Session

Weekend Session 6: Advanced Concepts - Research & Academia III

Location

Utah State University, Logan, UT

Abstract

When constructing a system-level fault tree to demonstrate device-to-system level radiation degradation, reliability engineers need relevant, device-level failure probabilities to incorporate into reliability models. Deriving probabilities from testing can be expensive and time-consuming, especially if the system is complex. This methodology offers an alternative means of deriving device-level failure probabilities. It uses Bayesian analysis to establish links between historical radiation datasets and failure probabilities. A demonstration system for this methodology is provided, which is a TID response of a linear voltage regulator at 100 krad(SiO2). Data fed into the Bayesian model is derived from literature on the components found within a linear voltage regulator. An example is presented with data pertaining to the device’s bipolar junction transistor (BJT)’s gain degradation factor (GDF). Kernel density estimation is used to provide insight into the dataset’s general distribution shape. This guides the engineer into picking the appropriate distribution for device-level Bayesian analysis. Failure probabilities generated from the Bayesian analysis are incorporated into a LTspice model to derive a system failure probability (using Monte Carlo) of the regulator’s output. In our demonstration system, a 96.5% likelihood of system degradation was found in the assumed environment.

Share

COinS
 
Aug 7th, 11:00 AM

Methodology for Correlating Historical Degradation Data to Radiation-Induced Degradation System Effects in Small Satellites

Utah State University, Logan, UT

When constructing a system-level fault tree to demonstrate device-to-system level radiation degradation, reliability engineers need relevant, device-level failure probabilities to incorporate into reliability models. Deriving probabilities from testing can be expensive and time-consuming, especially if the system is complex. This methodology offers an alternative means of deriving device-level failure probabilities. It uses Bayesian analysis to establish links between historical radiation datasets and failure probabilities. A demonstration system for this methodology is provided, which is a TID response of a linear voltage regulator at 100 krad(SiO2). Data fed into the Bayesian model is derived from literature on the components found within a linear voltage regulator. An example is presented with data pertaining to the device’s bipolar junction transistor (BJT)’s gain degradation factor (GDF). Kernel density estimation is used to provide insight into the dataset’s general distribution shape. This guides the engineer into picking the appropriate distribution for device-level Bayesian analysis. Failure probabilities generated from the Bayesian analysis are incorporated into a LTspice model to derive a system failure probability (using Monte Carlo) of the regulator’s output. In our demonstration system, a 96.5% likelihood of system degradation was found in the assumed environment.