VIETNAM NATIONAL UNIVERSITY HO CHI MINH CITY HO CHI MINH UNIVERSITY OF TECHNOLOGY -------------------- TRAN QUOC KIM MACHINE LEARNING IN PREDICTING MECHANICAL BEHAVIOR OF 3D PRINTED BEAMS WITH TRIPLY PERIODIC MINIMAL SURFACE (TPMS) SANDWICH CORES Major: Civil Engineering Major ID: 8580201 MASTER THESIS HO CHI MINH CITY, JANUARY 2023 VIETNAM NATIONAL UNIVERSITY HO CHI MINH CITY HO CHI MINH UNIVERSITY OF TECHNOLOGY -------------------- TRAN QUOC KIM MACHINE LEARNING IN PREDICTING MECHANICAL BEHAVIOR OF 3D PRINTED BEAMS WITH TRIPLY PERIODIC MINIMAL SURFACE (TPMS) SANDWICH CORES MÔ HÌNH MÁY HỌC TRONG DỰ ĐOÁN ỨNG XỬ CƠ HỌC CỦA DẦM IN 3D GIA CƯỜNG LÕI SANDWICH BỀ MẶT CỰC TIỂU TAM TUẦN HOÀN Major: Civil Engineering Major ID: 8580201 MASTER THESIS HO CHI MINH CITY, January 2023 THIS THESIS IS ACCOMPLISHED AT HO CHI MINH UNIVERSITY OF TECHNOLOGY – VNU HCMC Supervisors: Dr. Nguyen Thi Bich Lieu Signature: Assoc. Luong Van Hai Signature: Examiner 1: Dr. Thai Son Signature: Examiner 2: Dr.
Nguyen Phu Cuong Signature: The master thesis is defended at Ho Chi Minh University of Technology – VNU HCMC on 13th January 2023. The thesis defense grading committee consists of: 1. Do Nguyen Van Vuong 2. Nguyen Thai Binh 3.
Nguyen Phu Cuong 5. Council Member: Assoc. Luong Van Hai Confirmations of the Chairman of thesis defense grading committee and the Dean of faculty of thesis major after the thesis has been corrected (if any). CHARIMAN OF DEAN OF FACULTY THESIS COMMITTEE FACULTY OF CIVIL ENGINEERING Assoc.
Do Nguyen Van Vuong i VIETNAM NATIONAL UNIVERSITY HCMC SOCIALIST REPUBLIC OF VIETNAM HO CHI MINH UNIVERSITY Independence - Freedom - Happiness OF TECHNOLOGY MASTER THESIS ASSIGNMENTS Full name: Tran Quoc Kim Student ID: 2170979 Date of birth: 30/07/1999 Place of birth: Can Tho Major: Civil Engineering Major ID: 8580201 I. THESIS TITLE: Machine learning in predicting mechanical behavior of 3D printed beams with triply periodic minimal surface (TPMS) sandwich cores Mô hình máy học trong dự đoán ứng xử cơ học của dầm in 3D gia cường lõi sandwich bề mặt cực tiểu tam tuần hoàn II. THESIS ASSIGNMENTS AND CONTENTS: 1. Modeling the triply periodic minimal surface (TPMS) core reinforced beam, and comparing with experimental results; 2.
Collecting beam’s behavior data based on simulations while changing beam’s geometric properties; 3. Creating a machine learning model to predict mechanical behavior of the beams. DATE OF DELIVERING: 05/09/2022 IV. DATE OF COMPLETION: 27/12/2022 V.
Nguyen Thi Bich Lieu Assoc. Luong Van Hai Ho Chi Minh City, 6th March 2023 SUPERVISORS HEAD OF DEPARTMENT Dr. Nguyen Thi Bich Lieu Assoc. Luong Van Hai DEAN OF FACULTY FACULTY OF CIVIL ENGINEERING ii ACKNOWLEDGEMENT Having had the opportunity to study a master's program at the Vietnam National University Ho Chi Minh City – Bach Khoa University, I would like to express my sincere appreciation to the school administrators and departments for creating favorable conditions for me to complete the study program.
I also would like to express my gratitude to the lecturers of the Faculty of Civil Engineering who have always been dedicated to teaching and imparting useful knowledge. In addition, I would like to express my deep acknowledgements to Professor Nguyen Xuan Hung and the CIRTech Institute of Technology for giving me the opportunity to work with the excellent lecturers and colleagues at here. I would not be able to complete this thesis without the guidance of my supervisors Doctor Nguyen Thi Bich Lieu and Associate Professor Luong Van Hai. I would like to express my sincere gratitude to them.
