Testing, Diagnostics, and Condition Monitoring

Introduction and context

Electrical machines are expected to operate safely, efficiently, and continuously despite thermal, mechanical, electrical, and environmental stresses. A motor that fails unexpectedly can stop a production line, damage a driven process, create safety hazards, and impose substantial economic losses. Testing and condition monitoring provide the evidence needed to determine whether a machine is healthy, degrading, or already faulty.

Testing is usually performed during commissioning, acceptance, scheduled maintenance, or after repair. Condition monitoring is the repeated observation of machine variables during service. Diagnostics converts measurements into an assessment of fault type, severity, and likely location. Prognostics estimates how degradation may develop and supports decisions about maintenance timing.

A sound diagnostic process does not rely on a single measurement. It combines machine history, operating conditions, physical inspection, electrical data, vibration, temperature, insulation tests, acoustic measurements, and—where appropriate—lubricant or oil analysis. The central engineering principle is to compare measurements with a reliable baseline obtained under comparable speed, load, temperature, and environmental conditions.

Learning objectives

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

1. Explain the purposes and limitations of standard tests on electrical machines.
2. Select suitable sensors and measurement procedures for electrical, thermal, mechanical, acoustic, insulation, and oil-related observations.
3. Interpret trends and spectra without confusing symptoms with causes.
4. Diagnose common faults in induction machines, synchronous machines, transformers, and drives.
5. Apply basic reliability, statistical, and predictive-maintenance methods.
6. Explain the role and limitations of digital twins and explainable artificial intelligence.
7. Design a safe, traceable, and ethically responsible monitoring program.

1. A framework for machine health assessment

A machine-health program normally contains five linked activities:

• Measurement: acquire data using calibrated instruments and defined operating conditions.
• Validation: check sensor installation, signal quality, units, timestamps, and plausibility.
• Feature extraction: calculate quantities such as RMS current, crest factor, temperature rise, spectral amplitudes, or insulation resistance.
• Diagnosis: identify the most probable failure mechanism using physics, historical evidence, and expert rules.
• Decision: continue operation, increase monitoring, derate, schedule maintenance, or shut down.

A measurement is not a diagnosis. For example, high vibration may result from imbalance, misalignment, looseness, resonance, bearing damage, electrical magnetic forces, or a faulty sensor. Diagnosis requires the relationship between the observation and the machine's operating state.

Baseline and trend principles

For a measured variable x, a simple deviation from a baseline is

where x₀ is the reference value measured under comparable conditions. The difference has meaning only when speed, load, temperature, sensor location, and measurement method are sufficiently similar. A trend is generally more useful than one isolated value. A slowly increasing bearing temperature may be more important than a temporarily high value caused by an unusual ambient condition.

Measurements should be classified as:

• Absolute limits: values above which continued operation is unsafe or unacceptable.
• Alert limits: values requiring investigation or more frequent observation.
• Trip limits: values at which automatic protection or immediate shutdown is justified.
• Trend limits: rates of change that indicate accelerating degradation.

Limits should be based on manufacturer information, applicable standards, machine design, historical data, and risk assessment. Generic limits must not be transferred blindly between machines.

2. Standard electrical-machine tests

2.1 Nameplate and visual inspection

Before energized testing, record the machine rating, voltage, current, frequency, speed, connection, insulation class, enclosure, duty, bearing arrangement, and service factor. Inspect terminals, cooling passages, cable glands, earthing conductors, coupling guards, foundation bolts, shaft seals, and signs of contamination or overheating.

A visual inspection often detects loose connections, blocked ventilation, oil leakage, moisture ingress, cracked insulation, damaged fans, and incorrect rotation. These observations provide essential context for instrument data.

2.2 Resistance tests

The resistance of stator or rotor windings is measured using a low-resistance ohmmeter, preferably with four-wire Kelvin connections for low values. For a balanced three-phase winding, phase resistances should be comparable after correcting for temperature. Copper resistance changes approximately according to

where R_T is resistance at temperature T, R_ref is resistance at reference temperature T_ref, and α is the temperature coefficient of the conductor material, approximately 0.00393 per °C for copper near room temperature. The equation assumes a uniform conductor temperature and is an approximation over a moderate temperature range.

