Failure Data Sources

Recently some analysis organizations have compiled comprehensive failure data source books and computer databases. The information is formatted to give failure rate as a function of failure mode. Often additional information about the product, such as Type A versus Type B, is provided. Some example pages are shown in Figures 8-1,8-2 and 8-3 (Ref. 9). 

Reliability models and assumptions, for example down times and proof test intervals Failure data sources and reliability calculations Findings of the qualitative assessment of life-cycle activities A demonstration of rigor such as is described in Appendix 2 of this book Appropriate independence. 

Failure data sources Reliability calculations Findings of the qualitative assessment... 

QRA practitioners can use to satisfy some QRA objectives. Also, the American Institute of Chemical Engineers (AIChE) has sponsored a project to expand and improve the quality of component failure data for chemical industry use. And many process facilities have considerable equipment operating experience in maintenance files, operating logs, and the minds of operators and maintenance personnel. These data can be collected and combined with industrywide data to help achieve reasonable QRA objectives. However, care must be exercised to select data most representative of your specific system from the wide range available from various sources. Even data from your own plant may have to be modified (sometimes by a factor of 10 or more) to reflect your plant s current operating environment and maintenance practices. 

In doing this, a failure count obtained from different sources may not indicate clearly which I allures occurred during a test. This over-counting results in a conservative estimate. In addition to lest tiemands, the number of operational demands must be obtained which may be difficult to extract from available data sources. 

FAILURE SYMBOL FAILURE DESCRIPTION ESTIMATED RROBABILITY DATA SOURCE... 

Number and type of record The number of data points or tables of data presented in the resource or the number of events the data set reflects where available, the form in which the data are presented, such as failure rates or availability data, confidence intervals or error factors the raw data source used, sueh as surveys, plant records, tests, or judgment. 

Unknown number of records coded by oomponent, failure mode, and data source... 

The Department of Industrial Safety has been collecting and recording component failure data since 1978. For this purpose use is made of the following sources of information accessible (international) data-banks, literature and data... 

As failure data relating to mechanical components differ widely from source to source, TNO has set up a documentation system in which all relevant information is stored in one, uniform automated code called COMPI, which uses a component description code for the following information system of... 

Limited failure data from one plant and published data source failure rales... 

Only failures within the boundary indicated with the dash line in the Figure below are included in the reliability data source. 

Statistical Methods for Nonelectronic Reliability, Reliability Specifications, Special Application Methods for Reliability Prediction Part Failure Characteristics, and Reliability Demonstration Tests. Data is located in section 5.0 on Part Failure Characteristics. This section describes the results of the statistical analyses of failure data from more than 250 distinct nonelectronic parts collected from recent commercial and military projects. This data was collected in-house (from operations and maintenance reports) and from industry wide sources. Tables, alphabetized by part class/ part type, are presented for easy reference to part failure rates assuminng that the part lives are exponentially distributed (as in previous editions of this notebook, the majority of data available included total operating time, and total number of failures only). For parts for which the actual life times for each part under test were included in the database, further tables are presented which describe the results of testing the fit of the exponential and Weibull distributions. 

The LER data base served as the primary source of DG failure data, while a data base for DG successes was formed from nuclear plant licensees responses to a USNRC questionnaire (Generic Letter 84-15). Estimates of DG failure on demand were calculated from the LER data, DG test data, and response data from the questionnaire. The questionnaire also provided data on DG performance during complete and partial LOSP and in response to safety injection actuation signals. Trends in DG performance are profiled. The effects of testing schedules on diesel reliability are assessed. Individual failures are identified in an appendix. 

This report evaluates recent performance of DGs and all DG vendors with the exception of Transamerica Delaval, Inc. (TDl), because of the emphasis already being given to TDI diesels in other studies. For the period 1980 through 1983 inclusive, BNL reviewed and evaluated DG failure data, DG vendor inspection reports, the TDI lessons learned as they related to the other vendors, and previous pertinent studies. The data sources used for DG failure analysis were LERs, 10 CFR 50.55E, Part 21, NPRDS, and EPRI document files. The DG failures were classified relative to the DG component that failed (e.g., main bearings, starting system). The failures were also categorized and analyzed by mode, manufacturer, and cause. Manufacturers with significant failures are identified in the report. 

The failure rates and times-to-restore developed used a variety of data sources and data construction methodologies and are presented in Section 2. The principal methodology used is a kind of failure mode analysis for each component several principle modes of failure are analyed by characteristics including frequency of occurence, repair time, start-up time, and shut-down time. From these an average failure rate is developed and expressed as failures per million hours and mean time between failure(MTBF). 

Failure data from eight PWRs and nine BWRs for 1972 and from varied non-nuclear industry sources... 

