What is Surrogate Reliability Data?

Surrogate reliability data is summarized data from field testing that is not performed in-house. Often, the data will estimate the failures under specified operating hours/miles under certain conditions for particular components that are either the exact match or a similar part an engineer must assess to perform a reliability analysis.

This article is a quick summary of what to consider when making a decision regarding the utility/applicability of Quanterion’s surrogate data sets, including: NPRD, EPRD, FMD and the online version called ROADS. The end of the article discusses how to freely determine the applicability of these data sets to your current analysis. Additional industry-specific data sets are mentioned as well.
 

The Benefits of Surrogate Reliability Data

Engineers may choose to avoid surrogate data as it removes a semblance of control over the conditions tested to generate the failure data. However, the limited control is also a strength as it reduces any overconfidence in data generated solely in-house under very specific and limited conditions. Surrogate data, in contrast, averages field data from a variety of operating conditions and environments, which can provide a more accurate summary of failure data industry-wide. Additional benefits include:

• Reduced labor costs from performing numerous tests versus simply analyzing surrogate data already generated
• Reduced time to perform reliability analyses (compared to data generated in-house using lifetime testing)
• Faster reliability analyses starting with a set of data already generated

The Reliability Analyst’s Quantification Dilemma

The question of how to quantify the failure rate of a component comes up often when performing a reliability analysis. The best method to quantify the failure rate of a component is with field data or service data for the component in question from your own service records. Many organizations do not have reliable (pun intended) field data or service data.

If field data or service data is not used or available, the analyst is left with choices including:

• Physics of Failure (POF) Analysis
• Empirical Modeling (such as MIL-HDBK-217, NSWC-2011, 217Plus^TM, et al.)
• Life testing (HALT, HASS, etc.)

OR

• Surrogate Data

Surrogate Data: Good? Fast? Cheap?

Let’s view through the “Good, Fast or Cheap” lens; comparing surrogate data to the other three choices listed above.

• Good: Generally, of the above choices, surrogate data is the lowest quality. It is a truly rare circumstance where the available data is taken from the exact same part used in the exact same environment being considered. Also, surrogate data is generally “Interval Data,” meaning that it comes from a fleet of plants, vehicles and/or assets being maintained in their useful lifetime. In reliability, the non-unity Weibull shape parameter that helps determine Time TO Failure for an individual part is assumed to be randomized by repair and replacement within the fleet. Most surrogate data comes from annual maintenance, so it tells the analyst how often a part fails from one year to the next, but the information regarding when the part was initially fielded is lost. As such, surrogate data is generally considered good to estimate Time BETWEEN Failures.

• Fast: Surrogate data is extremely fast. Essentially, if there is surrogate data available for the component in question, the data is available as quickly as there is access to it and it can be looked up. The caveat here is whether the data is available or not.

• Cheap: Surrogate data can be extremely inexpensive. For example, an annual subscription to the ROADS reliability field failure rate and field failure mode data containing 3.7M data records can be accessed immediately for $800. Unless an organization has already done a POF analysis, Empirical Model or Life Testing for the components in question, the annual subscription is generally far less expensive than the cost of the tools, time spent in the learning curve and engineering labor required to do such an analysis.

• Wildcards: Working in an industry for which there is dedicated industry-specific surrogate data changes the “goodness.” For instance, the oil and gas industry has the OREDA database. The nuclear power industry has INL, NUREG and IAEA databases. Industry-specific surrogate data can be very good and applicable, nearly as good as field data. If an organization has already invested in Empirical Models or a Reliability Analysis software suite, surrogate data may have already been built into the model or it may already be built into the Reliability Analysis software as a lookup. In these situations, the additional costs to buy access to a surrogate data database may not be justified.

Is applicable data available? Use the part descriptors!

The ROADS reliability field failure rate and field failure mode database contains the following datasets: Nonelectronic Parts Reliability Data (NPRD, data on mechanical and electromechanical parts and assemblies), Electronic Parts Reliability Data (data on electronic components and assemblies), and Failure Mode / Mechanism Distributions (FMD, field failure mode and mechanism distribution data on a variety of electrical, mechanical, and electromechanical parts and assemblies).

Engineers can view all the part types included in NPRD, EPRD and FMD through the Parts Descriptors Lists available at no cost to help engineers determine if the datasets include data on an exact or similar part.

• • Nonelectronic Parts Reliability Data (NPRD-2023)
    NPRD-2023 Introduction
    NPRD-2023 Parts Descriptors List

• • Electronic Parts Reliability Data (EPRD-2024), Quanterion’s most recent dataset, Spec Sheet here
    EPRD-2024 Introduction
    EPRD-2024 Part Descriptors List

• • Failure Mode / Mechanism Distributions (FMD-2016)
    FMD-2016 Introduction
    FMD-2016 Part Descriptors List

Create a Quick Spreadsheet

Take a Bill of Material for your design, and “bounce” it against the part descriptor files above, using <CTRL> <F>. It can be assumed that there will be data available for all but the most novel electrical components in EPRD.

The example below is for a hypothetical collection of ten mechatronic components, which would be a mix of electrical and mechanical items.

I gave myself less than 30 minutes to do this and completed in about 20 minutes.

Cable Assemblies and “None Exact”

Wiring harnesses and cable assemblies are troublesome, due to “death by a thousand choices.” The reader will see “None exact” listed below for the motor wiring harness. This may, or may not be a dead end.