8. Condition Monitoring and Predictive Maintenance

Introduction and Context

Maintenance decisions have traditionally been based on fixed intervals, operator observation, or failure response. These approaches remain useful, but they can also create unnecessary maintenance, overlook developing defects, and expose production systems to unexpected stoppages. Condition monitoring provides a more evidence-based alternative. It involves measuring the physical, chemical, electrical, or operational condition of equipment and interpreting changes in those measurements to determine whether maintenance is required.

Predictive maintenance is the managerial and technical practice of using condition information, historical evidence, and analytical methods to forecast asset deterioration and plan intervention before functional failure occurs. It is therefore broader than the installation of sensors. It includes measurement strategy, data quality, diagnosis, prognosis, work planning, execution, and verification.

The central challenge is not simply detecting an abnormal signal. It is translating diagnostic evidence into an economically and operationally sound decision. A high vibration level may indicate imbalance, looseness, resonance, or a measurement problem. A temperature rise may result from overload, poor lubrication, blocked ventilation, or an instrument error. Effective maintenance professionals combine technology with equipment knowledge, process understanding, and risk-based judgment.

Learning Objectives

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

1. Explain the relationship between condition monitoring, predictive maintenance, and reliability-centred decision-making.

2. Evaluate vibration, thermography, lubrication analysis, ultrasonic, electrical, and process-data techniques.

3. Interpret diagnostic evidence while recognising uncertainty and measurement limitations.

4. Design a practical condition-monitoring workflow and select appropriate technologies.

5. Translate condition information into inspection, repair, derating, shutdown, or continued-operation decisions.

6. Assess organizational, ethical, digital, and managerial issues in predictive maintenance.

From Measurement to Maintenance Decision

Condition monitoring is best understood as a closed decision cycle rather than a collection of instruments. The cycle begins with identification of a failure mode and its observable symptoms. A measurement method is then selected according to the physics of that failure mode. Data are acquired under controlled and repeatable conditions, transformed into meaningful indicators, interpreted against a baseline or threshold, and connected to a maintenance action.

A useful distinction is between detection, diagnosis, and prognosis. Detection determines that an abnormal condition exists. Diagnosis identifies the likely defect and its severity. Prognosis estimates how the condition may develop and how much useful operating time remains. Detection is often relatively easy; diagnosis and prognosis require better data, engineering knowledge, and uncertainty management.

The P–F curve provides a conceptual foundation. The point P represents the earliest stage at which a potential failure can be detected. The point F represents functional failure, when the asset can no longer perform its required function. The interval between P and F is the available warning period. A monitoring technology is useful only when its detection capability, inspection frequency, and organizational response provide sufficient time for a planned intervention. A technically sophisticated system is ineffective if the warning period is shorter than the time required to obtain parts and schedule work.

Condition-based maintenance is not identical to predictive maintenance. Condition-based maintenance initiates work when evidence indicates that action is warranted. Predictive maintenance adds the use of trends, models, and forecasts to anticipate future condition. Both differ from preventive maintenance based solely on elapsed time or usage, and from corrective maintenance performed after failure.

Important Concepts and Terminology

A baseline is a reference measurement representing acceptable operation under defined conditions. An alarm limit indicates that a response or further investigation is required, while a trip limit is associated with automatic or immediate protective action. Trending examines the direction and rate of change rather than a single reading. Repeatability concerns consistency under the same conditions; reproducibility concerns consistency across different operators, instruments, or locations.

Diagnostic evidence is information supporting a conclusion about an equipment defect. False positive means that an alarm suggests a defect that is not present; false negative means that a real defect is missed. Remaining useful life (RUL) is an estimate of the time or usage before a defined performance or safety limit is reached. Data fusion combines information from different technologies, such as vibration and process data, to improve confidence.

A monitoring route is a planned sequence of measurement points and tasks. Online monitoring collects data while equipment operates, whereas offline monitoring requires periodic inspection or connection of a portable instrument. Criticality expresses the consequence and likelihood of asset failure in relation to safety, environment, quality, delivery, and cost.

The Principal Monitoring Methods

Vibration analysis

Vibration is particularly valuable for rotating equipment such as motors, pumps, compressors, fans, gearboxes, and turbines. Accelerometers and velocity sensors measure mechanical motion. Overall amplitude can reveal a general deterioration, while time waveform, spectrum, phase, and envelope analysis can provide clues about the failure mode.

