On this page
Two printed tracks do not automatically create two independent measurements. They may share a wiper carrier, supply, return, connector, harness, reference, analogue front end or calculation. A common disturbance can therefore move both outputs together and remain invisible to a simple agreement check. This guide defines channel boundaries, common causes and discriminating fault injections before any correlation statistic is interpreted. It supports drawing and system review; it does not assign a safety integrity level or claim diagnostic coverage without a complete application analysis.
Key design decisions
- Draw each sensing chain from physical input to reported value and mark every shared node.
- Specify faults that one channel must reveal without relying on the same shared element.
- Use correlation only after scale, range, timing and common-cause sensitivity have been characterized.
Draw both end-to-end measurement paths
Start at the physical position or force and follow each path through contact, resistive element, collector, terminal, connector, harness, excitation, return, analogue input, reference, conversion and software scaling. Give every node a channel label or mark it shared. Two separate printed arcs can converge at a common terminal; conversely, one substrate can carry genuinely separate conductors and returns. The physical drawing, schematic and connector pinout must agree. A block diagram that begins only at the electronic control unit hides the contact and card mechanisms most likely to affect both channels.
Define what the system expects from redundancy. The objective may be continued measurement after one open circuit, detection of an implausible pair, or bounded control during a single fault. Those are different architectures. State whether channels have equal slopes, opposite slopes, different ranges or a prescribed relationship. Also identify the safe response and which independent observer commands it. This page can organize those inputs but cannot determine system safety acceptability without the equipment-level hazard and diagnostic analysis.
Model channel response and residuals
Represent channel i as y_i = f_i(x, u_i, c), where x is the desired physical input, u_i contains disturbances unique to that channel and c contains common causes. Agreement between y_1 and y_2 can reveal some difference in u_1 and u_2, but cannot by itself identify a disturbance contained in c. For an expected affine relationship y_2 = a + b y_1, define residual r = y_2 - (a + b y_1). The coefficients, applicable input range, direction, temperature state and sampling alignment must be controlled before r is meaningful.
An illustrative calculation shows the limitation. If the expected relationship is y_2 = 5.0 - y_1 and measured values are y_1 = 1.80 V and y_2 = 3.16 V, then r = 3.16 - (5.0 - 1.80) = -0.04 V. That residual may fit a project-defined band, yet both values could still be wrong if a shared mechanical offset moved the physical input or a shared reference scaled both conversions. The result says the pair relationship is near expectation at that instant; it does not prove correct position or independence.
| Layer | Channel-specific evidence | Common-cause question |
|---|---|---|
| Card and contact | Separate tracks, collectors and continuity paths | Can one carrier, crack or contaminant affect both? |
| Harness | Separate pins and conductors | Where do returns or supplies join? |
| Readout | Separate inputs and defined scaling | Is the ADC reference or software common? |
| Diagnostic | Residual and range checks | What independent observer detects common motion error? |
Control range, direction and timing
Correlation can look poor when two valid channels have different transfer curves, and look excellent when a common fault moves both. Compare channels using their specified relationship, not raw equality. Partition the travel into regions if slopes or resolution change. Include electrical end points, overtravel and any deliberate dead zone. When one channel rises and the other falls, evaluate the transformed relationship and retain signs; taking absolute values can conceal an incorrect direction. Use the same physical coordinate and calibration state for every pair.
Time alignment matters during rapid motion. Different filters, ADC schedules or communication latency can create a transient residual even when both tracks are correct. Record sampling instants and filtering. If y_2 is delayed by delta t, a first-order motion-related disagreement is approximately (dy_2/dt) delta t. That term should not be mistaken for track mismatch. Conversely, excessive filtering can suppress a short intermittent contact fault. Verification should include relevant motion rates and both directions rather than only steady points.
Design fault injections that discriminate paths
Create a fault-injection matrix from the marked diagram. Open each channel-specific conductor, short it to its permitted adjacent nodes through controlled means, shift its excitation if separately supplied, and introduce a bounded input offset where safe and representative. For every injection, predict both outputs, residual, range flags and system response before testing. An open on channel one that also disables channel two reveals an undocumented shared dependency. A diagnostic that uses the failed channel's own reference is not independent merely because it runs in separate software.
Common-cause challenges require different observations. Move the shared mechanical datum, interrupt a shared return, perturb a common reference within a controlled bench setup, or compare against an independent position reference where the application permits. The purpose is not to simulate every field failure; it is to determine which classes are invisible to channel agreement. Record limitations and do not claim coverage for untested fault magnitudes, durations or combinations. Destructive or hazardous injections belong to an approved safety plan and suitable equipment.
Interpret channel-pair signatures
A step in one output with the other unchanged suggests a channel-specific contact, conductor, input or coefficient. Opposite residual signs around a crossover can indicate a scaling or zero mismatch. Simultaneous same-direction shifts with a stable residual point toward a common physical input, supply, reference or computation. Both channels pinned at an end may arise from true position, a shared stop, a common return fault or saturation; the pair alone cannot choose among them. Pair the signal record with independent position, supply and reference observations.
Noise coherence also needs care. Uncorrelated high-frequency interruptions can suggest individual contacts, whereas coherent ripple may be coupled electrically or mechanically. Correlation changes with bandwidth and filtering, so retain sample rate, filters and window length. A single coefficient is not a complete fault signature. Preserve raw paired traces, motion direction, environment and injected condition. When causes remain confounded, state the ambiguity and request a targeted observation rather than naming the printed track as the failure source.
Verify independence across relevant conditions
Run nominal mapping first so both transfer laws are known. Then repeat the planned channel-specific and common-cause injections at positions that challenge end regions, crossovers and slope changes. Include supply and temperature conditions supplied by the project, but do not accelerate or extrapolate without an approved mechanism. Verify startup, loss and restoration of excitation, because diagnostics can behave differently before filters settle. Confirm that stored coefficients and software channel assignments cannot be exchanged without detection.
Acceptance criteria should be linked to a stated function: detection time, residual band, allowed degraded output, or transition to a safe state. Include measurement uncertainty and fixture error so an apparent residual does not consume the whole band. Evidence from one prototype configuration applies only to its recorded schematic, software, harness and mechanical state. Connector, grounding, filter or calibration changes reopen the dependency map even when printed artwork is unchanged.
Package an auditable redundancy review
The release record should contain the two end-to-end diagrams, shared-node register, transfer relationship, residual definition, applicable range, time alignment, fault matrix, predicted responses, observed results and unresolved common causes. Link each drawing and schematic revision. Do not summarize this as dual-channel passed; state which injected conditions were detected, which independent observer was used and which dependencies remain outside the evidence. That record supports later system analysis without turning a component check into a safety certification.
For a project review, send card artwork and material stack, contact arrangement, mechanism drawing, schematic, connector and harness pinout, supply and return topology, analogue input details, software scaling, calibration procedure, sampling/filter timing, expected channel relationship and system reaction to disagreement. Include known failure modes and required standards or hazard-analysis constraints. The engineering response can propose a path-separation and fault-injection plan by application review; final diagnostic coverage and safety disposition remain with the complete system owner.
Review a dual-track resistive measurement architecture
Provide both complete signal paths and the required system response so shared dependencies and useful fault injections can be identified.
- Card artwork, contact geometry and material stack
- Mechanism, shared shaft or carrier and datum drawings
- Circuit schematic, connector and harness pinout
- Excitation, return, reference and ADC architecture
- Expected channel transfer relationship and tolerances
- Sampling, filtering, calibration and software assignment
- System fault response and required validation conditions
The drawing-upload form loads as you reach this section.