| Intelligent tolerance design for mechanical assemblies considering thermal gradients using supervised and unsupervised machine learning |
| کد مقاله : 1024-ICME2025 |
| نویسندگان |
|
Hossein Soroush، سعید خدایگان *، جواد همتی نیک Sharif University of Technology |
| چکیده مقاله |
| Tolerance design is one of the prominent tools to achieve the optimal performance of an assembly. Different factors like thermal deformations would make the initially assigned tolerances ineffective. Therefore, it is essential to consider the temperature variations during the tolerance design procedure. This study presented an intelligent tolerance design framework using machine learning (ML) models. First, the geometrical and dimensional tolerances influencing the performance of the assembly are identified. Next, tolerance design goals and the working temperature conditions are determined. In the third step, finite element modeling (FEM) is carried out to collect the required dataset. Then, supervised ML models predicted the assembly equation. In the following step, multi-objective optimization is performed to explore the optimal tolerances. Finally, unsupervised ML algorithms are employed to cluster the obtained optimized answers, and the classification model classifies the clustered dataset to provide various groups of tolerances according to the manufacturing costs and tolerance values, which helps to assign the optimal set of tolerances using an ML-based intelligent approach. A turbocharger assembly was investigated as a case study to evaluate the efficiency of the proposed approach. The random forest (RF) algorithm estimated the assembly equation with an 80% coefficient of determination. Next, the NSGA-II algorithm was applied to find the optimal tolerances. Subsequently, the K-Means model clustered the optimal answers into five groups according to their similarity. Finally, the RF algorithm was utilized to predict the class of the optimal solutions. The results indicated an accuracy of 92.31% for optimal answer classification. |
| کلیدواژه ها |
| Tolerance design, Thermal deformations, Random forest algorithm, K-Means model, Turbocharger |
| وضعیت: پذیرفته شده برای ارائه شفاهی |