Their orientations and suggestions are both the motivation and the objective to help me steady to complete the thesis. I am truly grateful to Vingroup Innovation Fund (VinIF) for the financial support during my research and implementation of the thesis under project code VINIF. Moreover, I would like to thank my family and friends for always supporting and encouraging me throughout the study and research process. Finally, I would like to wish my teachers, colleagues, family and friends good health, successfulness and happiness.
This thesis may have several shortcomings, so I would like to receive valuable comments from committee members and other students. My sincere thanks. Ho Chi Minh City, 6th March 2023 GRADUATE STUDENT TRAN QUOC KIM iii ABSTRACT (Presented in English) Bioinspired porous structures are highly porous structures with an outstanding strength-to-weight ratio. Their application has been applied in various fields such as aerospace and biomedical engineering, transportation, etc.
Recent research has indicated that 3D-printed plastic triply periodic minimal surfaces (TPMS) structure has tremendous impacts on cement beams, reducing maximum deflection, improving peak load, and enhancing ductility. This study proposes a machine learning (ML) surrogate model to predict beam behaviors subjected to a static bending load. To reinforce the considering beams, different combinations of core layer numbers and plastic volume fractions are adopted. Their influences are investigated using the Finite Element Method (FEM).
Consequently, the gathered data are used to develop the ML model through a three-phase assessment to achieve the most appropriate model for the present problem. This assessment consists of model hyperparameter tuning, first performance assessment, and overfitting handling with Deep Learning (DL) techniques. The results indicate a proportional relationship between the volume fraction and the beam peak load as well as the maximum deflection while increasing the number of TPMS layers enhances these properties nonlinearly. Additionally, from the model predictions, there might be a limit value that each trait cannot achieve at a specific volume fraction with any number of layers.
The final model developed in this study is verified by the maximum deviations between FEM and predictions for peak loads and maximum deflections, that are 2. A new early stopping condition can maximize the final model performances on both train and test data, therefore verifying the model's reliability in handling noisy data from FEM. iv TÓM TẮT LUẬN VĂN (Trình bày bằng tiếng Việt) Cấu trúc xốp được lấy cảm hứng từ sinh học là cấu trúc có độ xốp cao và tỉ lệ cường độ trên trọng lượng lớn. Chúng được ứng dụng trong nhiều lĩnh vực như hàng không vũ trụ, kỹ thuật sinh học, giao thông vận tải, vv.
Nghiên cứu gần đây đã chỉ ra rằng cấu trúc bề mặt cực tiểu tam tuần hoàn (TPMS) bằng nhựa in 3D có tác động to lớn đến các dầm xi măng, giảm độ võng cực đại, cải thiện tải trọng giới hạn, và tăng cường độ dẻo dai. Nghiên cứu này đề xuất một mô hình thay thế máy học (ML) để dự đoán hành vi của dầm chịu tải uốn tĩnh. Để gia cố các dầm, nhiều kết hợp khác nhau của số lớp lõi và tỉ lệ thể tích nhựa được áp dụng. Ảnh hưởng của chúng được khảo sát bằng phương pháp phần tử hữu hạn (FEM).
Từ đó, dữ liệu được thu thập được sử dụng để phát triển mô hình ML thông qua phương pháp đánh giá ba bước để tìm ra được mô hình phù hợp nhất cho vấn đề hiện tại. Đánh giá này bao gồm điều chỉnh siêu tham số của mô hình, đánh giá hiệu quả mô hình đầu và xử lý quá khớp bằng các kỹ thuật học sâu (DL). Kết quả cho thấy mối quan hệ tỉ lệ thuận giữa tỉ lệ thể tích và tải trọng giới hạn của dầm cũng như độ lệch tối đa trong khi tăng số lớp TPMS cải thiện các tính chất này một cách phi tuyến. Ngoài ra, các dự đoán của mô hình chứng minh được tồn tại giá trị giới hạn mà mỗi đặc điểm không thể đạt được ở một tỉ lệ thể tích cụ thể với bất kỳ số lớp nào.
