Accelerated development of soft magnetic materials

• Main Author: Padhy, Shakti Prasad
• Other Authors: Raju V. Ramanujan
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Materials::Metallic materials::Alloys; Engineering::Materials::Material testing and characterization; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/172227
• Institution: Nanyang Technological University

Soft magnetic materials have a wide range of applications in various fields, including electrical power generation and distribution, electric motors and generators, magnetic sensors, magnetic recording, electromagnetic shielding, and medical devices. Due to the rising electric energy demand, there is a pressing need to develop energy-related technologies with improved performance and efficiency. High-speed, high efficiency, low manufacturing cost, and small-sized electric machines, including electric motors for electric vehicles and generators for turbomachinery, are needed to be developed. Therefore, for such next-generation electric systems, it is required to develop superior magnetic materials possessing an attractive blend of magnetic, electrical, and mechanical properties. Using conventional techniques, the development of such new materials is slow, expensive, and limited to a few material compositions. Hence, this is addressed by accelerated materials development methodologies, which include combinatorial experiments, rapid characterization techniques, and machine learning (ML) methods. Two types of combinatorial experimental methods were implemented to generate Fe-Co-Ni alloy libraries. Firstly, a combinatorial magnetron co-sputtering method was used to synthesize compositionally graded Fe-Co-Ni film alloy libraries. Also, different processing conditions, such as substrate temperature and annealing temperature, were explored. The structural, magnetic, electrical, and mechanical properties of the libraries were evaluated rapidly. After studying the structure—processing parameter—property relationships, the library sputtered at 500 oC was found to display a superior combination of properties. The alloy composition of Fe12Co33.1Ni54.9 was discovered to exhibit an attractive combination of properties potentially suitable for next-generation electric machine applications. Secondly, a hyper-heuristic combinatorial flow synthesis device was designed and used in a semi-automated setup to produce a Fe-Co-Ni powder alloy library. The powders were then consolidated into a compositionally graded Fe-Co-Ni bulk alloy library using high-throughput spark plasma sintering (HT-SPS), followed by annealing in a hydrogen atmosphere. Rapid multi-property assessment of the libraries was performed. A large variation of properties was observed with respect to composition in an annealed bulk alloy library. Saturation magnetization (Ms) varied from 83.3 emu/g to 225.2 emu/g, coercivity (Hc) from 17.5 Oe to 78.4 Oe, resistivity (ρ) from 17.2 μΩ·cm to 986.7 μΩ·cm, and Vickers hardness (HV) from 41.9 HV to 281.7 HV. Fe49Co44.9Ni6.1 exhibited the highest Ms, Fe26.9Co22.3Ni50.8 exhibited the lowest Hc and highest HV, and Fe65.5Co11.7Ni22.8 exhibited the highest ρ. Further, a screening criterion was implemented to identify six promising compositions, out of which three compositions (Fe36.5Co55.1Ni8.4, Fe22.6Co73.4Ni4, and Fe5.7Co84.4Ni9.9) with a good balance of properties were identified as potentially suitable for next-generation electric machine applications. Apart from the combinatorial experimental methodologies, an ML framework was developed and implemented to discover multi-property optimized Fe-Co-Ni alloys. For this, a database of bulk Fe-Co-Ni-based alloys was curated from literature consisting of magnetic, electrical, and mechanical property data. A huge gap in the property data was identified and filled using a novel ML-based imputation strategy. Based on this imputed database, a multi-input multi-output predictive ML model was developed. Using multi-objective Bayesian optimization combined with the multi-property model, an inverse design strategy was deployed to predict promising compositions. Both the multi-property ML model and inverse design strategy were experimentally validated, and the predictions matched well with experiments for Ms, Tc, cost of material and HV. Using the inverse design strategy, two superior performance alloy compositions were designed – Fe65.84Co28.66Ni5.5 and Fe61.5Co23.14Ni15.35 with a targeted combination of Ms of 2.2 T and 2.4 T, Tc of 1173 K and 1173 K, and HV of 353.5 HV and 268.2 HV, respectively. The above ML-predicted compositions were synthesized by three different routes: arc melting, ball milling followed by SPS, and chemical synthesis followed by SPS.Further, they were annealed in a hydrogen atmosphere. Multiple properties of both synthesized and annealed samples were evaluated, and a comparative study of the properties was performed with respect to the synthesis routes. Arc melting was identified as the most suitable synthesis method for validating the ML-predicted Fe-Co-Ni alloy compositions, as the experimental property values of the synthesized and annealed samples were within 10% and 8% deviation of predicted property values, respectively. Overall, the experimental and ML strategies followed in this work prove to be very effective in exploring the composition space of the Fe-Co-Ni alloy system and identifying promising compositions with an optimal combination of properties for next generation electric machine applications.