In the past two years, the pace of new car launches has been accelerating, according to the monthly statistics of the new car listing of traditional car companies in the association, a total of 384 new cars have been launched in the passenger car market in 2021, with an average of 32 models per month.
The madness of a new model per day has caused a lot of competitive pressure on the research and development of car companies. Data show that the average development cycle of a new car has been shortened from the past 5 to 7 years to 2 to 3 years.
Under such "racing" pressure, car companies are seeking some cutting-edge scientific and technological means to improve the efficiency of the model research and development stage, such as the use of simulation for product development and test verification, so that the cloud simulation test to replace some of the physical test, to shorten the time of research and development verification and reduce costs, is a typical application case.
How simulation technology can reduce costs and speed up the development of new cars
Generally speaking, the traditional new car research and development adopts the empirical design-test verification method. However, the structure of the car is very complex, in the design process often can not know whether the local and overall strength can meet the design requirements, can only rely on round after round of tests to improve, which requires a long design cycle and high design trial production costs, so it is difficult to adapt to the needs of the market.
The large-scale application of simulation technology can use the basic computer analysis model to predict the strength, life and characteristics of the designed product in the early stage of product design, so as to guide the product design, ensure the product design indicators, effectively improve the reliability of the design product, shorten the research and development cycle, and reduce the research and development cost.
Sheng Yun, director of the CAE Center of Shanghai Huizhong Automobile Manufacturing Co., Ltd., once said in an interview: "In the past, the time period for automobile companies to develop a new chassis was about a few months, and the development cycle was shortened by about 1/3 after using simulation technology. The general development law of automobile companies is to simulate calculations first, and then use samples to verify, if problems are found in sample testing, then return to simulation technology to correct the model. After applying the simulation technology, about 2 to 3 samples are required. If simulation is not used, according to the traditional practice, about 8 to 10 samples are required. ”
At present, simulation technology has run through the concept design of automotive products, prototype production, performance optimization, and subsequent production and feedback on market problems. For example, in order to ensure product safety, new cars need to undergo rigorous crash tests before they are launched, and each crash test means that a new car is completely scrapped, and the auxiliary investment of each collision is also very high, which is a lot of money for auto companies.
Bu Xiaobing, section chief of the Collision Laboratory of China Automotive Technology and Research Center Co., Ltd., said: "Crash test is a destructive experiment, through the simulation technology for real restoration, can ensure that the assessment has more breadth and depth, such as active and passive safety integration, more ride mode conditions in the future, etc., with biomechanical dummy virtual evaluation test conditions Damage is closer to the real traffic accident." In addition, using simulation to reproduce the collision process, it is possible to quickly and cost-effectively find the key points to solve the problem, because some key points may require multiple collisions to be found if the traditional method is followed. ”
In addition to structural safety, automotive functional safety testing also makes extensive use of simulation technology, especially in the functional safety verification of autonomous driving, 11 billion miles of test mileage, relying on actual road testing is impossible to complete, and it is necessary to rely on simulation technology to meet the demand. For example, Tencent's virtual simulation platform TAD Sim, built with the help of 3D reconstruction, game engine and other technologies, can reconstruct the traffic scenarios with physical laws and running logic consistent with the real world based on real data, and then use cloud computing capabilities to run hundreds of test scenarios at the same time, with the ability to test 10 million kilometers per day, which can greatly improve test efficiency and reduce test costs.
Large-scale simulation is inseparable from high-performance computing on the cloud
Simulation has become an indispensable tool for research and development. In essence, simulation can be seen as a large-scale data processing and computing system, which also puts forward higher requirements for the computing power and performance of the platform. It needs to reproduce the physical motion rules of the real world through physical models in the virtual world, and then reproduce the whole life cycle process of the product with real data from the real world and quickly obtain results.
How many computational resources does the use of large-scale simulation require? Automobile research and development is extremely complex system engineering, and the data that needs to be processed in the simulation process is difficult to count, which is inseparable from the support of cloud computing and AI capabilities. Especially driven by the intelligence of the car, the type and complexity of simulation has increased significantly, and the computing resources required for simulation verification have exceeded people's expectations, and only dozens or hundreds of nodes were needed to meet the demand, but now tens of thousands of computing nodes and CPU Cores are needed to complete these tasks, which brings great challenges to the traditional architecture of car companies.
First, large-scale digital twin simulations make virtual projects more computationally complex every 5 years by a factor of 100. The ensuing hpc (high-performance computing) cluster resource demand is doubling every year, but many car companies choose to build their own HPC clusters can no longer meet the needs of today's market for R & D and manufacturing, and large-scale and sudden shortage of computing power has become a situation that car companies often and must face.
With the popularity of cloud computing, simulation services have also been using cloud computing services to quickly move to the cloud in recent years. It is becoming more and more common for car companies to use the latest computing power models and flexible utilization models of public clouds to carry out numerical simulation to assist in product design and analysis.
For example, Changan and Tencent Auto Cloud have launched a cooperation on HPC projects, and have obtained the latest and strongest computing power support every year, and Tencent Cloud has fully adapted to different CAE software in different scenarios to meet changan automobile's diversified R & D and manufacturing needs.
It is understood that at present, Tencent's high-performance computing cloud platform HPC has increased the main frequency to 3.4GHz, and the latest generation of RoCE V2 100G RDMA network can shorten the delay to microseconds. In addition, the massive elastic computing and storage resources can be reasonably elastically scaled according to the service valley peak of the car enterprise, providing A100/A10GPU resource pools, and with a parallel file system up to 100GB/s throughput, to improve the utilization rate of resources for the car companies and avoid waste.
Tencent HPC simulation cloud architecture
In addition to the research and development of automobiles, the cloud will run through the whole link of automobile design, production, manufacturing, sales and after-sales service, providing car companies with full-life cycle digital services. The use of the autonomous driving cloud platform can reduce the R&D threshold for car companies and improve R&D efficiency; the use of the "cloud + SaaS" digital marketing model can obtain a more accurate way for car companies to reach users. At present, more and more car companies prefer a flexible and open cloud development model, in the future, the cloud will use its unique flexibility, high elasticity, high capacity, low cost to provide a steady stream of energy for car companies.