A linear dielectric model assumes that response scales with excitation. If changing test voltage changes the normalized curve, a single small-signal model may no longer describe the measurement. First distinguish that effect from different temperature, prior charge or instrument range. [1]
Compare equivalent histories
Use only voltages and procedures permitted for the equipment and instrument. This comparison does not authorize raising a field-test voltage. Match the insulation boundary, charge history, measuring window and material condition before comparing amplitudes.
| Quantity | Linear comparison | Confounder to check |
|---|---|---|
| Time-domain current | Compare i(t)/U for identical histories | Residual charge, offset and range switching |
| Frequency-domain admittance | Compare I(f)/U(f) | Actual terminal voltage and phase uncertainty |
| Repeat at the same voltage | Assess repeatability | Temperature or condition drift |
| Return to an earlier voltage | Test history dependence | State changed by intervening excitation |
Interpret a difference within the uncertainty
Current divided by voltage has units of siemens. A repeatable difference beyond the combined measurement uncertainty can suggest field-dependent conduction, another nonlinear process or a boundary artifact. The comparison alone does not uniquely identify the mechanism.
A near-zero signal at a lower voltage may amplify the relative effect of background current. Apparent nonlinearity then can originate in the measurement chain. Inspect raw current, actual voltage and the instrument’s stated range performance.
Propagate uncertainty through i/U, including current, actual voltage and correlations. Then compare the normalized values on their common history and range; apparent extra displayed digits do not establish a resolved difference. [2]
Compare the voltage-normalized response
Assumptions: matched 100 s charging points at 20 °C give 20 nA at 200 V and 55 nA at 500 V. Independent standard uncertainties are 1% of each current and 0.5% of each voltage, with no shared correction. Normalized values are 100 and 110 pS; each has relative standard uncertainty √(0.01² + 0.005²) = 1.118%, giving u₁ = 1.118 pS and u₂ = 1.230 pS. Their 10 pS difference has uΔ = 1.662 pS. With assumed coverage factor 2, its interval is 6.676–13.324 pS and excludes zero. Reject one constant-normalized-response model over these permitted amplitudes until range/history artifacts are excluded. The result resolves a discrepancy and does not identify its insulation mechanism. [2]
References
- IEEE, IEEE C57.161-2018, IEEE Guide for Dielectric Frequency Response Test.
- B. N. Taylor and C. E. Kuyatt, NIST Technical Note 1297, Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results, 1994, sections 5 and 6.