Theses supervised by Prof. Dr. Demircan Canadinç
10 theses · Koç University
Shape memory alloy design by machine learning for biomedical and high-temperature applications
Shape memory alloys (SMAs) are of great importance due to their extensive usage in biomedical applications, aerospace engineering, or robotics. In recent years, although there has been a considerable amount of research to achieve optimum compositions of SMAs for these applications, due to the high experimental costs, the demand for alloys with optimum properties has not been met for many applications yet. In this research, using the predictive power of artificial intelligence and machine learning, a systematic approach to predict the optimum composition of the SMAs was proposed to address two problems related to the binary NiTi and NiTi-based SMAs. In particular, in chapter two, the optimum chemical composition was proposed to minimize the Ni ion release in the binary NiTi SMA. The method to do so was to gather a database from the existing literature and using it to train a special algorithm that provides the information for predicting the desired compositions. In chapter three, using the same approach, two models were developed to predict the phase transformation temperatures and thermal hysteresis of multi-component NiTi-based SMAs. These models were used to predict the optimum alloy with the highest Phase transformation temperatures with the least possible thermal hysteresis.
NiTiHf high temperature shape memory alloy design with the assistance of machine learning
The shape memory alloys demonstrate a unique property that allows them to recover enormous shape changes above certain temperatures; therefore, they are a notable option as a compliant actuator. Increasing demand in the aerospace and oil industry for high-temperature actuators motivates the search for high-temperature shape memory alloys. High operating temperature range, medium-ductility, and remarkably lower cost of NiTiHf alloys distinguish them from various high-temperature shape memory alloy systems. However, as a ternary alloy system, NiTiHf alloys have vast search space to be analyzed, which requires significant investment. On the other hand, the implementation of machine learning in material science has been proven promising and affordable alternative for experimental search, which gave inspiration to current work: designing NiTiHf shape memory alloy that can exhibit phase transformation beyond 400 ℃ with the assistance of machine learning. In this work, a comprehensive dataset was established from the available literature on NiTiHf alloy for training by a multilayer feedforward neural network algorithm. Via the optimized neural network model, phase transformation temperatures of unexplored NiTiHf search space were estimated. Two novel compositions with desired functions, namely the Ni49.7Ti26.6Hf23.7, and the Ni50Ti27Hf23 alloys, were selected to validate machine learning predictions. The former demonstrated an Af temperature of 403.5 °C with high cyclic stability, which verifies the success of machine learning in new alloy design.
A numerical and experimental design of high entropy alloys for biomedical applications
High entropy alloys are a relatively new type of multi-component alloys that attracted researcher's attention because of their superior properties, which mainly come from their high mixing entropy. Some of their main features are excellent mechanical and corrosion properties that lead them to be a potential candidate for biomedical applications. In this study, both numerical and experimental methods were used to enhance the biocompatibility of high entropy alloys. A machine learning approach was implemented to design a high entropy alloy with desired properties. Two unsupervised machine learning clustering techniques were performed to design a high entropy alloy with excellent biocompatibility properties. The characteristics required for these alloys to be used in biomedical applications were used as the input dataset of the models. These features were mainly pure metal's physical, chemical, mechanical, and biocompatibility properties, such as elastic modulus or cell viability measurements. Furthermore, a magnetron sputtering method was utilized to coat the high entropy alloys with silver for introducing antibacterial properties. The parameters of experiments were adjusted to achieve a coating film with desirable characteristics. It was shown that by increasing the deposition duration, the coating thickness and grain size increased, which can enhance the antibacterial characteristics. The ion release of coated samples was then measured after immersing in simulated body fluid for fourteen days. The measurements demonstrated that the Ag ion release increased by increasing the coating thickness, but the ion release of substrate elements stayed almost the same. Consequently, the higher release of Ag ions may lead to a better antibacterial effect since it is the main factor controlling antibacterial characteristics.
In-process monitoring and qualification of SLM produced nickel based superalloys using eddy current inspection method and a novel EC array probe design
Additive manufacturing gives high flexibility to part designers on the manufacturability of optimized and complex parts. This advantage brings a part inspectability challenge. In this study, the Eddy Current Array sensor application is investigated as an in-process monitoring/inspection method and discussed as an L-PBF layer acceptance technique for the volumetric qualification of aerospace parts. Laser Powder Bed Fusion (L-PBF) process produced Inconel 625 test samples are used for data collection. Intentionally defects are created by using a waterjet-guided laser system to trial the L-PBF in-process eddy current inspection capability.
