The Study Of Distributed Parallel Algorithms In Inverse Heat Conduction Problems

The inverse heat conduction problem(IHCP) is usually defined as theestimations of boundary/initial conditions, thermal parameters and heat source byutilizing the known temperature measurements inside the body or on the surface. Thestudy on IHCP is an interdisciplinary field related to the heat transfer, physics,methematics, computing, and experiment technique, etc. It has significantapplications in many engineering aspects, such as aerospace, nuclear engineering,metal casting and so on.Due to the ill-posedness and nonlinearity, solving IHCP is usually much moredifficult than solving direct heat conduction problem(DHCP). Although largeamounts of achievement has been made in this area all over the world, furtherinvestigation and effort are greatly demanded.The computational complexity of inverse problem is farther more than the directproblem's, so it's very practical significance to make researches on the algorithmswhich can compute very fast. Solving IHCP by parallel algorithm can pick upcomputing speed and have satisfying result. The hardware platform is PC connectedwith LAN. The software platform is MPI and LINUX. They together construct thewhole PC-cluster system. Based on the PC-cluster system, the thesis focuses ondetermination of thermal parameters in a two-dimension heat conduction equationusing the distributed parallel algorithms about Genetic Algorithm( ParallelGenetic-Neural Network Algorithm ).Firstly this thesis introduces IHCP and the research background andsignificances related with the subject; then shows basic theory of the parallelcomputing, introduces the cluster concept and the message passing system of MPI,after that, discusses how to establishes a parallel computing environment based on theMPI and Linux; following introduces the basic theory about GeneticAlgorithm(GA), Neural Network(NN) and Back Propagation (BP) Algorithmmostly, and then analyses the advantages and shortcoming of them; In order to solvethe thermophysical properties inverse problem of ceramic/metal combine material,the thesis combines BP neuaral network and GA to full advantages of both whichmakes new algorithms called the neural network algorithm based on GA, it has BPneural neural network's learning capability and robustness and GA's strong globalsearch capability; and then based on network background, integrates principle of parallel algorithm with parallel characteristic of NN, designs and implements theParallel Genetic-Neural Network Algorithm to solve the above IHCP;Finally, conclusions of this thesis and suggestions for further research are given.