An electronic design is not complete when its schematic simulates correctly or when a prototype produces the expected output once. It is complete only when evidence shows that the product meets its requirements over specified operating conditions, manufacturing variation, environmental stress, and an appropriate service life. Testing provides that evidence; troubleshooting converts abnormal observations into identified causes; reliability analysis estimates the probability of continued operation; and analytics turns measurements into defensible engineering decisions.

In mechatronics, these activities span sensors, signal-conditioning circuits, power electronics, embedded controllers, communication interfaces, and actuators. A faulty analogue measurement may appear as a software error, while electromagnetic interference may be mistaken for a defective sensor. Effective diagnosis therefore requires a disciplined relationship between requirements, measurement methods, physical models, and recorded evidence.

Learning objectives

Verification as an engineering process

Verification asks, “Was the product built according to its specified requirements?” Validation asks, “Does the product meet the user’s intended need in its real context?” A circuit can pass verification of voltage accuracy but fail validation because its sensor saturates in the actual mechanical environment.

A verification requirement must be observable and measurable. “The controller is stable” is incomplete; “the closed-loop position error shall remain below 1 mm for a 0.5 m step under a 2 kg load at 20–30 °C” is testable. Each requirement should have an identifier, acceptance limit, test method, equipment, environmental conditions, sample size, and pass/fail rule.

Verification-plan structure

Requirement	Metric and limit	Method	Conditions	Evidence
ADC accuracy	Error ≤ ±2 LSB	Calibrated voltage source and software capture	0–5 V, 25 °C	Raw readings and calculation
PWM frequency	20 kHz ±1%	Oscilloscope or frequency counter	Nominal supply	Waveform and statistics
Digital timing	Setup margin ≥ specified value	Logic analyzer	Worst-case bus traffic	Timing cursors and log
Protection	No damage after defined fault	Controlled fault injection	Current-limited supply	Test record and inspection

Acceptance criteria should be established before testing. Changing a limit after observing an unfavorable result is not engineering justification unless the requirement is formally revised and the reason is documented.

A test procedure should identify initialization, instrument configuration, connections, stimulus, measurement sequence, data format, safety controls, and recovery actions. The test environment must control variables that could confound the result, such as supply impedance, probe grounding, temperature, cable routing, load, firmware version, and calibration status.

Measurement instruments and measurement integrity

Digital multimeters and power measurements

A multimeter is appropriate for DC voltage, resistance, current, continuity, and often low-frequency AC quantities. Its input resistance can load a high-impedance node, and its current mode can create a significant burden voltage. A current measurement should therefore be connected in series and checked against the instrument’s maximum terminal rating. Never assume that a continuity range is safe on an energized circuit.

Oscilloscopes

An oscilloscope displays voltage as a function of time and is indispensable for observing ripple, switching transients, noise, overshoot, settling, duty cycle, and timing relationships. A probe is part of the measurement system. Probe capacitance can reduce bandwidth or alter an oscillator; a long ground lead can introduce ringing and false high-frequency noise. Use a short spring ground for fast signals, select an appropriate attenuation setting, compensate passive probes, and verify the channel’s bandwidth and sampling configuration.

The Nyquist condition requires the sampling frequency to exceed twice the highest frequency component for ideal band-limited reconstruction. In practical transient measurements, substantially higher sampling is desirable because edge shape, trigger position, and instrument filtering affect the result. Aliasing can produce a stable-looking but false waveform. Peak-detect, bandwidth-limit, averaging, and high-resolution modes change the measurement and must be documented.

An oscilloscope’s trigger makes a repetitive event appear stationary; it does not prove that the event always occurs. Single-shot acquisition, segmented memory, persistence, or statistical waveform functions may be needed for rare faults.

Logic analyzers

A logic analyzer samples thresholded digital channels and is useful for protocol decoding, state sequencing, and timing relationships. It does not measure analogue voltage shape accurately. A slowly rising signal may cross the threshold at an uncertain time, and incorrect logic thresholds can create false transitions. Sampling rate, channel skew, probe loading, clock recovery, and protocol settings must be checked. An oscilloscope is generally preferable for signal integrity; a logic analyzer is generally preferable for long digital transactions.

