A Practical Approach to System Reliability Growth Modeling and Improvement Om P. Yadav: Wayne State University: Detroit Nanua Singh: Wayne State University: Detroit Parveen S. Goel: TRW Automotive: Sterling Heights Key Words: Reliability improvement, Product development, Bayesian framework, Weibull distribution, Variance reduction. SUMMARY & CONCLUSION In a product development process, to develop appropriate design validation and verification program for reliability assessment, one has to understand the functional behavior of the system, role of components in achieving required functions and failure modes if component/sub-system fails to perform required function. The integration of these three issues will help design and reliability engineers in identifying weak spots in design and planning future actions and testing program. The existing system-level reliability predictions are generally developed based on a system model and component- level reliability prediction. These prediction methods are not of much help in pinpointing the exact location and nature of problem. Since time and budgetary constraints limit the extent of analysis and testing needed to estimate component reliability, it is necessary to utilize prior information available. System-reliability predictions should be updated iteratively as the design evolves and more information becomes available. In this paper, we propose a simple and practical two-stage approach of system reliability growth modeling considering components, functions, and failure modes. The consideration of these three dimensions will help in uncovering the weak spots in design responsible for low system reliability. The proposed method assumes Weibull distribution as failure time distribution and reliability model is based on the Bayesian framework (Ref 1) incorporating even fuzzy information (Ref. 2). The fuzzy logic model that has been developed for this purpose (Ref 3) is used to quantify the engineering judgment or fuzzy information of reliability improvement attributed to design changes or corrective actions. Uncertainty in data/ information at component levels propagates to system level reliability and makes system reliability prediction highly unreliable. The paper suggests a variance reduction strategy to give more accurate system reliability predictions. 1. INTRODUCTION The problem of demonstrating reliability growth of a company’s new product or system from testing is well known in manufacturing industry. The bogey testing philosophy has been widely applied in the automotive industry for reliability demonstration purposes and to evaluate the design early in the product development process. For bogey testing, observing failure and failure mechanism before bogey value is reached is usually not possible and hence does not give any insight about failure behavior or weakest link in the system. Reliability assessment based on bogey tests does not help in reliability improvement unless it takes us back to the system and guides in finding the critical spots and causes of criticality. The literature is full of claims that early reliability predictions are useful in highlighting potential weak spots in the design process, evaluating the effects of changes, and for planning of life testing and stress testing/screening activities (Ref. 4). Design verification and testing plan (sample size determination) models for bogey test (Ref. 5-7) do not give the sample size requirements for all test types to demonstrate reliability for each failure behavior or mechanism. These models do not consider the prior information or existing reliability level in determining the required sample size to demonstrate the reliability goal. Essentially, what is needed is a shift from reliability accounting tasks to reliability engineering analysis, the ability to decompose the predicted system reliability and allocate it to all elements of three dimensions (i.e. physical structure, functions, and time) and their interfaces to help in identifying weak spots (lower than target reliability) and nature of the problem. The paper proposes a two-stage framework, which considers all three dimensions of the product design to develop a design verification and reliability improvement plan. At the first stage, the goal is to develop a reliability growth and improvement plan to achieve reliability targets. In order to achieve that the system is decomposed into its components, functions, and damage behavior influencing the performance over a period of time. Since the damage behavior of any system aggravates with usage time, therefore, the time dimension of product design is captured through damage behavior or failure mechanisms of the system. The estimated system reliability is decomposed or allocated to each element of the system using criticality index. The criticality index for each element is calculated using criticality number or risk priority numbers (RPN) from failure modes and effects analysis (FMEA) document. The current reliability estimates and reliability targets are used to develop the design verification and testing plan for the next stage of design process. The second stage of the framework proposes a strategy to further improve the system reliability prediction once 351 2003 PROCEEDINGS Annual RELIABILITY AND MAINTAINABILITY Symposium 0-7803-7717-6/03/$17.00 © 2003 IEEE