Imbalance commonly produces a strong component at running speed. Misalignment may generate running-speed and harmonic components, often accompanied by coupling or bearing stress. Mechanical looseness can create harmonics and impacts. Bearing defects may generate repetitive impacts whose frequencies are related to bearing geometry and rotational speed. Gear defects may appear as gear-mesh components and sidebands.

Interpretation requires attention to mounting, sensor orientation, speed, load, resonance, and operating condition. A high reading at a structural resonance does not necessarily mean that the defect is proportionately severe. Vibration should therefore be combined with phase information, visual inspection, lubrication evidence, and process conditions.

Infrared thermography

Thermography detects infrared radiation and presents temperature patterns across equipment surfaces. It is useful for electrical panels, switchgear, cable connections, bearings, couplings, furnaces, insulation, steam systems, and rotating machinery.

A thermal anomaly may indicate high electrical resistance, phase imbalance, friction, inadequate lubrication, overload, blocked airflow, or insulation failure. The most meaningful comparison is often between similar components operating under similar load rather than against a universal temperature value. Emissivity, reflected radiation, distance, viewing angle, atmospheric conditions, and load must be considered.

Thermography identifies a thermal symptom, not automatically its cause. A hot connection may result from looseness, corrosion, overloading, or an undersized conductor. Electrical equipment should be inspected by competent personnel, with appropriate isolation and arc-flash controls.

Lubrication and oil analysis

Lubricants carry information about both the lubricant and the machine. Laboratory or field tests may examine viscosity, water content, particle count, elemental composition, oxidation, acidity, ferrous debris, and additive depletion. Wear metals can suggest component deterioration, while contaminant levels may reveal ingress of dust, water, or process material.

Oil analysis is most effective when samples are representative. Sampling from a drain point after prolonged shutdown may produce different results from sampling in a live circulation line. Sample location, cleanliness, frequency, labeling, and consistency are fundamental. A single abnormal result should normally be confirmed and interpreted with operating history. Changing oil without removing the contamination source may conceal rather than solve the problem.

Ultrasonic monitoring

Ultrasonic instruments detect high-frequency sound generated by turbulence, friction, impact, and leaks. They are used to locate compressed-air and steam leaks, inspect valves and traps, assess lubrication of bearings, and identify electrical discharge such as arcing or corona.

Ultrasound is advantageous in noisy industrial environments because high-frequency signals can often be isolated and converted into audible or numerical information. However, air leaks are affected by distance, pressure, background interference, and access. An ultrasonic indication should be verified before a repair decision, particularly where the cost of intervention is significant.

Electrical testing

Electrical condition monitoring includes motor current signature analysis, insulation-resistance testing, polarization measurements, winding-resistance tests, power-quality analysis, and partial-discharge assessment. These methods can identify phase imbalance, rotor-bar problems, insulation deterioration, loose connections, harmonics, and abnormal loading.

Electrical tests must be matched to the equipment state. Some tests are online and non-intrusive; others require isolation and may impose electrical stress. Results should be assessed against manufacturer guidance, asset history, temperature, humidity, load, and test voltage. A test value without context can lead to unjustified confidence or unnecessary replacement.

Process-data analysis

Many failures first appear as changes in process variables: pressure, flow, temperature, speed, power, cycle time, product quality, or energy consumption. Supervisory control and data acquisition systems, historians, and manufacturing execution systems can provide continuous operating data.

Process analysis is powerful because it connects equipment health with the function the asset performs. A pump may show normal vibration but declining flow at the same speed, indicating hydraulic deterioration or a process restriction. A compressor’s increased power per unit of output may reveal fouling or valve problems. Statistical process control, multivariable regression, anomaly detection, and machine-learning models can support such analysis.

Models must distinguish equipment deterioration from changes in product mix, ambient conditions, set points, and operator practice. A black-box alarm that cannot be explained or acted upon is unlikely to create sustainable value.

