Measurement Errors, Uncertainty, and Instrument Characteristics

Introduction

No measurement is perfectly exact. Every measured value is an estimate influenced by the measuring instrument, the observer, the environment, the measurement procedure, and the quantity being measured. Electrical engineering decisions nevertheless depend on measurements: protection systems must detect abnormal currents, power systems must maintain voltage within limits, and electronic products must satisfy performance tolerances. A sound engineer must therefore do more than read an instrument display. The engineer must identify possible errors, estimate uncertainty, assess instrument behaviour, and report results responsibly.

This chapter develops a systematic treatment of measurement errors and uncertainty. It distinguishes systematic, random, and gross errors; introduces statistical treatment of repeated observations; develops uncertainty propagation for calculated quantities; and explains static and dynamic characteristics. First-order and second-order instrument models are used to describe transient response, time delay, overshoot, damping, and bandwidth.

Learning Objectives

After studying this chapter, the student should be able to:

1. Distinguish accuracy, precision, error, uncertainty, sensitivity, resolution, and related terms.
2. Classify measurement errors as systematic, random, or gross and select suitable corrective actions.
3. Calculate mean, deviation, variance, standard deviation, and confidence-related quantities for repeated observations.
4. Propagate independent and correlated uncertainties through mathematical relationships.
5. Explain calibration, loading effect, hysteresis, dead zone, drift, repeatability, and reproducibility.
6. Model first-order and second-order instruments and interpret their time-domain and frequency-domain behaviour.
7. Select an instrument and measurement procedure appropriate to a stated engineering requirement.
8. Report measured results with appropriate significant figures and an explicit uncertainty statement.

1. Fundamental Measurement Concepts

A measurement compares an unknown quantity with an accepted reference. If the measurand is the voltage across a resistor, the measurand is not simply “the display reading”; it is the voltage defined under specified connection, loading, frequency, temperature, and time conditions.

The measured result is commonly written as

where x_m is the measured value, x_t is the true or reference value, and e is measurement error. In practice, the exact true value is generally unknown. A calibrated reference value or best estimate is used instead.

The absolute error is

where x_ref is an accepted reference value. The relative error is

and the percentage error is 100 e_r percent. These expressions require a nonzero reference value and consistent units.

Accuracy and precision

Accuracy describes closeness to the accepted value. Precision describes the closeness of repeated observations to one another. A set of readings may be precise but inaccurate when a constant calibration bias is present. Conversely, readings may have a correct average but poor precision because of noise.

A high-resolution display does not guarantee high accuracy. For example, a digital voltmeter may display 6.372 V while its calibration uncertainty is ±0.010 V. The extra digits should not be interpreted as exact information.

2. Classification of Measurement Errors

2.1 Gross errors

Gross errors are major mistakes caused by human or procedural failure. Examples include connecting an ammeter directly across a voltage source, selecting the wrong instrument range, misreading an analogue scale, recording 2.35 A as 23.5 A, or omitting a negative sign.

Gross errors are not usually treated statistically. They should be prevented, detected, and rejected through:

• careful circuit inspection before energization;
• correct range selection and polarity checks;
• independent verification of data entry;
• repeated measurements;
• comparison with physical limits and expected magnitudes;
• automated alarms and plausibility checks.

A value should not be discarded merely because it is inconvenient. Rejection requires a defensible reason, such as documented instrument malfunction or an identified procedural mistake.

2.2 Systematic errors

Systematic errors are repeatable biases that tend to remain constant or vary predictably. They affect accuracy and are not removed simply by averaging more readings. Important sources include:

1. Instrumental error: imperfect calibration, zero offset, friction, finite resolution, aging, or nonlinearity.
2. Environmental error: temperature, humidity, vibration, magnetic fields, supply-frequency variation, and electromagnetic interference.
3. Loading error: the instrument changes the circuit being measured. A voltmeter with finite resistance draws current and may reduce the voltage it is intended to measure.
4. Observational error: parallax, incorrect scale interpolation, or timing bias.
5. Method error: an imperfect measurement model, such as neglecting lead resistance or assuming a waveform is sinusoidal when it is not.

Systematic errors may be reduced by calibration, correction factors, substitution methods, shielding, guarding, environmental control, high-input-impedance instruments, and improved models. A correction should be applied only when its basis and sign are known. If the correction is uncertain, the remaining uncertainty must include that uncertainty.

2.3 Random errors

Random errors vary unpredictably from observation to observation. Thermal noise, quantization fluctuations near a transition, small environmental variations, and timing jitter are common examples. Repetition does not eliminate random error in an individual observation, but averaging can reduce the uncertainty of the mean under suitable assumptions.

