Quasi-Non-destructive Characterization of Carburized Case Depth by an Application of Centerless X-ray Diffractometers

Carburizing is a case-hardening process in which carbon is diffused into a thin steel surface layer at elevated temperatures, while the steel is typically austenitic. Carburizing is usually followed by a quenching and tempering process to produce tempered martensite at the surface (case depth), under which there is a more ductile bainitic core. The carbon gradient from carburizing creates a gradient in the steel surface hardness, strength, and wear resistance of the component. Hosford (Ref 1) and Réti (Ref 2) have shown that the main advantage of carburizing is that it also produces compressive residual stress on the surface of the component. Service life, especially wear and fatigue resistance, can be significantly improved by the well-controlled residual surface stress state of various components. Wang et al. (Ref 3) showed that the amount of carburizing-induced residual stress on a wind turbine gear leads to a bilinear curve in the gear's contact fatigue life. As the magnitude of the tensile residual stress increases, the rolling contact fatigue life decreases linearly.

In contrast, once the magnitude of the compressive residual stress exceeds a specific value, the contact fatigue life does not increase anymore. Wu et al. (Ref 4) highlighted the advantage of shot peening coverage on the carburized rollers. Not only the residual stress was improved, but the retained austenite was also transformed on the carburized layer after shot peening. Sugimoto et al. (Ref 5) applied severe plastic deformation to increase the compressive residual stress on a vacuum carburized steel.

Residual stress control often involves technology design supported by finite element simulation, as Hiremath et al. (Ref 6) and Li et al. (Ref 7) did on various carburized steel specimens and gears, respectively. O'Brien and Yeddu (Ref 8) used macroscale simulation and gained opposite results to Wu et al. (Ref 4) about the retained austenite. They showed that a carburized case with a high retained austenite volume fraction gives better fatigue than those with low retained austenite. Improper residual stress conditions can cause a variety of problems. Jo et al. (Ref 9) have shown that carburized steel exhibited cycling softening under both axial and torsional loading, while Freborg et al. (Ref 10) and Tewary et al. (Ref 11) focused on the distortion mechanisms and structural instability. Shoji et al. (Ref 12) pointed out that the heterogeneity in material chemistry and microstructure influences stress corrosion cracking. It can be concluded the efficiency of the carburizing process in terms of the lifetime of the component is mainly determined by the residual stress state. Despite this, the evaluation of carburizing process on a steel surface is traditionally accomplished by metallography and hardness measurements since the technology was introduced to the industrial practice in the last decades of the nineteenth century. The advantage of these methods is that carried out easily and cost-effectively in an industrial environment, and they are standardized and accepted. On the other hand, these are destructive methods (Ref 13).

However, a variety of non-destructive testing approaches have been investigated. One such technique was applied by Gogolinskii et al. (Ref 14), where portable hardness testing was used for surface characterization. Still, they concluded that this method does not meet the standards for the conventional hardness scale. Rodrigues et al. (Ref 15) successfully applied an ultrasonic test associated with signal processing and machine learning technique as a novel procedure for estimating the level of carburizing. Mercier et al. (Ref 16) proposed eddy current testing to characterize the extent of decarburization. They concluded that the eddy current technique is susceptible to surface modification, but it needs reference samples for industrial control. Amiri et al. (Ref 17) used an impedance plane for presenting results in eddy current testing for the determination of surface carbon content after carburizing.

One type of non-destructive, increasingly common device in industrial practice and on-site measurements is portable, centerless x-ray diffractometers (XRD) developed for residual stress testing. According to Mueller (Ref 18) the trend of the increasing demand is present in the automotive industry. Dive and Lakade (Ref 19) explained this need for residual stress measurement and presented the industrial usage of the portable residual stress measuring XRD systems among the other non-destructive methods. In Whitley's thesis (Ref 20) one can read about the portable residual stress measuring XRD systems utilization on military and marine fields from the 1980s and how these instruments helped to understand the structural failures of the different welded materials. The test principle is that the residual stress can be calculated from a change in the distance of a series of planes (hkl). Accordingly, the interference function of an (hkl) series is detected, and the Bragg angle is determined. By using these devices, of course, the residual stress state after cementation can also be characterized, a semi-destructive way like Filip et al. (Ref 21) did. Our proposal is to determine not only the value of the residual stress but also all the information that can be obtained from a diffraction signal. In the present case, this is the full width at half maximum (FWHM) parameter and the correlated carbon gradient and the resulting change in microstructure.

