The Cross-Check That Couldn't See a Shared Ground

The pedal sensor reported a position the driver never commanded, and both channels agreed on it. That agreement is the whole story. The diagnostic compared channel A against channel B, and by its design a comparison cannot catch a fault that moves both channels together. The harness revision that merged two ground pins was the trigger, but the root cause sits a level higher: the Detection rating assumed the two channels could not fail at the same time.

Trigger versus assumption

Write the trigger down first, because it is where everyone will want to stop. A cost-down revision moved both sensor returns onto one terminal. Under vibration that terminal developed fretting corrosion, its contact resistance climbed, and the local ground rose with it, so each channel's output rode up by the same offset. Nobody was negligent: the revision passed its own review, the terminal was within its rating when new, and fretting on tin-plated contacts is a known, slow mechanism. The assumption that stopped holding is a different kind of thing. The original DFMEA gave the plausibility check a low Detection score because it treated the two channels as independent, and that belief lived in the analysis, not in the harness drawing that someone changed.

Why the plausibility window was blind

The cross-check allowed 150 mV of disagreement between channels before setting a fault, a sensible window for gain drift and quantisation in either channel on its own. A common-mode ground shift produces zero disagreement, so the window was never reached however large the offset grew. Redundancy only keeps a system running through a single failure when the failure path of one element shares nothing with the other. Once both channels pass through the same terminal, the structure has one single point of failure that looks like two channels. The structure tree shows this plainly if the shared return is modelled as its own element with its own failure mode, and hides it completely if the return is folded into a generic harness line, which is what the original analysis did. That was reasonable while the grounds were separate and became wrong the day they were merged.

The rule to apply

Never give Detection credit to a comparison between two channels that meet at any node in the structure tree: either model that node explicitly with its own cause and treat the comparison as blind to it, or rate Detection as though the comparison did not exist.

Catalogs carry assumptions forward

Reusable catalogs are where this kind of belief travels unseen. The sensor entry was copied from a validated programme, and its Detection score was correct for the architecture it was written for; the copy took the score but not the condition it depended on. In FMEA Tool the structure, function and failure chain sits on one page, so a reviewer looking at a cross-check control can see, in the same view, which elements feed both channels. Action Priority recalculates as soon as Detection is re-rated honestly, and here that moved the item from low to high without anyone touching Severity. The harmonised method that automotive quality standards such as IATF expect already asks for this traceability; the tool simply makes the dependency hard to hide. What remains is a design trade-off. Complementary signal slopes would turn a common ground shift into a detectable sum error at no hardware cost, while separate returns would remove the shared node for the price of a connector pin, and the programme has to decide whether it wants to detect the common cause or eliminate it.

Isn't this the harness engineer's root cause, since their change introduced the shared terminal?

No. The change was the trigger, and it was reviewed against the criteria it was given. The root cause is that the DFMEA built channel independence into the Detection score without stating it, so no change review had a way to see that merging the grounds made that score invalid. The corrective action belongs to the analysis: make independence an explicit structure element, so any future change that touches it shows up in review.

Back to FMEA Tool · Domains

← Back to FMEA Tool
Engineering & Architecture Brief

The Cross-Check That Couldn't See a Shared Ground

The pedal sensor reported a position the driver never commanded, and both channels agreed on it. That agreement is the whole story. The diagnostic compared channel A against channel B, and by its design a comparison cannot catch a fault that moves both channels together. The harness revision that merged two ground pins was the trigger, but the root cause sits a level higher: the Detection rating assumed the two channels could not fail at the same time.

Trigger versus assumption

Write the trigger down first, because it is where everyone will want to stop. A cost-down revision moved both sensor returns onto one terminal. Under vibration that terminal developed fretting corrosion, its contact resistance climbed, and the local ground rose with it, so each channel's output rode up by the same offset. Nobody was negligent: the revision passed its own review, the terminal was within its rating when new, and fretting on tin-plated contacts is a known, slow mechanism. The assumption that stopped holding is a different kind of thing. The original DFMEA gave the plausibility check a low Detection score because it treated the two channels as independent, and that belief lived in the analysis, not in the harness drawing that someone changed.

Why the plausibility window was blind

The cross-check allowed 150 mV of disagreement between channels before setting a fault, a sensible window for gain drift and quantisation in either channel on its own. A common-mode ground shift produces zero disagreement, so the window was never reached however large the offset grew. Redundancy only keeps a system running through a single failure when the failure path of one element shares nothing with the other. Once both channels pass through the same terminal, the structure has one single point of failure that looks like two channels. The structure tree shows this plainly if the shared return is modelled as its own element with its own failure mode, and hides it completely if the return is folded into a generic harness line, which is what the original analysis did. That was reasonable while the grounds were separate and became wrong the day they were merged.

The rule to apply

Never give Detection credit to a comparison between two channels that meet at any node in the structure tree: either model that node explicitly with its own cause and treat the comparison as blind to it, or rate Detection as though the comparison did not exist.

Catalogs carry assumptions forward

Reusable catalogs are where this kind of belief travels unseen. The sensor entry was copied from a validated programme, and its Detection score was correct for the architecture it was written for; the copy took the score but not the condition it depended on. In FMEA Tool the structure, function and failure chain sits on one page, so a reviewer looking at a cross-check control can see, in the same view, which elements feed both channels. Action Priority recalculates as soon as Detection is re-rated honestly, and here that moved the item from low to high without anyone touching Severity. The harmonised method that automotive quality standards such as IATF expect already asks for this traceability; the tool simply makes the dependency hard to hide. What remains is a design trade-off. Complementary signal slopes would turn a common ground shift into a detectable sum error at no hardware cost, while separate returns would remove the shared node for the price of a connector pin, and the programme has to decide whether it wants to detect the common cause or eliminate it.

Technical FAQ: Isn't this the harness engineer's root cause, since their change introduced the shared terminal?

No. The change was the trigger, and it was reviewed against the criteria it was given. The root cause is that the DFMEA built channel independence into the Detection score without stating it, so no change review had a way to see that merging the grounds made that score invalid. The corrective action belongs to the analysis: make independence an explicit structure element, so any future change that touches it shows up in review.

Related Architecture Specifications (FMEATOOL): fmeatools.com · freefmea.com

More Technical Articles on fmeatool.com

FMEA for Edge Compute Offloading: Protocol Conformance and Production SLA Failure Modes →Zero-Trust Cryptography and FMEA for Fault-Isolated Failover →Analyzing Timing Failure Modes with FMEA: Deterministic Telemetry, Sub-Millisecond Latency Budgets and HIL/Digital-Twin Verification →FMEA Reference Architecture: Aligning DFMEA/PFMEA Data with AIAG-VDA, SAE J1739 and IEC 60812 for Interoperable Exchange →Using FMEA for Autonomous Diagnostics, Predictive Maintenance and Cross-Layer Security: A Reference Architecture →