Selecting and Combining Techniques

Technology selection should begin with the failure mode, not with the instrument available in the maintenance department. A practical selection sequence is:

1. Define the asset function and consequences of failure.

2. Identify credible failure modes and their physical symptoms.

3. Determine the warning period and required inspection frequency.

4. Select the least costly method that can reliably detect the relevant condition.

5. Establish measurement points, operating-state requirements, and data ownership.

6. Define alarm logic, escalation rules, and the required maintenance response.

7. Pilot the method, validate findings, and revise the programme.

No single technique covers all failure modes. Vibration may reveal a bearing defect while oil analysis reveals contamination. Thermography may identify a hot terminal while power-quality analysis explains the electrical cause. Combining independent evidence reduces ambiguity, but it also requires disciplined interpretation. More data do not automatically mean better decisions.

Thresholds may be absolute, statistical, model-based, or rate-of-change limits. Absolute limits are simple but may ignore asset-specific behaviour. Statistical limits compare a measurement with a stable population or baseline. Rate-of-change alarms can identify rapidly developing defects even when the absolute value remains moderate. Alarm settings should be reviewed after confirmed failures and false alarms.

Translating Diagnosis into Action

A condition report should answer four questions: What has changed? What is the probable cause? What is the consequence if no action is taken? What should be done, by when, and with what degree of urgency?

Possible actions include continued operation with increased monitoring, verification by a second method, adjustment of operating conditions, lubrication or cleaning, planned repair, immediate shutdown, or redesign. The decision should consider safety, environmental exposure, production loss, spare-part availability, repair lead time, redundancy, and the confidence of the diagnosis.

A useful decision matrix combines condition severity and consequence of failure. A moderate anomaly on a non-critical standby pump may justify a planned inspection. The same anomaly on a single boiler-feed pump may require immediate risk controls and accelerated intervention. Predictive maintenance is therefore a risk-management activity, not merely a condition-score exercise.

Maintenance work orders should preserve the evidence that generated them. The work package may specify the defect hypothesis, measurement trend, photographs or spectra, required tools, isolation requirements, acceptance criteria, and post-repair verification. After work is completed, the asset should be re-measured to determine whether the suspected cause was removed.

Applied Illustrative Case: Predictive Monitoring of a Process Pump

Background

A chemical-processing plant operated six centrifugal pumps serving a critical cooling circuit. One pump was a duty unit with limited standby capacity. The maintenance department had experienced repeated bearing replacements, but failures were usually discovered after elevated bearing temperature or abnormal noise had appeared.

Situation and analysis

A route-based programme collected monthly vibration, infrared, and oil samples. Over three months, the pump’s vibration velocity increased gradually, with a growing component at running speed and a moderate increase in the axial direction. Thermography showed a small temperature difference between the drive-end and non-drive-end bearing housings. Oil analysis showed rising ferrous debris but no significant water contamination.

The evidence suggested developing mechanical deterioration rather than simple lubricant contamination. Inspection of the coupling revealed angular misalignment caused by movement of the pump base. The misalignment increased bearing loading and generated the observed vibration pattern. The plant’s first reaction could have been to replace the bearing, but that would have treated the symptom.

Approach and solution

The reliability engineer classified the condition as requiring planned intervention within the available warning period. Operations temporarily reduced the pump load where process constraints permitted, and a replacement coupling element was ordered. During the scheduled window, technicians corrected the base and alignment, inspected the bearing, renewed contaminated lubricant, and recorded alignment measurements. Vibration and thermal measurements were repeated after return to service.

Outcome

The post-repair trend showed a substantial reduction in running-speed vibration and a stabilised temperature profile. The bearing was not replaced because inspection did not indicate unacceptable damage. Route frequency was temporarily increased, and alignment checks were added to the pump’s maintenance standard.

Learning points

The case demonstrates that a condition indicator rarely identifies the complete cause by itself. Trend information created planning time; multiple techniques increased diagnostic confidence; and root-cause correction prevented a repetitive bearing replacement. It also illustrates that predictive maintenance can reduce intervention rather than increase it when evidence supports continued service.

Implementation, Governance, and Professional Practice

A successful programme requires more than sensors. Asset criticality, failure-mode analysis, competent analysts, calibrated instruments, standard data definitions, and a functioning computerized maintenance management system are essential. The programme should assign ownership for data collection, diagnosis, work approval, and feedback.

