Smart Maintenance, Digital Transformation and Future Practice

Introduction and context

Maintenance management is changing from a mainly reactive, equipment-centred activity into a data-enabled business capability. Conventional preventive maintenance schedules work from elapsed time or operating cycles. Modern smart maintenance combines condition monitoring, connected assets, analytical models, mobile work execution and human expertise to decide what should be done, when it should be done and why.

This transformation is not simply the purchase of sensors or artificial-intelligence software. It is a socio-technical change involving asset strategy, work processes, information quality, cybersecurity, skills, governance and organisational culture. A technically advanced platform will not create value if technicians do not trust its recommendations, asset records are inaccurate, alarms are excessive, or production and maintenance departments pursue conflicting objectives.

For production managers, the central question is not “How can we automate maintenance?” but “How can digital information improve safe, reliable, sustainable and economically sound decisions?”

Learning objectives

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

1. Explain smart maintenance and distinguish CMMS, EAM, IIoT, digital twins and predictive maintenance.
2. Evaluate CMMS/EAM platforms using functional, technical, economic and governance criteria.
3. Describe the role of sensors, mobile systems, drones, augmented reality and artificial intelligence.
4. Assess data quality, cybersecurity, privacy, algorithmic bias and human oversight.
5. Select suitable digital-maintenance applications for different asset risks and operating contexts.
6. Recommend implementation practices that support sustainability, resilience and emerging workforce skills.

1. From planned maintenance to intelligent maintenance

Maintenance strategies are commonly classified as corrective, preventive, predictive and proactive.

Corrective maintenance restores an asset after a failure or after an unacceptable condition has been detected. It may be appropriate for low-consequence, inexpensive and easily replaceable items, but it is dangerous when failure threatens people, the environment or production continuity.

Preventive maintenance performs an intervention at a predetermined time or usage interval. It is useful when failure probability increases predictably with age, but excessive scheduled work can introduce maintenance-induced failures and waste components.

Predictive maintenance uses measurements such as vibration, temperature, oil condition, electrical current or acoustic emissions to estimate changing equipment condition and support intervention before functional failure.

Proactive maintenance seeks to remove the causes of failure, such as contamination, misalignment, poor lubrication, incorrect installation or unsuitable operating conditions. Digital tools are most valuable when they support this learning rather than merely generate more alarms.

Smart maintenance integrates these strategies. A critical pump may have statutory inspections, condition sensors, operator rounds, a spare-parts policy and an AI-supported anomaly model. The final decision remains a maintenance-management decision that considers safety, production schedules, repairability, cost and risk.

A useful reliability measure is mean time between failures (MTBF):

where T_operating is total operating time and N_failures is the number of relevant failures. MTBF is meaningful only when the observation period, failure definition and operating population are consistent. It is not automatically a probability of survival and should not be compared across assets with different duty cycles without adjustment.

Availability is often represented, for a repairable asset under stable conditions, as:

where A is inherent or operationally simplified availability, and MTTR is mean time to repair. The expression assumes that MTBF and MTTR are representative averages and that the chosen definition of downtime is consistent. In real plants, logistics, waiting for permits, spare parts and planned shutdowns may require a broader operational-availability model.

2. The digital maintenance architecture

A smart-maintenance system usually contains six interacting layers:

1. Physical assets: machines, tools, utilities, buildings and safety systems.
2. Sensing and identification: sensors, barcode or RFID tags, machine controllers and inspection devices.
3. Connectivity and edge processing: industrial networks, gateways and local computing that filter or analyse data near the asset.
4. Information platforms: CMMS, EAM, historians, data lakes and integration middleware.
5. Analytics and decision support: rules, statistical models, digital twins, machine learning and dashboards.
6. Human work execution: planners, technicians, operators, engineers, supervisors and managers using recommendations to act.

The architecture should support a closed loop: sense, interpret, decide, execute, verify and learn. A disconnected dashboard is not a closed-loop maintenance system.

Figure: The smart-maintenance loop from physical condition to verified work and organisational learning.

3. CMMS and EAM platform evaluation

A computerised maintenance management system (CMMS) manages work orders, preventive schedules, labour, materials, equipment history and maintenance reports. An enterprise asset management (EAM) platform normally covers a wider asset lifecycle, including specification, procurement, commissioning, operation, maintenance, renewal and disposal. The distinction varies between vendors, so functionality should be assessed rather than inferred from product labels.

Important capabilities include:

• asset hierarchy, functional locations and criticality classification;
• failure modes, job plans and preventive-maintenance libraries;
• work requests, approvals, permits, planning, scheduling and close-out;
• labour competencies, contractor control and time recording;
• inventory, bills of material, purchasing and stock optimisation;
• mobile access, offline operation, barcode/RFID and electronic signatures;
• integration with enterprise-resource-planning, supervisory-control, laboratory, energy and safety systems;
• dashboards, audit trails, configurable workflows and exportable data;
• role-based access, backup, recovery and cybersecurity controls.

A platform-selection process should begin with business and safety requirements, not demonstrations of attractive screens. A weighted decision matrix can make evaluation transparent:

Here S is the total score, w_i is the importance weight for criterion i, and r_i is the normalised rating of a supplier or platform for that criterion. Ratings must use a defined scale, such as 0 to 5. The result supports comparison but does not replace due diligence, reference visits, cybersecurity assessment or a pilot.

