Reliability Estimation of Classification Algorithms. I. Introduction to the Problem. Point Frequency Estimates

The study deals with estimating the reliability of classification algorithms. The present article (the first of three) examines the statistical methods used for point estimation of the reliability of classifiers in the frequency approach. The estimation results are applicable to the small-sample case, which is of considerable importance in practice.     

Reliability Estimation of Classification Algorithms. II. Integral Estimates

This article concludes the examination of reliability estimates of classification algorithms. It reviews the statistical methods used for interval estimation of the reliability of classifiers in the frequency and Bayesian approaches. “Hybrid” estimates combining both approaches are also considered. These estimates are particularly important as they are applicable to the small-sample case.__________Translated from Prikladnaya Matematika i Informatika, No. 17, pp. 112 – 128, 2004.     

无失效数据情形可靠性参数的估计和调整

本文在无失效取样情形下,提出了产品可靠性参数的一种估计和调整的方法———加权多层Bayes估计法.在无失效数据情形下失效率的多层Bayes估计和引进失效信息后失效率的多层Bayes估计的基础上,对可靠性参数进行了估计和调整———给出了失效率和可靠度的加权多层Bayes估计.最后,结合发动机的实际问题进行了计算,结果表明本文提出的方法可行且便于应用.     

Reliability and covariance estimation of weighted k-out-of-n multi-state systems

In the literature of reliability engineering, reliability of the weighted k-out-of-n system can be calculated using component reliability based on the structure function. The calculation usually assumes that the true component reliability is completely known. However, this is not the case in practical applications. Instead, component reliability has to be estimated using empirical sample data. Uncertainty arises during this estimation process and propagates to the system level. This paper studies the propagation mechanism of estimation uncertainty through the universal generating function method. Equations of the complete solution including the unbiased system reliability estimator and the corresponding unbiased covariance estimator are derived. This is a unified approach. It can be applied to weighted k-out-of-n systems with multi-state components, to weighted k-out-of-n systems with binary components, and to simple series and parallel systems. It may also serve as building blocks to derive estimators of system reliability and uncertainty measures for more complicated systems.     

产品可靠度的E-Bayes估计

提出了参数估计的一种新方法——E-Bayes估计法.对Pascal分布,给出了可靠度的E-Bayes估计的定义(在先验分布中有一个超参数情形),在此基础上给出了可靠度的E-Bayes估计,并给出了可靠度的E-Bayes估计性质——E-Bayes估计和多层Bayes估计的关系.最后,给出了模拟算例,结果表明本文提出的E-Bayes估计法可行且便于工程应用.     

失效率的综合E-Bayes估计

该文提出了可靠性参数的一种新估计方法综合E-Bayes估计法.在无失效数据情形下给出了失效率的E-Bayes估计的定义,并给出了失效率的E-Bayes估计。在引进失效信息后，给出了失效率的E-Bayes估计，并在此基础上给出了失效率和其它参数的综合E-Bayes估计。最后，结合实际问题进行计算，结果表明该文提出的方法可行且便于应用。     

Profust reliability of a gracefully degradable system

Research in traditional reliability theory is based mainly on probist reliability, which uses a binary state assumption and classical reliability distributions. In the present paper the binary state assumption has been replaced by a fuzzy state assumption, thereby leading to profust reliability estimates of a powerloom plant, which is modelled as a two unit gracefully degradable system. Results of Bowles and Palaez [Application of fuzzy logic to reliability engineering, Proc. IEEE 83(3) (1995) 435–449] have been deduced as a particular case of results presented here. It is also recognized that estimation of system parameters such as failure rates, is vital in reliability estimation. Available methods for such estimation do not cover the underlying uncertainty in the failure data collection involving human judgment, evaluation and decision. In this paper we introduce a new approach based on fuzzy set theory to estimate such system parameters.     

E-Bayesian estimation of the reliability derived from Binomial distribution

This paper introduces a new parameter estimation method, named E-Bayesian estimation method, to estimate reliability derived from Binomial distribution. The definition of E-Bayesian estimation of the reliability is proposed, the formulas of E-Bayesian estimation and hierarchical Bayesian estimation of the reliability are also provided. Finally, it is shown, through a numerical example, that the new method is much simpler than hierarchical Bayesian estimation in practice.     

无失效数据情形失效率的综合E-Bayes估计

给出了无失效数据情形失效率的E-Bayes估计和引进失效信息后的E-Bayes估计,并在此基础上给出了失效率和可靠度的综合E-Bayes估计。最后,结合实际问题进行了计算,对估计结果进行了分析,并将它们与未假定先验信息时的非Bayes置信限进行比较,结果表明该方法是合理的。     

K/n(G)系统可靠性评定的多源验前信息融合方法

在多源验前信息的情况下,以k/n(G)系统为例,讨论了如何对系统的可靠性指标进行估计的问题.在多个验前信息源给出了系统可靠性指标点估计的情况下,利用多层Bayes方法及经验Bayes方法对这些数据进行融合,并给出系统可靠性指标的Bayes点估计,同时对系统可靠度的置信区间也作了讨论.仿真算例表明这种处理方法是合理有效的.     

