When a bridge flexes, a bicycle frame carries a rider, or an aircraft wing meets turbulent air, the deformation can be far too small to see. A strain gauge turns that tiny shape change into a tiny resistance change. A Wheatstone bridge then turns the resistance change into a voltage that an instrument can measure.
This is a beautiful answer to the question “why is mathematics important?” Geometry defines strain. Materials connect strain with stress. Circuit algebra extracts a differential signal. Statistics and calibration decide whether the signal is trustworthy. No single formula does the whole job.
The examples here are educational. Real structural tests, weighing systems and safety-critical monitoring require qualified engineers, specified installation procedures, calibrated equipment and application standards.
Quick Reading Route
- Start with strain for the mechanical quantity.
- See the electrical link for gauge factor.
- Open the bridge for circuit mathematics.
- Follow a worked example for units and scale.
- Read calibration and uncertainty before using any result.
- Use the practice plan for a student route.
Why Mathematics Is Important in Strain Measurement
A resistance strain gauge is usually a thin conductive grid bonded to a surface. When the surface stretches in the gauge direction, the grid lengthens and its resistance changes. When the surface contracts, the sign reverses. The instrument never sees “strain” directly; it sees voltage after a chain of physical and electrical transformations.
Every transformation needs a model. The gauge must transfer the specimen’s deformation. Temperature can change resistance. Lead wires add resistance. Excitation can warm the gauge. Amplifiers add offset and noise. The structure may bend, twist or carry residual stress.
Mathematics makes these influences visible. It lets an engineer ask which variable produced a reading, how sensitive the circuit is, what range avoids overload, and how much uncertainty should accompany the final result.
Strain Is a Ratio
Normal engineering strain is
ε = ΔL / L,
where L is the original gauge length and ΔL is the change in length. Because both quantities use the same length unit, strain is dimensionless.
Small engineering strains are commonly written in microstrain. One microstrain, written με, is 10⁻⁶ strain. If a 100 mm reference length extends by 0.05 mm, strain is 0.05/100 = 0.0005 = 500 με.
The ratio matters more than the raw extension. A 0.05 mm change over 100 mm and a 0.10 mm change over 200 mm produce the same average strain. This lets results compare different sizes.
Sign and direction
Tension is often assigned positive strain and compression negative strain. A gauge responds mainly along its grid direction, so orientation matters. A surface can stretch in one direction while contracting in another.
In bending, fibres on one side of a neutral axis lengthen while those on the other side shorten. Two gauges placed symmetrically can therefore produce opposite mechanical signs. Bridge wiring can arrange those opposite resistance changes to reinforce the output.
Gauge Factor Connects Strain and Resistance
The gauge factor GF is defined approximately by
GF = (ΔR/R) / ε.
Rearranging gives ΔR/R = GF·ε. If a 120 Ω gauge has GF = 2.05 and experiences 500 με, then
ΔR/R = 2.05 × 500 × 10⁻⁶ = 0.001025.
The resistance change is 120 × 0.001025 = 0.123 Ω. This is small compared with the nominal resistance and illustrates why differential bridge circuits are useful.
Gauge factor is not an eternal universal constant. Use the manufacturer’s value and tolerance for the gauge batch and intended conditions. Transverse sensitivity, temperature and installation can influence the realised response.
Stress, Strain and the Material Model
Within a suitable linear elastic range, uniaxial stress and strain are related by Hooke’s law:
σ = Eε,
where E is Young’s modulus. A measured strain of 500 με in material modelled with E = 200 GPa gives an illustrative stress of 100 MPa.
That conversion is only as valid as its assumptions. The stress state may be multiaxial, the material may be anisotropic, local plasticity may occur, or the gauge may sit near a stress concentration. A strain reading is evidence at a location and direction; it is not automatically the maximum stress in an entire component.
Finite-element analysis can predict strain fields, but measured strain is still valuable for model validation. Agreement should be assessed across load cases and locations, not claimed from one matching point.
The Wheatstone Bridge
A Wheatstone bridge contains four resistive arms. Excitation voltage is applied across one diagonal and output is measured across the other. A general output relation is the difference between two voltage-divider fractions:
Vo/Vex = R2/(R1+R2) − R4/(R3+R4),
subject to the chosen arm labels and polarity.
If all four resistances are equal, both divider outputs are one-half the excitation and the ideal bridge output is zero. A small resistance change unbalances the bridge.
