Short answer
FMEA failure mode libraries are structured records, usually spreadsheets or database exports held by manufacturing quality teams, that list functions, failure modes, effects, causes, controls and risk ratings before or alongside real failures. They support defect reasoning and root-cause models. They contain no images unless linked, are heavily confidential, and cannot train robot control.
What an FMEA is, and why a buyer would want a library of them
Failure modes and effects analysis is a structured way to ask, before a product or process fails, what could go wrong, what would happen, why, and how it would be caught. IEC 60812:2018 describes how FMEA, including its criticality-oriented variant FMECA, is planned, performed, documented and maintained, and states that it applies to hardware, software, processes including human action, and their interfaces (IEC, 2018). The standard is generic and gives no safety-specific guidance on its own, which is one reason worksheets differ so much between companies.
For a buyer building defect-reasoning or root-cause models, FMEAs hold something defect images do not: expert knowledge of failure mechanisms stated before the failure. A single worksheet row ties a failure mode to its effects, its causes and the controls that detect it. Across many products this becomes a library of causal hypotheses, with the caveat that hypotheses are not outcomes.
Typical contents of an FMEA record
| Field group | What it holds | Notes for data use |
|---|---|---|
| Scope | Item, function or process step under analysis | Naming is free text and varies |
| Failure mode | How the function can fail | Often the most direct handle for taxonomy mapping |
| Effects | Consequence at local, next and end level | Often long free text |
| Causes | Mechanisms or conditions that produce the mode | May combine several in one cell |
| Current controls | Prevention and detection measures | Links to inspection steps |
| Ratings | Severity, occurrence, detection, and a priority measure | Scales differ by organization |
| Actions | Recommended and taken, with owner and date | Sparse follow-through is common |
IEC 60812:2018 lists alternatives to the traditional risk priority number among its informative content, so a buyer should not assume all libraries use the same priority arithmetic. Ask which scheme each worksheet applied and when it changed.
Design FMEAs (concerning product design) and process FMEAs (concerning manufacturing steps) are different in vocabulary. For inspection-model purposes, process FMEAs and the controls column point most directly at what inspection stations look for.
What these records can support
- Taxonomy bootstrapping. Failure modes across products suggest classes and synonyms for defect taxonomies.
- Cause and effect reasoning. Pairing symptoms with plausible causes can ground question-answering or root-cause suggestion, in the spirit of failure reasoning data.
- Control mapping. The detection column maps failure modes to inspection methods, which helps decide which image modalities to source.
- Prior knowledge for evaluation. Experts can compare a model's proposed causes with the listed ones.
These are possible uses to test, not demonstrated results. FMEA content is judgment recorded by teams, and is only as good as their experience and diligence.
Quality problems to check
Boilerplate rows. Many worksheets are cloned from earlier products and lightly edited. A large library may contain far fewer distinct ideas than it has rows.
Rating drift. Severity and occurrence ratings are subjective and calibrated differently across teams and years; they are weak labels at best.
Staleness. A worksheet written at launch may never be updated after real failures; compare with nonconformance reports and corrective action records to see whether reality reached the document.
Structure loss. A cell that says "cracks / porosity / short shot" holds three modes. Parsing is needed before the content works as structured data.
Missing links. FMEAs rarely point to images, measurements or part serials, so they stay isolated unless linkage is added.
Illustrative schema after parsing
# Illustrative only - synthetic content and ids
library: fmea_records_v0
source_type: process_fmea
record:
record_id: fm_000142
process_step: "step_name_redacted"
function: "hold seal alignment"
failure_mode: "misalignment"
effects:
local: "seal does not close"
end: "leak at customer"
causes: ["fixture wear", "operator placement variance"]
current_controls:
prevention: ["fixture check at shift start"]
detection: ["inline vision check"]
ratings: {severity: 8, occurrence: 4, detection: 3, scheme: "supplier-defined"}
actions: [{text: "replace fixture", status: closed}]
links: {defect_image_set: null, nonconformance_refs: []}
revision: {date: 2024-xx-xx, author_role: quality_engineer}
excluded: customer names, supplier names, part numbers (as agreed)
Every rating and text value above is synthetic. A real package would be defined by what the supplier can release and what the buyer needs, including whether to keep any rating at all.
What FMEA libraries cannot support
They cannot supply images, measurements or outcomes unless those are attached; the cleanest FMEA does not tell you what the defect looks like or how often it occurred. They are not evidence of actual failure rates, because occurrence ratings are estimates; observed rates sit in warranty and field failure data. They are not control data: nothing in a worksheet records robot joint states, actions or forces, and a humanoid cannot learn physical control from failure descriptions; that needs robot or teleoperation data. See failure taxonomies for embodied tasks for how robot-task failure categories are built from different evidence.
They also reflect a company's products and processes, so they transfer unevenly to other sites. And since the standard is process guidance, not a data format, there is no common schema to rely on.
Supplier types
Manufacturers' quality and reliability engineering teams hold the working files, often on shared drives, in quality-management software or in customer-portal exports. Tier suppliers also hold FMEAs they prepared for OEM customers, and those frequently belong to, or are restricted by, the customer. Contract manufacturers hold FMEAs for client products. Consulting and standards bodies may hold anonymized examples; check whether any redistribution right exists.
Questions to ask suppliers
- What FMEA types are held (design, process, other), and across how many products or lines?
- Which template and rating scheme are used, and did they change over time?
- Are records in structured databases or free-form spreadsheets, and how consistent are the columns?
- Have worksheets been revised after real failures, and is the revision history kept?
- Can failure modes be linked to nonconformance reports, defect images or part identifiers?
- Who owns each worksheet, particularly those prepared for or by customers and suppliers?
- Which fields hold customer names, supplier names, part numbers or process parameters that would need redacting?
- Were any worksheets prepared under a nondisclosure agreement?
- Could a subject-matter reviewer validate a sample of rows before purchase?
Permissions and provenance
FMEAs reveal how a company makes things, how it controls them and where it believes it is weak, which is commercially sensitive. They often name customers and suppliers, and some are contractually shared between parties. Ask who may license each file and what agreements apply. This is not legal advice. Stripping names and part numbers does not by itself resolve confidentiality or ownership questions, and process descriptions may still identify a company.
Discuss your requirements
If you want structured failure knowledge, describe the product areas, FMEA types, structure and permitted uses to SourceX. SourceX reviews the requirement and follows up about potential sources among US businesses; a request does not guarantee a match. Continue with the inspection data hub, the humanoid training data guide, or the catalog entry for manufacturing quality records.
Discuss your humanoid training-data requirements with SourceX
Sources
- IEC 60812:2018 Failure modes and effects analysis (FMEA and FMECA) — what the standard covers, its generic scope and tailoring guidance.