FACULTY OF MANUFACTURING ENGINEERING

BMFB 3723 MATERIAL CHARACTERIZATION

SEM II 2019/2020
3 BMFG 1

REPORT TITLE: A DEEP-LEARNING TECHNIQUE FOR PHASE

IDENTIFICATION IN MULTIPHASE INORGANIC COMPUNDS
USING SYNTHETIC XRD POWDER PATTERNS

LECTURER’S NAME: PM IR. DR. ASYADI AZAM BIN MOHD ABID

NO NAME MATRICS NO
1. MUHAMMAD KHAIRANY AFFAN BIN SAIMIN B051710134

2. MUHAMMAD SYAZWAN BIN SULAIMAN B051710127

3. NORMA SYAKILA BINTI ASOHA B051710156

4. NUR ATIRAH ASNA BINTI MOHAMAD RAUS B051710149

5. SHAHRIN HAKIMI BIN AHMAD B051710205

6. NOOR FATIN NABILAH BINTI ROSLEY B051710143

SUBMISSION DATE: 3rd APRIL 2020

Introduction
X-ray Powder Diffraction (XRD) is a rapid analytical technique primarily used for phase
identification of a crystalline material and can provide information on unit cell dimensions. We want to
stimulate the plausible powder X-ray diffraction (XRD) patterns for 170 inorganic compounds in the
Sr-Li-Al-O quaternary compositional pool, wherein promising LED phosphors have been recently
discovered. The XRD pattern is considered as nothing more than a one-dimensional image in the deep-
learning approach and a deep convolutional neural network (CNN) is employed and trained to learn
underlying features from a large number of XRD patterns. Convolutional neural network (CNN)
models are built and eventually trained using this large prepared dataset. The fully trained CNN model
promptly and accurately identifies the constituent phases in complex multiphase inorganic compounds.

Result

The phase identification result: The training loss/accuracy (dim pink lines) and the validation loss/accuracy (deep pink lines) for
both the CNN_2 (a, b) and CNN_3 (c, d) trained with Dataset_800k_org, CNN_3 (e, f) with Dataset_800k_rand, and CNN_3 (g,
h) with Dataset_180k_rand are plotted as a function of the iteration number. The table 1 shows the test accuracy for the simulated
and the real XRD test datasets for CNN_2 and CNN_3 architectures. The test accuracy was considered nearly perfect when the
simulated test dataset was employed both for the CNN_2 and CNN_3 architectures.

The three-level-phase-fraction prediction result trained with Dataset_800k_org. The training loss/accuracy (dim pink lines) and
the validation loss/accuracy (deep pink lines) plotted as a function of the iteration number up to the 10th epoch for the CNN_2F
(a, b), CNN_3F (c, d), CNN_4F (e, f), CNN_5F (g, h), and CNN_6F (i, j) architectures and the table 2 present the test accuracy
was slightly decreased in the constituent-phase fraction prediction.

Discussion
CNN models developed enables prompted phase identification as well as a rough measure of the relative
fractions of the constituent phases in a multiphase inorganic mixture sample consisting of Sr, Li, Al, and O.
XRD patterns stimulated for 170 inorganic compounds synthesized from quaternary compositional pool and
finally prepared three XRD pattern datasets (Dataset_800k_org, Dataset_800k_rand, and
Dataset_180k_rand) by combinatorically mixing them, and finally developed robust CNN models using
these large datasets. Although these results appear to be a small success accomplished in a limited
compositional space, i.e., the Sr-Li-Al-O quaternary compositional pool, the potential of the proposed deep-
learning approach for use in XRD analysis verified.
 UNIVERSITI TEKNIKAL MALAYSIA MELAKA
FACULTY OF MANUFACTURING ENGINEERING
Hang Tuah Jaya, 76100 Durian Tunggal, Melaka.

Tel: +606-3316419 Fax: +606 331 6411

LABORATORY DEMONSTRATION
ASSESSMENT FORM

Subject Code BMFB 3723

Subject Name MATERIALS CHARACTERIZATION
Session/Semester 2 – 2019/2020
Name of Student (1)
Learning Outcomes Program Outcomes
Able to communicate effectively on complex
engineering activities with the engineering community
Display good communication skill on matters and with society at large, such as being able to
related to materials characterization in a written comprehend and write effective reports and design
report and presentation documentation, make effective presentation, and give
LO3 PO9
and receive clear instructions.

STUDENTS’ NAME LIST: MATRIX NO.

1. MUHAMMAD KHAIRANY AFFAN BIN SAIMIN B051710134

2. MUHAMMAD SYAZWAN BIN SULAIMAN B051710127

3. NORMA SYAKILA BINTI ASOHA B051710156

4. NUR ATIRAH ASNA BINTI MOHAMAD RAUS B051710149

5. SHAHRIN HAKIMI BIN AHMAD B051710205

6. NOOR FATIN NABILAH BINTI ROSLEY B051710143

Score Weightage Marks

INTRODUCTION
Clear and brief explanation on the introduction. 4
Acceptable explanation about the introduction. 3 3
Serious deficiencies in the explanation of introduction. 1
No introduction in the report. 0
RESULTS
Clear explanation of the results. 4
Acceptable explanation of the results. 3 4
Serious deficiencies in the results. 1
No result. 0
DISCUSSION
Clear explanation of the discussion. 4
Acceptable explanation of the discussion. 3 3
Serious deficiencies in the discussion. 1
No discussion. 0
CONCLUSION
Clear directions and wrap up connections to objectives. 5
Acceptable conclusion. 3 1
Serious deficiencies in the conclusion 1
No conclusion 0
REPORT FORMAT AND WRITING SKILLS

Report is well structured and follows standard format. Report is very

clear and sentences are easy to understand. Minor grammatical
5
errors.

Report is well written but occasionally some points are not easy to
understand. Some grammatical errors are present. Standard format is
3 1
used.

Report is not well written with many grammatical errors and/or

1
standard format is not followed.

TOTAL

50

-1-