We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A FEM-guided data-driven machine learning model for residual stress characterization in ultrasonic surface rolling of lightweight alloys.
- Authors
Pradhan, Rahul; Altalbawy, Farag M. A.; Khan, Ahmed Raza; Rodriguez-Benites, Carlos; Sharma, M. K.; Asaad, Renas Rajab
- Abstract
This work proposes the development of a Bayesian-regularized machine learning (ML) model tailored for predicting residual stresses in compressive and tensile modes during ultrasonic rolling of lightweight alloys. Training data were acquired through finite element method (FEM) simulations covering various parameters and alloys. The model exhibited high performance, with R2 values of 0.973 and 0.926, and corresponding RMSE values of 0.038 and 0.070 for compressive and tensile modes, respectively. The successful outcome was attributed to the effective implementation of Bayesian optimization, showcasing its proficiency in scenarios with limited data volumes. Furthermore, a delicate balance between the relevance scores of material properties and rolling processing parameters was identified for optimal prediction performance. Specifically, higher tensile stress values correlated with elevated relevance scores of static pressure, frequency, amplitude of ultrasonic vibration, compressive strength, Poisson ratio, and material density. In contrast, higher compressive stress values were well-predicted with increased relevance scores of rolling depth, amplitude of ultrasonic vibration, yield stress, and Poisson ratio. The study also elucidates the rationale behind these relevance scores and provides a compelling case study demonstrating the fine-tuning of input parameters to achieve target residual stress levels.
- Subjects
MACHINE learning; RESIDUAL stresses; SURFACE analysis; POISSON'S ratio; YIELD stress
- Publication
Applied Physics A: Materials Science & Processing, 2024, Vol 130, Issue 6, p1
- ISSN
0947-8396
- Publication type
Article
- DOI
10.1007/s00339-024-07577-6