基于置信规则库的电路可靠性容差分析方法

发布时间：2018-11-03 16:01

【摘要】：电路可靠性容差分析是数/模电路设计与研发过程中的一个重要环节,对于提高产品生产中的成品率,以及使用中的可靠性是十分重要的。目前对电路可靠性容差分析主要涉及两大类方法,一是基于统计分析方法,二是基于几何逼近的确定性方法,这些方法通常具有计算量大、要求电路性能函数形式较为简单(其对应的设计域为单连通凸形)等缺点,这限制了它们在实际电路设计中的应用。实际上,在电路设计与电路仿真中,电路设计者或使用者会积累一些具有主观性和不确定性的经验知识,若将这些主观知识与电路仿真客观数据相结合进行容差分析,必将提升电路设计的效率并缩短产品的研发周期。为了实现该目的,本文基于置信规则库模型,将主观经验信息与电路仿真数据相结合,给出新的电路可靠性容差分析及模型优化方法,主要工作包括:(1)基于置信规则库(BRB)的电路容差设计方法。建立描述电路元件参数值与电路性能指标值之间非线性映射关系的BRB模型;在设计域呈现非连通和非凸形的复杂情况下,提出基于BRB的电路容差初值求解及容差域优化方法;通过Rosenbrock型的电路性能指标函数以及实际的轨道电路调谐单元共振性能指标函数为例,验证了在设计域非连通和非凸形情况下,所提方法在相对小的计算代价下,可以精确地从设计域中确定所有的元件容差及中心值。(2)基于置信规则库的电路中心值设计方法。当问题(1)中元件容差固定时,电路容差分析就是要找到元件的最优中心值(标称值)。建立基于BRB的电路可靠度估计模型,用其描述电路元件参数中心值与电路可靠度之间的非线性映射关系;然后,提出基于BRB电路中心值设计方法,其与传统的统计分析方法相比,在计算量较低的前提下,不仅能找到单一的最优中心值,更能确定满足可靠度要求的整个中心值的设计域,从而为设计者提供多个备选的中心值;最后,本文仍利用轨道电路调谐单元元件参数中心值设计为例说明所提方法的有效性。(3)基于序列线性规划的BRB模型参数优化方法。通过对前两个问题中BRB参数优化模型的目标函数与约束条件进行一阶泰勒展开,将参数优化的非线性规划问题转化较为简单的线性规划问题;然后给定参数寻优移动限,确定寻优区间,进行线性搜索,将搜索结果用作新基点,重新线性化,直至找到满足停止准则的BRB模型参数;该方法与传统其他非线性优化方法相比,线性规划问题是较容易解决且精确度较高的方法,且该方法可以从可行解空间中的任意一点出发对参数进行训练优化;最后本文仍利用轨道电路调谐单元元件参数中心值设计为例,验证本方法在训练BRB模型参数上,优化效率高的优势。
[Abstract]:Reliability tolerance analysis of circuits is an important part in the design and development of digital / analog circuits. It is very important to improve the product yield in production and the reliability in use. At present, the reliability tolerance analysis of circuits mainly involves two kinds of methods, one is based on statistical analysis method, the other is deterministic method based on geometric approximation. The simple form of circuit performance function (the corresponding design domain is a simple connected convex shape) is required, which limits their application in practical circuit design. In fact, in the circuit design and circuit simulation, the circuit designer or user will accumulate some subjective and uncertain experience knowledge. If the subjective knowledge is combined with the objective data of circuit simulation for tolerance analysis, It will improve the efficiency of circuit design and shorten the R & D cycle of products. In order to achieve this goal, a new method of reliability tolerance analysis and model optimization is presented by combining subjective empirical information with circuit simulation data based on the confidence rule base model. The main work includes: (1) the design method of circuit tolerance based on confidence rule base (BRB). A BRB model describing the nonlinear mapping relationship between circuit component parameter values and circuit performance index values is established, and an initial solution of circuit tolerance based on BRB and an optimization method in tolerance domain are proposed in the case of disconnected and non-convex design domain. Through the examples of the Rosenbrock type circuit performance function and the actual track circuit tuning unit resonance performance function, it is verified that the proposed method is of relatively small computational cost in the case of non-connected and non-convex design domain. All components tolerance and center values can be accurately determined from the design domain. (2) the circuit center value design method based on confidence rule base. When the component tolerance in problem (1) is fixed, the circuit tolerance analysis is to find the optimal center value (nominal value). The reliability estimation model based on BRB is established and used to describe the nonlinear mapping relationship between the central value of circuit component parameters and the reliability of the circuit. Then, the center value design method based on BRB circuit is proposed. Compared with the traditional statistical analysis method, the method can not only find the single optimal center value, but also find a single optimal center value. The design domain of the whole center value that meets the reliability requirement can be determined more, thus providing multiple alternative center values for the designer; Finally, an example is given to illustrate the effectiveness of the proposed method. (3) the BRB model parameter optimization method based on sequential linear programming (SLP). Through the first order Taylor expansion of the objective function and constraint conditions of the BRB parameter optimization model in the first two problems, the nonlinear programming problem of parameter optimization is transformed into a simple linear programming problem. Then given the parameter optimization moving limit, the optimization interval is determined, the linear search is carried out, and the search result is used as a new base point, and the search result is re-linearized until the parameters of the BRB model satisfying the stopping criterion are found. Compared with other traditional nonlinear optimization methods, the linear programming problem is easy to solve and has higher accuracy, and the method can train and optimize the parameters from any point in the feasible solution space. Finally, the center value design of track circuit tuning element is used as an example to verify the superiority of this method in training the parameters of BRB model with high optimization efficiency.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN702

【相似文献】