Centerless diffractometers have not yet been used to characterize the surface carbon content. However, the measured parameter signal shows a much stronger correlation with the carbon content than previously mentioned methods. In the literature, one can find several examples for using conventional diffractometers to determine the carbon content or its change in the different phases or microstructure features based on the interference peaks radial distribution. Ortiz et al. (Ref 22) studied the crystalline quality and carbon content in carbon steels with the help of infrared photocarrier radiometry. They analyzed the FWHM values of XRD measurements data. In another study, Lara et al. (Ref 23) presented a similar study complemented by thermal characterization. Garcia et al. (Ref 24) investigated the carbon content of retained austenite in bainitic steel by atom probe tomography and XRD, and they found a good match in the results. Luo (Ref 25) showed a Gaussian peak fitting method for XRD diffraction peaks and how to separately measure the volume fraction and tetragonal ratio of co-existing lath and plate martensite in ultra-high-strength steels. This method was applied to investigate the decarburization of martensite during a tempering treatment.

These portable diffractometers, due to their centerless arrangement, allow operation without destroying or altering the specimen. During our practical experience, Sepsi et al. (Ref 26) and Sepsi et al. (Ref 27) showed that with XRD on shafts and gears, the negative (disturbing) effect of decarburization (carbon loss on the surface) could be detected during residual stress measurement. Szobota et al. (Ref 28) showed that decarburization could cause a change not only in the stress state but also in the FWHM of the interference function of the ferrite (211) during residual stress measurement. Several authors report FWHM data with residual stress measurements. Suominen et al. (Ref 29) presented a study on high-strength steel weld toes and fatigue life. They found that relatively higher FWHM values indicate a hardened surface. Using shot peening, Fu et al. (Ref 30) showed a connection between hardness, residual stress, and microstructure. They also identified a linear relationship between the FWHM and hardness values. Similar results can be obtained after ultrasonic shot peening, according to Li et al. (Ref 31). Kelleher et al. (Ref 32) studied the effect of the roller straightening process in railway rails and worn ex-service rails. During their XRD measurements, they found a correlation between FWHM values and residual stresses. Dias et al. (Ref 33) investigated hot-rolled 42CrMo4 steels and their residual stress state after various heat treatments. They compared their XRD and Barkhausen noise measurement results and examined how the microstructure features influenced them.

The shape and the parameters of the ferrite interference function (FWHM) are influenced by several physical factors such as the amount of carbon in solid solution, lattice distortion, and crystallite size. These parameters can be precisely determined by analysis of the complete diffraction profile with a conventional XRD, such as a Rietveld analysis, as shown by Young (Ref 34). Centerless x-ray diffractometers, as they have been developed explicitly for residual stress testing, are suitable for recording only one or two interference functions in the large Bragg angle range. This means that a full-profile analysis cannot be performed, but the measured interference functions (one or two) data can be determined with high accuracy. This observation gave us the idea to use the diffractometer not only to show the residual stress but to find the depth where, in addition to martensite, other microstructure elements such as bainite, perlite, or ferrite also can form because of the lower carbon concentration. The carbon content in ferrite and other allotropic iron phases (especially its crystallite size) are very different in these microstructural elements. In the case of a mixed microstructure, the measured interference function of ferrite appears as a superposition of signals of a multi-component structure.

This work proposes a novel, non-destructive procedure for identifying different carburization levels and the case depth. Signal deconvolution of the interferential function during residual stress measurement was compared with the results of conventional destructive methods such as metallographic studies and hardness testing.

