Prediction Design Method And Application Based On Combination Surrogate Model

Product prediction design technology is developed as a raising science technology with the development of science technology and the extensive application of computer simulation technology. The core of prediction design technology is to build the surrogate model and predict its properties. The former establishment of mechanical production prediction design model is to make the simulation and analysis to the production based on CAE technology and the related software, and to trial-manufacture the samples by the obtained simulation design model. However, for the complex production, the application of simulation design model for the prediction design is calculation complex and time consuming, and it cannot employ the intelligent optimization method for the optimization design.Through the recent numerous researches, surrogate model technology can simplify the complex system or complex production effectively, reduce the expensive calculation costs of CAD/CAE and employ the heuristic optimization algorithms advantageously, which has a significant guidance and utilization meaning for the research and development of current mechanical production. However, there are still many problems to be solved when establishing the prediction design models by surrogate model. For the limited samples obtained by the expensive simulation, the selection principles of surrogate models types, the precision of surrogate models and the feasibility of optimization results of surrogate models are still the arduous works.Aiming this situation, this project takes the State Major Basic Research Foundation of China (No. A0920131001) and the military defense project of one institute of China North Industries Group Corporation (No.20140661) as the research background, studies the design issues of special vehicle parts and proposes a prediction and design method for the combinatorial surrogate models. Then it considers the conservative prediction of mechanical design and combines these two contents. The main works includes these three contents:(1) This paper proposes a combinatorial surrogate model method. We apply the surrogate model set rather than the unique surrogate model in building of prediction design model for the complex engineering problems. And we introduce the evolutionary multi-agent strategy into the combinatorial surrogate model and realize the evolution for the parameters of basic surrogate model. We minimize the root-mean-square error in fitting process and acquire the weight of basic surrogate model. Then we combine them and form the combinatorial surrogate model. Analytic function and engineering practical example show this model obtains the better prediction precision and robustness.(2) This paper proposes a conservative surrogate model prediction design method. The safety margin is introduced to predict the maximum stress in predicting production structure analysis, service life and the other related properties conservatively when building the conservative surrogate model, this can ensure the feasibility of optimization design result. We use cross-validation method to design the corresponding safety margin according to the specific problems. We proposes a small sample handling technology and obtain the conservative surrogate prediction design method based on the combinatorial surrogate model by fitting error of cumulative distribution function using combinatorial surrogate model. The analytic function and engineering practical example show our method has a better prediction precision with an assigned conservative level.(3) This paper builds the prediction design platform for the special vehicles based on the above research results, it manages the design data, simulation data and experiment data in the research and development process of vehicle parts. It designs the system by integrating the combinatorial surrogate model and makes the data analysis prediction research, which provides the data support for the prediction design and optimization design. The prediction design example of one kind of vehicle's driving axle housing validates the effectiveness of the prediction design platform based on the method in this paper.This paper takes the prediction and design for special vehicle's components and parts based on data-driven as the research background, proposes an evolutionary multi-agent combinatorial surrogate model and conservative prediction design method. The method in this paper helps to design the new type of vehicle's components and parts and is expected to provide the technique support and theoretical foundation for the research and development of complex mechanical production.