Quarter bridge
A quarter bridge has one active strain gauge and three nominally fixed arms. For a small resistance change in the active arm, a common first-order magnitude approximation is
Vo/Vex ≈ GF·ε/4.
The sign depends on which arm contains the gauge and the wiring polarity. The approximation neglects higher-order terms and assumes the other arms do not change.
Half bridge
A half bridge uses two active gauges. In a bending arrangement, one gauge may be in tension and one in compression. Wired in suitable arms, their changes add, increasing sensitivity and helping reject effects shared by both gauges.
Two gauges do not automatically compensate temperature. They need compatible characteristics and similar thermal conditions, and the bridge placement must make common thermal resistance changes cancel.
Full bridge
A full bridge uses four active gauges. Load cells often arrange gauges so that two increase resistance and two decrease, providing greater sensitivity and useful temperature compensation. Full-bridge output is still conditioned by bonding, geometry, excitation, wiring and amplifier behaviour.
NASA technical work on strain-sensing circuits documents full Wheatstone bridges with all four resistors active and illustrates why bridge configuration, rather than the physical gauge alone, determines electrical sensitivity.
Worked Example: From Microstrain to Millivolts
Take a quarter bridge with GF = 2.0, strain ε = 500 με and excitation Vex = 5.0 V.
First convert strain:
500 με = 500 × 10⁻⁶ = 0.0005.
Use the small-change approximation:
Vo/Vex ≈ 2.0 × 0.0005 / 4 = 0.00025.
Then
Vo ≈ 5.0 × 0.00025 = 0.00125 V = 1.25 mV.
If the instrumentation amplifier gain is 800, the ideal output becomes 1.00 V. But a 0.5 mV input offset would become 0.4 V at the output. High gain makes tiny signals usable and makes tiny errors important.
Ratiometric measurement
Bridge output is proportional to excitation. A ratiometric system compares output with the same excitation reference, reducing sensitivity to slow excitation changes. It does not cancel self-heating, wiring changes or amplifier drift.
Temperature Effects and Compensation
Temperature changes the gauge grid resistance, the specimen dimensions, the adhesive and lead-wire resistance. Even when there is no mechanical load, a bonded gauge can report apparent strain because the gauge and specimen expand differently.
A dummy-gauge arrangement places an unstrained gauge of the same type on similar material at the same temperature. Its thermal resistance change can cancel that of the active gauge in the bridge. The word “same” carries the burden: temperature gradients or different bonding can break the cancellation.
Self-temperature-compensated gauges are designed for specified thermal-expansion behaviour, but compensation has a range and residual error. A calibration record should identify gauge type, specimen material and temperature conditions.
NASA’s compensated high-temperature strain-gauge work describes adjacent bridge arms used to cancel thermally induced apparent strain, while preserving response to mechanical strain. It is a practical example of subtraction through circuit symmetry.
Lead Wires and Remote Gauges
In a two-wire quarter bridge, lead resistance sits in series with the gauge. Temperature changes along a long cable can therefore look like strain. Three-wire arrangements can compensate equal lead resistances under suitable bridge conditions. Four-wire sensing is used in other resistance measurements to separate current-carrying and voltage-sensing paths.
The algebra should include the leads rather than calling them “just wires.” If two supposedly equal leads differ or experience different temperatures, the residual appears at the output.
Shielding, grounding and cable routing affect noise. A shield connected carelessly at multiple points can create ground loops. These are system-level decisions, not properties of the strain gauge formula.
Bonding, Alignment and Strain Transfer
The gauge measures the strain transferred through its adhesive and backing. Surface preparation, adhesive thickness, curing, moisture protection and gauge length affect that transfer.
If a gauge intended for axial strain is installed at angle θ to the true principal direction in a simple uniaxial state, the measured value changes with orientation. For a basic plane-strain transformation, normal strain depends on cos²θ, sin²θ and the engineering shear term. Small alignment errors may be tolerable in one case and serious in another.
A short gauge reveals local gradients but is more sensitive to exact placement. A longer gauge averages strain over its grid. The “right” gauge length depends on the phenomenon and spatial scale.
Strain Rosettes and Directional Reconstruction
A single gauge measures one directional component. A rosette combines gauges at known angles, often 0°, 45° and 90°. Plane-strain transformation equations can reconstruct εx, εy and engineering shear γxy, then calculate principal strains and their directions.
