Thermomechanical Simulation of Manufacturing Processes using GPU-Accelerated Particle Methods

Abstract

Due to the tremendous growth of computer technology, numerical simulations have established themselves as an integral part of most scientific and engineering advances over the past 50 years. In essence, they are best appreciated in areas where the limitations of financial resources, theoretical developments, and experimental studies are regarded as serious challenges. Metal cutting, or manufacturing processes as a whole, is entangled with all of these limitations, hence a prime candidate for numerical investigations. Unsurprisingly though, the numerical analysis of metal cutting is complicated and fraught with pitfalls since there is a wide range of diverse physical phenomena to be modeled. In this view, several grand challenges to address include thermo-mechanical coupling, severe contact/friction conditions, generation of new surfaces, very large deformations, and extremely high deformation and temperature rates. Mesh-based methods such as FEM have proven to be phenomenally successful as a numerical tool for solving such problems, according to hundreds of thousands of scientific citations. Nevertheless, they face major difficulties in handling mesh distortions and numerical instabilities without particular remedies like remeshing and ALE formulations. Not only are these solutions time consuming, but some of them (e.g., the remeshing algorithm) usually leads to the degradation of computational accuracy. Lagrangian (meshfree) particle methods, on the other hand, can handle large deformations with no theoretical limit and without the caveat of mesh distortion (since there is no mesh). They relieve the burden of remeshing procedures, thus an attractive choice for metal cutting simulations. The main technical drawbacks of these methods, however, lie in: 1. Enforcing essential boundary conditions 2. Lack of interpolation consistency 3. Some numerical and tensile instabilities 4. Lack of explicit interface representation While numerous corrections have been made so far to rectify these four deficiencies in different applications, the use of particle methods in manufacturing simulations is still in its nuclei stage and requires further development. In this thesis, a new particle-based software tool that can accurately and efficiently simulate manufacturing processes is presented. A broad array of modern algorithms and state-of-the-art enhancements are incorporated to tackle the shortcomings of particle methods outlined above. Furthermore, algorithmic and computational measures are both implemented to optimize the runtime. Special care is taken to facilitate multi-resolution simulations, where dynamic refinement and coarsening procedures are enabled. While keeping the computational times manageable, the present particle-based code is additionally accelerated by the virtue of General-Purpose computing on Graphics Processing Units (GPGPUs) using the CUDA platform. As a result of this efficiency, one can adopt the code to gain valuable insights into various problems such as the modeling of complex 3D applications, performing several parametric studies, and running simulations in high resolution. Besides multiple reconstruction tests as well as structural benchmarks, the solver is applied to simulate various manufacturing problems such as laser drilling, ultra-precision machining, tribometer device, and metal cutting. The present methods are, nonetheless, more widely applicable to a range of problems in which geometrical, thermal, and mechanical aspects are substantially important. Indicatively, an enhanced Coulomb law whose coefficient of friction is a decreasing function of temperature is proposed for more realistic modeling of metal machining. Thanks to the remarkable speedup gained by GPU computing, the unknown parameters of this friction model are determined for the first time by employingan inverse identification method in iterative cutting simulations. Finally, a fully coupled geometrical-thermo-mechanical model is developed as a proof of concept to maintain the geometry of moving and/or newly generated interfaces encountered in complex multiphysics problems.