Low-rank and sparse approximations for contact mechanics

TL;DR

This paper investigates the effectiveness of low-rank and sparse approximation techniques for contact pressure modeling in contact mechanics, proposing new methods to address local nature challenges and improve simulation efficiency.

Contribution

It analyzes the limitations of low-rank methods for contact pressure and introduces sparse regression and non-linear interpolation strategies to enhance approximation accuracy.

Findings

Low-rank methods struggle with contact pressure due to local nature.

Over-complete dictionaries improve approximation by capturing local features.

Dynamic Time Warping enables efficient non-linear interpolation of contact data.

Abstract

(Rephrased) Non-conformance decision-making processes in high-precision manufacturing of engineering structures are often delayed due to numerical simulations that are needed for analyzing the defective parts and assemblies. Interfaces between parts of assemblies can only be simulated using the modeling of contact. Thus, efficient parametric ROMs are necessary for performing contact mechanics simulations in near real-time scenarios. Typical strategies for reducing the cost of contact models use low-rank approximations. Assumptions include the existence of a low-dimensional subspace for displacement and a low-dimensional non-negative subcone for contact pressure. However, the contact pressure exhibits a local nature, as the position of contact can vary with parameters like loading or geometry. The adequacy of low-rank approximations for contact mechanics is investigated and alternative routes based on sparse regression techniques are explored. It is shown that the local nature leads to loss of linear separability of contact pressure, thereby limiting the accuracy of low-rank methods. The applicability of the low-rank assumption to contact pressure is analyzed using 3 different criteria: compactness, generalization and specificity. Subsequently, over-complete dictionaries with a large number of snapshots to mitigate the inseparability issues is investigated. Two strategies are devised: a greedy active-set method where the dictionary elements are selected greedily and a convex hull approximation method that eliminates the necessity of explicitly enforcing non-penetration constraints in convex problems. Lastly, Dynamic Time Warping is studied as a possible non-linear interpolation method that permits the exploration of the non-linear manifoldm synthesising snapshots not computed in the training set with low complexity; reducing the offline costs.