For a gauge at angle θ,
εθ = (εx+εy)/2 + (εx−εy)cos2θ/2 + γxy sin2θ/2.
Three independent directions provide enough equations for three unknown plane-strain components. The geometry must be correct, and all gauges should sample essentially the same point.
Principal strain calculation involves square roots and inverse trigonometric functions. Quadrant-aware angle functions help prevent a direction error. Report the convention for shear strain and angle.
Signal Conditioning
Bridge output may be only millivolts. An instrumentation amplifier provides differential gain while rejecting common-mode voltage. Its input range, gain error, offset, noise, bandwidth and common-mode rejection all matter.
An anti-alias filter limits frequencies before analogue-to-digital conversion. Sampling faster does not recover a peak already attenuated by a slow amplifier, nor does it prevent aliasing without appropriate analogue filtering.
Excitation should remain within gauge and bridge limits. Increasing excitation raises signal but also increases power V²/R and self-heating. Sensitivity cannot be improved without checking the thermal consequence.
ADC counts and effective resolution
A 16-bit converter across ±5 V has a nominal code width of about 10 V/65,536 ≈ 153 μV. With amplifier gain 800, that represents about 0.191 μV at the bridge input. Converting that voltage to strain also requires excitation, bridge factor and gauge factor.
Noise and nonlinearity reduce effective resolution. Quoting 16 bits does not prove that all 65,536 levels carry independent information.
Calibration from Voltage to a Measurand
The theoretical bridge equation predicts electrical sensitivity, but a complete transducer is calibrated against known loads or displacements. Fit a line such as
y = ax + b,
where x is bridge output and y is the reference measurand. The slope is sensitivity and the intercept captures zero offset.
Residuals reveal nonlinearity. Loading and unloading paths reveal hysteresis. Repeated cycles show repeatability and zero return. Calibration should span the intended range and include enough points to expose curvature.
Shunt calibration connects a known resistor across a bridge arm to produce a calculable electrical unbalance. It checks wiring, excitation, amplifier and data acquisition. It does not test whether the gauge is bonded properly or whether specimen strain reaches the grid.
NASA records describe shunt-calibration systems that read full-bridge output and calculate strain. The technique is strong because it inserts a known electrical change, yet its boundary must remain explicit.
Uncertainty and Decision Rules
An uncertainty budget may include reference load, calibration fit, repeatability, temperature, alignment, gauge factor, excitation, amplifier gain and data-acquisition effects. Convert every component to the reported output unit using sensitivity coefficients.
If independent standard components in microstrain are 8, 5, 6 and 3, their root-sum-of-squares is sqrt(64+25+36+9) ≈ 11.6 με. With an illustrative coverage factor of two, expanded uncertainty is about 23 με. Correlated terms require covariance and cannot be combined blindly.
An alarm threshold also needs a decision rule. A 995 με estimate near a 1000 με limit is not the same decision when expanded uncertainty is 10 με versus 100 με. Duration, rate and consequence may matter in addition to peak value.
Bridge Nonlinearity and Exact Equations
The quarter-bridge relation Vo/Vex ≈ GFε/4 is a first-order approximation. It comes from expanding the exact voltage-divider expression for a small fractional resistance change. At larger strain or high required accuracy, the denominator change produces nonlinearity.
Let one arm be R(1+x) and the other three arms remain R. Depending on labeling, one divider becomes R/[R(1+x)+R] = 1/(2+x), while the reference divider remains one-half. Their difference is not exactly x/4. A series expansion gives the familiar linear term plus smaller higher-order terms.
This matters in two ways. First, software that converts voltage to strain can use the exact bridge equation. Second, calibration residuals can reveal combined nonlinearity from bridge algebra, amplifier and mechanical transducer.
For half and full bridges, each active arm may change with a different sign and magnitude. Memorised factors of two or four only apply to stated mechanical arrangements. Derive the divider voltages from the actual arm changes when precision matters.
Cyclic Strain, Fatigue and Rainflow Counting
Many structures experience repeated variable-amplitude loading. A time history may contain thousands of local maxima and minima rather than one simple sine wave. Peak strain, strain range and mean strain can all influence fatigue assessment.
Rainflow counting is a mathematical method that converts an irregular sequence into counted cycles of specified ranges and means. The resulting histogram can feed a material fatigue model such as a strain-life relation. It does not create material data or remove environmental and manufacturing effects.