The hot-rolled C45 grade steel bar specimen used for the experiments is shown in Fig. 1. This low-alloyed and tough structural steel with medium strength is frequently used in the automotive industry. The carburized layer was produced in an electric resistance furnace in an industrial environment, together with other industrial components. Surface-hardened layers were produced by two carburizing methods: the pack method for 20 hours or gas carburizing for 14 hours. In the case of gas carburizing, 1.2 carbon potential was used. Both carburizing processes were carried out at 920°C, and subsequent air cooling was applied.

After carburizing, the samples were cut (Fig. 2), and half of them received an additional hardening heat treatment with austenitization at 900°C for 30 minutes and subsequent oil quenching. A chemical surface cleaning process was used to remove the thin oxide layer. Figure 2 also shows the investigated surfaces during metallographic examination and XRD test values as a function of depth (electrochemical polishing spot).

Optical microscopic composite images were prepared using a Zeiss Axio Imager optical microscope. Image analysis was used to determine the martensite volume fraction based on Cprob software by Póliska et al. (Ref 35). It was carried out with a 45-pixel wide (8.25 µm) sliding window, and a three-period moving average was applied as smoothing. Hardness measurements were taken using an Instron Tukon 2100B (Wilson Instruments) microhardness tester with a load of 100 g. The dwell time was 10 seconds. The first measuring point was at 20 µm from the edge of the sample and then at 100 µm increments toward the piece's core up to 3 mm. Macrohardness was also performed with the load of 1000 g with 300-µm increments, and the first point was at 200 µm from the edge of the sample. The dwell time was 10 seconds. For each specimen, the measurements were taken in 4 parallel lines from the edge to the core. During the evaluations, the average values with the error bars are presented.

All four specimens were subjected to XRD measurements, where residual stress and interferential function parameters were detected through the depth. The depth measurements were taken with electrochemical polishing (Struers MoviPol-5 machine). The first measurement point for all samples was made on the surface and then with 100-µm increments toward the core of the sample, up to a depth of 3 mm. The residual stress measurements were taken with a Stresstech Xstress Robot centerless x-ray diffractometer (Fig. 3), and the data were evaluated using the sin²PSI method, as discussed by Krawitz (Ref 36). Table 1 summarizes the parameters used during the XRD residual stress measurement and calculation. The FWHM data are recorded as an average of the FWHM (without deconvolution) data of the 16 signals determined at eight different tilt angles at a given depth. Additional analysis with parabolic background extraction and Gaussian peak deconvolution of the ferrite (211) interferential function was used to prove the presence of fractions of different crystallite sizes, i.e., different microstructures. The volume fraction of martensite was determined from the deconvolution of the diffraction signal. After peak deconvolution, the martensite fraction is given by the ratio between the area of the peaks with higher and lower FWHM values.

