Distributed Finite Element Analysis

Abstract

A lot of advancement has taken place in computer generation in last five decades. And this advancement has led to the development of computers having processing speeds of teraflop range and memory of terabyte range.

As the computing power available increased, attempt was made to solve more and more difficult problems. Some of the problems related to field such as artificial intelligence, numerical analysis, Quantum chromo dynamics, Climate modeling, Fluid turbulence, Vehicle dynamics, Ocean circulation etc, require such a high computational power and also memory requirement that present day’s computers may found slower to solve these problems. These limitations have led to develop some other solutions. High performance computing is result of such attempts.

Various possible ways are available for high performance computing i.e. supercomputers, massive parallel machines, cluster computing etc. But as these options are costly from initial cost and maintenance point of view, an alternative option of Parallel Processing on network of computers, also known as distributed computing can been used. As distributed computing is carried out on network of computers, no special hardware is needed and also no special means of maintenance is needed. All these aspects make distributed computing a favorable option for high performance computing. In present study, distributed computing is carried out using WebDedip environment, developed using JAVA technology based on client server approach.

As far as structural field is concerned, problems include difficult geometry, boundary conditions and loading conditions, which makes the problem complex. Also, as the size of problem goes on increasing, requirement of computational power and storage memory goes on increasing. So, present single computers may found slower to tackle such problems. Hence analysis of such problem becomes a tedious and time consuming task. In such cases some other alternative is needed which can serve better in such conditions. High performance computing is one of the solutions for such problems.

In present study an attempt has been made to implement distributed computing in field of structural engineering. Finite Element Analysis method has been used for analysis. Substructure analysis technique has been incorporated to implement analysis process on distributed computing. To observe the advantage of high performance computing, four problems have been solved using sequential and parallel processing. Of these problems, one is the plane stress problem, two are plate bending problems and one laminated plate problem.

First chapter includes the introduction part. Topics such as need of high performance computing, efforts made in India as well as in world for high performance computing and object of study are included in this chapter.

Second chapter includes the literature survey. Scope of work is also included in the same chapter.

Third chapter includes the issues related to parallel processing. These issues are computer architecture, parallelism in sequential as well as in multiprocessor computers, dynamic load balancing, operating system characteristics, parallel processing models, performance measurement and software tools for parallel processing.

Forth chapter deals with the aspects related to WebDedip environment. Details of various components to be installed on server as well as client machine, minimum software and hardware requirement for distributed computing and procedure of problem implementation on parallel computers are some of the topics, discussed in this chapter.

Fifth chapter includes the distributed plane stress analysis problem. The problem is a square plate having hole in the center and subjected to uniform tensile force on both sides. For analysis of plate, CST element is used. Plate is divided into 27504 number of elements. So the problem is having total 13990 number of nodes (i.e. 27980 DOFs). To observe the effect of distributed computing, plate analysis is carried out using 3, 4, 5 and 6 computers.

Sixth chapter deals with the distributed plate bending problem. This chapter includes two problems. For analysis of both problems, eight nodded isoparametric plate bending element is used. Of two problems, first problem is the annular plate subjected to uniform lateral loading. Plate is divided into 1800 elements. So problem is having total 5725 number of nodes (i.e. 17175 DOFs). Plate analysis is carried out by diving it into 3, 4, and 6 number of substructures. The second problem is the simply supported skew plate subjected to uniform pressure. Plate is divided into 3000 number of elements. So, total number of nodes are 9221 and DOFs are 27663. To implement the problem on parallel computers, plate is divided into 3, 4, 5 and 6 number of substructures.

Seventh chapter covers the distributed laminated plate analysis. The problem is a square plate having four laminates of same thickness and subjected to sinusoidal loading. For analysis of laminated composite plate, eight nodded isoparametric element is used. The plate is divided into 1600 elements. So problem is having total 4961 number of nodes and 29766 DOFs. To observer the effect of distributed computing, the plate analysis is carried out using 3, 4 and 6 number of computers.

Chapter eight includes the summary, conclusion and further scope of work.