Machine Design #37: Sensor State Confirmation — Presence Does Not Prove Correct Seating
A sensor detecting a workpiece does not prove that the workpiece is correctly seated. A cylinder-end sensor does not prove that the clamp has generated the required holding force. A PLC output does not prove that the actuator completed its motion.
Many machine faults begin with a gap between the state the control system assumes and the state that physically exists. The sensor may be working exactly as specified while measuring the wrong property. Good sensing design starts with the claim that must be proven, not with the sensor already familiar to the designer.
This article provides a conceptual engineering framework. It does not replace risk assessment, measurement-system analysis, component specifications, functional-safety design, or validation for the actual machine and environment.
1. The measurand ladder
1. Presence
Something exists inside a detection region. Presence says nothing by itself about part type, orientation, seating, or process quality.
2. Identity
The detected object is the intended model, variant, material, or uniquely tracked item. Identity requires a distinguishing feature such as a code, profile, color, RFID tag, or verified dimension.
3. Orientation
The part faces the correct direction or rotational angle. A presence sensor may detect an upside-down part just as reliably as a correct one.
4. Position
The relevant feature lies within a defined spatial tolerance. A switching point gives a binary boundary; it does not automatically provide a position measurement.
5. Seating or contact
The part has reached and contacted the functional datum or support. This is stronger than merely being nearby.
6. State
The mechanism has achieved the condition required by the sequence—for example, locked, clamped, engaged, pressure established, or tool retracted.
7. Process complete
The intended result has been achieved: a fastener has reached torque and angle, a press fit reached the valid force–displacement window, adhesive was dispensed, or a leak test passed.
Moving upward on this ladder requires evidence closer to the engineering claim. Do not use a lower-level signal to assert a higher-level result without a justified model and validation.
2. Write the claim before selecting the sensor
Use a sentence that can be challenged: “The left locating face of model A contacts datum B within 0.15 mm before clamping,” rather than “Sensor X12 is on.”
Then identify the most direct measurand, expected range, tolerance, response time, environment, failure modes, and required diagnostic coverage. Only after that should the team compare inductive, capacitive, photoelectric, laser, ultrasonic, pressure, vacuum, force, vision, encoder, or identification technologies.
This order prevents a common design failure: choosing an inexpensive sensor first and rewriting the requirement to match what it can see.
3. Presence sensors: understand the limit of a binary signal
A proximity switch can confirm that a target entered its switching region. Its result depends on target material, size, approach direction, temperature, supply voltage, mounting metal, contamination, and manufacturing tolerance.
The nominal sensing distance is not the guaranteed switching point for every target and condition. Design with assured operating distance and margin from both the guaranteed-on and guaranteed-off boundaries. If the part can be present but tilted, doubled, or resting on debris, one presence bit is insufficient.
4. Identity needs a distinguishing feature
If two models share the same external silhouette near the sensor, another proximity switch at that location does not create identity. Use a deliberate poka-yoke feature, encoded target, barcode or data matrix, RFID, color pattern, vision feature, or dimensional combination that truly distinguishes the variants.
Define behavior for unreadable, duplicate, stale, and mismatched identifiers. A code read successfully is not enough if the recipe loaded in the machine belongs to another model.
5. Orientation must be verified from asymmetric information
An orientation check needs a feature that changes with orientation. Suitable evidence may be an offset hole, notch, chamfer, profile, key, or a combination of two sensors arranged so that valid states are mutually exclusive.
Avoid relying on a feature that can be hidden by oil, reflection, or part-to-part variation. Verify all credible wrong orientations, including a part rotated 180 degrees, flipped over, or placed one pitch away.
6. Position: a switching point is not a measurement
A binary sensor answers whether the target crossed a threshold. It does not reveal distance from the threshold, remaining margin, or drift over time. When position tolerance is narrow or gradual wear matters, use an analog distance sensor, encoder, linear scale, vision measurement, or multiple independent boundaries as appropriate.
If a binary sensor is sufficient, document the tolerance stack from the functional datum through the target, bracket, sensor repeatability, temperature, vibration, and adjustment method. The sensor position on the CAD model is not the final detection capability.
