| Car body structure design is a key part of automobile production.The large-scale finite element calculation problem involved is the key in automotive CAE analysis.It is also a very complicated and difficult process.With the development of computer hardware,numerical simulation technology has also been widely applied to the finite element analysis of the car body structure.Researchers can adjust the design parameters at any time to complete highprecision numerical simulation.The application of these technologies greatly improves the design efficiency and precision of the car body structure.However,computer simulations face a major problem with bringing convenience,that is,as the scale of the model increases,the computational efficiency becomes more and more low.Therefore,it is necessary to study the parallel computing method of finite element.At present,many parallel computing methods for finite elements use coarse-grained parallel strategies.The dependence on computer cluster nodes is very high.On the computer cluster with CPU as the core,the number of cluster nodes directly affects the computational efficiency.A coarse-grained parallel computing method relies heavily on cluster size.So implementing low-cost finite element parallel computing based on coprocessor is increasingly critical.The Intel MIC-based coprocessor is a multi-core coprocessor with highly parallelism.It is ideal for handling high-complexity parallel computing.When developing parallel programs based on the MIC architecture,single-node resources can be maximized to achieve a good performance-to-power ratio.Moreover,the MIC architecture has a large number of programming modes.The development method is very flexible.At the same time,the CPU and MIC have a large number of common codes,and the program reusability is very high.The many advantages of the MIC architecture provide a new idea for the re search of parallel computing.Based on the practical engineering application,this paper proposes a fine-grained parallel computing method based on MIC architecture for the large-scale computing problems involved in the body design.This method gets rid of the dependence on large-scale computer clusters and realizes the high precision and high efficiency nonlinear finite element parallel computing with low hardware cost.The main work of this paper includes:(1)Based on the MIC architecture,the OpenMP programming model is used to realize the parallel computing of the plate-shell unit nodal force.First,the dependencies between nodes and elements are established during the preprocessing stage,avoiding data competition during parallel execution.And by the loop queue control,the computing task is decomposed into multiple parallel blocks.A single parallel block is executed serially through a single thread to balance the thread load.Then,a mapping method between threads and elements,nodes,and degrees of freedom is established to fully exploit the threading advantages of the MIC architecture.Finally,this paper implements the vectorization execution of the MIC architecture.By reconstructing the multidimensional array pointer into a one-dimensional pointer,linearized storage is realized.The aligned memory access problem is also solved.(2)Based on the MIC parallel calculation of the plate and shell element,the explicit finite element MIC full-process fine-grained parallel calculation method for the plate-shell structure is proposed.The dynamic memory allocation strategy saves the time and cost of space development and data transmission by reusing memory space during the iterative process.The “nocopy” method is used to optimize data transmission.The high-efficiency data interaction between the MIC and the host is realized,and the asynchronous execution strategy is adopted.The parallel calculation using synchronization is used to mask the process of returning the node array and node force from the MIC end,and avoiding the process.The MIC side stores a large number of intermediate results,reducing its memory storage pressure.(3)Based on the explicit finite element MIC full-flow fine-grained parallel calculation method of the plate-shell structure proposed in this paper,the relationship between calculation scale and algorithm performance is analyzed by using a locomotive example.The results show that with the increase of calculation scale,The parallel efficiency of the algorithm is also more ideal.The absolute acceleration ratio can be as close as 12 times.Then a num erical example with 147,264 elements and 147,700 nodes is used to analyze the scalability of the algorithm.The finite element parallel calculation of the model is completed by calling different numbers of threads.The results show that the parallel efficiency of the algorithm is more improved with the increase of the number of threads.Superior,the relative acceleration ratio can reach up to 80 times,so the parallel method can be used to flexibly deploy threads in practical applications,and the algorithm has excellent scalability.(4)This paper presents a complete MIC parallel calculation method for sheet metal forming.Based on the parallel calculation of the explicit finite element MIC of the plate shell element and the shell structure,the mold position update,the slip speed and the MIC parallel calculation of the contact force and the blank holder force are further realized.The calculation accuracy of the algorithm is evaluated by the numerical example of the roof cover.The results show that the accuracy loss of the algorithm is extremely low.At the same time,the parallel efficiency of numerical examples with different computational scales is analyzed.The results show that the parallel efficiency of the algorithm is better with the increase of the numerical example scale.For 490,000 elements simulation models,the absolute acceleration ratio can reach 48 times.Finally,the influence of the number of threads on the parallel efficiency is analyzed.For the numerical example of 250,000 elements,the relative acceleration ratio of calling 224 threads can be 52 times.Using this algorithm,a MIC-based sheet metal forming parallel analysis software CADEM-MIC was developed.(serial number: 2019R11L014673). |
PDF Full Text Request