGHAZ of the simulated samples were investigated by using transmission electron microscopy (TEM) (JEOL JEM 2200FS EFTEM/STEM) at 200 kV on different carbon extraction replicas. The carbon extraction replicas were prepared as illustrated in [48]. ImageJ software was employed to analyze 10 TEM images with the total investigated area of 100 µm 2 in order to determine the number density and mean ECD of the precipitates. Additionally, the precipitate size at 95% in the cumulative ECD distribution (D90% ppt ) was determined. The chemical composition and types of the precipitates were determined using EDS by analysis of at least 40 precipitates per condition on different carbon extraction replicas to obtain a reliable analysis. Furthermore, the crystallographic structure was determined using a TEM diffraction pattern.

Retained Austenite Determination
Rigaku SmartLab X-Ray diffractometer (XRD) with Co Kα radiation was employed to determine the volume fraction of retained austenite. The operating conditions were an accelerating voltage of 40 kV, a current of 135 mA, a scan speed of 7.1945 • /min, a step size of 0.05 • and a range of 45 • > 2θ > 130 • . XRD data were treated by using whole profile Rietveld refinement analysis. The carbon content of retained austenite in wt.% (Cγ) was determined from the lattice parameter, i.e., a = 0.3578 + 0.0033Cγ [49], where a (nm) is the lattice parameter of retained austenite.