Unequal resistance may indicate loose terminals, poor joints, shorted turns, incorrect connections, or damaged conductors. Resistance alone cannot reliably detect every inter-turn fault because the resistance change may be very small.

2.3 Insulation resistance and polarization index

Insulation resistance is measured by applying a DC test voltage with the machine isolated, discharged, and safely grounded before and after the test. The value depends strongly on temperature, humidity, contamination, winding size, and test voltage. Results should therefore be recorded with these conditions.

The polarization index (PI) is commonly defined as

where R₁₀ is insulation resistance after 10 minutes and R₁ after 1 minute of a constant-voltage test. PI is dimensionless. A low or falling PI can indicate moisture, contamination, surface leakage, or deteriorated insulation, but interpretation depends on winding type and test method. A high PI does not prove that the insulation is free from all defects, especially localized inter-turn weaknesses.

High-potential and surge tests can reveal insulation weaknesses, but they impose electrical stress and must be performed only by competent personnel using specified procedures. Testing must follow applicable safety rules and manufacturer limits.

2.4 No-load, blocked-rotor, and load tests

For an induction motor, the no-load test supplies rated frequency and approximately rated voltage while measuring input power, current, and speed. It helps estimate core loss, friction, windage, and magnetizing current. The blocked-rotor test applies reduced voltage at standstill and is used to estimate equivalent-circuit parameters and starting behavior. Because the rotor does not turn, heating can be rapid; the test must be brief and controlled.

A load test directly measures torque, speed, input power, and output power. Mechanical output power is

where T is shaft torque in newton-metres and ω is angular speed in radians per second. For rotational speed n in revolutions per minute,

Efficiency is

where P_in and P_out are input and output power in consistent units. These equations assume that measured torque represents useful shaft torque and that losses are included through the measured input-output relationship.

For a balanced three-phase sinusoidal system, active input power is

where V_L and I_L are line voltage and line current, and cosφ is the power factor. The formula assumes balanced phases and sinusoidal quantities; power analysers should be used when distortion is significant.

2.5 Transformer tests

The open-circuit test, usually performed at rated voltage on the low-current side, estimates core loss and magnetizing current. The short-circuit test, performed at reduced voltage and rated current, estimates copper loss and equivalent series impedance. Temperature correction and instrument accuracy are important because the measured losses may be relatively small.

Turns ratio, winding resistance, insulation resistance, excitation current, dielectric response, and frequency-response analysis may be used to assess transformers. A change in frequency-response signature can indicate winding movement after a through-fault or transport event.

3. Vibration monitoring

Vibration is commonly measured with accelerometers mounted on bearing housings in horizontal, vertical, and axial directions. Velocity is often useful for general machine severity over a broad frequency range, acceleration is sensitive to high-frequency bearing impacts, and displacement is useful at low rotational frequencies. The selected quantity must match the fault and frequency range.

The RMS value of a sampled vibration signal x_i is

where N is the number of samples and x̄ is the sample mean. RMS represents overall energy-like magnitude, but it does not identify the fault source.

The crest factor is

where x_peak is the largest absolute peak. An increasing crest factor can indicate impulsive bearing damage even when overall RMS has not yet risen substantially. It is sensitive to sampling, filtering, and isolated interference.

Frequency interpretation

Fast Fourier Transform (FFT) analysis separates a time waveform into frequency components. Typical indicators include:

• 1× running speed: imbalance, although other faults can also produce it.
• 2× running speed: angular or parallel misalignment, bent shaft, or structural effects.
• Several harmonics of running speed: looseness or distortion.
• Bearing characteristic frequencies with harmonics and sidebands: rolling-element bearing damage.
• Twice-line-frequency electromagnetic components: electrical asymmetry, eccentricity, or magnetic-force effects.
• Broadband high-frequency energy: lubrication problems, bearing impacts, or friction.

The bearing frequencies depend on bearing geometry, shaft speed, number of rolling elements, pitch diameter, ball diameter, and contact angle. Manufacturer data or a validated bearing database should be used rather than approximate formulas when making a maintenance decision.