Sampling rate matters. If a logger misses short peaks, the counted spectrum becomes artificially gentle. Filtering can also reduce ranges. The acquisition chain should be designed for the expected structural dynamics before cycle counting begins.
For a student exercise, generate a slow sequence of turning points and apply a simplified cycle-counting rule. The important idea is that damage evidence lives in the sequence, not only in the largest displayed value.
Zeroing, Drift and Baseline Management
Zero is a measurement decision. A bridge may be balanced electrically before load, but the structure can already contain assembly stress, gravity load or temperature strain. Taring the display removes an offset; it does not erase the physical state.
Long tests need reference checks. Plot zero output during unloaded opportunities, environmental temperature and excitation. A gradual drift may reflect adhesive creep, temperature, electronics or structural change.
Baseline correction must be documented. Subtracting a fitted trend can improve analysis when the trend is demonstrably instrumental, but it can also remove real slow structural response. Preserve raw data and explain every correction.
Common Misconceptions
- The gauge measures force directly. It measures a resistance change associated with local strain; force requires a structural calibration model.
- A balanced bridge has no error. Zero output can coexist with offset cancellation, wiring errors or equal unwanted changes.
- More excitation is always better. Signal increases, but so do self-heating and possible drift.
- A full bridge is four times better in every way. Sensitivity depends on the mechanical strain signs and wiring; complexity and installation also grow.
- A perfect calibration line proves field accuracy. Field temperature, mounting and load paths can differ from calibration.
- Finite-element agreement proves the model. One matching gauge is weak validation; multiple cases and locations are stronger.
A Four-Week Learning Plan
Week 1: strain and units
Convert extension to strain and microstrain. Draw tension, compression and the neutral axis for simple bending.
Week 2: bridge algebra
Calculate voltage-divider outputs, then derive quarter-, half- and full-bridge small-change sensitivity for stated wiring.
Week 3: signals and calibration
Choose gain, ADC range and sampling rate for simulated data. Fit a calibration line and inspect residuals and hysteresis.
Week 4: uncertainty and communication
Build an uncertainty budget, evaluate an alarm near its limit and write a result that includes conditions and limitations.
Guidance for Students
Draw both the structure and circuit. Mark which gauges stretch, which resistances rise, and which bridge node voltage changes. Many sign mistakes disappear when the physical and electrical pictures share labels.
Keep microstrain conversion visible. Write 500 × 10⁻⁶ before using 500 in an equation. Track volts, millivolts and microvolts explicitly.
When a model and experiment disagree, inspect the chain rather than forcing the data. Load path, bonding, temperature, alignment, wiring, gain and sampling are all candidates.
Guidance for Parents and Educators
Use paper beams, low-voltage bridge simulators and manufacturer data rather than loading real structures. The educational goal is to connect ratios, circuits and evidence safely.
Ask students what the instrument actually observes. “It measures voltage, then we infer strain” is more mature than saying the screen directly sees stress.
Encourage complete graphs: axes, units, uncertainty and loading direction. A neat conclusion cannot repair missing conditions.
Did You Know?
The bridge is powerful because it measures a difference between ratios. Large common voltages can coexist with a tiny useful differential signal. This same pattern appears in many sensors because subtraction can expose small changes while rejecting shared influences.
The same bridge idea can measure more than strain. Load cells, pressure transducers and torque sensors convert their mechanical quantity into strains at carefully chosen locations. Calibration then maps the combined bridge output to force, pressure or torque. The sensor’s mechanical geometry is part of its mathematics: it concentrates useful strain while staying elastic and rejecting unwanted load directions.
Cross-talk tests apply one load component at a time and measure outputs on every channel. A calibration matrix can correct small coupled responses, but large cross-talk may indicate a poor mechanical design. Software correction should not become an excuse to ignore overload strength, fatigue or temperature stability.
Clear evidence wins.
Frequently Asked Questions
What does a strain gauge measure?
It responds to deformation along its grid direction through a change in electrical resistance. Strain is inferred using gauge factor and the measurement system.
Why use a Wheatstone bridge?
It converts small resistance changes into a differential voltage and can increase sensitivity or compensate shared effects through its configuration.
What is microstrain?
One microstrain is 10⁻⁶ strain. A reading of 500 με means a length change of 500 parts per million.
Why is temperature compensation needed?
Temperature changes resistance and can produce apparent strain through mismatch among gauge, specimen and adhesive.