Analysis of Toughness Testing
The CVN absorbed energies of the CCHAZ of the studied steels are presented in Figure 1, and a summary is provided in Table 2. As can be observed, at −40 • C, testing was limited to specimens of t 8/5 = 24 s. At this temperature, all specimens had relatively high impact toughness, and the fracture occurred in a ductile mode. However, the absorbed energy in Ref specimens was statistically higher than in either of the Ti-deoxidized steels.
of the other variants. The lowest energies were observed again in Ti-deoxidized specimens with t8/5 = 64 s. However, the differences were not statistically significant due to inherent scatter in values in the ductile-brittle transition range.
The CVN results were largely unexpected since, based on our previous work [37], the Ti-deoxidized steels and especially Tilow were expected to have an increasing fraction of AF in the microstructure along with increasing cooling time, which, in turn, is often associated with improved toughness performance. Conversely, here, the highest toughness was consistently achieved in the Al-deoxidized reference steel and Tilow, with the longest cooling time showing the lowest toughness. The subsequent chapters aim to elucidate the reasons for this behavior. Figure 1. Impact toughness of the coarse-grained heat-affected zones (CGHAZ) in the studied steels tested at three temperatures (Sub-size Charpy V-notch (CVN) specimens, 5 × 10 × 55 mm). Figure 1. Impact toughness of the coarse-grained heat-affected zones (CGHAZ) in the studied steels tested at three temperatures (Sub-size Charpy V-notch (CVN) specimens, 5 × 10 × 55 mm). At −60 • C, the Ref specimens, regardless of the cooling time, showed higher impact toughness values compared to those of Ti-deoxidized variants. In particular, the Ti low with t 8/5 = 64 s showed poor performance. Similar trends were observed at −80 • C, i.e., the Ref specimens had gradually lower values, but they were still higher on average than those of the other variants. The lowest energies were observed again in Ti-deoxidized specimens with t 8/5 = 64 s. However, the differences were not statistically significant due to inherent scatter in values in the ductile-brittle transition range.
The CVN results were largely unexpected since, based on our previous work [37], the Ti-deoxidized steels and especially Ti low were expected to have an increasing fraction of AF in the microstructure along with increasing cooling time, which, in turn, is often associated with improved toughness performance. Conversely, here, the highest toughness was consistently achieved in the Al-deoxidized reference steel and Ti low , with the longest The typical load-displacement curves obtained from the instrumented Charpy-V tests are presented in Figure 2. The graphs illustrate the differences between the materials, with the Ref specimens having higher displacement, and the total area of the curves compared with the Ti-deoxidized specimens. Three distinct types of curves were observed in the results: (i) completely brittle behavior, where the onset of brittle fracture occurs when the F m is reached; (ii) mixture of ductile and brittle fractures, in which the ductile propagation phase occurs after the F m and is followed by brittle fracture; and (iii) ductile and brittle propagation followed by crack arrest and ductile fracture. All the materials exhibited the ability for crack arrest after brittle propagation at −40 • C (t 8/5 = 24 s), as observed in Figure 2c. However, at −60 • C, the fracture type shifted to brittle fracture in the Tideoxidized specimens. At t 8/5 = 5 s, the ductile propagation phase was virtually absent, whereas a clear ductile region was observed in the specimens with cooling times of 24 s and 64 s. At −80 • C, brittle fracture occurred as F m was reached in the Ti low and Ti high specimens, and it was more pronounced in the t 8/5 = 64 s specimens where extremely low values of displacement were obtained. The Ref specimens demonstrated similar behavior with t 8/5 = 5 s, while a clear ductile zone was observed in the t 8/5 = 64 s specimens, as evidenced by the higher total absorbed impact energy.
The distribution of total energy absorbed during the impact test at -60 • C is presented in Table 3. The fractions of energy absorbed in the crack initiation, crack propagation and crack arrest phases are denoted as E i , E p and E a , respectively. E i is determined from start to F m , E p from F m to the onset of brittle crack propagation and E a from crack arrest to complete fracture. The majority of the Ref specimens demonstrated E i values between 25 and 35 J. The corresponding values for the Ti-deoxidized specimens were similar (23-33 J), except for the Ti low specimens with t 8/5 = 64 s showing initiation energies in the range of 11-28 J. More noticeable differences were observed in the propagation energy E p . In Ref steel, E p was higher than E i in most cases. On the contrary, E i was higher than E p in every specimen of the Ti-deoxidized steels. Notable amounts of E a were only observed on Ref specimens.  )   5  29  19  4  26  7  0  37  10  1  5  13  0  0  26  9  0  29  6  1  5  32  34  8  25  9  1  26  13  1  24  37  52  0  28  17  0  33  19  0  24  33  40  5  29  24  0  33  25  0  64  25  24  0  24  15  0  11  0  0  64  30  46  4  27  21  0  16  0  0  64  30  33  1  23  14  0  14  0  0  64  29  43  4  29  6  0  28  22  0 Clear differences are observed between Ref and Ti-deoxidized steels in the loaddisplacement graphs. The Ref specimens have higher total displacement due to a higher fraction of ductile fracture, whereas crack initiation in the Ti-deoxidized steels resulted in brittle failure more often immediately after reaching the peak load. In addition, Ref specimens showed better crack arrestability and, therefore, considerably higher total displacements. The results for Ti high and Ti low were similar for t 8/5 of 5 and 24 s, but it was considerably lower for Ti low at 64 s. These results suggest that the CGHAZ microstructure of the Ti-killed steels is less effective in impeding the propagation of the crack than that of the Ref steel. Microstructure characteristics such as grain boundary misorientations can have a decisive effect on the propagation of the crack. A fine grain size with a large fraction of high angle grain boundaries provides more obstacles to hinder the propagation of the crack. Additionally, with the differences between the deoxidizing treatments, the inclusion structures may differ from one another. In particular, if the density of large inclusions is higher, there exist more potential sites for brittle fracture initiation. The distribution of total energy absorbed during the impact test at -60 °C is presented in Table 3. The fractions of energy absorbed in the crack initiation, crack propagation and crack arrest phases are denoted as Ei, Ep and Ea, respectively. Ei is determined from start to Fm, Ep from Fm to the onset of brittle crack propagation and Ea from crack arrest to complete fracture. The majority of the Ref specimens demonstrated Ei values between 25 and 35 J. The corresponding values for the Ti-deoxidized specimens were similar (23-33 J), except for the Tilow specimens with t8/5 = 64 s showing initiation energies in the range of 11-28 J. More noticeable differences were observed in the propagation energy Ep. In Ref steel, Ep was higher than Ei in most cases. On the contrary, Ei was higher than Ep in every specimen of the Ti-deoxidized steels. Notable amounts of Ea were only observed on Ref specimens.
Clear differences are observed between Ref and Ti-deoxidized steels in the load-displacement graphs. The Ref specimens have higher total displacement due to a higher fraction of ductile fracture, whereas crack initiation in the Ti-deoxidized steels resulted in brittle failure more often immediately after reaching the peak load. In addition, Ref spec-