Vibration diagnosis should include phase, coherence, time waveform, shaft speed, and structural response. Resonance can amplify a harmless excitation at a particular speed, so changing speed during a run-up or coast-down test can be highly informative.

4. Temperature and thermal diagnostics

Temperature sensors include resistance temperature detectors, thermocouples, thermistors, infrared cameras, and fibre-optic sensors. Measurements may be taken at stator windings, bearings, housings, oil, terminals, and cooling-air paths.

The winding temperature rise is

where T_w is winding temperature and T_a is ambient temperature. Insulation ageing is strongly temperature dependent. A commonly used engineering approximation is that a 10 °C increase in operating temperature approximately doubles the rate of thermal ageing within an applicable temperature range. This is a rule of thumb, not a universal law; insulation systems, duty cycle, hot spots, and moisture alter the result.

Thermal patterns provide clues:

• A hot terminal suggests high contact resistance or a loose connection.
• One unusually hot phase may indicate unbalance, poor connection, or winding damage.
• A hot bearing may indicate lubrication failure, excessive preload, misalignment, or overload.
• A hot motor centre with poor airflow suggests blocked ventilation or fan failure.
• A localized hot spot in a transformer may indicate a winding, connection, or oil-circulation problem.

Infrared thermography requires correct emissivity settings, suitable viewing angles, and awareness of reflected radiation. Shiny metal terminals can produce misleading readings. Thermal data should be correlated with electrical load and ambient temperature.

5. Current-signature and power-quality analysis

Motor current-signature analysis (MCSA) measures phase currents and examines their RMS values, imbalance, harmonics, and sidebands. It can be performed non-invasively in some installations, but interpretation is affected by inverter switching, load torque oscillations, supply distortion, and variable speed.

For three phase currents I_a, I_b, and I_c, a simple current-unbalance indicator is the maximum deviation from the average divided by the average:

where I_avg = (I_a + I_b + I_c)/3. This is a practical indicator, not a universal standard definition. Negative- and positive-sequence components provide a more rigorous analysis for unbalanced systems.

Current signatures can indicate broken rotor bars, air-gap eccentricity, stator winding faults, overload, and supply problems. However, a spectral line is not proof of one fault: load variation and drive modulation can create similar components. Data should be collected at known speed and load, preferably over time.

Power-quality assessment includes voltage unbalance, harmonics, sags, swells, transients, and interruptions. Variable-frequency drives require instruments with adequate bandwidth and correct procedures for PWM waveforms. Incorrect measurement can damage instruments or produce false conclusions.

6. Acoustic and ultrasonic monitoring

Airborne sound can reveal cooling-fan problems, arcing, mechanical looseness, bearing damage, and abnormal friction. Ultrasound instruments detect high-frequency energy that may be difficult to hear and can be useful for compressed-air leakage, electrical discharge, and early bearing lubrication problems.

Acoustic measurements are strongly affected by background noise, reflections, microphone position, machine enclosure, and production activity. Measurements should use repeatable locations and, where possible, order tracking or frequency analysis. Acoustic evidence is most reliable when supported by vibration, temperature, or electrical evidence.

7. Insulation, oil, and lubricant diagnostics

Insulation diagnostics

Beyond insulation resistance and PI, specialist tests include dissipation factor, partial-discharge measurement, surge comparison, dielectric absorption, and tan-delta analysis. Partial discharge is a localized electrical discharge that does not completely bridge the insulation. Its phase-resolved pattern can help distinguish internal voids, surface discharge, slot discharge, and corona, but interpretation requires appropriate sensors and expertise.

Oil and lubricant analysis

For oil-filled transformers, dissolved-gas analysis (DGA) examines gases such as hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, and carbon dioxide. Gas generation depends on thermal and electrical stress. Ratios and total dissolved combustible gas can assist diagnosis, but results must be interpreted using recognized standards, sampling quality, transformer design, loading history, and trends.

For bearings and gearboxes, lubricant analysis may include viscosity, water content, particle count, ferrous debris, elemental spectroscopy, oxidation, and additive depletion. A rise in iron particles can indicate wear, but sampling location and timing determine whether the sample represents the critical component. Oil analysis identifies wear products; it does not replace mechanical inspection.