Design optimization of plate heat exchanger by utilizing artificial intelligence
The plate heat exchangers are components that provide heat transfer between two or more mediums; therefore, they have crucial importance in the industry. While the plate heat exchangers are produced, sheet metal forming process is employed. Growing interest in manufacturing lightweight, cost effective, and robust products in HVACR industry motivates the search for compact plate heat exchangers used in combi-boilers. Lightweight compact plate heat exchangers distinguish them from other plate heat exchangers in case of raw materials shortage periods. However, manufacturing lightweight plate heat exchangers requires time consuming simulation works and significant investment for production trials. On the other hand, implementation of machine learning assisted optimization process has been proved promising and reasonable alternative for traditional production methods which inspired current work: "design optimization of plate heat exchangers by utilizing artificial intelligence" that satisfies requirements stated in material order specifications of Bosch Home Comfort. In this work, an extensive dataset was constructed from finite element analysis simulations on sheet metal forming process by considering the design limits of compact plate heat exchanger produced in Bosch Home Comfort Manisa Plant to train multilayer feed-forward neural network algorithm. The maximum thinning amount of metal plates after forming process was predicted by using optimized neural network algorithm. Six cases with desired outputs were selected to validate neural network algorithm predictions. The demonstrated thinning amounts verify the success of machine learning algorithm in design optimization of plate heat exchangers.
Experimental and computational assessment of high-temperature properties in ti-based high entropy alloys
During the last two decades, high entropy alloys (HEAs) have gained noticeable attention due to their outstanding properties and potential for various applications. Within the prospective applications, due to their high melting point and high thermal stability, refractory high entropy alloys (RHEAs) are considered an alternative to Ni-based superalloys for high-temperature applications. Therefore, in order to shed light on the high-temperature properties of RHEAs, two aspects of their properties were investigated. After a brief introduction to the concepts discussed in this thesis, the oxidation behavior of four different HEAs was investigated via static oxidation experiments utilizing a variety of surface and structural characterization methods such as X-ray photon spectrometry (XPS), X-ray diffraction (XRD), scanning electron microscopy (SEM). The results suggested that three different oxidation mechanisms in these alloys are dependent on the composition, temperature, and crystal structure of the HEAs. In addition, each element's effect on these alloys' oxidation behavior was discussed. In the second part of the thesis, using atomistic simulations, the effect of the formation of chemical short-range order (CSRO) on mechanical properties was investigated in two different compositions of RHEAs. The results showed that the CSRO formation in these alloys can increase the materials' strength while maintaining the HEAs' ductility. In addition, the effect of CSROs on deformation-induced phase transformation (DIPT) was captured via atomistic simulations. To validate these results, an extensive amount of in-situ compression experiments under SEM were performed and these results validated the formation of HCP phases due to DIPT. Overall, the results and discussions in this thesis provide a better view of the high-temperature properties of the HEAs with the aim of enabling their applications at high temperatures.
Machine learning assisted design of biomedical high entropy alloys with low elastic modulus for orthopedic applications
High entropy alloys (HEAs) have received considerable attention from the scientific community since the 2000s due to their excellent properties and potential to be used in various structural and functional applications. HEAs consist of multi-principal elements governing their final properties in addition to manufacturing methods and heat-treatment processes compared to traditional alloys, whose properties are governed by one main principal element. Therefore, understanding the effect of each element on the properties of HEAs is a complex process. Due to the "cocktail effect", one of the core four effects of HEAs, HEAs can attain unpredictable and outstanding performance superior to the performance of all constituent elements. Because HEAs consist of multiple elements, thousands of possible compositions can be developed, giving rise to different properties for the same family of HEAs. Hence, conventional trial-and-error methods become costly and inefficient in discovering new HEAs. A solution to this issue is using computational methods, such as density functional theory (DFT), molecular dynamics (MD), or machine learning (ML). However, DFT and MD are computationally expensive and time- consuming. On the contrary, ML is an efficient tool for establishing complex and non- linear relations between inputs and target property, making the new HEA discovery process faster and cheaper. In this thesis, three new biomedical HEAs, namely, Hf27Nb12Ta10Ti23Zr28, Hf30Nb14Ta10Ti28Zr18, and Hf12Nb16Ta35Ti29Zr8 were designed and developed utilizing ML. In the first chapter, Hf27Nb12Ta10Ti23Zr28 and Hf30Nb14Ta10Ti28Zr18 HEAs with low elastic modulus, closer to that of the bone, were predicted to reduce the "stress shielding" effect between the bone and implant material. Predictions were validated through experimental methods. In the second chapter, in order to enhance the antibacterial properties of HEAs developed in the previous chapter, they were coated with Ag via Physical Vapor Deposition (PVD). Specifically, the effect of PVD process parameters on Ag coatings' mechanical and ion release behavior was investigated. In the following chapter, the microstructure, surface oxide layer properties, and corrosion behavior of Hf27Nb12Ta10Ti23Zr28 and Hf30Nb14Ta10Ti28Zr18 HEAs were studied, revealing that they exposed superior corrosion behavior in simulated body fluid (SBF) and artificial saliva (AS) compared to conventional implant material, CoCrMo. Lastly, a new corrosion-resistant biomedical Hf12Nb16Ta35Ti29Zr8 HEA was developed utilizing ML in the fourth chapter. It was found that the produced ingot had a dendritic microstructure in the center and a homogeneous microstructure around the circumference. Samples cut from the homogenous part of the ingot showed outstanding corrosion resistance as opposed to the samples with dendritic microstructure and conventional implant material, CoCrMo. Overall, the findings of the thesis prove that ML methods can be utilized to discover new HEAs with desired properties.
