Measurement Transducers: Static & Dynamic Characteristics

A measuring transducer provides an output quantity with a specified relation to an input quantity. The word specified carries most of the engineering: the relation is valid only over stated ranges, environmental conditions, loading, excitation, and time behavior. Current metrology vocabulary prevents familiar but incompatible words such as accuracy, precision, sensitivity, and resolution from collapsing into one vague notion of "quality."

Start with the measurand #

"Temperature" is not a complete measurand. "The surface temperature of the bearing outer ring at the marked point, after thermal equilibrium, referred to the stated scale" is much closer. The sensor may disturb it through self-heating, thermal conduction, mechanical stiffness, electrical loading, or extracted flow. That disturbance belongs in the measurement model rather than being dismissed as an instrument detail.

Static characteristic #

After transients settle, represent the nominal conversion as

$$y = f(x),$$

or, near an operating point $x_0$,

$$y \approx y_0 + S(x_0)(x-x_0), \qquad S(x_0)=\left.\frac{df}{dx}\right|_{x_0}.$$

The slope $S$ is sensitivity. It may vary across the range; a single gain is justified only over the region where that variation is acceptably small.

Error, uncertainty, and accuracy #

For a reference input $x_r$, an indicated or inferred value $x_m$ has error

$$e = x_m - x_r.$$

Error is the difference for a particular result relative to a reference value. Measurement uncertainty characterizes the dispersion of values that could reasonably be attributed to the measurand. Accuracy is qualitative in VIM: it is closeness of agreement, not a number to be written as "±0.5%." A datasheet should state a maximum permissible error, uncertainty, accuracy class, or another defined quantity instead.

Common contributions are:

• calibration residuals and reference uncertainty;
• repeatability and noise;
• temperature, supply, humidity, mounting, and cross-axis effects;
• loading and lead resistance;
• hysteresis and creep;
• analog filtering, sampling, quantization, and computation;
• time alignment when several channels describe one event.

Do not add worst-case percentages blindly when their denominators differ. Convert each contribution to a common output or input-referred quantity, state the coverage model, and propagate it through the measurement equation described in measurement error and uncertainty.

Repeatability is not reproducibility #

A Gauge R&R study separates variation observed under repeatability conditions—same procedure, operator, equipment, location, and short time—from variation introduced by reproducibility conditions such as different operators or setups. It answers whether the measurement process can distinguish part-to-part variation in the studied process. It does not establish traceability, calibrate bias, or prove suitability outside the sampled range and conditions. A system can be highly repeatable and consistently wrong.

Dynamic characteristic #

When $x$ changes on the same time scale as the transducer, the static curve is insufficient. A common first-order model is

$$\tau \frac{dy}{dt} + y = Kx,$$

with transfer function

$$G(s)=\frac{Y(s)}{X(s)}=\frac{K}{1+s\tau}.$$

For a step input, $y$ reaches $63.2\%$ of its final change at one time constant, $95.0\%$ at about $3\tau$, and $99.3\%$ at about $5\tau$. The right-hand plot above contrasts this monotonic lag with an underdamped response. Step response time must include a stated settling band; "response time" without the band and input step is incomplete.

A second-order sensor model,

$$G(s)=\frac{K\omega_n^2}{s^2+2\zeta\omega_n s+\omega_n^2},$$

adds natural frequency $\omega_n$ and damping ratio $\zeta$. Low damping produces overshoot and resonance; high damping removes ringing but slows the response. Mechanical accelerometers, pressure systems with tubing, and elastic sensing elements often need this model.

Frequency response and sampling #

Bandwidth is not simply the highest frequency that produces any output. It is the frequency interval over which a declared gain and phase error remain acceptable. A first-order low-pass reaches $-3$ dB and $-45^\circ$ at $f_c=1/(2\pi\tau)$, which may already be too much error for a control loop.

The complete chain includes sensor dynamics, conditioning filters, sample-and-hold, ADC, digital filtering, and timestamping. The analog stages must limit out-of-band energy before sampling; otherwise aliasing converts it into false in-band motion that calibration cannot remove.

A defensible characterization procedure #

1. Define the measurand, range, environmental conditions, mounting, excitation, load, and output representation.
2. Calibrate at multiple points in both directions; retain raw repeats, not only averages.
3. Fit the declared model and plot residuals against input, direction, time, and influence quantities.
4. Apply steps or sweeps fast enough to identify the dynamic model without violating the sensor's ratings.
5. Repeat reference points after the test to reveal drift and irreversible change.
6. Report the calibration relation, uncertainty, bandwidth or response time, and every condition needed to reproduce them.