| With the help of the encouragement from policies and markets,automotive industry has developed boomingly.Its volume and sales are still in the first place in 2014.At meantime,we should pay a special attention to the auto quality.As known,people have realized that the product quality is designed,and the uncertainty is the root cause for quality issues.An effective way to improve the quality is to enhance the design’s robustness by considering the uncertainty from the design phase.Thus,it is important to study the robust design methodology and its applications in the presence of uncertainty to improve the product quality.Probability and fuzzy based design methods require a large quantity of data to establish the distribution and membership function when describing the uncertainty.However,it is hard to be satisfied in practical.Although interval analysis doesn’t need the data as much as the two theories,it has non-reversible computations and limited semantic expressing such that it has limitations when describing the uncertainty,especially in applications.As a new method to quantity the uncertainty,generalized interval makes up the restrictions of classical interval by manipulating the orders between the two bonds of an interval,and has reversible computation with dual operator.However,the study of generalized interval based design methods has not been started yet,especially in sematic describing,modeling,feasibility analysis and evaluation of dynamic system in the presence of uncertainty.Robust design is a popular design method in the present of uncertainty,which is to de-crease the quality characteristic the variation due to the uncontrollable inputs by adjusting the controllable ones without eliminating the uncertain source.Robust design is defined as an op-timization problem which the optimum target is wanted.However,the design process is fulfill with the improving the information such that the optimization model should vary with it.It will be a waste if the model is solved repeatedly before it is fixed.Therefore,the feasible one which satisfies the constraints should be discussed first before the optimum is to be found.Searching the feasible results which satisfies all the design constraints is to provide completed information of the problem for the engineer to make a robust decision.Generalized interval based uncertain design methods are studied in the thesis,which is to discuss the practicability of the proposed methods in the framework of constraint satisfaction problems.The main contents and novelties are as follows:(1)A feasible design space searching model based on quantified constraint satisfaction problems is proposed,in which generalized interval is applied to quantify the episematic uncer-tainty,and an interval can be expressed as controllable or uncontrollable with logic quantifiers.A branch-and-prune algorithm is developed to find feasible design spaces for three types of so-lutions based on semantic analysis.In addition,a hybrid stratified Monte Carlo method is also proposed to verify the solutions produced by the branch-and-prune algorithm.The proposed method builds a QCSP based design model,find out robust and feasible solutions with the help of universal or existential quantifiers and constraints.(2)A global sensitivity analysis method is proposed for generalized interval valued vari-able,which is developed based on QCSPs modeled design problems.The indeterminacy of an uncertain variable is analyzed by a generalized Hartley like measure based indicator.The infeasibility is used to evaluate the impact of an uncertain variable on the quality output.A sen-sitivity zone is formed for a variable with respect to each quality output.The proposed method can analyze the impacts of numerical uncertainty and design intent of a variable on each of quality performance with evaluating the lower and upper bounds for this variable,but without assuming its probability distribution.(3)An extended Kalman filtering is proposed base on generalized interval probability,considering with system error and disturbance simultaneously by using generalized interval probability.The statistical properties of generalized random intervals are analyzed and defined,based on which the extended Kalman filtering is reasoned,including the predict and update steps.The proposed method extends Kalman filtering from real number to interval probability.It can be applied to general design problems of estimating the hidden state by observations.In summary,the proposed methods provide alternatives for traditional design methods to deal with interval valued variables.A transmission parameters design problem is used as an illustration,which states that the proposed methods are applicable and effective. |