Appendix III of this report provides a detailed description of the reliability data used in event tree and fault tree quantification. Because of its extensive operating experience and the uniqueness of the BRP design, BRP plant-specific data was used whenever possible. Plant-specific data sources included plant maintenance orders, control room log books, surveillance tests, LERs, event reports, deviation reports, plant review committee meeting minutes, and USNRC correspondence. The plant-specific data used spanned the period from 1970 to 1979. Data before 1970 did not include maintenance orders or surveillance tests and therefore were excluded. The plant-specific data collected for BRP is presented in detail in Appendix XIII. Table III-4 summarizes 30 plant-specific component failure rates and Table 11-06 contains plant-specific maintenance unavailabilities for 20 components. These tables are a summary of the BRP component failure and maintenance outages. 

Opening segments of the IP2 PRA data analysis section describe the definitions of terms and concepts employed, the assumptions made, and limitations recognized during the data base construction. A set of 39 plant-specific component failure mode summaries established the basis for component service hour determinations, the number of failures, and the test data source for each failure mode given for each component. Generic data from WASH-1400, IEEE Std 500, and the LER data summaries on valves, pumps, and diesels were combined with plant-specific failure data to produce "updated" failure information. All the IP2 specialized component hardware failure data, both generic and updated, are contained in Table 1.5.1-4 (IP3 1.6.1-4). This table contains (by system, component, and failure mode) plant-specific data on the number of failures and service hours or demands. For some components, it was determined that specifications of the system was warranted because of its impact on the data values. 

Marti, H. F. and R. A. Waller. An Exploratory Comparison of Methods for Combining Failure Rate Data from Different Data Sources. Report No. LA-7556-MS, Los Alamos Scientific Laboratory, 1987. 

It is important to distinguish the types of time-related and demand-related equipment failure rate data that can be found and used in afire risk assessment. Basically, four types of data and corresponding data sources provide both time-related and demand-related failure data as shown in Table 6-3 and discussed in the following sections. 

Cynthia is using a secondary data source to conduct her cost-benefit analysis. After she has identified her variables of interest, she asks the HMO to provide her with baseline information on all heart failure patients prior to implementing her service. The HMO is able to stratify the data by diagnosis, so Cynthiaasks for annual numbers and costs of hospitalizations, emergency room visits, and medications for their heart failure patients (see Table 27-3). 

Table 1.2 lists major data sources by property for both pure components and mixtures. It is also useful to organize the information by database that is the purpose of this section. We will describe a number of major databases and compilations and how they may be accessed. It is impossible for such a list to be complete we have tried to list the most comprehensive sources we are aware of. Sources that are confined to one or two properties can be found in the section corresponding to that property (and in Table 1.2). Failure to mention a particular data source should not be taken as an indication that the source is not valuable.

Comparable wood failure data (hard maple) for resorcinol-formaldehyde and epoxy resins are 92 2 and 30%, respectively (from same sources as given under footnote 5). 

In spite of their limitations, industry databases can be extremely valuable especially when no other data source exists. If the failure rate data is too high, the result will be a higher PFH/PFDavg. If this occurs and too much safety integrity is designed into a safety instrumented function, that is tolerable. 

Many companies have an internal expert who has studied these sources, as well as their own internal failure records, and maintains the company failure rate database. Some use failure data compilations found on the Internet. While the data in industry databases is not product specific or application specific, it does provide useful failure rate information for specific industries (nuclear, offshore, etc.) and a comparison of the data provides information about failure rates versus stress factors. 

Although product specific FMEDA reports offer superior data sources when compared to industry databases, they still do not account for application specific stress conditions that may affect actual failure rates. Ideally in the future manufacturers will be able to provide not only point estimates of failure rates but perhaps even equations with application specific variables to more precisely calculate the needed numbers. That wiU happen if there is demand and the needed data is collected. 

Quantitative A failure rate is assigned to each failure mode. Whenever possible, failure rates should (ARP4761 para G.3.2) be determined from historical (i.e. in-service) failure data of similar (but preferably identical) equipment already in field use. These must be justified by including accumulated flight hours, number of occurred failures and justification of similarity. Alternatively, industry sources of failure modes and failure rates include ... 

Table 11.3 lists some data (average values) used in the study NUREG 1150 (NUREG, 1987). Other sources of failure data are described in Fullwood (1999), Taylor (1994) and Smith (1997), as well as many other sources. 

For years, simulator data collection has been seen and proposed as rich and good additional data source helping in HRA quantitative part. Although simulator exercises observation can help significantly in qualitative analysis of some factors influenciDg failure potential (team work), the idea of grounding HEP quantification just on simulator data is fairly naive, because the statistics is clearly insufficient to produce direct estimates and because of common simulator fidelity problems. 

---