Common implementation failures include collecting data without a response process, selecting alarm limits from generic tables without local validation, ignoring process context, measuring at inconsistent locations, and treating vendor-generated scores as unquestionable diagnoses. Excessive alarms produce alert fatigue. Poorly maintained sensors create false assurance. A predictive system should therefore be audited through indicators such as valid alarm rate, avoided failures, planned versus emergency work, response time, and post-maintenance confirmation—not merely the number of sensors installed.

Safety and ethics are central. A predictive model must not encourage continued operation beyond safe limits merely to protect production output. Personnel should be trained in electrical hazards, rotating equipment isolation, confined-space requirements, and data integrity. Cybersecurity is increasingly relevant because connected monitoring systems may provide pathways into operational technology networks. Access control, network segmentation, secure updates, and backup procedures should be incorporated into the programme.

Data-driven maintenance also raises questions of accountability. If an algorithm recommends continued operation and a failure occurs, responsibility cannot be transferred to the software. Human experts must understand the model’s scope, uncertainty, assumptions, and failure cases. Personal data from operator systems should be managed according to organizational policy and applicable law.

Digital Development and Emerging Practice

Industrial Internet of Things platforms increasingly connect wireless sensors, historians, mobile applications, and maintenance systems. Edge computing can process data near the machine, reducing communication requirements and enabling rapid alerts. Cloud platforms support fleet-level comparison and remote expertise. Digital twins combine engineering models, sensor data, and operating history, although their value depends on model quality and disciplined asset data.

Artificial intelligence and machine learning are being applied to anomaly detection, image interpretation, remaining-useful-life estimation, and work-order prioritization. These methods are useful where relationships are complex, but they require representative training data and controls against concept drift—the condition in which operating behaviour changes over time. Explainable outputs, human review, and periodic model validation are preferable to unexamined automation.

The emerging direction is not technology replacement of maintenance professionals but augmented decision-making. The most effective systems integrate physical understanding, statistical analysis, operator knowledge, and maintenance execution.

Best-Practice Principles

A mature condition-monitoring programme:

• begins with criticality and failure modes;
• chooses methods according to defect physics;
• establishes repeatable measurement conditions;
• uses trends and multiple evidence sources where appropriate;
• links every alarm to an explicit response;
• records uncertainty and verifies diagnoses;
• corrects root causes rather than merely replacing damaged parts;
• integrates findings with planning, stores, safety, and production;
• reviews false alarms, missed failures, and economic outcomes; and
• protects monitoring systems and maintains professional accountability.

Chapter Summary

Condition monitoring provides evidence about the developing condition of an asset, while predictive maintenance uses that evidence to anticipate deterioration and plan intervention. The P–F interval explains why detection timing, inspection frequency, and organizational response are inseparable. Vibration, thermography, lubrication analysis, ultrasonics, electrical testing, and process-data analysis each reveal different physical symptoms and have specific limitations. Sound decisions require diagnosis in context, not blind reliance on thresholds or software scores.

An effective programme connects failure-mode analysis, reliable measurement, trend interpretation, risk-based action, work execution, and post-maintenance verification. Digital technologies, analytics, and AI can extend capability, but they do not remove the need for engineering judgment, safety governance, data quality, and accountability.

Key Terms

Condition monitoring: Systematic measurement and interpretation of asset condition.

Predictive maintenance: Maintenance planning based on condition evidence, trends, and forecasts.

P–F interval: Time between detectable potential failure and functional failure.

Baseline: Reference condition measurement under defined operating circumstances.

Alarm limit: Value or pattern requiring investigation or response.

Trip limit: Protective threshold associated with immediate automatic or manual action.

Detection: Recognition that an abnormal condition exists.

Diagnosis: Identification of the likely defect or cause.

Prognosis: Estimation of future condition or remaining useful life.

Trending: Analysis of measurements over time.

Vibration spectrum: Representation of vibration amplitude across frequencies.

Thermography: Infrared imaging used to evaluate temperature patterns.

Oil analysis: Examination of lubricant properties, contaminants, and wear debris.

Ultrasonic monitoring: Detection and interpretation of high-frequency sound.

Motor current signature analysis: Evaluation of electrical current patterns to assess motor condition.

Process-data analysis: Use of operating variables and production data to identify asset abnormalities.

False positive: Alarm indicating a defect that is not actually present.

Data fusion: Combination of evidence from multiple monitoring sources.

Remaining useful life: Estimated operating time before a defined failure or performance limit.