Random errors are commonly modelled as independent, approximately normally distributed variations. This model is useful but not universal. Heavy-tailed disturbances, drift, correlated noise, and occasional outliers require additional analysis.

3. Statistical Treatment of Repeated Observations

Suppose n observations of the same measurand are obtained: x₁, x₂, …, x_n. The arithmetic mean is

The mean is an estimate of the central value and is especially useful when random errors have zero mean and observations are independent.

The deviation of observation x_i from the mean is

The deviations satisfy ∑dᵢ = 0 apart from rounding.

The sample variance is

and the sample standard deviation is s = √s². The denominator n − 1 is used because the mean is estimated from the same data; it gives an unbiased estimate of the population variance under independent, identically distributed sampling.

The standard uncertainty of the mean, often called the standard error, is

This expression assumes independent observations with a stable variance. If observations are correlated, increasing n may provide much less improvement than 1/√n suggests.

Worked Example 1: Repeated voltage observations

Five measurements of a nominal 10 V source are 10.02, 9.98, 10.01, 10.00, and 9.99 V.

The mean is

x̄ = (10.02 + 9.98 + 10.01 + 10.00 + 9.99)/5 = 10.000 V.

The deviations are 0.020, −0.020, 0.010, 0, and −0.010 V. Their squared sum is

0.0004 + 0.0004 + 0.0001 + 0 + 0.0001 = 0.0010 V².

Therefore,

s² = 0.0010/(5−1) = 0.00025 V²,

s = 0.0158 V,

and

s_x̄ = 0.0158/√5 = 0.00707 V.

The result based only on repeatability may therefore be reported as approximately 10.000 ± 0.007 V for one standard uncertainty, provided other uncertainty sources are negligible. A calibrated instrument specification may increase the final uncertainty.

Outliers and statistical judgement

An outlier is an observation inconsistent with the assumed measurement process. It may result from a gross error, a transient disturbance, or a legitimate but rare event. Before rejecting it, inspect the circuit, instrument record, time sequence, and environmental conditions. Robust methods such as the median or median absolute deviation can be useful when normality is doubtful, but they do not replace engineering investigation.

4. Uncertainty Evaluation and Propagation

Error is a difference from a reference; uncertainty is an estimate of the doubt surrounding a measurement result. A result should include both an estimate and an uncertainty statement, for example, R = 1.000 kΩ ± 0.005 kΩ at an approximately stated coverage level.

Uncertainty components are often grouped as:

• Type A: evaluated by statistical analysis of repeated observations;
• Type B: evaluated from calibration certificates, manufacturer specifications, previous data, resolution, environmental limits, or engineering judgement.

Type A and Type B describe evaluation methods, not different physical kinds of error. Both contribute to the combined standard uncertainty.

If independent standard uncertainties u₁, u₂, …, uₖ contribute to a result, the combined standard uncertainty is

This root-sum-square form applies when the components are uncorrelated and expressed in the same output units.

For a result y = f(x₁, x₂, …, xₙ), the first-order propagation law is

Here u_y is the standard uncertainty of y, partial derivatives are evaluated at the best estimates, and cov(xᵢ,xⱼ) is covariance. If inputs are independent, covariance terms are zero.

For a product or quotient, relative uncertainty is often convenient. For y = AB/C, independent small uncertainties give

For y = A^p,

These approximations require sufficiently small uncertainties and a locally linear model.

Instrument resolution as an uncertainty component

If a digital display rounds to increments of width q and no better information is available, the rounding error is often modelled as uniform over −q/2 to +q/2. Its standard uncertainty is

This model applies to rounding uncertainty, not necessarily to total instrument accuracy. Manufacturer accuracy specifications must be included separately when applicable.

Worked Example 2: Resistance from voltage and current

A resistor is determined from R = V/I. Let V = 12.00 ± 0.06 V and I = 2.00 ± 0.02 A, with independent standard uncertainties.

The resistance estimate is

R = 12.00/2.00 = 6.00 Ω.

The relative standard uncertainty is

u_R/R = √[(0.06/12.00)² + (0.02/2.00)²] = √(0.005² + 0.010²) = 0.01118.

Thus u_R = 6.00 × 0.01118 = 0.0671 Ω. The result is appropriately reported as approximately R = (6.00 ± 0.07) Ω for one standard uncertainty. The uncertainty should not be reported with excessive digits.

Expanded uncertainty

For practical reporting, an expanded uncertainty is often written

where k is a coverage factor. For an approximately normal result and a large effective number of degrees of freedom, k ≈ 2 is commonly associated with roughly 95% coverage. This is not universally exact; small samples and non-normal distributions require an appropriate coverage analysis.