Automated measurement

Automated test systems commonly contain a device under test (DUT), programmable source or load, switching matrix, measurement instruments, controller, and results database. Automation improves repeatability and throughput but can reproduce a wiring or software error rapidly. A sound system includes instrument self-tests, connection checks, calibration records, unique DUT identification, controlled firmware versions, limits stored under revision control, timestamps, raw-data preservation, and safe shutdown.

A measurement script should separate instrument communication, stimulus control, data validation, calculations, and reporting. It should detect out-of-range values, communication timeouts, stale readings, and impossible states rather than silently substituting zeros or previous results.

Uncertainty, error, and evidence

Measurement error is the difference between a measured value and the unknown true value; uncertainty expresses incomplete knowledge of that value. Error is not necessarily observable, whereas uncertainty can be evaluated from calibration data, specifications, repeatability, resolution, and environmental effects.

For independent input quantities, a first-order combined standard uncertainty is

where y = f(x₁, x₂, …, x_n) is the measured result, x_i are input quantities, u(x_i) are their standard uncertainties, and the partial derivatives describe sensitivity. This equation assumes a sufficiently linear model near the measured values and uncorrelated inputs. For correlated inputs, covariance terms must be added:

If a specification gives a rectangular distribution with half-width a, a common estimate is u = a/√3, provided all values in the interval are considered equally likely. For a normally distributed calibration uncertainty, the stated standard deviation may be used directly. These assumptions must be recorded rather than applied automatically.

For a resistance measurement of 10.00 kΩ with instrument standard uncertainty 0.02 kΩ and lead/contact standard uncertainty 0.01 kΩ, assuming independence,

A reported result might be 10.00 kΩ ± 0.04 kΩ at approximately 95% coverage if a coverage factor near 2 is justified. The number of digits must not imply greater certainty than the uncertainty supports.

Repeatability describes variation under the same conditions; reproducibility describes variation when conditions such as operator, instrument, or laboratory change. A calibration traceability chain connects a result to recognized standards through documented calibrations and stated uncertainties.

Guard bands and decisions near a limit

Suppose a specification is 5.00 V ± 0.05 V and a result is 5.047 V with expanded uncertainty ±0.020 V. The measured value is inside the limit, but the uncertainty interval extends beyond 5.05 V. A responsible decision rule may classify the result as indeterminate rather than pass. Guard bands intentionally create an internal acceptance limit to control the probability of accepting a nonconforming item. The rule must be agreed before testing.

Fault isolation and troubleshooting

Troubleshooting is an evidence-based search through possible causes. A useful sequence is:

Signal tracing follows energy or information from input to output. If an analogue chain contains a sensor, amplifier, filter, ADC, and software conversion, measure each interface under the same stimulus. A missing signal at the amplifier output narrows the search; a correct signal there shifts attention to the ADC reference, sampling, grounding, or software.

Fault isolation should distinguish symptom, proximate cause, and root cause. “Motor stops” is a symptom; “overcurrent protection trips” is a proximate cause; “undersized connector contact creates intermittent resistance and local heating” may be the root cause. The five-whys method can help, but it should be supported by physical evidence rather than used as a substitute for analysis.

A fault tree starts with an undesired event and combines contributing events using logical OR and AND relationships. An OR gate means any listed event can produce the higher event; an AND gate means all required events must occur. Failure mode and effects analysis (FMEA) instead begins with component or process failure modes and evaluates their effects, causes, controls, and actions. Risk-priority scoring can organize attention, but numerical rankings are not probabilities and should not conceal severe low-frequency hazards.

Tolerance and worst-case analysis

Component tolerances, reference accuracy, temperature coefficients, supply variation, ageing, and manufacturing processes cause output variation. For a linear quantity

where Δx_i is the maximum magnitude of deviation of input x_i. This method assumes the specified deviations can occur simultaneously in the adverse direction, so it is conservative and suitable for safety limits.

Statistical analysis is different. If independent variations are represented by standard deviations σ_i, the standard deviation of a linear output is

This is appropriate when the distributions and independence assumptions are credible. It does not guarantee an absolute limit. Monte Carlo simulation can model nonlinear circuits and non-normal distributions, but its conclusions depend on valid distributions, correlations, sample size, and implementation.

