A FMEA manufacturing document listed 'label misapplication' as a failure mode with a Detection score of 3, reflecting the team's confidence in their end-of-line visual check twice per shift.
The label error that caused the recall occurred at 2:47 AM on a Wednesday, between those checks. The FMEA manufacturing process did not fail to identify the failure mode. It failed to account for the gap between when the failure occurred and when the detection control would find it.
This guide explains how that gap closes.
FMEA in manufacturing is a structured method for identifying potential failure modes in a product or process, scoring each by Severity, Occurrence, and Detection, and prioritizing corrective action based on those scores. The risk priority number calculation (S x O x D) determines where resources go. AI vision systems close the gap static FMEA leaves open: they replace estimated Detection scores with real inspection data, and feed live Occurrence rates back into the risk register continuously.
What Is FMEA in Manufacturing and Why Does the Classic RPN Method Fall Short?
FMEA manufacturing methodology identifies every failure mode a product or process could produce, scores each by Severity (S), Occurrence (O), and Detection (D), and ranks them by risk priority number calculation (S x O x D = RPN). The FMEA then drives corrective actions, preventing failures from reaching the customer or triggering production shutdowns.
FMEA originated in US military failure analysis manufacturing practice in the 1940s and was adopted by automotive (IATF 16949), aerospace, medical devices (ISO 13485), and general manufacturing as a mandatory or recommended quality discipline (ASQ FMEA methodology overview). The methodology is sound. The scoring system has a known flaw.
The AIAG and VDA FMEA Handbook introduced an Action Priority (AP) system to replace the flat RPN threshold precisely because an RPN of 90 with S=9, O=2, D=5 (a safety-critical failure) is more important than an RPN of 120 with S=2, O=6, D=10 (a cosmetic issue). The classic formula scores the cosmetic issue higher. The standard's own governing body acknowledged the problem and rewrote the guidance. (Symestic FMEA Manufacturing Guide, April 2026)
The Three Components That Build Every RPN Score
Every risk priority number calculation is the product of three independent ratings, each scored 1 to 10, each carrying a different assumption about what the production team can control.
1. Severity (S)
Severity measures the impact of the failure on the customer or downstream process if it occurs, not how likely the failure is. Rating anchors run from 1 (no effect) through 5 to 6 (degraded product function) to 9 to 10 (safety impact or regulatory violation).
The critical point: Severity cannot be reduced by adding more inspection. A high Severity score requires an engineering change to the product design or process. Adding a camera at the end of the line does not lower the impact of the failure if it occurs. Only removing the root cause does.
2. Occurrence (O)
Occurrence scores how frequently the failure mode occurs under current prevention controls. A score of 1 means the failure is so rare it is considered unlikely to occur at all. A score of 10 means the failure occurs repeatedly and current prevention controls are not functioning. Occurrence is reduced by implementing preventive actions, not detection controls.
The structural problem with Occurrence scoring in most facilities: the score is an estimate based on team experience, not derived from a defect frequency log. When KOMPASS generates a classified inspection log for every production cycle, the Occurrence score becomes a data-derived figure rather than an educated guess.
ML-integrated FMEA manufacturing approaches have achieved R squared values of 0.985 in RPN prediction accuracy when trained on structured inspection data (ScienceDirect, January 2026).
3. Detection (D)
Detection scores the likelihood that current controls will find the failure before it reaches the customer. A score of 1 means the failure is almost certain to be detected. A score of 10 means the current controls cannot detect it at all.
The 2:47 AM scenario from this blog's opening is a Detection score problem. The team rated Detection at 3, reflecting confidence in their end-of-line check. That confidence was valid for the check that happened.
It was not valid for the production that ran between checks. A manual inspection interval is not the same as a Detection control. Risk assessment quality depends on the gap between those two concepts being understood.
What Is the Difference Between Process FMEA and Design FMEA in Manufacturing?
A process FMEA guide analyzes failure modes in the production process itself: the steps that could produce a defective product. Design FMEA manufacturing analysis identifies failure modes in the product's design before tooling is committed. Both are required in a complete quality system. Skipping design FMEA manufacturing analysis causes recurring production defects with no root cause identifiable in the process FMEA.
When DFMEA Is Conducted and What It Produces
Design FMEA is conducted before the first tool cut, by a team that includes product design engineers, reliability engineers, and customer application engineers. Its input is the set of design requirements and customer functional specifications. Its output is design change recommendations, specification revisions, and validation test requirements that prevent the identified failure modes from appearing in the production process.