5. Static Characteristics of Instruments

Static characteristics describe behaviour when the measured quantity is constant or changes slowly enough that dynamic effects are negligible.

Range, span, and calibration curve

The range is the interval of input values for which the instrument is intended to operate. The span is the difference between upper and lower range limits. A calibration curve relates output to input:

where x is input and y is output. For a linear instrument,

where a is zero offset and b is sensitivity. Sensitivity is

or, for a linear instrument, S = b. Its units are output units per input unit.

Nonlinearity is the maximum deviation of the calibration curve from a specified reference straight line, commonly expressed as a percentage of full-scale span. The reference line must be identified because different choices produce different nonlinearity values.

Hysteresis, dead zone, and threshold

Hysteresis occurs when the output for a given input depends on whether the input is increasing or decreasing. It may result from magnetic effects, mechanical friction, or material memory. Dead zone is an input interval over which output does not respond. Threshold is the minimum input required to produce a detectable output change.

Drift and stability

Drift is a gradual change in output or calibration with time, temperature, supply, or other conditions. Zero drift changes the output at zero input; sensitivity drift changes the slope of the calibration curve. Stability describes the ability to maintain characteristics over time.

Loading effect

A measuring instrument is part of the circuit. If a voltmeter with input resistance R_m is connected across a source with Thevenin resistance R_s, the measured voltage is

where V_th is the open-circuit Thevenin voltage. The fractional loading error relative to V_th is

Thus high input resistance reduces voltage-measurement loading. For current measurement, a low-resistance ammeter is preferred, but its burden voltage can disturb the circuit.

Static sensitivity and resolution are different

Sensitivity is a slope; resolution is the smallest distinguishable input change. An instrument can have high sensitivity but poor resolution if its output is noisy, or fine resolution but poor accuracy because of calibration bias.

6. Dynamic Characteristics

Dynamic characteristics describe the response of an instrument to time-varying inputs. They matter whenever the input changes on a time scale comparable with the instrument response.

An instrument may be represented by a differential equation relating input x(t) to output y(t). With zero initial conditions, the transfer function is

where s is the complex Laplace variable. The transfer function describes amplitude and phase changes as functions of frequency.

6.1 First-order instruments

A first-order instrument is modelled by

where τ is the time constant, K is static sensitivity, and x(t) and y(t) are input and output. Its transfer function is

For a step input x(t) = X₀u(t), with zero initial output,

The output reaches 63.2% of its final value at t = τ, approximately 95% at 3τ, and approximately 99.3% at 5τ. The 10–90% rise time is approximately 2.2τ.

For sinusoidal input, the magnitude and phase are

where ω is angular frequency in radians per second. The −3 dB cutoff angular frequency is ω_c = 1/τ, or f_c = 1/(2πτ) in hertz.

6.2 Second-order instruments

A standard second-order model is

where ω_n is natural angular frequency, ζ is damping ratio, and K is static sensitivity. The transfer function is

The damping classification is:

• ζ = 0: undamped;
• 0 < ζ < 1: underdamped, with oscillatory response;
• ζ = 1: critically damped, fastest response without oscillation;
• ζ > 1: overdamped, non-oscillatory but slower than critical damping.

For an underdamped unit step, define the damped natural frequency

and the peak overshoot ratio

where M_p is expressed as a fraction of the final value. This formula applies only to the underdamped case. Increasing damping reduces overshoot but generally changes speed and bandwidth.

The second-order model is important for moving-coil mechanisms, accelerometers, galvanometer systems, sensor mounts, and control-system measurement channels. A measurement system designed with insufficient bandwidth may attenuate fast signals; one with excessive resonance may amplify certain frequencies and distort the result.

Worked Example 3: First-order temperature sensor

A temperature sensor has τ = 4 s and static sensitivity K = 1. A step change in temperature of 20 °C is applied. After 8 s,

y(8) = 20[1 − e^(−8/4)] = 20(1 − e^(−2)).

Since e^(−2) ≈ 0.1353, the output is 17.29 °C above its initial value. The sensor has reached about 86.5% of the final change. Approximately 20 s, or 5τ, is required to reach about 99.3%.

7. Case Study: Current Measurement in a Motor Drive

Background

Consider an illustrative industrial case in which a variable-speed dc motor drive is tested. The drive controller requires current feedback with a nominal value of 25.0 A. Engineers initially use a bench ammeter whose full-scale range is 30 A and whose stated accuracy is ±1% of full scale. The current waveform contains switching ripple at several kilohertz.

Problem

The measured average current is 25.1 A, but the motor occasionally trips on overcurrent. The team must decide whether the trip is caused by actual current excursions or by measurement error and inadequate dynamic response.

