Custom Humanoid Actuator Current Sensing Testing

Screen phase-current range, shunt heat and sampling timing. Use the evidence below to turn the result into a bench test plan.

Custom Humanoid ActuatorCreated · Reviewed · Illustrative engineering model

Current sensing screening tool

Edit current, inspect span and heat, then copy a draft bench plan. All defaults are illustrative.

Required current inputs

0.5–120 A RMS. Phase RMS.

1–250 A peak. Instantaneous, not RMS.

Within entered screening budgets

Estimated current span±37.0 A peak
Nominal ADC step20.14 mA/LSB
Shunt heat / element0.648 W
Timing margin800 ns
Architecture and example selection
Review electrical, thermal & timing assumptions (required)

Every field is required. Replace example values with datasheet limits and measurements. Temperature uses a 25°C resistance reference. Low-side heat uses conduction factor 1 as an upper bound.

Electrical, thermal and timing inputs

0.1–50 mΩ. Current sense resistor value installed in the phase or low-side bridge.

0.05–5 %. Resistor manufacturing resistance tolerance at reference 25°C.

0–250 ppm/°C. Shunt resistance thermal drift rate per degree Celsius.

1–200 V/V. Voltage gain configured on the differential sense amplifier or CSA IC.

0.05–5 %. Worst-case gain non-linearity and resistor network mismatch.

0–2000 μV. Absolute worst-case input offset including temperature drift; zero only for a proven residual budget.

10–18 bits. Native conversion bit resolution of the embedded FOC DSP/MCU.

1.8–5 V. Clean analog reference voltage feeding the microcontroller ADC converter.

5–120 °C/W. Steady-state element-to-local-ambient thermal resistance for the actual PCB; confirm by measurement.

-40–95 °C. Operational ambient air temperature inside the sealed joint cavity.

10–100 kHz. Switching frequency of GaN/MOSFET three-phase power inverter stage.

50–1500 ns. Hardware interlock dead time preventing bridge shoot-through during switching.

0–10000 ns. Time required for differential analog signal to settle within 0.1% after PWM edges.

1–10000 ns. Use the configured MCU acquisition time and source-impedance requirements; not inferred from PWM frequency.

1–100000 ns. Shortest interval from the commanded switching transition to the next disruptive edge, including dead time and analog recovery; verify the selected channels.

0–1 V. Reserve at each ADC rail for amplifier swing, reference error and transient margin. Confirm with the selected part.

40–170 °C. Illustrative budget, not a solder or component rating. Use the part derating curve and qualified assembly limit.

0.01–50 W. Allowed continuous power at the actual terminal/board temperature; verify the manufacturer derating curve.

Span / required peak
1.23×
First-order element temperature
64.44 °C
TCR magnitude vs 25°C
0.197%
Input-offset equivalent current
75.00 mA
Single-channel gain error budget
1.197%
Total shunt heat (installed count)
1.944 W
Conservative timing budget
1200 ns
Timing budget / PWM period
3.00% (not a blind angle)
  • These are screening budgets only. Noise, ADC INL/ENOB, reference drift, parasitics, peak pulse energy, actual torque ripple and safety integrity require measured evidence.

Next: verify the entered budgets, measure worst-case operating points, and agree acceptance limits using the checklist below. A green budget does not establish motor, assembly or safety suitability.

Four decisions before releasing a current-sensing design

Decision summary: evidence and required next action
DecisionBasisNext action
Reserve actual output swingADC rails alone overstate usable span; amplifier output swing and reference bias matter [1].Inspect the selected part and scope the largest phase-current transient.
Track temperature from the resistance referenceAmbient temperature and self-heating both change shunt resistance [4].Measure the element and terminal temperatures on the assembled PCB.
Prove sample validity across operating pointsIn-line recovery and low-side vector duration both affect valid sampling [2,3].Capture ADC timing and amplifier output at the worst modulation.
Keep screening separate from safety validationA budget result does not establish an application performance level [7].Agree acceptance limits and validate the complete control system.

Supporting references: [1] output limits, [4] temperature effects, [2] PWM recovery, [3] low-side design, [7] safety scope.