Hardness Differences in Simulated CGHAZ
The average hardnesses measured from the simulated CGHAZ of each studied steel using different cooling times are presented in Figure 3. Supplementary hardness measurements were also carried out for CGHAZ specimens simulated using t 8/5 = 17 s, as well as immediate water quenching from peak temperature (t 8/5 ≈ 2-3 s). It can be observed that the hardness decreases monotonously with increasing cooling time in each steel. Additionally, the hardness of the simulated CGHAZ for all the conditions is the lowest in Ref and highest in Ti low . As the value of hardness is related to that of the tensile strength,

Microstructural Characterization
General microstructural images of the studied steels with the shortest and longest applied cooling times are presented in Figure 4. With a short cooling time (t8/5 = 5 s), the microstructure transformation of each steel appeared to mainly have relatively fine lathlike bainitic features, while with a longer cooling time (t8/5 = 64 s), the microstructures were coarse, less hardened and consisted of plate-like bainitic features as well as ferrite. Acicular ferrite can be observed to have nucleated from nearby inclusions. The hardness differences originate from the compositional differences. Carbon content often has a direct correlation to hardness, but these steels had comparable carbon contents (0.05%). Therefore, the differences are related to other alloying elements affecting hardenability. Vanadium, a potent hardening agent in the microalloying range was 0.07% in Ti high and Ti low , while Ref only had a trace amount (0.01%) of V. Additionally, a minor increase in hardenability could be attributed to silicon content, which was higher in Ti low (0.23%) compared to Ti high (0.03%) and Ref (0.01%).

Microstructural Characterization
General microstructural images of the studied steels with the shortest and longest applied cooling times are presented in Figure 4. With a short cooling time (t 8/5 = 5 s), the microstructure transformation of each steel appeared to mainly have relatively fine lathlike bainitic features, while with a longer cooling time (t 8/5 = 64 s), the microstructures were coarse, less hardened and consisted of plate-like bainitic features as well as ferrite. Acicular ferrite can be observed to have nucleated from nearby inclusions.

Phase Fractions
The detailed fractions of different microstructural phases in the studied steels with t8/5 = 5 s and 24 s have been presented in a previous paper and are not reproduced here [37]. However, the phase fractions of t8/5 = 64 s that were obtained similarly by IQ analysis are presented in Figure 5. The phase fractions of all the studied variants are gathered in Table 4.
When the cooling time from 800 °C to 500 °C was extended to 64 s, AF was detected also in Tihigh in small amounts and even in Ref. In Tilow, AF was also observed with shorter cooling times, but at 64 s, its fraction had already increased to 46% of the microstructure. The reasons for the occurrence of AF in this steel and not in the other two have been discussed in prior works and were concluded to be due to coarser prior austenite grains and overall favorable inclusions for AF formation. The increasing fraction of AF could be generally expected to improve the toughness. However, in the present study, Tilow had the weakest impact toughness and even the increasing fraction of AF along the increasing cooling time did not appear to be beneficial. It is still possible that, even in the case with the highest fraction of AF, there still was not enough AF in order to improve impact toughness.

Phase Fractions
The detailed fractions of different microstructural phases in the studied steels with t 8/5 = 5 s and 24 s have been presented in a previous paper and are not reproduced here [37]. However, the phase fractions of t 8/5 = 64 s that were obtained similarly by IQ analysis are presented in Figure 5. The phase fractions of all the studied variants are gathered in Table 4. Table 4. Fractions of different microstructures (%) in the simulated CGHAZ of the studied steels obtained by IQ analysis as well as the volume fraction of retained austenite (RA) (%) and C-content in austenite (wt%) obtained by X-ray diffraction (XRD).    When the cooling time from 800 • C to 500 • C was extended to 64 s, AF was detected also in Ti high in small amounts and even in Ref. In Ti low , AF was also observed with shorter cooling times, but at 64 s, its fraction had already increased to 46% of the microstructure. The reasons for the occurrence of AF in this steel and not in the other two have been discussed in prior works and were concluded to be due to coarser prior austenite grains and overall favorable inclusions for AF formation. The increasing fraction of AF could be generally expected to improve the toughness. However, in the present study, Ti low had the weakest impact toughness and even the increasing fraction of AF along the increasing cooling time did not appear to be beneficial. It is still possible that, even in the case with the highest fraction of AF, there still was not enough AF in order to improve impact toughness.

