Study On Parallel Computing And Multi-fields Coupling Of 3D SDDA In Cloud Environment

The rapid development of modernization promotes the development of geotechnical engineering technology,and numerical simulation has a wide range of applications in the engineering field because of its advantages of repeatability,controllability,and nondestructive.As a discontinuous numerical method with a strict mathematical foundation,Discontinuous Deformation Analysis method(DDA)is indispensable in the study of large deformation and displacement of rock masses.However,the contact algorithm of 3D DDA is complex and time-consuming.The further developed the Three Dimensional Spherical Discontinuous Deformation Analysis method(3D SDDA)uses the rigid sphere as the basic unit,simplifies the contact algorithm,and the freedom unit is only 6,which reduces the difficulty of the contact algorithm and calculation cost.After derivation,compilation,verification,and optimization of the 3D SDDA program,there are still some problems that need to be solved.For example:(1)With the increase of pellet units in the serial program,the computing time and memory occupation increase rapidly.(2)Simulated rock fracture under multi-field coupling needs to be developed.To deal with the above problems,this paper reconstructs the calculation module part of the serial program from several aspects such as code writing,algorithm,and principle.The multi-fields coupling algorithm of the 3D SDDA program is further derived.The technical concepts related to cloud computing are studied,and the corresponding 3D SDDA parallel program is written based on the cloud environment.The advantages of reliability and efficiency of the new program were demonstrated through the simulation of verified examples and engineering case analysis,which made a small contribution to the development of the 3D SDDA program into practical engineering.The main conclusions of this paper are as follows.(1)The 3D SDDA particle bonding model is improved to limit the relative rotation between bonding particles;the 3D SDDA program is optimized in four aspects: data structure of the calculation module,contact pair storage,sparse matrix compression storage,and iterative solution method,which solves the problems of time-consuming and large memory occupation in the serial program.The new serial program is conducive to the development of related parallel programs.(2)The thermal-mechanical coupling model applicable to the 3D SDDA program is proposed,the heat conduction equation between spherical particles and the thermal coupling realization are derived,and the accuracy of the heat conduction model is verified.At the same time,the thermal cracking problems in the relevant tests are simulated.(3)The hydraulic-mechanical coupling model applicable to the 3D SDDA program is proposed.The hydraulic network in the regular particle accumulation model is established.The hydraulic network update algorithm,fluid flow,and fluid pressure calculation are proposed.The accuracy and precision of hydraulic pressure evolution and hydraulic network update under a complex fracture network are further verified,and the hydraulic fracture propagation in horizontal wells is simulated.(4)We discuss the mainstream cloud computing service platform,build and configure the running environment of 3D SDDA program,and propose three loadbalanced parallel computing models: MPI parallel model,Open MP parallel model,and hybrid MPI+Open MP parallel model for the calculation module.Further,we elaborate the points of each parallel model implementation,and verified the accuracy and reliability of the 3D SDDA parallel computing program in the cloud environment.Finally,we test the computational efficiency and speedup ratio of the three parallel computing models under different sphere particle size and number of running processes/threads.we obtained the highest speedup ratio of more than 70 times.(5)In the cloud environment,the parallel computing model and the multi-fields coupled model are used to simulate engineering examples to verify the correctness and effectiveness of the method proposed in this paper.Firstly,the hybrid MPI+Open MP model is used to simulate the sliding process of the million-particle pile,and it took only36 hours to complete the simulation.Similarly,the formation process of the Huang Yang Gou landslide is simulated under this parallel computing model,and the landslide model is built with 10286 triangular surfaces and 284874 spheres,and the simulation takes only 27 hours.Then,an Open MP parallel model is taken to rewrite the multi-fields coupling modules,and the parallel computing efficiency of the multi-fields coupling programs is tested respectively.