Does a full bridge eliminate temperature error?
It can reject common thermal changes under suitable conditions, but gradients, mismatched gauges and installation differences leave residual error.
What is shunt calibration?
It introduces a known electrical resistance change to check the bridge and signal chain. It does not verify mechanical strain transfer.
Can strain be converted directly to stress?
Only with an appropriate material and stress-state model. Simple σ=Eε is limited to suitable uniaxial linear-elastic conditions.
Is this enough to test a safety-critical structure?
No. Safety-critical work requires qualified design, procedures, calibrated equipment, appropriate standards and reviewed decisions.
Useful Next Reading
- NASA: compensated high-temperature strain gauge for an official bridge-compensation example.
- NASA: Wheatstone full-bridge strain-sensing circuit for recent technical documentation.
- Gauge Blocks, Calibration Chains and Measurement Uncertainty for traceability and uncertainty.
- Spectrophotometry, Beer–Lambert Law and Calibration Curves for another measurement chain.
- Manufacturing Tolerances, Measurement and Quality Control for decisions near limits.
Closing Perspective
Strain measurement joins mechanics, materials, circuits, data and judgement. The gauge makes a small physical change electrical; the bridge makes it visible; calibration makes it useful; uncertainty makes it honest.
For a student, the transferable lesson is powerful: follow the complete chain from what happened in the world to what the instrument observed and what the final number is allowed to claim.
A Deeper Case Study: Designing a Bicycle-Crank Measurement
Imagine a student team wants to compare how smoothly two riders apply force through a bicycle crank. The goal is not to certify a real bicycle or make a safety claim. It is to design a defensible measurement chain on a laboratory crank or safe test beam. That chain begins long before a voltage appears on a screen.
Define the measurand before choosing the sensor
“Pedalling force” is too vague. A strain gauge bonded to a crank measures local surface strain along the gauge grid. Converting that result into crank torque requires a mechanical model and a known location. If the crank is approximated as a beam, surface strain depends on bending moment, cross-section and elastic modulus. A real crank may also twist, curve and carry a force that changes direction.
A better measurand statement is: “the bending moment about this axis, inferred from a calibrated bridge over this load and temperature range.” That sentence identifies the quantity, method, range and limitation. It also prevents a common mistake: treating a local strain reading as a direct universal force reading.
Choose gauge locations from the strain field
Two gauges placed on opposite faces can experience strains of similar magnitude and opposite sign during bending. A half bridge can wire those changes so their electrical effects add. This increases bending sensitivity and rejects some common influences. It can also reduce response to an unwanted uniform temperature change if the gauges are well matched and share the same thermal environment.
Location matters. Near a hole, fillet or bonded joint, strain may change rapidly with position. A short grid reveals a more local value; a longer grid averages the field. The team should sketch the expected tension and compression zones, then justify why each gauge direction and grid length match the measurement question.
Predict the signal before selecting electronics
Suppose calibration loading is expected to produce plus or minus 700 microstrain, the gauges have nominal factor 2.0 and excitation is 5 V. For a simple bending half bridge with two equally active gauges in an appropriate configuration, a first-order output magnitude is roughly twice the comparable quarter-bridge response. The approximate full-scale bridge signal is therefore on the order of 3.5 mV, subject to the actual circuit convention.
This estimate guides the amplifier. A gain of 500 would map 3.5 mV to about 1.75 V. That may suit a bipolar acquisition range, but the design must leave room for bridge offset, overload and transients. Choosing gain from the expected signal is better than increasing gain until the graph looks large.
Calibrate the complete chain
Clamp the test article consistently and apply several known loads in increasing and decreasing order. For each load, record bridge output after a defined settling time. Fit a line only after plotting the data. The slope becomes sensitivity, the intercept reveals zero offset, and residuals show where the line fails to describe observations.
If the loading and unloading curves separate, report hysteresis rather than hiding it inside one fitted line. Repeat the sequence to estimate repeatability. A shunt-calibration resistor can check bridge and electronics response, but it does not test the adhesive bond, structural load path or strain transfer. Mechanical calibration remains essential for the intended measurand.
Convert a time series into a fair comparison
Once voltage has been converted to torque, the team still needs statistics. Peak torque alone rewards one brief spike. Mean torque depends on the selected interval. Root-mean-square torque gives extra weight to large deviations. An angularly resolved plot can reveal where torque rises and falls through each crank revolution.
