Research On Data Mining Techniques For Automotive Structural Optimization

With the rapid development of the automotive industry and the continuous improvement of people's living standards,the number of automobile ownership in China is increasing rapidly.The consequent problems of excessive energy consumption and environmental pollution are also becoming worse and worse.As a key issue in the development of automotive industry,automotive lightweight technology is undoubtedly one of the most effective means to solve this series of problems.The precipitation of time brings not only the advances of technology,but also the richness and accumulation of data.Data mining techniques can extract valuable information and knowledge hidden in the data to achieve fuller use of historical data.This paper combines the emerging data mining technology with the traditional automotive lightweight issue to conduct a systematic and in-depth study of automotive lightweight design.The research of this dissertation are summarized as follows:Firstly,select traditional support vector regression model as the research foundation and make it clear that the performance of the SVR model is determined by the choice of each parameter in the SVR model construction.The obtaining of the optimal SVR model is converted into the obtaining of the optimal combination of parameters.Particle swarm optimization algorithm is utilized to obtain the SVR model with the best prediction performance.Secondly,by using the mutual information feature selection method in data mining,the mutual dependence between each design variable and the target response is determined by calculating the mutual information value.Considering of the practical application in automobile design optimization problem,the prediction accuracy of the surrogate model constructed based on the screened dataset is chosen as the control parameter to finish screening the optimal design variable subset.The dimensionality of data can be reduced reasonably while important information is preserved.The complexity of the system is reduced and the optimization efficiency can be improved.Then,the K-means clustering algorithm in data mining is used to analyze the training sample set.The relationship between the clustering result and the number of clusters is investigated and an index is proposed to determine the optimal number of clusters.After the feature information of original design space is effectively recognized,original design space can be reduced appropriately and the feasible design domains are obtained.The optimization efficiency is improved and the likelihood of finding the global optimal solution is significantly increased.Finally,an overall framework of lightweight design of body structure which integrates the key design variables selection,high-precision approximation model construction and design domain recognition is established.A real-world application of a vehicle is implemented.In the practical engineering application,the rolling over condition is considered,the lightweight design of the vehicle body is conducted while the relevant performance indicator meets the design requirement.The research results of the dissertation show that combining the emerging data mining techniques with the traditional body lightweight design issue can effectively achieve the goal of weight reduction.