Comparison of Microscopy and Image Processing Techniques to Quantify Recrystallization Fraction in HSLA Steel Grades

High-strength, low-alloy (HSLA) steel is widely used in automotive reinforcements and structural components due to its low cost, formability, and weldability. Grain boundary strengthening, a key mechanism in HSLA steels, is achieved by refining grain size, increasing grain boundaries, and restricting dislocation motion [1,2]. Cold rolling of the sheet steel introduces microstructural texture (i.e., grains are elongated in the rolling direction (RD)) and elevates dislocation density, which is relieved through an annealing treatment—a process comprising recovery, recrystallization, and grain growth. During the annealing process, elongated grains become ‘equiaxed’ or more circular, changing the microstructure and thereby changing properties as shown in Figure 1. Strength of the material progressively reduces with increasing annealing temperatures, whereas ductility is improved. Due to the influence of microstructure on properties, obtaining an efficient best practice method for measuring recrystallization in HSLA steel grades is valuable to develop processing-structure-property relationships in production of this grade. The extent of recrystallization can be quantified by assessment of the microstructure by many methods: manual assessment of optical micrographs; through image analysis software, such as MIPAR, which identifies recrystallized grains based on morphology; or with electron backscatter diffraction (EBSD) point-to-point misorientation angles with grain average misorientation (GAM) [3]. All these methods are comparatively investigated on a Cliffs-produced HSLA Grade 380 (380 MPa minimum yield strength requirement) 0.0765” gauge coil. Longitudinal cross sections from the sample in the central position were prepared for optical microscopy and SEM-EBSD using conventional metallographic techniques. The same optical micrograph was used for both manual and image processing techniques. SEM-EBSD scans were performed with a JEOL JSM-7200F scanning electron microscope equipped with a field emission source operating at 20kV on 100x100µm areas with a step size of 0.3 µm, and post-processed with EDAX OIM (v.8) software. A minimum confidence index of 0.1 was considered, and misorientation threshold was given to be 3.2°. For analyzing the optical micrographs, MIPAR V3.4.1 was used by creating a recipe that segments and quantifies features of individual grains in the micrograph. The first method involved quantifying fraction of non-recrystallized grains manually using three concentric circles overlaid on an optical image as shown in Figure 2(a). The number of elongated grains intersecting the circle is divided by the total number of grain intersections. This value is averaged between the three circles and reported as the estimate of fraction of non-recrystallized grains. This method can be time consuming, as each grain is manually identified and counted. Furthermore, the measurement is highly dependent on the viewers interpretation of whether a grain is recrystallized or not. Parameters such as the degree of etching can be variable between samples and this method may not be repeatable. The second method uses MIPAR image processing software to identify and segment all grains in the microstructure. It calculates various features of the grains like roundness, aspect ratio, area fraction, and X-Y bounding box values. Roundness measurement was determined to be the best estimate of recrystallization fraction. A perfectly round or ‘equiaxed’ grain is said to have a roundness value closer to 1. Recrystallization fraction is then estimated by setting a threshold roundness value; and determining number fraction of all grains above that threshold as shown in Figure 2(b). Since MIPAR relies so strongly on quality of the micrograph, etching trials were carried out with differing concentrations of nital and it is determined that either a deep etch with 3% nital, or regular etch with 10% nital; combined with focus enhancement techniques during imaging resulting in improved identification with the software. Optical micrographs were chosen over scanning electron micrographs as the area of surface imaged is larger, and it is much more efficiently obtained. The third analysis technique is carried out with OIM EBSD software. All raw data is cleaned using a neighbor confidence index of 0.1, in which questionable data below the confidence index is reassigned based on the neighboring kernels. Analysis is performed using grain average misorientation (GAM) area fractions with misorientation angle threshold of 3.2° were determined as non-recrystallized grains as shown in Fig 2(c). Results for each of the three methods are summarized in Table 1. The concentric circle and EBSD-GAM methods yield similar values while the MIPAR method is low. It likely results from MIPAR dividing non-recrystallized grains into smaller grains as shown in Figure 3 that, in some cases, fall below the threshold. This can be fixed by using MIPAR’s deep-learning module in conjunction with hand-drawn recipes to properly segment elongated grains as seen in Figure 4[4]. The concentric circle method is time consuming and not repeatable. EBSD-GAM is a very reliable method, as it resolves operator and resolution errors. However, it is expensive and time consuming between sample preparation and scan time. The MIPAR method is a relatively fast and inexpensive option for measuring recrystallization. The current recipe needs to be reinforced with machine learning methodologies to accurately predict the fraction of recrystallized grains. Metallographic sample preparation and etching followed by optical microscopy is a relatively easy and efficient method to obtain information about the microstructure. Improved segmentation techniques will improve understanding of process-structure-property relationships for mitigating quality issues in commercial production and accelerating alloy development efforts. (a) Effect of microstructure on increasing annealing temperature for HSLA grades; and (b) strength-ductility trade-off with increasing annealing temperature. (a) Concentric circles overlaid on optical image with intersecting grains highlighted. Non-recrystallized grains highlighted in blue; (b) and MIPAR-segmented image shaded according to relative roundness values (c) EBSD GAM map (left); GAM map with grains above misorientation threshold highlighted in white (middle); GAM distribution with misorientation angles above 3.2° highlighted (right). Summary of Fraction of Non-Recrystallized Grains Summary of Fraction of Non-Recrystallized Grains MIPAR identification of grain features with elongated grain recipe. (a) represents the optical micrographs and (b) represents segmentation with MIPAR. Some examples of misidentified grains are highlighted. (a) Optical nital-etched micrograph of an HSLA Steel grade with hand-drawn recipe; and (b) roundness parameter distribution with segmented grains.