In the Nuclear Security Enterprise (NSE), many high reliability components must be stored for long periods of time before being called on to function a single time. During dormant storage, changes in the performance of these components may occur due to environmental exposures. These exposures may enhance the natural degradation of materials or result in shifts in the performance of electronics. Ongoing assessment of these components is necessary to inform the need for upgrades or replacements to ensure high reliability requirements are being maintained. This paper presents several assessment methodologies that are used and have been proposed for this problem. We also present methods that we believe to be most appropriate for the assessment of nuclear weapons components subjected to dormant storage.
The Bernoulli CUSUM (BC) provides a moving window of process performance and is the quickest control chart to detect small increases in fraction defective. The Bernoulli CUSUM designs presented here require 2, 3, or 4 failures in a moving window to produce a signal. The run length distribution provides insight into the properties of the BC beyond the Average or Median Run length. A retrospective analysis of electronic component pass/fail data using the BC suggested that a problem may have been present during previous production. Subsequent production used the BC for real time process performance feedback.
This document outlines a statistical framework for establishing a shelf-life program for components whose performance is measured by the value of a continuous variable such as voltage or function time. The approach applies to both single measurement devices and repeated measurement devices, although additional process control charts may be useful in the case of repeated measurements. The approach is to choose a sample size that protects the margin associated with a particular variable over the life of the component. Deviations from expected performance of the measured variable are detected prior to the complete loss of margin. This ensures the reliability of the component over its lifetime.
With short production development times, there is an increased need to demonstrate product reliability relatively quickly with minimal testing. In such cases there may be few if any observed failures. Thus, it may be difficult to assess reliability using the traditional reliability test plans that measure only time (or cycles) to failure. For many components, degradation measures will contain important information about performance and reliability. These measures can be used to design a minimal test plan, in terms of number of units placed on test and duration of the test, necessary to demonstrate a reliability goal. Generally, the assumption is made that the error associated with a degradation measure follows a known distribution, usually normal, although in practice cases may arise where that assumption is not valid. In this paper, we examine such degradation measures, both simulated and real, and present non-parametric methods to demonstrate reliability and to develop reliability test plans for the future production of components with this form of degradation.
Proposed supplement I to the GUM outlines a 'propagation of distributions' approach to deriving the distribution of a measurand for any non-linear function and for any set of random inputs. The supplement's proposed Monte Carlo approach assumes that the distributions of the random inputs are known exactly. This implies that the sample sizes are effectively infinite. In this case, the mean of the measurand can be determined precisely using a large number of Monte Carlo simulations. In practice, however, the distributions of the inputs will rarely be known exactly, but must be estimated using possibly small samples. If these approximated distributions are treated as exact, the uncertainty in estimating the mean is not properly taken into account. In this paper, we propose a two-stage Monte Carlo procedure that explicitly takes into account the finite sample sizes used to estimate parameters of the input distributions. We will illustrate the approach with a case study involving the efficiency of a thermistor mount power sensor. The performance of the proposed approach will be compared to the standard GUM approach for finite samples using simple non-linear measurement equations. We will investigate performance in terms of coverage probabilities of derived confidence intervals.
Sandia is undergoing tremendous change. Sandia's executive management recognized the need for leadership development. About ten years ago the Business, Leadership, and Management Development department in partnership with executive management developed and implemented the organizational leadership Success Profile Competencies to help address some of the changes on the horizon such as workforce losses and lack of a skill set in the area of interpersonal skills. This study addresses the need for the Business, Leadership, and Management Development department to provide statistically sound data in two areas. One is to demonstrate that the organizational 360-degree success profile assessment tool has made a difference for leaders. A second area is to demonstrate the presence of high performing leaders at the Labs. The study utilized two tools to address these two areas. Study participants were made up of individuals who have solid data on Sandia's 360-degree success profile assessment tool. The second assessment tool was comprised of those leaders who participated in the Lockheed Martin Corporation Employee Preferences Survey. Statistical data supports the connection between leader indicators and the 360-degree assessment tool. The study also indicates the presence of high performing leaders at Sandia.
The design of experiments (DOEx) approach was used to characterize the Precision Laser Beam Welding Process with respect to four processing factors: Angle of Attack, Volts, Pulse Length, and Focus. The experiment was performed with Lap Joints, Nickel-Wire Joints, and Kovar-Wire Joints. The laser welding process and these types of welds are used in the manufacture of MC4368A Neutron Generators. For each weld type an individual optimal condition and operating window was identified. The widths of the operating windows that were identified by experimentation indicate that the laser weld process is very robust. This is highly desirable because it means that the quality of the resulting welds is not sensitive to the exact values of the processing factors within the operating windows. Statistical process control techniques can be used to ensure that the processing factors stay well within the operating window.