Research On Data Mining Based Multidisciplinary Design Optimization For Vehicle Passive Safety

Due to the strong competition in the global automotive industry,safety,fuel efficiency,and environment protection have become three hotspot issues.In order to adapt to and meet the rapidly changing market demand,multidisciplinary design optimization(MDO)has brought huge changes in new product development and collaborative design.However,for some largescale and complicated engineering problems,it usually takes too much time to solve them using traditional evolutionary optimization algorithms.Besides,these engineering problems generally involve multiple design variables and multiple constraints,increasing the difficulty to obtain the optimal solution.Therefore,a high-performance multidisciplinary optimization design method becomes critical to vehicle passive safety design.Meanwhile,large amounts of data exist or are generated during the development process.These data generally contain informative knowledge.Applying data mining to vehicle passive safety design,to extract underlying and useful information and knowledge that can describe the overall data characteristics,has won much concern and attention from many vehicle research and development departments,and will have more importance and significance to vehicle safety and lightweight research.However,current retrieve methods and statistical analysis only acquire the ostensible knowledge and cannot fully utilize these data.How to use them in a high level and to better achieve vehicle passive safety design optimization is becoming an objective requirement for the research of vehicle safety.This dissertation focuses on data mining based multidisciplinary design optimization for vehicle passive safety design.This study is supported and funded by Ford Motor Company University Research Programs(URP)and the State Key Laboratory of Mechanical Systems and Vibration Open Project.Combined with the author's two-year research experience in the Passive Safety Department,Research and Innovation Center,Ford Motor Company(Dearborn,MI,USA),this dissertation summarizes two research directions for data mining based MDO for vehicle passive safety design: optimization design problem reduction and efficient MDO algorithm development,and a flowchart combined with design constraint reduction,design variable reduction,MDO search strategy,algorithm parameter setting and Pareto optimal solution evaluation metric,followed by industrial real-world application research.The main contributions and conclusions of this dissertation can be summarized as follows:(1)Design constraint reduction for vehicle passive safety designAn improved discrete rough set based constraint reduction approach is presented to decrease the problem complexity and reduce the development time cost.In the method,an enhancement for attribute reduction algorithm in rough set is proposed,after the analysis of the characteristics of the data in vehicle body design,and on the basis of traditional rough set.This enhancement can be inserted into all kinds of rough sets and improve them.Several mathematical examples and real-world industrial problems are performed to demonstrate the advantages of the enhancement.This presented constraint reduction method is demonstrated by a real-world vehicle body design problem for safety.The results show that it has great potential to speed up product development process by prioritizing all safety requirements.(2)Design variable reduction for vehicle passive safety designTo reduce the number of design variable,a new fuzzy rough set integrating with fuzzy theory,inconsistency matrix and the above enhancement that can make the attribute reduction more suitable for those data in the vehicle passive safety design process,is presented for multivariate dimension reduction,thus decreasing the problem complexity and increasing the capacity of acquiring the optimal solution.A real-world vehicle body design problem and an occupant constraint system problem are exploited to demonstrate the new approach.The results show that this approach can effectively improve the quality of design solution.(3)Efficient search strategy for multidisciplinary design optimizationDue to the fact that a large number of low-quality designs and repetitive designs exist during the MDO process using heuristic optimization algorithms,such as NSGA-II.A data mining based strategy is developed to increase the search efficiency and accelerate the development process.Clustering is used to identify the near-duplicate or imperfect designs and generated in the optimization process based on the real-time updating database.On the other hand,the historical data of the search process are used to build an approximation model,and optimization is conducted based on this model,then the current non-dominated solutions acquired using the approximation model and finite element model respectively are combined.Finally,this problem is optimized until the stopping criterion is satisfied.Such optimization strategy can efficiently employ the simulation data,speed up the optimization solving process,and increase the possibility of obtaining the globally optimal solutions.(4)Parameter setting for optimization algorithm and a new distribution metric forcomparing Pareto optimal solutionsParameter values have a great influence on the optimum search result and efficiency of NSGA-II.This dissertation employs theoretical analysis and massive case studies to give a proper range of values of distribution indexes for crossover and mutation.Classification and Regression Tree,and another several examples are used to verify this recommendation,and results show its validity.Meanwhile,a new distribution metric for comparing Pareto optimal solutions is proposed as the stopping criterion for NSGA-II,thus avoiding the unnecessary time-consuming design evaluations,and improving the optimization efficiency.Comparisons between opinions from domain experts and judgments by this proposed and other metrics on several mathematical and engineering examples are done,and shows the proposed metric is more effective than others.(5)Methodology and application of data mining based MDO for vehicle passive safetydesignA methodology for vehicle passive safety design optimization is proposed by integrating improved discrete rough set based design constraint reduction method,new fuzzy rough set based design variable reduction method,the clustering and approximation model based optimization strategy,properly recommended algorithm parameter values,and the proposed new distribution metric for comparing Pareto optimal solutions.The detailed flowchart of data mining based MDO for vehicle passive safety design is presented.A real-world engineering application of an occupant restraint system is used to verify the feasibility and effectiveness of the proposed methodology.Totally six crash scenarios with full frontal impact,three impact speeds,two sizes of dummies are considered in this real-world application.The result shows that the obtained solutions using the proposed method are more optimal and better distributed than conventional method.