A machine learning based model to find optimal tolerances in non-rigid assemblies under thermal gradients

Tolerance allocation is one of the key tools to achieve the minimum manufacturing cost and the best performance for mechanical assemblies. Variation in dimensions and geometry shape due to different working temperature for various components within an assembly can make initially assigned tolerances ineffective, thereby impacting the system’s optimal performance. Therefore, it is necessary to consider the component’s deformation caused by thermal gradients. In this research, a new method for tolerance allocation in mechanical assemblies is presented. First, the system’s working conditions and tolerance design goals are specified, followed by finite element simulation of the assembly at desired temperature. In the next step, the assembly equation is determined according to the complexity of the problem by using different machine learning algorithms such as random forest and regression. Finally, the tolerance accumulation, problem constraints, and objective functions are calculated and identified. Multi-objective optimization is then performed considering thermal gradients by utilizing the NSGA-II algorithm to obtain optimal tolerances. A comparison between the presented method and conventional approaches, using an internal combustion engine assembly as a case study, demonstrates reduced manufacturing costs and improved satisfaction with tolerance limits in the new method.

Keywords:

Tolerance allocation , Thermal gradients , Finite element simulation , Random forest model , NSGA-II algorithm