A fair rider comparison should align cycles consistently, state cadence, preserve sign convention and show variability across several revolutions. Confidence comes from repeat observations, not from an impressive number of decimal places.
Build an uncertainty statement
Possible contributors include calibration-load uncertainty, fitted-slope uncertainty, bridge nonlinearity, repeatability, temperature drift, amplifier offset, alignment and fixture variation. Put all contributions into the same output quantity before combining them. If one contribution dominates, improving a tiny secondary term will not materially improve the result.
The final statement might read: “Within this fixture, temperature interval and calibration range, the inferred bending moment is X with an expanded uncertainty of Y under the stated coverage convention.” This is more useful than calling the instrument “accurate” without a number or condition.
Three Reasoning Traps Worth Catching
A stable zero is not proof of correct sensitivity
A bridge can return to zero yet still have the wrong gain because of poor strain transfer, an incorrect gauge factor or a calibration error. Zero, sensitivity and linearity are different properties.
More excitation is not free sensitivity
Bridge output rises with excitation, but power dissipation also rises and can heat the gauge. Self-heating depends on gauge resistance, heat sinking and environment. Excitation is an optimisation variable, not a volume knob.
A digital filter cannot repair a biased model
A low-pass filter can reduce high-frequency noise, but it cannot correct a misaligned gauge, a temperature bias or an incorrect force-to-strain relationship. Mathematics helps distinguish random variation from systematic error so the remedy targets the cause.
Fifteen Applied Investigations
These concise investigations turn the article's mathematics into controlled questions. Use safe models, change one variable at a time, label units, graph the evidence and keep every conclusion inside the tested conditions.
| Investigation | What to vary or calculate | Transfer insight |
|---|---|---|
| 1. Paper-beam strain map | bend a paper strip gently and mark tension and compression sides; compare sign and distance from the neutral axis. | Strain depends on position even under one bending moment. |
| 2. Gauge-factor calculation | use GF=2.05 and strains of 100, 500 and 1000 microstrain; calculate fractional resistance change GF times strain. | A tiny mechanical strain produces a still smaller electrical change. |
| 3. Quarter-bridge sensitivity | use a 5 V excitation and the small-change quarter-bridge approximation; estimate millivolts of output for 500 microstrain. | Bridge output must be amplified without losing sign or stability. |
| 4. Half-bridge bending | place one active gauge in tension and one in compression; sum their signed bridge effects. | Opposite mechanical strains can reinforce electrically. |
| 5. Full-bridge comparison | assign four active gauges to two tension and two compression arms; compare sensitivity with quarter and half bridges. | Configuration controls sensitivity and compensation. |
| 6. Temperature dummy gauge | give active and dummy gauges the same thermal resistance change; subtract the common effect in adjacent bridge arms. | Temperature compensation works only when conditions are genuinely shared. |
| 7. Lead-wire error | add equal and unequal lead resistances to a remote quarter bridge; compare zero shift and sensitivity. | Wiring belongs inside the measurement model. |
| 8. Shunt calibration | place a known resistor across one bridge arm; calculate the equivalent fractional resistance change. | An electrical check tests much of the signal chain but not structural strain transfer. |
| 9. Amplifier gain | map a 2 mV bridge signal into a 2 V acquisition range; calculate gain and headroom. | High gain magnifies offset and noise as well as signal. |
| 10. ADC resolution | compare 12-bit and 16-bit conversion across the same voltage span; convert one code into volts and then microstrain. | Nominal bit depth is not the same as effective resolution. |
| 11. Calibration line | fit load against bridge output for increasing and decreasing loads; calculate slope, intercept and residuals. | Calibration reveals sensitivity and nonlinearity. |
| 12. Hysteresis check | compare readings at the same load on loading and unloading paths; plot their difference. | Adhesive, structure and fixture behaviour can make history matter. |
| 13. Uncertainty budget | combine calibration, repeatability, temperature and alignment terms; convert all components into microstrain before root-sum-of-squares. | Unit discipline makes uncertainty contributions comparable. |
| 14. Rosette transformation | use three gauge directions on a simulated plane-strain state; compare normal strains under a rotated axis. | Directional measurements reconstruct a two-dimensional state. |
| 15. Safe decision | interpret a strain alarm near its threshold; include uncertainty, drift and transient duration. | A threshold needs a decision rule rather than one displayed digit. |