Release evidence has three separate gatesScreenSpan / heat / timing budgetsMeasureNoise / drift / transientsApproveAgreed limits + recordsThe interactive result covers the first gate only.
Release evidence has three separate gates. Explanatory diagram, not measured device data. Scroll horizontally on small screens.

Choose topology using testable trade-offs

In-line sensing measures the phase conductor; low-side sensing needs a suitable inverter switch state. Neither architecture is automatically qualified for a humanoid joint. The intended control envelope and measured feedback quality determine suitability. PWM recovery and low-side reference designs provide distinct engineering constraints.

Topology comparison: what must be demonstrated
ArchitectureStrength / cost driverBoundaryTest evidence
In-line shuntDirect phase measurement; amplifier common-mode capability adds costSwitching recovery, output swing, shunt heatScope settling across bus voltage, temperature and modulation
Low-side three-shuntGround-referenced sensing; extra channel and selection logicValid sample channels depend on switching sequenceVerify shortest window and channel selection in every sector
Low-side two-shuntOne fewer resistor and sensing channelReconstructed third phase is not independent evidenceCheck simultaneous channel validity and reconstruction error
Magnetic / isolated ICCan reduce insertion resistance; isolation and packaging add constraintsConductor loss, noise, external fields and isolation are part-specificRequest sensitivity, primary resistance, noise spectra and exact isolation rating

No current-price survey or supplier quotation is available here. Compare the complete BOM, layout area, calibration effort, thermal integration and firmware validation rather than a claimed universal dollar saving. Magnetic device selection needs the exact ordering code. Manufacturer catalog

Measurement chain and error entry pointsPhase shuntR, tolerance, temperatureAmplifierGain, offset, PWM recoveryADC + firmwareReference, noise, timingMeasure the complete chain; ADC bit count alone does not specify accuracy.
Measurement chain and error entry points. Explanatory diagram, not measured device data. Scroll horizontally on small screens.

Calculation methods, assumptions and known unknowns

The tool is deterministic and runs locally. It assumes a mid-reference bidirectional output, a 25°C nominal shunt, symmetric rail reserve and positive worst-case error magnitudes. Defaults are scenarios, not component recommendations. Equations below define this screening model; they do not replace datasheet limits.

Reproducible model equations and limitations
OutputEquation / conventionLimitation
Nominal shunt heatP = I_RMS² × R25; N = 2 for two-shunt, otherwise 3Low-side conduction factor is conservatively 1; actual waveform and phase heating may differ
First-order element temperatureT = T_local + P × RθIgnores resistance feedback, mutual heating and transient thermal impedance
TCR magnitudeeT (%) = |T − 25°C| × TCR / 10,000Magnitude budget; not signed compensation or post-calibration residual
Bounded current spanRmax = R25(1 + (tolerance + eT)/100); Gmax = G(1 + gain_error/100); Imax = max(0, ((Vref/2 − rail_reserve)/Gmax − |Vos|)/Rmax)First-order hot resistance estimate; reserve must cover real output swing and reference/transient errors
Nominal code widthΔI = Vref / (2^bits × G × R25)Code width is not noise, effective resolution or total accuracy
Input offset equivalentI_offset = |Vos| / R25Single-channel input-referred error; no mechanical torque amplitude implied
Single-channel gain budgettolerance + amplifier gain error + eTArithmetic magnitude budget; phase mismatch and correlated errors need channel data
Conservative timing sumt_required = dead_time + settling/filter_delay + ADC_acquisitionCompare with the entered transition-to-next-edge interval, including recovery; overlap and trigger placement depend on hardware

The 15% span-reserve warning, 95°C temperature budget and 1 W per-shunt budget are illustrative project choices. Replace them with approved design limits. Missing noise, ADC errors, pulse duration, motor saliency, mechanical compliance and safety analysis stay unknown; the tool reports no torque-ripple amplitude or blind-sector angle.

Physical thermal and sense-tap considerations are described by ROHM and TI. The thermal estimate uses nominal cold resistance, so it can understate hot dissipation; confirm the full thermal model before accepting a near-limit result.