Mô hình cuối cùng được phát triển trong nghiên cứu này được kiểm chứng bằng các sai số tối đa giữa FEM và các dự đoán cho tải cực đại và độ lệch tối đa, lần lượt là 2,5% và 3,5%. Một điều kiện dừng sớm mới có thể tối đa hóa hiệu quả mô hình trên cả hai tập dữ liệu huấn luyên và kiểm thử, qua đó chứng minh độ tin cậy của mô hình đối với các dữ liệu phức tạp từ FEM. v COMMITMENT I hereby declare that this thesis entitled "Machine learning in predicting mechanical behavior of 3D printed beams with triply periodic minimal surface (TPMS) sandwich cores” is my research work. All sources referenced are properly and fully cited.
The research data and results in this thesis are guaranteed to be honest and have never been used to defend any other theses. I will take full responsibility for this statement. Ho Chi Minh City, 6th March 2023 GRADUATE STUDENT TRAN QUOC KIM vi TABLE OF CONTENTS MASTER THESIS ASSIGNMENTS .v TABLE OF CONTENTS. vi LIST OF ABBREVIATIONS.
viii LIST OF TABLES AND CHARTS. ix LIST OF FIGURES. Research objective and contents. Research object and scope.
Triply periodic minimal surface structures. Triply periodic minimal surface. Applications of TPMS. TPMS-reinforced beam.
Cement beam with 3D printed TPMS core. Effectiveness of TPMS core. Finite element analysis simulation. Machine learning model.
Introduction to machine learning. Artificial neural networks. RESULTS AND DISCUSSIONS. Finite element method process.
Mesh convergence study. Impact of TPMS-core properties. Machine learning process. Model hyperparameter tuning.
Best model predictions. CONCLUSION AND RESEARCH DEVELOPMENT DIRECTION. Research development direction .64 LIST OF PUBLICATIONS .70 viii LIST OF ABBREVIATIONS Abbreviation Meaning MS Minimal Surface TPMS Triply Periodic Minimal Surface SCDP Simplified Cementitious Damage Plasticity AM Additive Manufacturing FEM Finite Element Method FEA Finite Element Analysis FSDT First-order Shear Deformation Theory AI Artificial Intelligence ML Machine Learning DL Deep Learning ANN Artificial Neural Networks ix LIST OF TABLES AND CHARTS Table 3. The investigating TPMS beams’ geometric descriptions and labels [1].
The core parameters of the investigating beams [1]. The mixture of the cementitious mortar [19]. Mechanical characteristics of the cement material [19]. The SCDP model parameters of the cement material [19].
Mechanical characteristics of the ABS plastic material [19]. The properties of C3D4 and S3/S3R elements. The dataset adopted in the proposed three-phase process for conducting the final surrogate model [1]. The model properties that were used in this thesis ML process [1].
The FEA results of the investigating TPMS-reinforced beams with both peak loads and maximum deflections [1]. The ANN hyperparameters investigated in the model architecture tuning and the optimization tuning processes [1]. The first three greatest ANN results for the validation loss with the ‘Adam’ optimizer [1]. The first three greatest ANN results for the computational time along with a validation loss being less than 0.01 and ‘Adam’ optimizer [1].
Validation losses and training times of various models with different stop patience values and optimizers [1]. Assessment values of the hyperparameter tuned model [1]. The assessment values of the final model [1].57 x LIST OF FIGURES Figure 3. Classical minimal surfaces: a) Catenoids, and b) Helicoids (from https://wikipedia.
Typical TPMS structures: a) Primitive, b) Gyroid, c) I-graph and wrapped package-graph, and d) Diamond. a) Network-based, b) skeletal-based, and c) sheet-based Primitive solids. The nature-inspired TPMSs, a)-d) the butterfly wing with Gyroid geometry, e) the micro Fischer-Koch structure of nano-porous gold, and f) the sea urchin microstructure with the appearance of Primitive TPMS [31]. The specimen of the plastic 3D printed TPMS-reinforced beam before and after filled with cement [20].
The stress-strain curves of a) one-TPMS-layer reinforced beam and b) two-TPMS-layer reinforced beam [19]. The stress-strain curves of various beam schemes including the plain cement beam, the ABS-mold cement beam, and the TPMS-reinforced cement beam [19]. The crack propagations of a) one, and b) two-layer TPMS -reinforced beam with 25% and 100% bending load [19]. The configuration of the three-layer TPMS-reinforced beam.
Parameters of a Primitive sheet-based TPMS unit. The three-layer TPMS-reinforced beam’s simulation under the three- point bending test [1].