Microstructure
The volume fraction of retained austenite (RA) in the studied steels obtained by XRD is also presented in Table 4. The fraction of RA was relatively low in all variants, and it was the lowest in the Ref specimens. The presence of RA may indicate martensite-austenite (M-A) islands, which are known to deteriorate toughness in HAZ [31,34,50].
For the sake of comparison, the fraction of RA was also determined by EBSD. The results (in %) were 0. and Ti low , respectively. Due to the difference in the characterization method, the fraction of RA is clearly lower than was observed with XRD. However, the trend is the same that the slower cooling time slightly increases the fraction of RA; moreover, in Ti-deoxidized steels, there is a slightly higher fraction of RA than in the Al-deoxidized reference steel. Figure 6, based on the EBSD data, shows the occurrence and shape of RA in the simulated CGHAZ of Ti low with t 8/5 = 64 s where the highest fraction of RA was detected.
The volume fraction of retained austenite (RA) in the studied steels obtained by XRD is also presented in Table 4. The fraction of RA was relatively low in all variants, and it was the lowest in the Ref specimens. The presence of RA may indicate martensite-austenite (M-A) islands, which are known to deteriorate toughness in HAZ [31,34,50].
For the sake of comparison, the fraction of RA was also determined by EBSD. The results (in %) were 0.2, 0.3 and 0.2. at t8/5 = 5 s and 0.4, 0.5 and 1.0 at t8/5 = 64 s in Ref, Tihigh and Tilow, respectively. Due to the difference in the characterization method, the fraction of RA is clearly lower than was observed with XRD. However, the trend is the same that the slower cooling time slightly increases the fraction of RA; moreover, in Ti-deoxidized steels, there is a slightly higher fraction of RA than in the Al-deoxidized reference steel. Figure 6, based on the EBSD data, shows the occurrence and shape of RA in the simulated CGHAZ of Tilow with t8/5 = 64 s where the highest fraction of RA was detected.

Role of Prior Austenite Grain Size
Reconstructed prior austenite grains in the CGHAZ were presented in the authors' previous paper [37] where, after holding samples at 1350 °C for 2 min and water quenching, the PAGS in Ref, Tihigh and Tilow was 75 µm, 77 µm and 125 µm, respectively. The coarse PAGS is known to decrease toughness [31], even if also promoting AF formation [5,6]. However, the coarser PAGS in Tilow compared to the other studied steels may partly explain the deteriorated impact toughness.

Effective Grain Size
The grain sizes had non-normal distributions, so the median, 95th percentile and 80% cumulative grain sizes were used in the comparisons in Table 5. Generally, the median varied from 1.04 to 2.44 µm, and the longer cooling time coarsened the largest grains. With t8/5 = 5 s, both the coarsest grains and the median were finer in Tilow compared to other steels (p < 0.05, Mann-Whitney). In intermediate t8/5 = 24 s, there were statistical differences, except between Ref and Tilow. In the slowest cooling scenario, t8/5 = 64 s, the grain sizes were finest in Tilow, and no statistical difference was observed between Ref and Tihigh. Interestingly, the median grain size of the latter was the finest overall. However, the 95th percentile and D80% cumulative grain sizes were on the large side, with both t8/5 = 24 s and 64 s.

Role of Prior Austenite Grain Size
Reconstructed prior austenite grains in the CGHAZ were presented in the authors' previous paper [37] where, after holding samples at 1350 • C for 2 min and water quenching, the PAGS in Ref, Ti high and Ti low was 75 µm, 77 µm and 125 µm, respectively. The coarse PAGS is known to decrease toughness [31], even if also promoting AF formation [5,6]. However, the coarser PAGS in Ti low compared to the other studied steels may partly explain the deteriorated impact toughness.