Design of an Efficient Endodontic File Material Utilizing Artificial Intelligence
Nickel-titanium (NiTi) alloys have led to a revolutionary development in endodontics due to their flexibility and shape-memory properties. However, despite this advancement, one of the most critical issues limiting the clinical success of NiTi rotary instruments is their fracture within the root canal. The most common cause of these fractures is the material's limited resistance to cyclic fatigue. In this study, machine learning was utilized to determine the ideal NiTi composition that would maximize the cyclic fatigue resistance of NiTi rotary instruments. For this purpose, thermal and structural properties were first obtained through SEM-EDX and DSC analyses conducted on various commercial endodontic files, and these features were included as inputs in the model. Feature selection was performed using Pearson Correlation Coefficient (PCC) and Random Forest (RF) importance-based ranking methods. Subsequently, six different regression algorithms (RF, GBDT, SVR, LR, KNN, XGB) were trained, and their performances were evaluated using R², RMSE, and MAE criteria. Preprocessing steps, including outlier detection, standardization, and 10-fold cross-validation, were applied to the dataset. As a result of analyses conducted using a stepwise feature addition strategy, the highest generalization performance was achieved with the K-Nearest Neighbors (KNN) and Support Vector Regression (SVR) models. To evaluate the effect of training/test ratios, various data splitting scenarios were tested, and a 10% testing ratio was determined to be optimal. Using the best-performing model, synthetic NiTi compositions were generated in the range of 48–57 at.% Ni with a resolution of 0.03 at.%. Based on the predictions, the three compositions with the highest NCF values were identified and selected for experimental validation. The results obtained demonstrate the effectiveness of machine learning in material design and contribute to the development of next-generation endodontic instruments.
Designing corrosion‐resistant HEAs using machine learning and fabricating corrosion‐resistant hea thin films through controlled coating conditions for biomedical applications.
High-entropy alloys (HEAs) have emerged as a promising class of biomaterials due to their unique microstructural stability, mechanical strength, and superior corrosion resistance. This thesis presents a comprehensive investigation into the development, characterization, and performance of novel HEAs for biomedical applications, with a specific focus on orthopedic implants. Four interconnected studies were conducted, each addressing critical aspects of alloy design, thin film deposition, and performance evaluation. In the first study, TiTaNbZrMo HEA thin films were sputtered on NiTi shape memory alloy substrates to mitigate nickel ion release—a factor that limits the clinical use of NiTi-based implants. The results demonstrated that deposition at low chamber pressure yielded dense, crack-free coatings with significantly enhanced corrosion resistance, bioactivity, and adhesion strength. The second study leveraged machine learning (ML) techniques to design novel HEA compositions by correlating compositional and electrochemical parameters with corrosion potential. Two predicted alloys, Ti34.8Ta17Nb21.4Zr14.2Mo12.6 (HEA1) and Ti35Ta23Nb20.8Zr14.2Mo7 (HEA2), were fabricated and experimentally validated, showing superior corrosion resistance compared to conventional alloys. In the third study, the microstructure, bioactivity, and corrosion behavior of the ML-designed alloys were comprehensively evaluated. Both alloys exhibited a refined microstructure with well-distributed constituent phases, improved passivation behavior and stable oxide film formation, confirming their potential for biomedical applications. The final study extended this evaluation by combining electrochemical and non-electrochemical methods to investigate the corrosion mechanisms of the two ML-designed alloys under simulated harsh environments, including corrosive water and salt-spray chamber tests. The results revealed that the alloys not only resisted localized corrosion but also maintained their stability in aggressive media, highlighting the role of corrosion-resistant elements (such as Ta, Nb, and Mo) in their composition. Collectively, these studies highlight the synergistic integration of thin film engineering and machine learning-driven alloy design in advancing next-generation biomedical HEAs. The findings provide a solid foundation for optimizing alloy compositions and deposition conditions, paving the way for durable, biocompatible, and corrosion-resistant implants.
Biyomedikal uygulamalar için özgün yüksek entropi alaşımlarının geliştirilmesi
High entropy alloys are solid solution alloys that contain multi-principal elements in nearly equiatomic percent. This unique concept introduces a superior combination of enhanced mechanical properties that cannot be exhibited by conventional alloys consisting of one or two principal elements. The superior properties such as high strength, high fracture toughness, excellent wear, and corrosion resistance have led to the investigation of whether the novel materials could be included in the field of medicine as biomaterials. Microstructural properties, mechanical properties and bioactivities of the high entropy alloys Ti25Ta25Hf25Nb25, Ti20Ta20Hf20Nb20Zr20, Ti20Ta20Hf20Mo20Zr20, and the medium entropy alloy CoCrMo have been evaluated to observe the potential for the fruitful use of these materials in the biomedical field.