51 / 78

Lesson 51 of 78 · Sensors, Actuators & Embedded Behavior

Sensors, Metrology & Uncertainty

Choose a sensor by first defining the measurand: “shaft angle relative to housing at the control-loop update instant,” not merely “position.” NIST’s manufacturing robotics work emphasizes metrics and test methods because capability must be expressed and verified consistently 1.

A physical measurand passes through sensor, conditioning, conversion, scaling, estimate, and decision, with uncertainty sources entering each stage
A measurement is a chain of transformations and uncertainty, not a number that appears directly from reality. Credit: StudyCorner original diagram · CC BY 4.0 · Source

Static and dynamic characteristics

Consider range, resolution, sensitivity, offset, linearity, hysteresis, repeatability, drift, cross-sensitivity, bandwidth, delay, environmental rating, mounting influence, and failure behavior. A precise sensor mounted to a flexible bracket measures bracket motion too.

Combine uncertainty honestly

Suppose a distance result has independent standard uncertainties of 0.08 mm from calibration, 0.05 mm from repeatability, and 0.06 mm from temperature compensation. Root-sum-square combined standard uncertainty is

\[ u_c=\sqrt{0.08^2+0.05^2+0.06^2}=0.112\ \text{mm}. \]

An expanded uncertainty might multiply by a stated coverage factor. Do not RSS bounds that are biased, correlated, or not statistical.

Sensor diversity

  • encoders measure angular or linear position but need referencing and alignment;
  • strain gauges infer load from deformation and require bridge conditioning and thermal compensation;
  • inductive/capacitive/photoelectric sensors detect presence with material and environment dependencies;
  • force-torque sensors enable interaction but can saturate under impacts;
  • vision measures many features but depends on optics, lighting, calibration, occlusion, and computation.

Fail detectably

Design plausible diagnostics: out-of-range, stuck value, implausible rate, disagreement between redundant channels, missed heartbeat, broken-wire state, reference-target check, or limit-switch inconsistency. A numeric value is not necessarily valid data.

Sensor selection dossier

For a part-present and position task, compare three sensing principles. State measurand, target variability, environment, required uncertainty, update rate, false-positive/negative consequence, mounting, calibration, diagnostics, and acceptance test. Select the sensor plus its fixture, lighting or target, wiring, and software check as one measurement system.

Source trail

References

  1. 1
    Robotic Systems for Smart Manufacturing Program. National Institute of Standards and Technology. verifiedMeasurement science, performance metrics, test methods, interoperability, planning, agility, and collaborative workcell integration. Cited at: measurement approach.
Further reading

Check your understanding

  1. What is the measurand?
  2. Repeatability alone proves what?