AMSAA-BISE可靠性增长模型不能成立

阐述周源泉、翁朝曦提出的AMSAA-BISE模型存在的问题,提出一个近似计算模型,并给出数值例加以说明.     

参数的E Bayes估计法及其应用

提出了参数的一种估计方法—— E Bayes估计法 ,对寿命服从指数分布的产品 ,在失效率的先验分布为 Gamma分布时 ,给出了失效率的 E Bayes估计和多层 Bayes估计 ,并在此基础上给出了失效率和可靠度的 E Bayes估计的性质 .结合实际问题进行了计算 ,结果表明提出的 E Bayes估计法可行且便于应用 .     

二项分布可靠度E-Bayes估计的性质

作者以前提出了一种新的参数估计方法——E-Bayes估计法,对二项分布的可靠度,给出了E-Bayes估计的定义、E-Bayes估计和多层Bayes估计公式,但没有给出E-Bayes估计的性质.该文给出了二项分布可靠度F-Bayes估计的性质.     

基于测试数据的装备贮存可靠性评估

针对无失效数据情形下装备贮存可靠性估计问题,提出了一种利用性能测试数据进行估计的方法.首先利用测试数据估计装备在不同测试时的失效概率,然后利用配分布曲线法估计装备贮存寿命分布函数中的未知参数.由于方法充分利用了装备性能测试数据中所隐含的可靠性变化趋势,使其估计结果具有一定的可信性.     

Numerical method for estimating multivariate conditional distributions

Summary  A computational framework for estimation of multivariate conditional distributions is presented. It allows the forecast of the joint distribution of target variables in dependence on explaining variables. The concept can be applied to general distribution families such as stable or hyperbolic distributions. The estimation is based on the numerical minimization of the cross entropy, using the Multi-Level Single-Linkage global optimization method. Nonlinear dependencies of conditional parameters can be modeled with help of general functional approximators such as multi-layer perceptrons. In applications, the information about a complete distribution of forecasts can be used to quantify the reliability of the forecast or for decision support. This is illustrated on a case study concerning the spare parts demand forecast. The improvement of the forecast error due to using non-Gaussian distributions is presented in another case study concerning the truck sales forecast.     

威布尔分布或极值分布可靠度置信下限的回归表示

对于定数截尾样本，给出了基于极值分布的位置和尺度参数的最好线性无偏估计（BLUE），获得了威布尔分布的可靠度的点估计和置信限之间的回归模型，从而可由威布 尔可靠度的点估计根据回归方程得到可靠度的置信下限，省去了大量的用表，为实际工作者带来了极大的方便，计算结果表明，回归方程有很高的精度。     

一种真空荧光屏的可靠性分析

基于真空荧光屏(VFD)寿命服从对数正态分布的条件,本文在逐次定数截尾样本下研究了VFD的平均寿命、失效率和可靠度的点估计及其置信下限.并利用Monte-Carlo方法对估计结果的优劣进行了分析比较,验证了结论的正确性和可行性.     

A markov process model for software reliability analysis

Software reliability is a rapidly developing discipline. In this paper we model the fault-detecting processes by Markov processes with decreasing jump intensity. The intensity function is suggested to be a power function of the number of the remaining faults in the software. The models generalize the software reliability model suggested by Jelinski and Moranda (‘Software reliability research’, in W. Freiberger (ed.), Statistical Computer Performance Evaluation, Academic Press, New York, 1972. pp. 465–497). The main advantage of our models is that we do not use the assumption that all software faults correspond to the same failure rate. Preliminary studies suggest that a second-order power function is quite a good approximation. Statistical tests also indicate that this may be the case. Numerical results show that the estimation of the expected time to next failure is both reasonable and decreases relatively stably when the number of removed faults is increased.     

E-Bayes估计渐近性质的证明

随着科学和技术的进步需要有效地改进产品可靠性的评估方法,相关文献提出了参数估计的一种新方法——E-Bayes估计法.对Pascal分布,在两个超参数情形给出了分布参数的E-Bayes估计的定义、E-Bayes估计公式,并提出了E-Bayes估计的两个渐近性质,但没有给出证明.给出了这两个渐近性质的证明.     

A Wiener process model with truncated normal distribution for reliability analysis

The use of degradation model to perform the reliability analysis has drawn much attention due to the fact that the performance of numerous highly reliable systems degrades over time. To describe the unit-to-unit variability for a population of systems, the random effect has been incorporated into the degradation model that plays an important part in assessing the reliability of deteriorating systems. In the existing literature, the normal distribution is commonly adopted to represent the random effect, but the assumption can be unsuitable for some practical applications, such as the degradation process of train wheels. In this paper, we present a degradation modeling and reliability estimation approach by using truncated normal distribution to characterize the unit-to-unit variability. A Wiener process with truncated normal distribution is firstly applied to model the degradation process of the deteriorating system, and the analytical expressions of probability density function and reliability function are derived. Expectation maximization algorithm is then used to estimate the model parameters. The effectiveness and feasibility of the presented approach are illustrated through a numerical example and practical case studies for laser devices and train wheels. The results indicate that the presented approach can obtain better reliability estimation results by considering the truncated normal distribution when the unit-to-unit variability has significant difference.