Worked tolerance example

An inverting amplifier has gain G = −R_f/R_in. Let R_f = 100 kΩ ±1% and R_in = 10 kΩ ±1%. Nominal gain is −10. For worst-case magnitudes, the maximum gain magnitude occurs when the feedback resistor is high and input resistor low:

Thus the gain range from resistor tolerance alone is approximately −9.802 to −10.202. This excludes op-amp finite gain, input offset, bias current, temperature drift, loading, and measurement uncertainty; a complete design must include them.

Reliability engineering

Reliability is the probability that an item performs its required function for a stated time under stated conditions. It is not the same as quality, which concerns conformance at manufacture, although poor quality can reduce reliability.

where R(t) is reliability at time t. The failure probability by time t is F(t) = 1 − R(t). The mean time to failure is

for a non-repairable item under this model. For a repairable item, mean time between failures (MTBF) is often used, but its definition depends on whether repair, logistics, and scheduled downtime are included. MTBF is not a guaranteed lifetime and should not be used to claim that every unit will operate for that duration.

These equations assume clearly defined independence and functional architecture. Common-cause failures, shared power supplies, thermal coupling, and dependent control logic can invalidate the apparent redundancy.

Failure rates are rarely constant throughout life. Early failures may arise from manufacturing defects, useful life may show an approximately stable rate, and wear-out may produce an increasing rate. Burn-in, screening, derating, thermal design, vibration testing, and component qualification address different portions of this life pattern.

Accelerated life testing applies increased temperature, voltage, humidity, vibration, or cycling to obtain information more quickly. Acceleration must preserve the relevant failure physics. A stress that creates an unrealistic failure mechanism can produce misleading results. Test plans should define censoring, sample size, failure criteria, confidence bounds, and extrapolation assumptions.

Analytics for engineering decisions

Data analytics begins with a measurement question, not with a preferred plot. Useful descriptive quantities include mean, median, range, standard deviation, quantiles, and control charts. A sample mean is

where n is sample size. The n−1 denominator estimates population variance from a sample when observations are independent and identically distributed. It does not correct for bias, drift, or autocorrelation.

Plots should include units, limits, sample identity, and time order. A histogram can reveal spread and multiple populations; a time plot can reveal drift; a scatter plot can reveal dependence on temperature or supply voltage. Correlation does not establish causation. Confounding variables, selection bias, small samples, and repeated measurements from the same DUT can produce overconfident conclusions.

Statistical process control distinguishes common-cause variation from signals requiring investigation. Control limits are estimated from process data and are not automatically the same as engineering specification limits. A process can be statistically stable but consistently outside specification, or unstable while its current values happen to fall within specification.

AI and machine learning can assist anomaly detection, waveform classification, predictive maintenance, and test optimization. However, training data must represent operating conditions and failure modes, labels must be audited, and false positives and false negatives must be evaluated against safety consequences. An opaque classifier should not replace physical verification in a safety-critical protection function. Data provenance, cybersecurity, privacy where human-related data are involved, and model-change control are part of engineering governance.

Illustrative case: intermittent position error in a robotic axis

Background and problem

A laboratory robot intermittently reports a position error of approximately 4 mm. The error occurs after several minutes of operation, but the system passes a static commissioning test.

Analysis

The team first reproduces the failure while recording motor current, encoder A/B signals, 5 V supply, controller reset status, and cabinet temperature. The oscilloscope shows that the encoder supply falls from 5.02 V to 4.62 V during motor acceleration. The logic analyzer records extra encoder transitions during the same interval. A thermal camera shows heating at a removable connector.

Three hypotheses are considered: software counting error, electromagnetic interference, and connector voltage drop. Replacing the controller does not remove the fault. A temporary direct cable from the supply to the encoder eliminates the failure, while adding software filtering does not. Four-wire voltage measurement then identifies a high-resistance crimp contact. The contact resistance increases with temperature and motor current, causing supply droop and distorted threshold crossings.

Solution and outcome

The connector is replaced with a correctly crimped, locking contact; the supply wiring is separated from high-current switching paths; local decoupling is improved; and an undervoltage diagnostic is added. The verification plan is expanded to include acceleration at maximum load, elevated cabinet temperature, connector movement, and a 24-hour endurance run. No position errors occur in the defined test, and the captured evidence is linked to the hardware revision and firmware version.

Learning points

The symptom appeared in software but originated in power integrity. Measuring only the nominal encoder waveform at rest concealed the fault. The decisive tests were time-correlated measurements, controlled substitution, and a physical inspection guided by evidence. The corrective action addressed both the immediate defect and detection of recurrence.