7. Seating: measure close to the datum
To prove seating, sense the feature whose gap to the datum matters. A sensor on the opposite side of a flexible, warped, or tolerance-sensitive part may report presence while the functional face remains lifted.
Useful methods include multiple seating points, air-gap sensing, displacement measurement, vacuum or pressure behavior, force response, and vision of a datum-related edge. Select the method based on required confidence and the consequences of false acceptance.
Debris is a key challenge case. A chip only 0.2 mm thick may keep a large workpiece visibly “in place” while destroying the machining or assembly datum.
8. Clamp closed does not prove correct clamping
A cylinder reed switch or end-position sensor confirms piston position. It may not detect a missing workpiece, undersized part, broken linkage, bent clamp arm, pressure loss, or contact on the wrong feature.
For critical clamping, combine geometry, pressure or force, and workpiece evidence. Examples include a clamp-position window instead of one end signal, pressure confirmation after contact, displacement monitoring, or a force–travel signature. Define what happens when signals disagree; do not simply choose the bit that allows the cycle to continue.
9. Process complete requires evidence of the process result
“Tool extended” does not prove that a screw was tightened, a hole was drilled, adhesive was dispensed, or a press fit was accepted. Measure the process variable or result that matters.
For tightening, use torque, angle, program result, and traceable tool status. For pressing, use force–displacement windows and characteristic points. For dispensing, monitor flow, pressure, weight, image, or another justified indicator. For machining, consider spindle load, tool monitoring, dimensional inspection, or explicit process completion from the equipment.
The stronger the quality or safety consequence, the closer the evidence should be to the final result.
10. Target design is part of the sensing system
The sensor, target, bracket, cable, connector, and surrounding structure form one measurement system. Design a target with adequate size, material, contrast, angle, stiffness, and cleanability. Provide adjustment features without allowing uncontrolled adjustment.
Avoid targets that can bend, loosen, rotate, collect chips, or be replaced in the wrong orientation. Protect the sensor from impact without blocking its field. Include cable routing and connector access in the mechanical design rather than treating them as late electrical details.
11. Design sensing margin
Do not place the nominal switching point exactly at the process limit. Allocate margin for part tolerance, mounting tolerance, repeatability, thermal expansion, vibration, contamination, aging, and service adjustment.
Document the worst-case target-to-sensor distance for valid and invalid states. The two distributions must remain separated under credible conditions. If they overlap, a more accurate sensor cannot always solve the problem; the mechanical datum or target geometry may need redesign.
Margin should be measurable during commissioning. A sensor that only shows ON/OFF may hide a shrinking margin until production stops.
12. Mechanical mounting and protection
Use a rigid reference from the functional structure, not a thin cover that vibrates independently. Prevent rotation and loss of adjustment. Provide guarded access for replacement, clear indication of the set position, and enough space for the cable bend radius.
Consider coolant, oil, dust, welding spatter, electromagnetic noise, cleaning chemicals, ambient light, temperature, and impact. The IP rating alone does not prove suitability for every exposure or cleaning method.
13. Transition diagnostics: the signal must change at the right event
A signal permanently ON may satisfy a static permissive even when the sensor or wiring has failed. Monitor the expected transition during the sequence.
When the actuator leaves home, the home signal should turn off within an expected interval. When it reaches work position, the work signal should turn on. On return, the reverse transitions should occur. Failure to change is diagnostic evidence even if the final bit appears acceptable.
14. Timing diagnostics
Timeouts should come from physics: stroke, speed, flow, acceleration, PLC scan, network update, sensor response, and validated margin. A timeout that is too short creates nuisance stops; one that is too long hides degrading motion and extends exposure to a fault.
Track actual transition time as condition data. A cylinder that normally moves in 0.6 seconds and gradually rises to 1.4 seconds is telling the maintenance team something before it reaches a 2-second alarm limit.
15. Plausibility between multiple signals
Define combinations that are physically possible. A cylinder should not be at both end positions. A guard cannot be locked while its open sensor is valid, unless the mechanical design explicitly permits that intermediate state. A workpiece cannot be identified as two mutually exclusive models.
Plausibility checks help detect shorts, stale network data, swapped connectors, damaged targets, and sequence assumptions. They are diagnostics unless implemented as part of a validated safety function.