Thermal drift and Kelvin routing must be tested together

A 20 A RMS current through a 2 mΩ shunt gives 0.8 W nominal per phase. With 25°C/W element-to-local-ambient resistance and 30°C ambient, the first-order element estimate is 50°C. A 50 ppm/°C TCR then gives a 0.125% magnitude relative to 25°C. These are derived example values, not measured product ratings.

Illustrative 50 ppm/°C resistance drift from a 25°C reference25°C / 0%55°C / 0.15%85°C / 0.30%Derived: |ΔR/R| (%) = TCR × |T_element − 25°C| / 10,000
Illustrative 50 ppm/°C resistance drift from a 25°C reference. Explanatory diagram, not measured device data. Scroll horizontally on small screens.

Sense at the manufacturer-specified element taps. Copper and solder included in a two-wire pickup can introduce additional temperature-dependent resistance. Validate the actual PCB because a nominal resistor specification does not characterize the complete path. PCB/TCR evidence · shunt layout guidance

Separate power and voltage-sense pathsPower conductorCarries phase currentShunt elementSense at specified Kelvin tapsDifferential inputMatched short sense routingKeep power-path copper and switching coupling out of the sensed voltage.
Separate power and voltage-sense paths. Explanatory diagram, not measured device data. Scroll horizontally on small screens.

Measure a valid sampling interval, not a guessed blind angle

At 25 kHz the PWM period is 40,000 ns. An example 300 ns dead time, 650 ns recovery and 250 ns acquisition gives a conservative 1,200 ns budget, or 3% of a period. Against an entered 2,000 ns available interval, the margin is 800 ns. This ratio alone says nothing about electrical blind angle or maximum modulation.

Conservative timing budgetSwitching edgeDead time / blankingAnalog recoveryAmplifier + filter settlingADC acquisitionConfigured apertureCompare the sum with the interval from the switching command; timing overlap depends on hardware.
Conservative timing budget. Explanatory diagram, not measured device data. Scroll horizontally on small screens.
Timing evidence to request at each operating extreme
Operating pointCaptureDecision
High modulation / field weakeningPhase channel selection, PWM state, amplifier output, ADC apertureAre at least the required channels simultaneously valid?
High bus voltage / fast edgesSwitching node and sense output with suitable probesDoes common-mode recovery settle inside the actual window?
Hot and cold electronicsSame acquisition settings, source impedance and filterDoes the worst-case residual stay within the agreed error budget?
Changed firmware or PWM modeTrigger phase, blanking and reconstruction logsRepeat the validation; timing does not transfer automatically

In-line measurement also needs recovery after PWM transitions. TI explains this limitation. Low-side sample availability depends on vector duration and the implementation. TI reference design

Bench protocol: zero, gain, temperature and switching recovery

Traceable calibration workflowKnown zeroConfirm no residual currentBipolar referencesFit offset and slopeIndependent checkRecord residual + uncertaintyRepeat at temperature; retain raw data and the original calibration coefficients.
Traceable calibration workflow. Explanatory diagram, not measured device data. Scroll horizontally on small screens.
Draft bench procedure: agree numeric limits before testing
StepMethodRecord / acceptance basis
1. Establish real zeroEnsure no motor or regenerative current; allow settlingPer-channel mean, noise spread, temperature and uncertainty; agreed offset budget
2. Apply electrical referencesInject positive and negative current points with a calibrated current referenceRaw ADC codes, slope/intercept, reference uncertainty and independent residual points
3. Repeat at temperatureEvaluate cold start, thermal soak and rated dutyShunt/terminal temperatures, gain drift and device derating against project limits
4. Validate dynamic samplingCapture switching edges, amplifier output and acquisition intervalSettling residual in each switching sector and worst operating point
5. Check overload separatelyUse a controlled pulse plan within qualified hardware limitsPulse width, repetition, protection path and resistor pulse-energy limits
6. Verify final assemblyRead back coefficients; repeat an independent checkSerial, PCB/firmware versions, calibrated instruments, limits and deviations

Electrical current calibration and current-to-torque verification answer different questions. Use the electrical reference to calibrate sensing; use a calibrated shaft-torque sensor and controlled load to study mechanical response. A torque measurement alone cannot isolate current-channel gain error.