Effective Grain Size
The grain sizes had non-normal distributions, so the median, 95th percentile and 80% cumulative grain sizes were used in the comparisons in Table 5. Generally, the median varied from 1.04 to 2.44 µm, and the longer cooling time coarsened the largest grains. With t 8/5 = 5 s, both the coarsest grains and the median were finer in Ti low compared to other steels (p < 0.05, Mann-Whitney). In intermediate t 8/5 = 24 s, there were statistical differences, except between Ref and Ti low . In the slowest cooling scenario, t 8/5 = 64 s, the grain sizes were finest in Ti low , and no statistical difference was observed between Ref and Ti high . Interestingly, the median grain size of the latter was the finest overall. However, the 95th percentile and D80% cumulative grain sizes were on the large side, with both t 8/5 = 24 s and 64 s.
Since the finer grain size is known to improve the toughness, the deteriorated impact toughness of Ti-deoxidized steels with t 8/5 = 64 s may be due to the number of coarse grains. However, in Ref, the coarser grain size in the case of t 8/5 = 64 s compared to shorter cooling times does not appear to lower impact toughness, and especially at −80 • C, the best impact toughness was measured in Ref with t 8/5 = 64 s.  When it comes to the sizes of the inclusions, it is evident that in the Ref and Ti low samples, less than one-third of inclusions are larger than 3 µm, whereas the corresponding fraction in the Ti high samples is almost 50%. The area fractions of inclusions in Figure 7b suggest that both coarse inclusions and all inclusions have the highest area fraction in Ti high samples, followed by Ti low and Ref. This is in line with the total oxygen contents of the steels that were 23 ppm in the Al-deoxidized Ref, whereas the Ti-deoxidized steels had levels of 47 and 80 ppm in Ti low and Ti high , respectively. Between the samples of different cooling times, there were no remarkable differences, which was expected since inclusions have already formed during the manufacturing processes, and in all cases, the time occurrence at a high temperature is relatively too short to cause changes in inclusion contents.

Role of Inclusions
Inclusions coarser than 3 µm are presented separately since coarser inclusions are known to be more harmful to the toughness and ductility of steels than smaller ones. The beneficial inclusions regarding AF formation should be those consisting of MnO-TiO x together with MnS, regardless of the occurrence of TiN, i.e., the class MnO-TiO x -MnS (+TiN) in Figure 7, as was suggested in a prior study [37]. However, the inclusion size should remain modest enough to prevent their detrimental effect on toughness and ductility. In Ti-deoxidized steels, the majority of the inclusions coarser than 3 µm were MnO-TiO x (+TiN). However, in Ti high , their number was approximately twice that of Ti low . In Ti high , among the coarse inclusions, there was also TiN, which is known to be extremely harmful to toughness [24,26]. In the Ref material, the coarse inclusions consisted mainly of Al 2 O 3containing complex inclusions as well as MnS, but some TiN was also observed. The number of coarse inclusions in Ref was approximately the same as in Ti low , but the area fraction of coarse inclusions was slightly higher in Ti low than in Ref, indicating the inclusion mean size as being coarser in Ti low .
Since the fractography studies (presented later in Section 3.6) indicated the fracture initiators to be TiN in Ref and mainly TiO x -containing inclusions in Ti-deoxidized steels, it can be understood that these are the most detrimental inclusion types in these steels regarding impact toughness. In Ti-deoxidized steels, it appears to be logical since TiO xcontaining inclusions form the majority of all the coarse inclusions. The higher number of these inclusions in Ti high compared to the other studied steels could be attributed to the relatively low impact toughness of this steel. On the other hand, the weakest impact toughness was observed in the Ti low that actually has less coarse inclusions than Ti high and approximately the same number as in the Ref that had the best impact toughness of the studied steels. When it comes to the sizes of the inclusions, it is evident that in the Ref and Tilow samples, less than one-third of inclusions are larger than 3 µm, whereas the corresponding fraction in the Tihigh samples is almost 50%. The area fractions of inclusions in Figure 7b suggest that both coarse inclusions and all inclusions have the highest area fraction in Tihigh samples, followed by Tilow and Ref. This is in line with the total oxygen contents of the steels that were 23 ppm in the Al-deoxidized Ref, whereas the Ti-deoxidized steels had levels of 47 and 80 ppm in Tilow and Tihigh, respectively. Between the samples of different cooling times, there were no remarkable differences, which was expected since inclusions have already formed during the manufacturing processes, and in all cases, the time occurrence at a high temperature is relatively too short to cause changes in inclusion contents.
Inclusions coarser than 3 µm are presented separately since coarser inclusions are known to be more harmful to the toughness and ductility of steels than smaller ones. The beneficial inclusions regarding AF formation should be those consisting of MnO-TiOx together with MnS, regardless of the occurrence of TiN, i.e., the class MnO-TiOx-MnS (+TiN)