Concept drift: Change in operating relationships that reduces the accuracy of an analytical model.

Review Questions

Short Answer Questions

1. Distinguish condition-based maintenance from predictive maintenance.

2. What do the P and F points represent on a P–F curve?

3. State two factors that affect the interpretation of thermographic images.

4. Why is representative sampling important in oil analysis?

5. Define a false negative in condition monitoring.

Long Answer Questions

1. Explain how vibration analysis can support diagnosis of rotating-equipment faults and discuss its limitations.

2. Describe a complete workflow for converting a condition-monitoring alarm into a maintenance work order.

3. Compare thermography, ultrasonics, and electrical testing in terms of applications and constraints.

4. Discuss how process data can complement physical condition-monitoring techniques.

5. Explain the organizational and governance requirements for a mature predictive-maintenance programme.

Analytical / Application Questions

1. A motor shows rising vibration, stable temperature, and increasing power consumption under the same production load. Which additional evidence would you collect, and how would you decide between continued operation and planned shutdown?

2. An oil sample from a gearbox shows increased iron particles but normal viscosity and water content. Develop a diagnostic and verification plan.

3. A plant has many wireless sensors but continues to experience emergency failures. Analyse the likely organizational and technical causes and propose corrective actions.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes predictive maintenance from condition-based maintenance?

A. Predictive maintenance is performed only after functional failure.

B. Predictive maintenance uses trends or forecasts to anticipate future condition.

C. Condition-based maintenance never uses measurement data.

D. Predictive maintenance is based exclusively on fixed calendar intervals.

Correct Answer: B. Predictive maintenance uses trends or forecasts to anticipate future condition.

Question 2

On a P–F curve, the interval between P and F primarily represents:

A. The time available for planned intervention after potential failure is detectable

B. The time required to install a sensor

C. The period between two preventive maintenance tasks

D. The duration of a production shift

Correct Answer: A. The time available for planned intervention after potential failure is detectable

Question 3

A strong running-speed vibration component in a rotating machine most commonly suggests which initial hypothesis?

A. Water contamination in lubricant

B. Imbalance

C. Insulation breakdown

D. Compressed-air leakage

Correct Answer: B. Imbalance

Question 4

Which factor is especially important when interpreting an infrared image?

A. Emissivity and reflected radiation

B. Bearing geometry only

C. The colour of the machine paint only

D. The number of maintenance staff on shift

Correct Answer: A. Emissivity and reflected radiation

Question 5

Why should an oil sample be collected consistently from a representative location?

A. To ensure the result reflects the machine’s operating lubricant condition

B. To eliminate the need for trend analysis

C. To guarantee that all wear has stopped

D. To avoid recording the machine’s operating history

Correct Answer: A. To ensure the result reflects the machine’s operating lubricant condition

Question 6

Ultrasonic monitoring is particularly suitable for detecting:

A. Compressed-air leaks and certain electrical discharges

B. Only low-frequency shaft imbalance

C. Chemical composition of lubricants

D. Product demand fluctuations

Correct Answer: A. Compressed-air leaks and certain electrical discharges

Question 7

A predictive-maintenance alarm should normally lead first to which action?

A. Automatic replacement of the entire asset

B. Interpretation and verification in operating context

C. Deletion of the measurement from the database

D. Immediate shutdown in every case

Correct Answer: B. Interpretation and verification in operating context

Question 8

What is a major danger of using generic alarm limits without local validation?

A. They may create false alarms or miss asset-specific deterioration

B. They always eliminate the need for skilled analysts

C. They prevent all forms of data collection

D. They guarantee accurate remaining-useful-life estimates

Correct Answer: A. They may create false alarms or miss asset-specific deterioration

Question 9

In the illustrative pump case, replacing the bearing alone would have been inadequate because the underlying issue was:

A. A compressed-air leak

B. Base movement and coupling misalignment

C. Excessive water in the oil

D. A process historian failure

Correct Answer: B. Base movement and coupling misalignment

Question 10

Which practice best supports responsible use of AI in predictive maintenance?

A. Allowing the model to override all safety limits

B. Using representative data, human review, and periodic model validation

C. Removing all engineering explanations from alarm reports

D. Assuming model accuracy remains constant despite process changes

Correct Answer: B. Using representative data, human review, and periodic model validation