Prepare a traceable bench-test RFQ

Torque spectra: identify contributions before assigning a cause

A stationary phase-current offset becomes a rotating disturbance after the Park transform; unequal channel gains can introduce a second-harmonic component. Their actual torque amplitudes also depend on channel signs, transformation conventions, current control dynamics, motor properties and gearing. This page therefore avoids converting a single offset or tolerance number into a certified ripple figure.

Separate electrical errors from mechanical rippleCurrent logsSynchronized phase channelsAngle / speedElectrical frequency referenceShaft torqueCalibrated independent sensorOffsets may produce 1× and imbalance 2× components; amplitude needs system measurements.
Separate electrical errors from mechanical ripple. Explanatory diagram, not measured device data. Scroll horizontally on small screens.
Separate known electrical quantities from unmeasured system effects
QuantityKnown from this toolEvidence still required
Nominal ADC code widthComputed from Vref, bits, gain and shuntNoise spectrum, ENOB, INL, reference drift and sample settling
Input-offset equivalent currentVos divided by nominal resistancePer-channel residuals across temperature and startup conditions
Phase mismatchNot calculatedMatched channel measurements, covariance and calibration residuals
Shaft torque rippleNot calculatedSynchronized torque/current/angle logs, sensor bandwidth and uncertainty
Force / collision sensitivityNot calculatedTransmission friction, compliance, contact geometry and validated system response

Use electrical frequency from measured rotor angle and pole pairs when comparing spectra. Distinguish sensing errors from cogging, inverter dead time, gearing and load vibration; an isolated spectral peak is insufficient evidence of a sensing fault.

Worked decision examples: assumptions, result and next action

All five cases below are derived illustrations. None is a customer test record or a released actuator specification. Presets in the tool are separate editable examples and may intentionally reveal a budget conflict.

Current sensing screening tool with 18 A phase RMS and 30 A instantaneous peak example inputs and calculated budgets
Actual software execution captured 2026-11-03: default in-line example, 18 A RMS, 30 A instantaneous peak, 2 mΩ, gain 20 and 3.3 V reference. This shows reproducible screening arithmetic, not measured actuator performance or a customer test. The capture contains only the tool; site navigation is outside the image and floating site controls are hidden for this isolated capture.
Illustrative cases with reproducible arithmetic
Scenario / assumptionsDerived resultDecision / next measurement
Case 1: Wrist, 8 A RMS, 5 mΩ shuntNominal power = 8² × 0.005 = 0.320 W per in-line elementCheck noise and conductor loss; small code width alone does not establish gentle-touch sensitivity
Case 2: Elbow, two low-side shunts, 16 A RMS, 2.5 mΩDuty-factor-1 heat bound = 0.640 W each, 1.280 W for twoMeasure actual conduction and channel validity; no unconditional suitability rejection
Case 3: Knee, 75 A instantaneous peak, 1 mΩ, gain 20, 3.3 V referenceIdeal rail span = ±82.5 A; 0.15 V reserve reduces nominal span to ±75 A before error allowancesReserve is already consumed; revisit gain/resistance and verify pulse overload
Case 4: Hip holding, 35 A phase RMS, 0.75 mΩ, 55°C local ambient, 18°C/WP = 0.91875 W; first-order T = 71.5375°C; at 20 ppm/°C, drift magnitude = 0.0931% vs 25°CFor a stationary rotor, measure actual per-phase RMS; do not assume equal three-phase heating
Case 5: Finger, 15 mΩ, gain 20, 3.3 V, 16-bit ADCNominal code width = 3.3 / (65536 × 20 × 0.015) = 0.1678 mA/LSBCompare with measured noise and calibration residual; no sub-gram force claim follows

Misuse, cost and safety limits

For an ideal three-wire motor, ia + ib + ic = 0. An independently measured residual may support diagnostics, but reconstructing ic = −ia − ib makes that sum zero by construction. It cannot independently detect a failed channel.

DC bridge current is a different quantity. With ideal upper-switch states sa, sb, sc, iDC = sa·ia + sb·ib + sc·ic. A source-side measurement may also include DC-link capacitor and auxiliary currents. Compare synchronized switching-state models and independent measurements rather than equating bus current with the phase sum.