8. Diagnostic reasoning and fault isolation

A useful fault-isolation process is:

1. Establish the symptom and operating conditions.
2. Confirm the measurement and eliminate sensor or wiring errors.
3. Compare with baseline and peer machines.
4. Separate electrical, mechanical, thermal, and process causes.
5. Seek corroborating evidence from an independent measurement.
6. Estimate severity, consequence, and confidence.
7. Recommend an action and define a follow-up measurement.

The distinction between symptom, mechanism, and root cause is important. High current is a symptom; overload or a shorted turn may be a mechanism; a seized pump, incorrect process setting, or degraded insulation may be the root cause.

A diagnostic matrix can support, but not replace, engineering judgment.

9. Reliability and predictive maintenance

Reliability is the probability that an item performs its required function for a specified time under stated conditions. For a constant failure rate λ, the exponential reliability model is

where t is time and λ has units of failures per unit time. This model assumes a constant failure rate and is often unsuitable for a complete machine over its entire life because real equipment commonly exhibits infant mortality, useful life, and wear-out regions.

The mean time to failure under this model is

For repairable equipment, mean time between failures and mean time to repair are often more useful. Availability can be approximated by

where MTBF is mean time between failures and MTTR is mean time to repair. The approximation assumes stable, repeated operating and repair cycles and excludes some logistics and planned-maintenance effects.

Predictive maintenance uses condition indicators to estimate when intervention is needed. Common methods include trend extrapolation, control charts, anomaly detection, remaining-useful-life models, and failure-mode-based rules. The objective is not to predict an exact failure date with false precision; it is to reduce risk while avoiding unnecessary maintenance.

A maintenance decision should consider consequence, detectability, lead time for spares, safety, production cost, and the uncertainty of the diagnosis. A critical low-confidence alarm may justify immediate inspection even if the measured value is moderate.

10. Digital twins and explainable AI

A digital twin is a maintained digital representation of a physical asset, connected to data and informed by a model. For an electrical machine, it may combine nameplate data, finite-element or equivalent-circuit models, thermal models, vibration history, drive data, maintenance records, and operating context.

A useful twin must define:

• the asset identity and configuration;
• sensor locations, calibration, and data quality;
• model assumptions and valid operating range;
• update frequency and version history;
• uncertainty and confidence intervals;
• actions permitted from its outputs.

A simulation model is not automatically a digital twin. A twin requires traceable linkage to the physical asset and lifecycle data.

Artificial intelligence can classify vibration spectra, detect anomalies, estimate temperature, or combine heterogeneous data. Supervised models require representative labelled failures, which are often scarce. Unsupervised methods can detect novelty but may confuse a legitimate change in load or process with damage. Data drift, sensor replacement, and changes in machine configuration can invalidate a model.

Explainable AI is essential when an algorithm influences a safety or maintenance decision. Explanations may include influential features, similar historical cases, reconstructed signals, confidence estimates, and counterfactual questions such as: “Would the alarm disappear if load and speed returned to the baseline?” Explanations must be technically faithful, not merely persuasive visualizations.

AI should support, not conceal, human responsibility. Operators must be able to challenge an alarm, inspect the evidence, and apply a safe fallback procedure. Training data, model versions, overrides, and decisions should be auditable. Personal data, cybersecurity, access control, and protection of operational technology are also relevant governance issues.

11. Illustrative case: recurrent induction-motor bearing alarms

Background

A 110 kW, 400 V induction motor drives a process pump in a continuous manufacturing plant. A wireless accelerometer reports intermittent high-frequency alarms on the drive-end bearing. Overall velocity remains below the plant's normal alert level.

Problem

Maintenance personnel initially suspect a defective sensor because the alarm is intermittent and the motor continues to operate. Replacing the bearing immediately would require a production stoppage.

Analysis

The monitoring engineer first verifies sensor attachment, battery condition, timestamp alignment, and calibration. The alarm occurs repeatedly at high load and disappears during unloaded operation. The time waveform shows impacts, and the crest factor increases while broadband acceleration rises. Envelope analysis reveals components consistent with the installed bearing geometry. The drive-end bearing temperature is 8 °C higher than the non-drive-end bearing, although ambient temperature and load are stable.