Analysis

The stated full-scale uncertainty is approximately ±0.30 A, because

0.01 × 30 A = 0.30 A.

This is not equivalent to ±1% of the reading. In addition, the ammeter has a finite bandwidth and may attenuate switching-related peaks. The measurement connection also introduces burden voltage, which can alter the drive current. Repeated readings of the displayed average current have a small standard deviation, but this demonstrates repeatability rather than absence of systematic error.

The team installs a calibrated shunt and differential oscilloscope probe. The shunt resistance is selected to limit burden voltage, and the oscilloscope bandwidth is verified to exceed the significant ripple frequency. The shunt voltage is converted to current using I = V_sh/R_sh. Calibration uncertainty, oscilloscope gain uncertainty, shunt tolerance, and noise are included in the uncertainty budget.

Approach and solution

The engineers compare three quantities: average current, peak current, and protection-threshold current. They also verify probe polarity, grounding, sampling rate, and anti-alias filtering. The dynamic record reveals brief current peaks of 31 A lasting several milliseconds, although the average is only 25.0 A. The original ammeter did not display these peaks because its response was too slow.

Outcome

The protection circuit is found to operate correctly. The motor controller is retuned to reduce transient demand, and the protection threshold is reviewed against the motor manufacturer's safe operating limits. The final report states both average and peak current, measurement bandwidth, calibration information, and uncertainty.

Learning points

1. Repeatability of a slow display does not establish dynamic accuracy.
2. Full-scale accuracy specifications must be interpreted correctly.
3. Loading and burden voltage can alter the measured circuit.
4. Sampling rate, bandwidth, and probe connection are part of the measurement system.
5. A defensible engineering decision requires an uncertainty budget and a clearly defined measurand.

8. Best Practices, Limitations, and Professional Responsibility

A reliable measurement procedure begins by defining the measurand and acceptable uncertainty. Select an instrument whose range, bandwidth, resolution, input impedance, isolation, and environmental rating meet the requirement. Calibrate against traceable standards when decisions depend on accuracy. Record instrument identification, range, settings, environmental conditions, wiring, date, operator, and raw data.

Do not confuse a manufacturer tolerance with a statistical standard deviation. Do not average away a suspected systematic error. Do not report more significant figures than the uncertainty supports. When a measurement affects safety, protection, compliance, or financial responsibility, preserve raw records and document corrections.

Electrical measurements also involve safety and ethical obligations. Instruments must have suitable voltage and transient categories, fusing, insulation, and grounding. An engineer must not manipulate data to obtain a desired result or silently discard inconvenient observations. Automated and AI-assisted measurement systems can help detect drift, classify anomalies, and estimate uncertainty, but their outputs require validation, traceable data, and human review. An algorithm cannot compensate for an incorrectly defined measurand or an unsafe connection.

Contemporary systems increasingly combine sensors, digital signal processing, networked acquisition, and self-calibration. These systems improve diagnostic capability but introduce sampling, quantization, synchronization, cybersecurity, and software-verification issues. The complete measurement chain—not merely the sensor—must be characterized.

Chapter Summary

Measurement error is the difference between a measured value and an accepted reference, whereas uncertainty quantifies the doubt associated with a result. Accuracy concerns closeness to the reference, while precision concerns agreement among repeated observations. Gross errors arise from mistakes and must be prevented or investigated. Systematic errors produce bias and require calibration, correction, improved methods, or environmental control. Random errors vary unpredictably and can often be characterized statistically.

For repeated observations, the mean estimates the central value, the sample standard deviation describes spread, and the standard uncertainty of the mean is approximately s/√n under independent, stable conditions. Uncertainty propagation uses sensitivity coefficients and covariance terms; the root-sum-square formula is a special case for independent components. Expanded uncertainty requires an appropriate coverage factor.

Static characteristics include range, span, sensitivity, resolution, linearity, hysteresis, drift, loading, and repeatability. Dynamic behaviour is modelled with differential equations and transfer functions. A first-order instrument is governed by its time constant and exhibits exponential step response and low-pass frequency behaviour. A second-order instrument is governed by natural frequency and damping ratio; underdamping produces overshoot and ringing, while critical damping gives the fastest non-oscillatory response.

Professional measurement requires a defined measurand, suitable instrumentation, calibration, uncertainty analysis, safe practice, complete documentation, and honest reporting.

Key Terms

Accuracy: Closeness of a measurement to an accepted reference value.

Precision: Closeness of repeated measurements to one another.

Measurement error: Difference between a measured value and a reference value.

Uncertainty: Quantified doubt associated with a measurement result.