From diagnostic idea to validated safety functionHazard assessmentSpecify required functionArchitectureIndependence + fault analysisValidationDemonstrate achieved integrityA current-sum check or a fast comparator alone does not establish a performance level.
From diagnostic idea to validated safety function. Explanatory diagram, not measured device data. Scroll horizontally on small screens.

ISO 13849-1 defines a control-system design methodology; its public scope explicitly leaves application safety functions and required performance levels to application-specific determination. No diagnostic coverage percentage, stop time, PL, SIL or certification is established by this page. Official ISO scope

Risk and trade-off register with practical mitigation
RiskConsequenceMitigation / alternative
Using within-budget as product acceptanceUnmodeled thermal feedback, noise or overload can dominateRequire measured component/assembly limits and a signed acceptance protocol
Reducing shunt resistance only to cut heatSignal-to-noise and offset-equivalent current worsenCompare total error at the required bandwidth; consider an appropriate magnetic sensor
Selecting a cheaper topology without validation costFirmware and calibration effort may exceed BOM savingsQuote channels, test time, fixture costs, thermal integration and lifetime validation separately
Reusing one sampling plan across operating modesHigh modulation, regeneration or changed filtering invalidates measurementsRevalidate operating extremes and firmware revisions; limit the envelope until proven
Assuming sensing alone ensures safe torque limitingCommon-cause and mechanical hazards remain unaddressedComplete the application hazard assessment and independent safety validation

Frequently asked questions

Tool scope

Is a within-budget result a hardware pass?

No. It compares entered values with illustrative budgets. It cannot establish component derating, noise performance, control stability, pulse survival or safety integrity.

Which current convention does the tool use?

Rated current is phase RMS over the evaluated waveform. Maximum instantaneous current is the largest absolute phase current including transients. A sinusoidal amplitude is RMS × √2; do not apply that conversion to a DC holding phase.

Why does the isolated sensor mode show no calculated result?

A magnetic sensor uses sensitivity in V/A, offset, conductor loss and delay from its own datasheet. Applying a shunt resistance and amplifier gain would give an unrelated result. Use the manual RFQ checklist instead.

Does a 16-bit ADC guarantee better torque precision?

No. Nominal code width excludes noise, reference error, INL, effective resolution and motor mechanics. Measure the complete acquisition chain at the intended bandwidth.

Topology

Is in-line current sensing mandatory for humanoid joints?

No universal topology requirement is established here. In-line measurement avoids low-side switch-conduction restrictions but still needs PWM transient recovery. Select topology using measured sampling availability, losses, cost and the required control performance.

When can two low-side shunts be used?

For a three-wire motor, the third current can be reconstructed from two synchronized valid measurements. Validate which channels are observable at every switching sector and operating point. Reconstruction is not independent third-channel redundancy.

Are magnetic current sensors lossless?

No. Devices with an integrated primary conductor have insertion resistance and I²R heat, and the electronics draw supply power. External-field and isolation limits also depend on the selected component.

Thermal

Why is thermal drift referenced to 25°C?

The tool assumes shunt nominal resistance and tolerance are specified at 25°C. It uses the estimated element temperature relative to that reference, including local ambient and self-heating. Use the actual calibration reference for a calibrated model.

Is 95°C a universal shunt temperature limit?

No. It is the default example project budget. The allowed temperature depends on the resistor, terminals, PCB, solder assembly, derating curve and qualified service profile.

Does the heat result cover short current peaks?

No. The first-order estimate uses rated RMS and nominal resistance. Pulse duration, repetition, transient thermal impedance and pulse-energy derating must be assessed separately.

Timing

Can a blind-sector angle be calculated from PWM frequency alone?

No. Switching sequence, modulation, current waveform, channel selection, blanking and ADC timing are needed. The tool compares an entered interval including switching recovery with a conservative timing sum; it does not calculate blind-sector angles.

How should the available sampling interval be measured?

At the worst modulation, speed and bus voltage, measure from the commanded switching transition to the next disruptive edge for the selected channels. This available interval includes dead time and analog recovery. Compare it with the complete timing sum; do not enter only the already-settled portion.