Precipitates
Nanoscale precipitates existing in the studied steels were studied by using carbon extraction replicas and TEM. In all the variants, TiN precipitates were found to also contain other elements, such as C, Nb, V, Fe and/or Mn. Additionally, cementite (Fe 3 C) was commonly found. Figure 8 presents a typical TiN precipitate. Interestingly, in Ti low with t 8/5 = 5 s, TiN-Al 2 O 3 and MnO-Al 2 O 3 -TiN precipitates were also found, which were not present in the other studied samples.
The numerical data of the precipitates in the studied steels are summarized in Table 6. The highest number of precipitates was found in Ti high , where the highest number of microscale inclusions was also found (Figure 7). The lowest number of precipitates was found in Ti low with t 8/5 = 5 s, which may explain the coarsened prior austenite grain size in the CGHAZ of Ti low . On the other hand, in Ti low with t 8/5 = 64 s, there were more precipitates than in Ref and close to that of Ti high , which does not support making conclusions about the role of precipitates on PAGS. Obviously, the relatively small investigated area partly explains the inconsistencies. The numerical data of the precipitates in the studied steels are summarized in Table  6. The highest number of precipitates was found in Tihigh, where the highest number of microscale inclusions was also found (Figure 7). The lowest number of precipitates was found in Tilow with t8/5 = 5 s, which may explain the coarsened prior austenite grain size in the CGHAZ of Tilow. On the other hand, in Tilow with t8/5 = 64 s, there were more precipitates than in Ref and close to that of Tihigh, which does not support making conclusions about the role of precipitates on PAGS. Obviously, the relatively small investigated area partly explains the inconsistencies.

Fractography
The fracture surfaces of the CVN specimens were studied by SEM in order to analyze crack initiation and propagation, as well as overall appearance. Fracture surface morphology and brittle fracture initiators of all materials with t 8/5 = 5 s at −60 • C are presented in Figure 9. The Ref steel had considerably larger areas of ductile tearing at the root of the notch, which corresponds to the amount of ductile propagation on load-displacement graphs. The main fracture mode was cleavage fracture on all steels with varying amounts of ductile regions. The cleavage facets seemed to be larger in Ti low specimens compared with the other steels, which is likely connected to the larger PAGS. At −80 • C, the fracture mode was completely brittle at both 5 and 64 s cooling times, the latter having some ductile tearing in the root of the notch in Ref specimens. The fracture surfaces of the 24 s specimens at −40 • C were not investigated, since the main focus of this study was in the ductile-brittle transition zone.

