Motivation and Objective

Aluminum alloy Al6061 has wide range of application due to its properties. This alloy is a precipitate hardening alloy in which strengthening happens during aging process. In this process new phase start forming from supersaturated solid solution. Many efforts have been done for last few decades to find the effect of aging process on precipitating and to connect final properties of materials to volume fraction, size, and interface nature of precipitates. However, properties of Al6061 is a combination of different structural features which exist at different length scales. Information about each of these microstructural features can be measured through different experimental techniques. Therefore, finding structure, property, processing linkages requires data analysis on multi-scale experimental datasets.

Work flow

Considering above-mentioned objective, we performed different microscopic techniques to collect data for each structural feature. The structural features are :

  • Coherent GP zones and semi-coherent precipitate particles
  • Al-Fe-Si-type constituent particles
  • Grain structure

A list of parameters that need to be quantified was provided from our collaborator in NIST and ASM international. This parameters is basically a quantifying values for each structural features. These parameters are listed as below:

  • Volume fraction, Size (nm), Aspect ratio, Spacing (nm) of precipitate particles
  • Volume fraction, Size (nm), Aspect ratio, Spacing (nm) of constituent particles
  • Size (microns), Aspect ratio of grains and the strengths of different texture orientations.

Microscopic techniques used in this project are:

  • Optical Microscopy (OM)
  • Scanning Electron Microscopy (SEM)
  • Electron Back Scattered Diffraction (EBSD)

Project plan falls into following directions:

  • Performing aging treatment on materials at different temperature conditions.
  • Collecting datasets from each condition using OM, SEM, and EBSD techniques.
  • Extracting parameter listed above for each structural feature.
  • Performing mechanical testing on each condition, and calculating property information.
  • Running PCA on entire datasets to find structure, property linkage.

Challenges and Trials

OM datasets

  • Data collection: the main challenge in this part was etching of samples. We tried etching techniques to reveal precipitates in OM but they didn’t work well. So we decided to use OM only for constituent particles. Also we tried to collect images at different magnification and run segmentation to see what magnification is the best in terms of the area size, and pixel resolution. The images collected at 50X provide a good resolution, however, few constituent particles exist in each image. Magnification of 10X covers large area on the sample but the resolution is not enough to resolve particles which are close to each other. Finally, after collecting couple of images at different magnifications, we decided to use 20X magnification, and I collected 10 images at random regions of the sample.
  • Image correction: I found a systematic problem in the optical microscope which is related to the illumination part in the microscope. This problem causes a gradient of intensity cross the image from right to left. We are trying to make a executable file on microscope computer as a post-processing tool to remove the shadow from each image that taken on microscope and save it as a new file. The code fits a surface to the average intensity of each image using regression and change background intensity to an even level for entire sample.
  • Image segmentation: I tried few segmentation codes that I’ve found online for image segmentation. Most of them are based on simple threshold cutoff and they didn’t provide a promising segmentation parameters for all images. We did try and error to find the best threshold but it didn’t work on all images. In one of the class report session, one group introduced an image segmentation “algorithm”., and that worked fine for OM datasets. Here is the link to the image segmentation post: “Multimodal Data Segmentation using Peak Fitting Algorithm and Gaussian Likelihood Maximization”.
  • Extract particle information: Particle volume fraction, size, and spacing were calculated on each segmented image. Since the segmented image is a binary image, the volume fraction and particle size are the number of pixels for the particle phase in entire image and in each particle respectively. Spacing of particles were calculate based on Chord length in both horizontal, and vertical directions in the image.

