Efficient phase-field simulation for linear superelastic NiTi alloys under temperature gradients

Engineering the martensitic transformation (MT) with exceptional and controllable properties is essential for the innovative application of shape memory alloys (SMAs) in advanced technologies. Herein, we combine high-throughput (HTP) phase-field simulations and machine learning approaches to propose the concept of temperature-controlled mechanics and demonstrate that it is possible to design NiTi alloys with outstanding mechanical properties that integrate ultra-low modulus, linear superelasticity, and no hysteresis in environments with temperature gradients. These nontrivial mechanical properties originate from continuous variations in the critical stress for the MT, which contributes to a gradual and continuous MT rather than a sharp first-order transition as occurs in common SMAs. An active learning workflow based on uncertainty sampling is employed to guide phase-field simulations to efficiently clarify and optimize the temperature in the environment for different NiTi alloys with the desired properties. Furthermore, the SISSO (Sure Independence Screening and Sparsifying Operator) algorithm is applied to the datasets from the HTP simulations to establish an explicit expression for the Young's modulus, which is verified by additional phase-field simulations and is instructive for the inverse design of the temperature field. The present study not only provides fundamental insights into the effects of temperature gradients on MTs and overall mechanical properties, but also offers a promising computational approach for developing advanced materials with extraordinary properties.

Automated summary

This study proposes the concept of temperature-controlled mechanics and demonstrates the design of NiTi alloys with exceptional mechanical properties by combining high-throughput phase-field simulations and machine learning approaches. The nontrivial mechanical properties, including ultra-low modulus, linear superelasticity, and no hysteresis, are achieved through continuous variations in the critical stress for the martensitic transformation. Active learning workflow and the SISSO algorithm are employed to optimize the temperature environment and establish an explicit expression for the Young's modulus, providing guidance for the inverse design of the temperature field. This research not only provides insights into the effects of temperature gradients on martensitic transformations and mechanical properties, but also offers a promising computational approach for developing advanced materials.