Fractography
The fracture surfaces of the CVN specimens were studied by SEM in order to analyze crack initiation and propagation, as well as overall appearance. Fracture surface morphology and brittle fracture initiators of all materials with t8/5 = 5 s at −60 °C are presented in Figure 9. The Ref steel had considerably larger areas of ductile tearing at the root of the notch, which corresponds to the amount of ductile propagation on load-displacement graphs. The main fracture mode was cleavage fracture on all steels with varying amounts of ductile regions. The cleavage facets seemed to be larger in Tilow specimens compared with the other steels, which is likely connected to the larger PAGS. At −80 °C, the fracture mode was completely brittle at both 5 and 64 s cooling times, the latter having some ductile tearing in the root of the notch in Ref specimens. The fracture surfaces of the 24 s specimens at −40 °C were not investigated, since the main focus of this study was in the ductile-brittle transition zone. The fracture surface morphology of the low impact energy (13 J) Ref specimen with t 8/5 = 5 s is presented in Figure 9a. Fracture morphology shows a relatively small cleavage facet size comparable to that of the Ti high specimens. However, the fracture pattern shows no dimples, unlike in the Ti high and Ti low specimens displayed in Figure 9c,e, respectively. The cleavage facet size of the Ti low specimen was considerably larger than that of the other steels. Nevertheless, the absorbed energy of the Ti low specimen (36 J) was similar to the Ti high specimen (33 J). The fracture surface of the Ti low specimen revealed more ductile tearing, which may explain the similar results despite having a larger cleavage facet size. However, the t 8/5 = 64 s specimens of Ti low lacked the ability to impede crack propagation, as evidenced by the lower E i and E p in Table 3.
Fractographic analysis showed that the brittle fracture was initiated by coarse inclusions or clusters of inclusions. In the case of the Ref material, fracture initiators were coarse TiN inclusions, as shown in Figure 9b. On Ti-deoxidized specimens, the brittle fracture initiations were located at TiO x and MnO-TiO x inclusions or inclusion clusters containing TiO x particles and other Ca-based inclusions. However, the failure initiators were not found in all of the investigated specimens.
Overall, the impact toughness of the Ref steel was slightly better than that of the Ti-deoxidized steels. The fracture surfaces at −60 • C showed similar features between all materials for the most part. However, the Ti low specimens with t 8/5 = 64 s demonstrated highly brittle behavior compared with the other steels. The PAGS of Ti low was larger than on the other steels and was observed at a larger cleavage fracture facet size. However, the deteriorated impact toughness could not be completely attributed to the larger PAGS since higher energies were observed with other cooling rates that had similar coarse fracture morphology. Instead, impact toughness seemed to correlate with the amount of ductile areas in the fracture surface. Interestingly, even the highest amount of AF (46%) produced by this cooling rate seemed to be inadequate in effectively hindering crack propagation. The work from Xiong et al. showed that the AF in Ti-deoxidized steels can be effective in improving the impact toughness of CGHAZ by increasing the number of high angle boundaries [22]. The fraction of the AF was not determined in the aforementioned study, but the cooling time of t 8/5 = 80 s was used. Therefore, there are likely differences between the AF fractions compared to the present study. Figure 10 shows, as an example, the macroscale fractographies of the variants with the shortest and longest cooling time (t 8/5 = 5 s and 64 s) tested at −60 • C. It is evident that most deformations can be observed in the Ref samples (a and d) that showed comparably good impact toughness. Generally, the fractographies of t 8/5 = 64 s samples (d,e,f) appeared coarser compared to those of t 8/5 = 5 s, especially in the case of Ti low with t 8/5 = 64 s (f), where fracture had occurred at nearly 100% of the area in brittle mode.

Regression Modeling
Multivariate linear regression models were prepared for the CVN and characterized microstructural features as presented in Table 7. The parameters used contained the toughness, inclusion data, phase fractions and grains sizes that were determined. The significance of all the variables was considered, and those that had p > 0.05 were excluded from the models. The variance inflation factor (VIF) was used to assess the multicollinearity of the independent variables. The presented models had no VIF > 5 instances. In model B, two outliers (±3SD) from the residuals were subtracted from the analyses in order to mitigate their influence. Models C, D and E examined the factors contributing to crack initiation, propagation and arrest, respectively.
The models generally show beneficial attributions of plate-like bainite relative to absorbed energy. Large inclusions or retained austenite were generally observed as harmful. Other phases and precipitates had occasional significance, but their effects varied. In particular, the modelling for the instrumented CVN data showed high variance and varying correlation coefficients. Metals 2021, 11, x FOR PEER REVIEW 19 of 23

Regression Modeling
Multivariate linear regression models were prepared for the CVN and characterized microstructural features as presented in Table 7. The parameters used contained the toughness, inclusion data, phase fractions and grains sizes that were determined. The significance of all the variables was considered, and those that had p > 0.05 were excluded from the models. The variance inflation factor (VIF) was used to assess the multicollinearity of the independent variables. The presented models had no VIF > 5 instances. In model B, two outliers (±3SD) from the residuals were subtracted from the analyses in order to mitigate their influence. Models C, D and E examined the factors contributing to crack initiation, propagation and arrest, respectively.

Conclusions
Three experimental laboratory steels were studied in order to measure the effect of acicular ferrite on impact toughness. The studied steels were two Ti-deoxidized steels with a varying fraction of acicular ferrite and Al-deoxidized reference steel, with otherwise comparable chemistry. In contrast to other studies, acicular ferrite was not found to improve the impact toughness of CGHAZ in this study, and the best impact toughness was achieved in conventional Al-deoxidized reference steel that did not have any or a remarkable fraction of acicular ferrite. The possible reasons for the weak impact toughness in the acicular ferrite-containing steels were a coarser prior austenite grain size and effective grain size, a marginally higher hardness/tensile strength and unbeneficial coarse inclusions. It is also possible that a higher fraction of acicular ferrite in the microstructure would be needed in order to observe the beneficial effect on impact toughness than was achieved in this study.