The significant variation in hardness data is consistent with the variation in carbon concentration during carburization and the resulting formation of heterogeneous microstructural features. The decreasing nature of the macrohardness data is consistent with the variation in the amount of martensite determined by XRD or image analysis. There is no direct correlation between hardness and martensite volume fraction because not only one but several types of microstructural features are formed besides the martensite, and their quantity varies. From the point of view of the application, the residual stress provides complex information. In the pack carburized specimen, the residual stress changes sign, from +300 MPa tensile stress to −300 MPa compressive stress in the 1000-µm layer, while the macrohardness remains practically unchanged (800 HV1). The decrease in hardness coincides with the transition in residual stress from compressive to the tendency of tensile stress for both samples. It is well correlated with the results after shot peening (Ref 21, 36). This point occurs at a depth of about 900-1100 µm for the pack carburized specimen and 300 µm near the surface for the gas carburized specimen. The transition of the residual stress data to compressive stress and then to tensile stress results in an increase and a decrease in the FWHM data, respectively. The excellent correlation of the FWHM data with the microstructure is the most important result. The thickness of the fully martensitic microstructure (case depth) can be determined most accurately from the FWHM data for the 20-hour pack carburized sample. A clear maximum is indicated on the curve. The maximum is also smaller (but definite) for the gas carburized sample, which is explained by the presence of a mixed martensitic-bainitic microstructure. It can be concluded that the FWHM data indicate microstructural changes as reliably as the hardness data. In addition to martensite, the appearance of other structures such as bainite, perlite, or ferrite can be detected with the greatest certainty by peak deconvolution analysis. The (211) interference function of the ferrite phase cannot be fitted by only one peak in some cases. The need for additional peaks during the fitting is possible for two reasons. One is that the reflections of Kα₁, Kα₂ are separated; this is a standard characteristic of diffraction. Another reason is that the ferrite phase is present in different microstructures with different compositions or grain sizes. For the first, ferrite and martensite, and for the second, the combined presence of ferrite and perlite are a good example. This means that in Fig. 16, curve B is added to describe separate ferrite, which should have a bigger grain size than the ferrite in pearlite, represented by curve A. Here, curve B appears to have a higher FWHM value that does not correlate with the expected larger grain size. But the fact the reflection of the ferrite with smaller grain size (in perlite) is split into Kα₁ and Kα₂ peaks, A and C, respectively, must be considered. Thus, the FWHM value of curve B should be compared to the FWHM value of the composite curves (A + C). The advantage of the new method is that it is susceptible to these processes. Hence, it is very suitable for detecting other microstructural elements in addition to martensite, which is a significant problem for the carburization process. The detection of other microstructural elements in addition to martensite is of practical importance for carburization. In this case, the martensite and other microstructural features appear at peak deconvolution and can be denoted by peaks with a higher FWHM value and a lower FWHM value, respectively. The curve with a higher FWHM value represents the martensite with distorted lattice. From the signal-deconvolution data, the ratio of the different peaks shows the volume fraction change of different microstructures. This was also confirmed by the metallography images and quantitative analysis. That is, microstructure changes can be well inferred from the signal-deconvolution data.

It can be observed that the residual stress curve in the cemented layer is minimal, i.e., the maximum compressive stress is below the surface, and tensile stress is developed near the surface and in the deeper layers. In the vicinity of the compressive stress maximum (900-1500), the scatter of the results and the uncertainty of the measurement are significant, which is associated with an increase in the FWHM value.

This effect also reflects the presence of different scattering centers with low-volume fractions adjacent to the martensite. In the present case, the diffraction signal consists of a mixed signal of the ferrite phase from martensite to bainite to perlite, with a poor signal-to-noise ratio. The FWHM values are further enhanced because the volume fraction of new scattering centers adjacent to martensite is typically small. The image analysis suggests that they are less than 10% in total.

The disadvantage of centerless diffractometers is in this case the angular range detected is small compared to detected FWHM values and same with the signal/noise ratio. This problem can be improved by developing detectors.

Overall, it can be concluded that the variation of the residual stress signal is sensitive to the changes in the carburized layer due to the carbon gradient. The appearance of microstructural features other than martensite, as long as their total volume fraction is small, causes the increase of FWHM value despite the decrease in carbon content. When the amount of microstructural features exceeds about 10%, the FWHM starts to decrease and fitting the interference function with two or more curves is possible only.

The most important conclusion from this work was that evaluation of the interference function during residual stress measurement could be used as a non-destructively characterization of the microstructure of the carburized layer. Conventional methods such as hardness testing and optical microscopy cannot detect these changes with such sensitivity.

This promising preliminary work on two samples should be repeated in several other case studies. If validated, this non-destructive method could be used in future cases to provide information on (1) case depth, (2) case hardness, and (3) case microstructure, in addition to the residual stress state of the carburized components. This procedure would be particularly advantageous for quality control. It is non-destructive and could even be carried out with in situ testing integrated into the production line. The non-destructive nature is realized because the depth residual stress measurement can be performed at a part of the component that is not involved in the load or where the tested volume is machined anyway during subsequent processing.