DFMEA answers: 'Could this design, as drawn, produce a product that fails to perform its function in the field?' It does not address what happens after tooling is committed and production begins. That is PFMEA's role.
When PFMEA Is Conducted and What It Produces
Process FMEA manufacturing is conducted before production start, typically by a cross-functional team including quality engineers, manufacturing engineers, process engineers, and production operations. Its input is the process flow diagram and the control plan draft. Its output is control plan updates, detection control upgrades, CAPA assignments, and SOP revisions.
PFMEA answers: 'In what ways could this production process produce a defective product, and what controls prevent that?' It is the living quality document most facilities maintain and, in most cases, fail to update after the initial launch review.
Why FMEA Should Be Cross-Functional and Not Owned by Quality Alone
Fmea manufacturing is a structured risk assessment quality method that identifies potential failure modes in a product design or production process, scores each by severity, likelihood of occurrence, and detectability, then prioritizes corrective actions.
It should be conducted by a cross-functional team including quality, manufacturing engineering, process engineering, and operations. A single quality engineer working from a spreadsheet cannot have accurate knowledge of all process failure modes.
How Do You Build a Process FMEA in Seven Steps?
A process FMEA guide follows six to seven sequential steps: define the process scope, identify failure modes per step, analyze effects on customer or downstream process, identify causes, assess current prevention and detection controls, calculate RPN or apply the AIAG-VDA Action Priority table, and assign corrective actions with named owners and deadlines.
Steps 1 to 3: Scope, Failure Modes, and Effects
Step 1 defines the boundary: which production steps are in scope, what each step does, and what output it is expected to produce. Step 2 identifies failure modes per step: in what ways could this step produce the wrong output or fail to perform its function. The goal is to identify all plausible failure modes, including low-probability ones, not just the obvious failure history.
Step 3 analyzes the effect of each failure mode on the customer or the next downstream process. This is where the Severity score is set. A failure mode with a safety impact must be scored 9 to 10 regardless of how rarely it occurs. The Severity score in FMEA manufacturing practice is the non-negotiable input; it cannot be argued down by frequency data.
Steps 4 and 5: Causes and Current Controls
Step 4 identifies the specific cause of each failure mode. '5 Whys' and Ishikawa fishbone diagrams are both used here. The cause must be physical and verifiable, not 'operator error,' which is a symptom description rather than a cause. A valid cause is 'seal head temperature exceeds tolerance because the temperature control relay has no drift alarm.'
Step 5 assesses current prevention controls (what stops the cause from occurring) and current detection controls (what would find the failure after it occurs). This is where the Detection score is set. A manual check every 30 minutes has a Detection score of 5 to 7 in most scoring tables. Fmea manufacturing practice requires the Detection score to reflect the actual probability of finding the failure with the current control, not the probability of finding it if the control is functioning perfectly.
Steps 6 to 7: Risk Evaluation and Action Closure
Step 6 calculates the RPN (S x O x D) and applies the AIAG-VDA Action Priority table, which sorts failure modes into High (act immediately), Medium (plan action), and Low (document and monitor) categories based on a severity-weighted evaluation rather than a flat RPN threshold. This step identifies where engineering resources go.
Step 7 assigns corrective actions with named owners and fixed completion dates. After completion, scores are re-evaluated to confirm the RPN has reduced to an acceptable level. A FMEA manufacturing documents that are not re-scored after corrective action completion is a compliance record, not a risk management tool.
How Does AI Vision Close the Detection and Occurrence Gap in FMEA?
AI vision systems convert FMEA manufacturing from a periodic compliance exercise into a continuous risk management system. Failure mode effects analysis scoring has always required accurate Occurrence and Detection data. AI inspection systems provide it continuously, automatically, and at the scale that manual data collection cannot reach.
How KOMPASS Replaces Estimated Occurrence Scores with Real Data
KOMPASS runs 100% inline inspection and logs every inspection event: defect type, lot, line, shift, and timestamp. Over 30 days of production, that log contains the actual frequency of every failure mode KOMPASS is configured to detect. The Occurrence score in the FMEA register can be updated from that data rather than from the team's memory of how often they 'think' the failure occurs.
This change is the difference between an estimated risk score and a data-derived one. The failure mode effects analysis method has always assumed accurate Occurrence data. AI delivers it. Jidoka's KOMPASS AI vision inspection logs every defect event with classification, generating the structured record that FMEA Occurrence scoring requires.
