Examination of a Priori Simulation Process Estimation on Structural Analysis Case

Matthew Spinazzola, Purdue University

Abstract

Computer Aided Design (CAD) technology enables users to translate ideas into 3 dimensional digital models. 3D modeling has proved invaluable for engineers and architects as it allows them to quickly generate realistic and accurate visual representations of any product or structure. In addition to their value as a communication tool, CAD systems also provide simulation and analysis capabilities to study the behavior of a product when it interacts with the physical environment. Designers can use this information to improve and optimize designs. For example, computational fluid dynamic analysis can help designers understand the airflow distribution around a jet model and the aerodynamic properties of a particular aircraft, so informed decisions can be made during the design phase. Simulation is a computationally intensive process. CAD models must often be prepared and/or simplified in various ways before they can be used in simulations. Though CAD models can be used as a direct input for the simulation software, the geometric complexity of the models must be reduced in order to decrease the computational costs of the simulation. Interoperability can also be a problem, as design engineers have to work with independent software packages for modeling and analysis, which are not always fully compatible. The simulation’s performance depends on the size and complexity of the CAD model. An assembly with thousands of parts will generally be difficult and time consuming to evaluate. Furthermore, the computational techniques used to analyze the model are more efficient on simple geometry such as polyhedral shapes than they are on more complex geometry (e.g. complex surfaces, etc.). Therefore, a designer may remove specific parts of the assembly or certain features from a model which do not add relevant information to the process to perform simulations more quickly with less computational power. Naturally, the simulation results must be valid and equivalent. However, an error is introduced through the simplification, since the prepared model differs from the original model. The simplification error can be minimized if the correct preparation processes are utilized. Preparing a CAD model for simulation generally consists of three steps: simplification, adaptation, and meshing. Simplification involves the modification of geometry and the removal of certain elements. The adaptation step involves identifying surfaces supporting the boundary conditions and extracting faces for meshing. Boundary conditions define the inputs of a simulation model and connect the model to its surroundings. Lastly, meshing approximates the geometry with small complex elements like triangles or tetrahedron which ultimately allows for the numerical analysis. Without simplification, there are many situations where meshing is impossible and the computing time is too large. There are multiple simplification methods. According to Thakur et al., simplification methods can be classified as Surface Entity operators (SE), Volume Entity operators (VE), Explicit Feature operators (EF), or Dimension Reduction operators [1]. To perform one simulation objective, an analyst will combine many preparation processes together. They may adjust the number of features suppressed, the level of simplification, or the size of the mesh elements. Moreover, the situations in which analyses are performed can greatly differ, having many different loads, stresses or boundary conditions in different locations. With the great amount of preparation processes available, the distinctiveness of each analysis scenario, and the overall complexity of the system, it is difficult to optimize the simplification process. Automating the simplification process is challenging due to the lack of formalization of preparation processes. This occurs because it is difficult to predict the impact a particular type of simplification will have on the simulation. Analysts remove features based on their expertise and experience as well as estimations of the impact on the simulation and computational time. Additionally, different simulations require different simplifications. For example, features that are normally removed in a thermal analysis will not necessarily be removed in a computational fluid dynamics (CFD) simulation. Additionally, preparation and simulation costs must also be considered, further complicating the situation. Providing engineers with knowledge of the impact of a particular simplification would provide more certainty in the CAD to CAE process, and would allow for an easier balance of simplification and simulations costs. Despite the large amount of research dedicated to this problem, a formalization process has not been developed. However, some methods have been proposed that provide accurate predictions of the simplification impact and reduce time and analysis costs. One of these methods involves the use of Artificial Intelligence (AI) techniques. The utility of artificial intelligence is to establish relationships and identify rules without users directly inputting these rules. The deep, informal knowledge from experts can be learned by an AI algorithm using established examples, and then applied to a new example. An AI algorithm can be trained on part models, simplified models, and analysis results, and then predictions of simplification impact can be performed on new models (Danglade et al 2014) [2].

Degree

M.Sc.

Advisors

Hartman, Purdue University.

Subject Area

Artificial intelligence

COinS