SEM datasets

  • Data collection: *SEM is used to image precipitates in the microstructure. I collected all SEM images with 100 micron view field with 48nm resolution. However, this resolution was not enough to image small precipitates in as-received, 400, 525 samples. Higher resolution images required from TEM.
  • Image segmentation: I had one main problem in this step. Most of the images for as-received sample and the ones aged at 400F, and 525F don’t show precipitates in the matrix. So I don’t know whether if precipitates exist and we can’t see it or they don’t exist. I tried different combinations of various settings in SEM and etching methods in order to reveal precipitates but they didn’t work well. In addition, for images with precipitates in matrix, I tried the same segmentation techniques for OM on SEM datasets. The segmenting code works well as long as the precipitates are small and distinct from constituent particles. For larger precipitates such as ones form at 775F, the segmentation code doesn’t work perfectly. Also I tried a new segmentation “algorithm”, which find the center of small precipitates well. The algorithm find the center by multiplying the inverted image to its gradient.
  • Extract precipitate information: I didn’t include precipitates statistics in final result for two reasons. First, not all the images show precipitates in the matrix, second so many precipitates have size of one or two pixels which makes segmentation with high uncertainty. I’m collecting new datasets with a higher resolution to have enough information about size and morphology of precipitates.

EBSD datasets

  • Data collection:
  • The main challenge in EBSD is to collect data from a large area. Each EBSD data contains 75 smaller scans with a resolution of 4 micron. EBSD details for different condition are presented in a “post”.
  • Extract grain information: Grain size, and aspect ratio were calculated. You can find more details of how they are calculated in the post. For now, I didn’t include grain orientation and boundary information in statistics. but eventually all these data needs to be added to the list of parameters.

Mechanical Property Measurement

In order to find structure-properties linkage, mechanical uniaxial test were done on all heat treatment conditions (5 samples for each condition). The full summary of tests and results are presented in a “post”. You can find a “Comparison” of stress-strain plots for different conditions.

Data Analysis

At the beginning, all measured parameters are used in PCA plots. It has shown in a “post” that all measured parameters (Size, volume fraction, chord length) are embedded in 2-point statistics so applying PCA on both combined data are redundant. In “final presentation” it is described that since the optical, SEM images are from random smaller areas compared to EBSD, the 2-point statistics averaged out and then combined with EBSD statistics. In addition, Tony help me on having “interactive PCA plots with associated images, and also a post about “A Comparison of Supervised and Unsupervised Learning, which I’m still trying to understand the concepts. In the post you can see PCA on segmented images as well as 9 SVE’s with dimension of 200X200 extracted from each image.

Collaborations

This woks done through a collaboration with the following people: Ali Khosravani, Jordan Weaver, Tony Fast, Yuksel Yabansu, Ahmet Cecen, Andrew Castillio, Rashin Khodaei

Experimental part

  • Heat treatment (Ali Khosravani)
  • Sample preparation (Ali Khosravani & Rahsin Khodaei)
  • EBSD data collection, and analysis (Ali Khosravani)
  • OM (Ali Khosravani)
  • SEM (Ali Khosravani)
  • Mechanical Tests, and analysis (Jordan Weaver, Andrew Catillio)
  • Nanoindentation and analysis (Ali Khosravani)

Image segmentation, and data analytic

  • Image Correction, removing shadow from OM images (Ahmet Cecen)
  • Image segmentation (Tony Fast, Ali Khosravani, Jordan Weaver)
  • Chord length measurement (Yuksel yabansu, Ali Khosravani, Ahmet Cecen, )
  • Programming codes for 2-point tatistics, PCA, calculation of particle parameters (Tony Fast, Yuksel Yabansu, Ali Khosravani)

Closure and Future Work

Based on the experience in the class, following future works requires to establish structure, property, processing linkages

  • High throughput experiments is required to have more sample conditions with less amount of time and effort. Based on the experiences in this project, I will collect OM, EBSD and SEM datsets at the same region, so no need for averaging of statistics for data from each technique.
  • Nanoindentation I’m collecting Nanoindentation data on 1500 micron indenter tip. The analyzing of 100 micron indenter is almost done. You can see most of the NI results for 100 micron indeter in this “post”. The post will be updated soon with more details.
  • ** Statistics on grain orientation, and grain boundary information** In the class project, only grain size, and aspect ration were considered in data analysis. Further work needs to include grain orientation, and grain boundary information as well (2-point statistics on EBSD data)
  • Image segmentation. In order to include precipitates statistics in analysis, we need a robust image segmentation algorithm for SEM images.