16. IO-Link and diagnostic data
IO-Link can expose signal quality, measured value, threshold margin, contamination warnings, temperature, device identity, and event codes. This data can improve commissioning and predictive maintenance.
Do not collect data without an action plan. Decide which values create an alarm, maintenance warning, trend, or setup check. Preserve device parameters and replacement procedures so a new sensor receives the approved configuration.
17. Standard sensors and safety sensors are not interchangeable
A standard sensor may provide excellent process diagnostics but cannot automatically serve as a safety input. A safety function needs an architecture, component suitability, diagnostic coverage, fault response, and validation consistent with the required performance.
Conversely, a safety-rated interlock may confirm a guard condition but say nothing about workpiece quality. Keep safety claims and process claims explicit, then select the appropriate evidence for each.
18. Use challenge tests
A challenge test deliberately presents states that must be rejected: no part, wrong model, reversed part, tilted part, debris under the datum, incomplete clamp, low pressure, damaged target, disconnected sensor, shorted input, stale communication, and borderline distance.
Run tests across tolerance extremes, temperature, contamination, minimum and maximum speed, and realistic lighting. Record which condition was presented, what the sensor reported, what the control decided, and whether the resulting action was correct.
19. Build an acceptance plan
Define before commissioning:
- The claim and acceptance tolerance.
- Valid and invalid samples.
- Worst-case mounting and target conditions.
- Required detection margin and response time.
- Diagnostic and fault-response tests.
- Evidence to record for FAT and SAT.
- Revalidation triggers after adjustment, replacement, recipe change, or mechanical modification.
Acceptance should prove the complete sensing function, not just that the sensor indicator lights.
20. Plan for maintenance
Make sensors accessible without forcing technicians into an unsafe posture. Mark the approved set position, use keyed connectors where practical, protect parameters, and provide a reference target or test procedure.
Cleaning and inspection intervals should match the failure mechanism. Optical windows need contamination checks; mechanical targets need looseness and damage checks; cables need flex-life inspection; pressure and vacuum sensing needs leak and calibration consideration.
21. Sensor-selection process
Step 1 — Write the claim
State exactly what must be proven and why it matters.
Step 2 — Choose the most direct measurand
Select presence, identity, orientation, position, seating, force, pressure, or process result as appropriate.
Step 3 — Select sensor and target
Compare technology, target design, environment, response, tolerance, and lifecycle.
Step 4 — Design the mounting
Reference the functional datum, protect the device, and control adjustment.
Step 5 — Design diagnostic logic
Use transitions, timing, plausibility, and device-health data.
Step 6 — Run challenge tests
Present credible wrong and borderline states and verify rejection.
Step 7 — Manage the lifecycle
Document replacement, parameter restoration, inspection, and revalidation.
22. Review checklist
Claim
- [ ] The physical state to prove is written clearly.
- [ ] Presence is not being used as unsupported evidence of identity, seating, or completion.
- [ ] Safety and process claims are separated.
Hardware
- [ ] Sensor technology and target suit the environment and tolerance.
- [ ] Mounting references the functional structure and cannot drift unnoticed.
- [ ] Detection margin includes tolerance, temperature, vibration, and contamination.
Logic
- [ ] Expected signal transitions and physical time limits are monitored.
- [ ] Impossible signal combinations are detected.
- [ ] Unknown or contradictory states stop controlled operation safely.
Validation
- [ ] Wrong model, orientation, seating, and borderline samples are challenged.
- [ ] Wiring, target, communication, and contamination failures are tested.
- [ ] FAT/SAT evidence and revalidation triggers are defined.
Safety
- [ ] A standard sensor is not credited as a safety device without justification.
- [ ] Required safety functions use a suitable architecture and validation.
- [ ] Diagnostic messages do not encourage bypass.
Conclusion
A sensor is useful only when it provides evidence for the state the machine must know. Presence is not identity. Position is not seating. End of stroke is not clamping force. A completed command is not a completed process.
Write the claim first, measure as close as practical to the functional datum or process result, design margin and mounting, then verify transitions, timing, plausibility, and failure response.
The MINATA rule is simple: a command expresses intent; feedback provides evidence. Machine logic should advance on evidence strong enough for the consequence of being wrong.
Public references
View all MINATA technical articles