Calibration

Is a dynamometer a current calibration reference?

Use a calibrated electrical current reference for current-channel gain and offset. A dynamometer with a traceable torque sensor independently checks current-to-torque behavior, friction and harmonics.

How should startup zero calibration be tested?

First establish that actual current is zero; disabled gates alone may not ensure this during motion or regeneration. Record mean offset, spread, temperature and settling. Reject unexplained drift before enabling the drive.

Can electrical-frequency peaks identify sensing faults by themselves?

No. Offset and channel imbalance can contribute fundamental and second-harmonic components, but torque spectra also include cogging, mechanics, inverter nonlinearity and controller dynamics. Correlate synchronized electrical and mechanical records.

Safety and RFQ

Does a current-sum check establish PL d?

No. A safety function needs application-specific requirements, architecture analysis, reliability evidence, diagnostic validation and common-cause assessment. ISO 13849-1 does not prescribe the required performance level for every application.

Is DC-bus current the sum of phase currents?

No. For an ideal three-wire motor, ia + ib + ic = 0. With upper-switch states sa, sb and sc, ideal instantaneous bridge DC current is sa·ia + sb·ib + sc·ic. DC-link capacitor current and measurement position must also be considered.

What should accompany a supplier acceptance report?

Provide serial number, PCB and firmware revisions, instrument calibration and uncertainty, current convention, raw zero/gain data, thermal operating point, sampling scope captures, agreed limits and deviations. A calculator export is only a draft test request.

Primary sources and evidence boundaries

Links checked 2026-11-03. Product availability, pricing and certification are not inferred from these sources. Manufacturer examples explain design constraints; numeric examples elsewhere are calculations under stated assumptions. The ISO citation uses its public scope, not a claim of access to the full paid standard.

Next scheduled review: . Recheck earlier when the selected sensor, PCB layout, firmware, PWM mode, calibration procedure or source revision changes.

  1. TI INA241x datasheet, Rev. D

    Part-specific gain, output swing, common-mode limits and PWM recovery. These values are not universal acceptance criteria.

  2. TI SBOA160D: Low-Drift, Precision, In-Line Motor Current Measurements With PWM Rejection

    Why in-line signals still suffer common-mode switching disturbances and settling delay.

  3. TI TIDA-00778 current-sensing reference design

    Examples of one-, two- and three-shunt sensing and active-vector timing. Reference-design results do not qualify this actuator.

  4. ROHM: Effect of PCB Design on Temperature Coefficient of Resistance

    Ambient, self-heating and copper sense-tap placement affect resistance error.

  5. TI SBAA460: Shunt Resistor Selection for Isolated Data Converters

    Shunt selection, Kelvin layout, thermal effects and calibration considerations.

  6. Allegro integrated current sensor product catalog

    Sensor-specific sensitivity, bandwidth, primary resistance and isolation variants. Request the exact ordering-code datasheet.

  7. ISO 13849-1:2023 official scope and abstract

    Safety-related control design methodology; the public abstract does not prescribe application PLr or sensor trip limits.

Turn the screening into an agreed current-sensing test plan

Send the joint operating waveform, phase-current convention, bus voltage, sensing topology, exact components, thermal envelope, PWM/ADC configuration and required uncertainty. Attach the copied draft plus drawings and available scope captures. Acceptance values, test capability, cost and lead time require project review.

Minimum RFQ evidence package
PackageIncludeIf unavailable
Electrical and thermalRMS/instantaneous peak, pulse profile, component codes, PCB layout and temperature limitsMark unknown; request a characterization plan before quoting acceptance
AcquisitionSwitching sequence, trigger, filters, ADC settings, firmware revision and raw capturesRestrict the evaluated operating envelope and measure the missing windows
Acceptance and traceabilityAgreed limits, current/torque references, instrument uncertainty, serial and revision fieldsStart with a draft protocol; no performance or certification claim

Inquiry Email

[email protected]

Include target torque/speed, quantity, and delivery location.

Application Engineer

WhatsApp Engineering Support

Talk directly about drawings, torque-speed targets, and RFQ data gaps.

Related engineering checks: continuous torque testing · compliance testing · contact engineering.