Gross error: Large error caused by a mistake or abnormal procedure.

Systematic error: Repeatable or predictable bias in measurement.

Random error: Unpredictable variation between observations.

Calibration: Comparison with a reference to establish measurement characteristics.

Sensitivity: Change in output per unit change in input.

Resolution: Smallest distinguishable input or output increment.

Loading effect: Change in the measured system caused by instrument connection.

Hysteresis: Dependence of output on the direction of input change.

Drift: Gradual change in instrument behaviour with time or conditions.

Time constant: First-order response time required to reach 63.2% of a step change.

Bandwidth: Frequency range over which an instrument responds within specified limits.

Natural frequency: Characteristic frequency of a second-order system.

Damping ratio: Parameter describing oscillatory tendency in a second-order system.

Type A uncertainty: Uncertainty evaluated using statistical analysis.

Type B uncertainty: Uncertainty evaluated from non-statistical information.

Expanded uncertainty: Combined standard uncertainty multiplied by a coverage factor.

Review Questions

Short Answer Questions

1. Distinguish accuracy from precision.
2. Why does averaging repeated measurements not normally remove systematic error?
3. Define sensitivity and resolution.
4. What is meant by a first-order instrument time constant?
5. State the four damping classifications of a second-order system.

Long Answer Questions

1. Explain gross, systematic, and random errors, giving electrical measurement examples and suitable reduction methods.
2. Derive the mean, sample variance, and standard uncertainty of the mean for repeated observations, and state their assumptions.
3. Explain Type A and Type B uncertainty evaluation and show how independent components are combined.
4. Discuss loading error in voltage measurement using a Thevenin equivalent circuit.
5. Compare first-order and second-order instrument responses in both time and frequency domains.

Analytical / Application Questions

1. A voltmeter with input resistance 1 MΩ measures a source having Thevenin voltage 10 V and resistance 100 kΩ. Calculate the indicated voltage and the percentage loading error.
2. A sensor has time constant 2 s and is subjected to a step input. Determine the percentage of final response after 4 s and estimate its 10–90% rise time.
3. A resistance is calculated from voltage and current measurements. Explain how you would construct an uncertainty budget if the voltage and current uncertainties were correlated because both instruments share a reference standard.

Multiple Choice Questions with Answers

Question 1

Which type of error is most directly reduced by taking many independent repeated observations and averaging them?

A. Systematic error

B. Random error

C. Gross connection error

D. Calibration offset

Correct Answer: B. Random error

Question 2

For independent uncertainty components, the combined standard uncertainty is calculated by

A. Adding all components algebraically

B. Taking the largest component only

C. Taking the square root of the sum of squared components

D. Multiplying all components together

Correct Answer: C. Taking the square root of the sum of squared components

Question 3

Five voltage readings have a mean of 10.000 V and sample standard deviation 0.0158 V. What is the standard uncertainty of the mean?

A. 0.00316 V

B. 0.00707 V

C. 0.0158 V

D. 0.0790 V

Correct Answer: B. 0.00707 V

Question 4

A first-order instrument reaches what percentage of its final step response after one time constant?

A. 36.8%

B. 50.0%

C. 63.2%

D. 86.5%

Correct Answer: C. 63.2%

Question 5

The transfer function of a first-order instrument is

A. K/(τs + 1)

B. Kτs

C. K/(s² + τs + 1)

D. K(s + τ)

Correct Answer: A. K/(τs + 1)

Question 6

For a second-order system with damping ratio between zero and one, the response is

A. Critically damped

B. Overdamped

C. Undamped and non-oscillatory

D. Underdamped and oscillatory

Correct Answer: D. Underdamped and oscillatory

Question 7

A voltmeter has input resistance 1 MΩ and is connected to a source with Thevenin voltage 10 V and resistance 100 kΩ. What voltage does it indicate?

A. 0.909 V

B. 9.09 V

C. 10.0 V

D. 11.0 V

Correct Answer: B. 9.09 V

Question 8

If a digital display rounds to increments of width q, and rounding is modelled as a uniform distribution, the standard resolution uncertainty is

A. q

B. q/2

C. q/√12

D. √12q

Correct Answer: C. q/√12

Question 9

Which characteristic describes dependence of an instrument output on whether the input is increasing or decreasing?

A. Hysteresis

B. Sensitivity

C. Resolution

D. Bandwidth

Correct Answer: A. Hysteresis

Question 10

For an underdamped second-order system, increasing the damping ratio generally

A. Increases overshoot

B. Reduces overshoot

C. Makes the natural frequency zero

D. Eliminates all measurement delay

Correct Answer: B. Reduces overshoot