As part of the U.S. Department of Energy's Hydrogen, Fuel Cells & Infrastructure Technologies Program, Sandia National Laboratories is developing the technical basis for assessing the safety of hydrogen-based systems for use in the development/modification of relevant codes and standards.  Sandia is developing benchmark experiments and a defensible analysis strategy for risk and consequence assessment of unintended releases from hydrogen systems. This work includes the performance of quantitative risk assessments (QRA) of hydrogen facilities.   The QRAs are used to identify and quantify scenarios for the accidental release of hydrogen and to identify the significant risk contributors at different types of hydrogen facilities.  The results of the QRAs are envisioned as one input into a risk-informed codes and standards development process that identifies a minimum set of facility design and operational requirements needed to establish an acceptable level of risk.   

A key input into a QRA is the data required to quantify the frequency of potential accident scenarios that involve the release of hydrogen and subsequent ignition.  The data required includes initiating event frequencies (e.g., the frequency of leaks from different components), mitigating component failure probabilities (e.g., hydrogen detectors), the probability of human errors that may lead to or exacerbate a hydrogen release, and the probability of important phenomenological events such as the potential for auto-ignition of a hydrogen jet. Unfortunately, data specific to hydrogen facilities are currently not readily available; thus, other sources of data must be utilized in the QRAs of hydrogen facilities. 

This paper describes one effort to establish the data needed for the performance of QRAs of hydrogen facilities.  Generic data sources, including those specific to compressed gas, were surveyed and a range of values were obtained for the required parameters.  Bayesian methods have been utilized to merge the data from these different sources.  In addition, limited hydrogen-specific information was obtained from the literature and from industry sources.  Two different methods were used to generate parameter values from this hydrogen-specific data.  In the first method, classical statistical approaches were used to estimate parameter values utilizing only the hydrogen-specific data.  In the second approach, Bayesian methods were utilized to merge the generic and hydrogen-specific information to obtain parameter values.  As more hydrogen data become available, the parameter estimates can be updated such that they will reflect true hydrogen-specific failure probabilities.  The resulting values will not only support the QRAs of hydrogen facilities but will also identify the components that are the major contributors to facility outages.