How 100% Inspection Coverage Changes the Detection Score
A manual inspection check every 30 minutes has a known structural gap: failures that occur between checks are not found at the detection point. The Detection score assigned to that control should reflect the probability of detection across the full production interval, including the unmonitored minutes between checks.
KOMPASS inspects every unit. There are no unmonitored intervals. The Detection score for a failure mode covered by KOMPASS should be 1 to 2 in any FMEA manufacturing scoring table. AI converts failure mode effects analysis from a periodic to a continuous risk system (MAD AI, August 2025). The 2:47 AM gap in this blog's opening scenario closes not because the FMEA changed, but because the detection control it listed became real.
How NAGARE Verifies That Corrective Actions Are Being Followed
FMEA corrective actions often require process changes: a new SOP step, a revised operator sequence, a poka-yoke modification. The FMEA assumes those changes will be followed consistently. Without a verification mechanism, that assumption is the same as the Detection score assumption before AI: optimistic.
Jidoka's NAGARE process monitoring digitizes the corrective SOP and monitors every operator action against it. When the FMEA's corrective action requires 'verify seal temperature within tolerance before proceeding,' NAGARE confirms that the verification step is completed on every production cycle, every shift. The FMEA manufacturing risk register's assumed Detection improvement becomes a verified one.
Jidoka Technologies and FMEA Integration
KOMPASS and NAGARE address the two inputs that determine whether FMEA manufacturing scoring reflects reality or estimate: Occurrence data from 100% inspection coverage, and Detection verification through continuous process monitoring.
- KOMPASS: 100% inline defect detection generates structured defect logs by type, frequency, lot, and shift. Feeds real Occurrence data into FMEA registers. Improves Detection scores to 1-2 range.
- NAGARE: Digital work instruction enforcement verifies corrective action SOP compliance every production cycle. Converts assumed corrective action effectiveness into verified data.
- Industries: Active across automotive, pharmaceuticals, electronics, and FMCG environments where FMEA is a regulatory or customer audit requirement.
See how KOMPASS and NAGARE feed real production data into your risk assessment quality program at jidoka-tech.
Conclusion
The label error at 2:47 AM was not a failure of the FMEA manufacturing process. It was a failure of the detection control the FMEA assumed was adequate. When KOMPASS runs 100% inspection and feeds that data back into the FMEA register, the 2:47 AM gap closes. Not because the failure mode effects analysis changed, but because the detection control it listed became real.
See how KOMPASS and NAGARE feed real production data into your risk assessment quality program at jidoka-tech.ai.
Frequently Asked Questions
1. What Is FMEA in Manufacturing?
FMEA in manufacturing is a structured risk assessment method that identifies potential failure modes in a product design or production process, scores each by severity, likelihood of occurrence, and detectability, then prioritizes corrective actions. It should be conducted by a cross-functional team including quality, manufacturing engineering, process engineering, and operations, not by a single quality engineer working from a spreadsheet.
2. How Is the Risk Priority Number Calculated in FMEA?
The risk priority number calculation in FMEA multiplies three scores: Severity (S) x Occurrence (O) x Detection (D), each rated on a 1 to 10 scale, producing an RPN ranging from 1 to 1,000. The AIAG-VDA FMEA Handbook now recommends using an Action Priority table alongside or instead of the flat RPN threshold, because equal RPNs can represent very different risk profiles when Severity scores differ significantly.
3. What Is the Difference Between Process FMEA and Design FMEA in Manufacturing?
Design FMEA analyzes potential failure modes in a product's design before tooling is committed; Process FMEA analyzes failure modes in the production process itself. Both are necessary in a complete quality system. Design FMEA manufacturing catches design-driven failures before production; process FMEA guide work catches process-driven failures at the line. Skipping DFMEA causes recurring production defects with no root cause found in the PFMEA.
4. How Does AI Improve the Detection Score in an FMEA?
AI vision systems improve the FMEA Detection score by replacing periodic manual inspection with 100% continuous inspection coverage, eliminating the gap between inspection intervals where undetected failures accumulate. KOMPASS AI vision logs every inspection event, generating the real occurrence and detection data that FMEA manufacturing scoring should be built on, replacing the estimated scores that most facilities use today.
5. How Often Should an FMEA Be Updated in a Manufacturing Environment?
An FMEA should be reviewed and updated whenever a design change, process change, new failure mode, or customer complaint occurs, not only on a fixed annual schedule. The AIAG-VDA standard treats FMEA manufacturing documents as living quality records. AI systems that continuously log defect events and flag deviation patterns can trigger FMEA reviews automatically, replacing the quarterly-or-never update cycle with event-driven reviews.