A lubrication review shows that grease was being added at intervals shorter than the supplier's recommendation, with an incorrect quantity. The grease sample contains evidence of over-lubrication and incipient wear debris. Alignment measurements also show a small coupling offset within the machine's historical tolerance but close to the plant's action threshold.

Approach and solution

The plant reduces load temporarily, corrects the lubrication procedure, and schedules a bearing inspection during the next planned outage. Alignment is corrected, and the bearing is replaced because inspection confirms early raceway damage. The sensor remains installed to verify the outcome.

Outcome

After repair, envelope acceleration and crest factor return toward baseline, bearing temperature decreases, and no further alarms occur over the observation period. The final diagnosis is early bearing damage accelerated by poor lubrication and marginal alignment—not a sensor fault.

Learning points

1. A low overall vibration value does not exclude an early localized bearing fault.
2. Sensor validation should precede expensive mechanical intervention.
3. Independent evidence—waveform, temperature, lubrication, and inspection—raises diagnostic confidence.
4. Maintenance procedures are part of root-cause control; replacing a bearing alone would not prevent recurrence.

12. Practical implementation and limitations

A monitoring program should begin with a failure-mode and effects analysis. For each important failure mode, define the observable symptom, sensor, sampling interval, alert rule, responsible person, and required action. Critical assets may require continuous monitoring; less critical machines may be assessed during route-based inspections.

Good practice includes calibrated instruments, consistent sensor locations, synchronized timestamps, documented operating conditions, protected data storage, and quality checks. Alarm limits should be reviewed after machine modification, speed changes, process changes, or sensor relocation.

Important limitations include:

• fault signatures can overlap;
• early degradation may produce weak signals;
• variable-speed drives change spectral relationships;
• environmental noise contaminates acoustic and vibration data;
• poor sampling can make oil results unrepresentative;
• models trained on one fleet may not transfer to another;
• false alarms cause alarm fatigue and may lead personnel to ignore genuine warnings;
• maintenance actions based on uncertain predictions can create new failures.

Safety has priority over data collection. Lockout/tagout, discharge of stored energy, guarding, arc-flash controls, intrinsically safe equipment where required, and competent supervision are mandatory considerations. Test records should be retained with traceability to the instrument, operator, date, machine state, and procedure.

Chapter Summary

Testing and condition monitoring transform machine behavior into evidence for maintenance decisions. Electrical tests assess winding resistance, insulation condition, power, efficiency, supply quality, and transformer characteristics. Vibration analysis uses time, frequency, phase, and envelope information to distinguish mechanical and electromagnetic problems. Temperature measurements reveal thermal loading, cooling defects, connection problems, and bearing distress. Current-signature, acoustic, oil, and dissolved-gas data provide complementary evidence but require careful attention to operating conditions and sampling quality.

Reliable diagnosis is a structured process: establish a baseline, validate the measurement, compare trends, form competing hypotheses, seek corroboration, quantify uncertainty, and select a proportionate action. Reliability models and predictive maintenance help prioritize resources, while digital twins and AI can integrate complex data. Neither technology removes the need for engineering judgment, cybersecurity, explainability, safety controls, and accountable human decisions.

Key Terms

Baseline: Reference measurement obtained under defined operating conditions.

Condition monitoring: Repeated observation of asset condition during operation.

Crest factor: Ratio of signal peak magnitude to RMS magnitude.

Digital twin: Data-connected digital representation of a specific physical asset.

Dissolved-gas analysis: Analysis of gases dissolved in transformer oil.

FFT: Algorithm that estimates frequency components from sampled data.

Insulation resistance: Resistance measured between insulation systems and conductors.

Machine diagnostics: Identification of probable fault mechanisms from evidence.

MCSA: Analysis of motor-current magnitude and frequency content.

MTBF: Mean time between failures for a repairable item.

MTTR: Mean time to repair.

Partial discharge: Localized electrical discharge that does not completely bridge insulation.

Polarization index: Ratio of ten-minute to one-minute insulation resistance.

Predictive maintenance: Maintenance scheduled using evidence of developing condition.

