Research On Reliability Assessment Method For Long-life Products With Zero-failure Data

The case of zero-failure data refers to the no failure of the sample within the specified test time(because the exact life data cannot be known),which belongs to a special case of timing truncation test in reliability test.Earlier,zero-failure data were often rejected as abnormal data or simply treated conservatively.For high-reliability and long-life products,the cost of obtaining failure data through experiments is very high.With the wide application of high-quality long-life products in aerospace,marine engineering,military,communication engineering and power engineering,more and more zero-failure data will be encountered in time-limited tail-off test.Therefore,it is of great practical significance to conduct a reasonable:statistical analysis of the test results without failure data and to establish an effective productreliability assessment model.It is still a difficult problem to establish a reliability evaluation model which can not only meet the requirements of test time and cost,but also obtain reasonable life prediction results.It is necessary to use zero-failure data to evaluate the reliability of products.In this paper,the key technologies in the process of reliability evaluation and modeling of high-quality and long-life products without failure data are explored,and the methods of parameter estimation of product life distribution,reliability modeling,reliability evaluation and life prediction are studied in depth and systematically.The specific research content includes:(1)Reliability evaluation using samples with long truncation time will get relatively"rash advance" estimation results through confidence limit analysis method under the condition that the sample size of zero-failure data has been determined.In order to correct the result that the reliability index is relatively high,a modified one-sided confidence limit evaluation method is proposed by introducing failure information into confidence limit analysis.By using the new method to evaluate the reliability of product deployment,the reliability and MTBF(Mean Time Between Failure)estimates of a certain type of torque motor are obtained under the condition of exponential distribution.By comparing it with other unilateral confidence limit methods,it is proved that the new method can effectively prevent the relative "rash advance" phenomenon in reliability evaluation and life prediction when a few samples have a long Tail-cutting time.At the same time,the sensitivity of reliability index to parameter change is analyzed,and the calculation results of reliability and MTBF sensitivity to failure rate are given.(2)The problem of inaccurate results and sometimes deviating from the real situation is the reliability evaluation using zero-failure data confidence limit analysis method.The limitations of one-sided confidence limit method and optimal confidence limit method in zero-failure data processing are pointed out.The reason why the exact shape parameter information of product life distribution can improve the accuracy is analyzed.Two kinds of confidence limit methods are compared and studied,and some improvements are made.The optimal confidence limit method,which can only deal with single group of zero-failure data,is extended to deal with grouped zero-failure data.The feasibility of the extended method is discussed and the concrete implementation process is given.The example analysis shows that introducing the exact information of shape parameters of product life distribution can effectively improve the accuracy of evaluation results under Weibull distribution conditions.(3)The point estimation and interval estimation of parameters are obtained by using different methods in the reliability evaluation process of zero-failure data products,which will result in inconsistency of results.A new model for point estimation and confidence interval estimation of reliability parameters without reducing the credibility of reliability evaluation results is proposed.In the new model,Bootstrap method is applied to zero-failure data distribution curve method,so that distribution curve method can be extended from point estimation of product reliability to point estimation of reliability and confidence interval estimation.The comparison between the new model and other confidence interval estimation models proves that the new model has higher precision of parameter interval estimation.The results of case study show that under Weibull distribution,the new model can simultaneously obtain the point estimation and confidence interval estimation of reliability parameters,and improve the accuracy of reliability interval estimation.(4)In the process of reliability evaluation of zero-failure data,there is a problem that the reliability of products will be obviously underestimated based on two-parameter Weibull distribution condition.The reliability evaluation of zero-failure data of high-quality long-life products is proposed by introducing three-parameter Weibull distribution.In the three-parameter Weibull distribution,it is easy to obtain the estimation of scale and shape parameters by the matching distribution curve method,and the estimation of position parameters is relatively difficult.At present,the location parameter estimation method in Weibull distribution is only applicable to failure data analysis.In order to obtain the location parameter estimation from zero-failure data,the relationship between sample size of zero-failure data and unknown parameters is given based on the median rank estimation of failure probability,and the location parameter estimation method is given.The reliability estimates of products under Weibull distribution with two and three parameters are compared based on zero-failure data of a certain type of container hull.The results show that the introduction of three-parameter Weibull distribution into the reliability evaluation of zero-failure data can effectively improve the reliability evaluation of zero-failure data products and prevent the problem of underestimation of product reliability.