Limitations, safety, and professional practice

Testing can damage equipment and expose personnel to hazardous voltage, stored energy, motion, heat, or rotating machinery. Use isolation where appropriate, current limiting, interlocks, emergency stops, protective equipment, and a documented energized-work procedure. Never connect an earth-referenced oscilloscope probe to a point that is not safely referenced to protective earth; use a properly rated differential probe or isolation arrangement.

A test result is valid only within its stated conditions. Calibration does not guarantee correct setup, and a passed test does not prove untested behavior. Maintain configuration control for schematics, component substitutions, PCB revisions, firmware, test scripts, instrument settings, and environmental conditions. Preserve raw data and failed results; deleting inconvenient observations destroys traceability and can create safety and regulatory exposure.

Best-practice workflow

A robust workflow is to define measurable requirements, identify hazards and failure modes, design the verification matrix, characterize instruments, calibrate and check the setup, run nominal tests, test boundaries and stresses, analyze uncertainty and variation, isolate failures systematically, perform regression tests after changes, and issue a report that distinguishes observation from interpretation. Design decisions should cite evidence: measured margin, uncertainty, sample size, failure mechanism, reliability assumption, and remaining risk.

Chapter Summary

Verification converts requirements into repeatable tests with predefined acceptance rules. Oscilloscopes reveal analogue behavior and transients, while logic analyzers efficiently examine digital sequences; both require appropriate sampling, thresholds, grounding, and configuration. Uncertainty quantifies incomplete knowledge and must be separated from bias and tolerance. Troubleshooting is a controlled process of symptom characterization, hypothesis testing, signal tracing, and regression verification. Worst-case analysis protects against simultaneous adverse variations, whereas statistical analysis estimates likely spread under distributional assumptions. Reliability depends on time, environment, failure mechanisms, and architecture; MTBF is not a guaranteed lifetime. Analytics is valuable only when data are representative, traceable, and interpreted with awareness of bias and causation. Safe, ethical engineering preserves failed evidence, states limitations, and makes decisions supported by documented measurements.

Key Terms

Uncertainty: Quantified incomplete knowledge associated with a measurement result.

Aliasing: False frequency or waveform information caused by inadequate sampling.

Trigger: Condition that selects the acquisition reference point on an oscilloscope.

Monte Carlo analysis: Repeated random sampling used to estimate output variation.

Reliability: Probability of performing a required function for a stated time and condition.

Failure rate: Frequency of failure under specified conditions and model assumptions.

MTBF: Mean time between failures for a repairable system, under a stated definition.

Guard band: Internal acceptance margin used to account for measurement uncertainty.

Review Questions

Short Answer Questions

Long Answer Questions

Analytical / Application Questions

Multiple Choice Questions with Answers

Question 1

A. Verification checks specified requirements; validation checks fitness for intended use.

B. Verification is performed only after field deployment; validation is performed only in simulation.

D. Verification replaces the need for acceptance criteria; validation replaces calibration.

Correct Answer: A. Verification checks specified requirements; validation checks fitness for intended use.

Question 2

What is a principal risk of using a long oscilloscope ground lead on a fast digital signal?

Correct Answer: B. It can introduce ringing and apparent high-frequency noise.

Question 3

For independent uncertainty contributions, which expression gives first-order combined standard uncertainty?

Correct Answer: C. The root-sum-square of sensitivity-weighted standard uncertainties

Question 4

Question 5

For a constant failure rate λ, which expression represents reliability at time t?

Question 6

Which instrument is generally most suitable for decoding a long digital communication transaction?

Question 7

A measured value lies inside a specification limit, but its uncertainty interval extends beyond that limit. The most defensible classification may be:

Question 8

Which action is most appropriate immediately after reproducing an intermittent fault?

Correct Answer: B. Preserve the original state and record the symptom and conditions

Question 9

Question 10

Why should an AI anomaly detector not automatically replace physical verification in a safety-critical circuit?

B. A classifier may have unrepresentative training data and unquantified false decisions

Correct Answer: B. A classifier may have unrepresentative training data and unquantified false decisions

Chapter 14 Testing, Troubleshooting, Reliability and Analytics

Testing, Troubleshooting, Reliability and Analytics

Introduction