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

RMS: Root-mean-square measure of signal magnitude.

Root cause: Underlying condition that produces a failure mechanism.

Thermal ageing: Degradation caused by temperature and time.

Trend: Direction and rate of change of a measured variable.

Vibration spectrum: Representation of vibration amplitude versus frequency.

Review Questions

Short Answer Questions

1. Why is a baseline necessary in condition monitoring?
2. Define polarization index and state what it may indicate.
3. Distinguish RMS vibration from crest factor.
4. Give two causes of unequal three-phase motor currents.
5. Why must oil samples be collected consistently?

Long Answer Questions

1. Explain a complete diagnostic workflow from measurement validation to maintenance decision.
2. Describe how vibration, temperature, and current data can be combined to diagnose an induction-motor fault.
3. Explain the purposes and limitations of no-load, blocked-rotor, and load tests.
4. Discuss how reliability models support predictive-maintenance planning.
5. Explain the governance requirements for using AI in electrical-machine diagnostics.

Analytical / Application Questions

1. A motor has stable overall vibration but a rapidly increasing crest factor and bearing-envelope components. What hypotheses should be considered, and what additional evidence should be collected?
2. Phase currents are 98 A, 101 A, and 106 A. Calculate the practical current-unbalance indicator defined in this chapter and discuss possible causes and confirmation tests.
3. A transformer shows a sudden increase in dissolved combustible gas after a load change. Design a cautious investigation plan that addresses sampling, operating history, electrical tests, safety, and the possibility of a false alarm.

Multiple Choice Questions with Answers

Question 1

What is the principal purpose of establishing a machine baseline?

A. To eliminate the need for calibration

B. To compare later measurements under comparable operating conditions

C. To guarantee that the machine will not fail

D. To replace all manufacturer limits

Correct Answer: B. To compare later measurements under comparable operating conditions

Question 2

The polarization index is defined as:

A. R₁ ÷ R₁₀

B. R₁₀ ÷ R₁

C. R₁₀ − R₁

D. R₁₀ × R₁

Correct Answer: B. R₁₀ ÷ R₁

Question 3

Which measurement is especially sensitive to impulsive early bearing damage?

A. Crest factor

B. Average ambient temperature

C. Nameplate voltage

D. Power factor alone

Correct Answer: A. Crest factor

Question 4

For a rotating shaft, mechanical output power is calculated using:

A. P = V/I

B. P = Tω

C. P = I²R only

D. P = mgh

Correct Answer: B. P = Tω

Question 5

A hot motor terminal is most directly suggestive of:

A. A loose or high-resistance connection

B. Perfect phase balance

C. Low environmental humidity only

D. Correct lubrication quantity

Correct Answer: A. A loose or high-resistance connection

Question 6

Why can a vibration component at 1× running speed not by itself prove imbalance?

A. Imbalance never produces vibration

B. Other faults, including misalignment and resonance, can produce it

C. 1× frequency cannot be measured

D. Vibration spectra contain no frequency information

Correct Answer: B. Other faults, including misalignment and resonance, can produce it

Question 7

Under the constant-failure-rate model, reliability is:

A. R(t) = 1 + λt

B. R(t) = λt

C. R(t) = e^(−λt)

D. R(t) = e^(λ/t)

Correct Answer: C. R(t) = e^(−λt)

Question 8

Which statement best describes a digital twin?

A. A static drawing of any motor

B. A data-connected digital representation of a specific physical asset

C. A replacement for all physical inspections

D. A sensor without a mathematical or historical model

Correct Answer: B. A data-connected digital representation of a specific physical asset

Question 9

Why is explainability important in AI-based machine diagnostics?

A. It guarantees perfect predictions

B. It allows evidence, uncertainty, and decision logic to be examined

C. It removes the need for cybersecurity

D. It prevents all sensor failures

Correct Answer: B. It allows evidence, uncertainty, and decision logic to be examined

Question 10

A motor has phase currents of 98 A, 101 A, and 106 A. Using the chapter's practical indicator, with average current 101.67 A, the approximate current unbalance is:

A. 1.64%

B. 4.26%

C. 8.00%

D. 10.67%

Correct Answer: B. 4.26%