Typical criteria may include reliability of offline mobile operation, integration capability, total cost of ownership, usability, vendor viability, data portability, safety workflows and support for local regulations. Procurement should also clarify ownership of operational data, exit arrangements, service-level commitments, software-update responsibilities and access to audit evidence.

4. Sensors, IIoT and data quality

The industrial Internet of Things (IIoT) connects industrial assets, people and software so that information can be collected and used across operational processes. Sensors may measure vibration, speed, pressure, flow, temperature, current, humidity, corrosion, position or acoustic energy. Sensor choice must follow a failure-mode analysis. Measuring a variable merely because it is easy to measure can create cost without decision value.

A practical sensing decision asks:

• What failure mode is being detected?
• Which physical variable changes before functional failure?
• What sampling rate, accuracy, range and environmental protection are required?
• Is the sensor calibrated and traceable?
• What action follows an abnormal value?
• Can the measurement be maintained throughout the asset life?

Data quality has several dimensions. Accuracy concerns closeness to the true value; completeness concerns missing records; timeliness concerns whether information arrives when needed; consistency concerns agreement across systems; validity concerns conformance to permitted formats and ranges; and uniqueness concerns avoidance of duplicate assets or events.

A maintenance-data quality programme should establish naming conventions, asset identifiers, units, time synchronisation, master-data ownership, validation rules and procedures for correcting historical records. Sensor data should retain metadata: location, calibration status, sampling frequency, engineering unit, transformation and quality flag. A value without context may be impossible to interpret safely.

Data cleansing cannot correct a fundamentally unsuitable measurement. Nor should missing data automatically be replaced by a mean value; imputation can conceal a developing fault. Missingness itself may indicate a failed sensor, network problem or unsafe operating condition.

5. Predictive analytics and AI-enabled maintenance

Predictive maintenance may use threshold rules, statistical process control, regression, survival analysis, signal processing or machine learning. A useful distinction is between:

• diagnosis: what has changed or failed;
• prognosis: how the condition may develop and, where supportable, the remaining useful life;
• prescription: what action should be taken, considering risk, cost and production constraints.

Supervised learning requires labelled examples, such as confirmed bearing failures. Unsupervised or semi-supervised methods identify departures from normal behaviour and are useful where failure labels are scarce. Deep learning can process complex signals but often requires substantial data, computing resources and explanation methods.

A model should be evaluated using data separated by time or asset, not merely random records from the same operating episode. Otherwise, information leakage can make performance appear better than it will be in practice. Relevant measures include precision, recall, false-alarm rate, missed-failure rate, calibration and economic value. In safety-critical settings, the cost of a missed failure may be far greater than the cost of an inspection, so accuracy alone is inadequate.

For a binary alert, precision and recall are defined as:

TP denotes true positives, FP false positives and FN false negatives. Precision answers, “Of the alerts issued, how many were correct?” Recall answers, “Of the failures that occurred, how many were detected?” These measures depend on the selected threshold and on reliable labels.

AI should be treated as decision support unless its safety case, validation, monitoring and authority boundaries justify greater automation. Human experts must be able to review evidence, reject an implausible recommendation, record the reason and escalate uncertainty. Models also require drift monitoring because equipment, products, operating regimes and sensor behaviour change over time.

6. Digital twins, mobile work, drones and augmented reality

A digital twin is a digital representation connected to a physical asset or process through data and models. It is more than a three-dimensional drawing. A useful twin contains an explicit purpose, identity linkage, relevant state variables, model assumptions, update mechanisms and defined users. A design model may support commissioning, while an operational twin may support anomaly detection, energy optimisation or scenario testing.

Digital twins should not imply perfect real-time knowledge. They are abstractions with uncertainty. Engineering judgement is required when sensor coverage is incomplete, models are extrapolated beyond tested conditions or operating data are unreliable.

Mobile maintenance systems allow technicians to receive work, view procedures, scan assets, capture readings, attach photographs and close tasks at the point of work. Offline capability is essential in remote or shielded locations. Interfaces should minimise typing, display critical safety information clearly and avoid turning technicians into clerical data-entry operators.

Drones can inspect roofs, stacks, storage tanks, power lines and hazardous or inaccessible areas. They may reduce exposure to work at height or confined-space entry. Their use requires flight planning, competent operators, airspace compliance, privacy controls, weather limits, battery management and a process for converting imagery into an actionable defect record.

Augmented reality (AR) overlays digital instructions, component identification or remote expert guidance onto the technician’s view. AR is valuable for complex procedures and training, but inaccurate spatial registration, poor lighting, network failure or distracting displays can increase risk. A conventional controlled procedure must remain available as a fallback.

7. Cybersecurity, privacy and governance

Connected maintenance expands the attack surface. Threats include unauthorised remote access, ransomware, manipulation of sensor values, theft of maintenance records, denial of service and compromise of safety or control networks. Cybersecurity should follow a defence-in-depth approach:

• identify assets, data flows, dependencies and critical functions;
• segment operational technology from enterprise and internet-facing networks;
• apply least privilege, strong authentication and controlled remote access;
• patch and manage vulnerabilities according to operational risk;
• maintain secure backups and tested recovery procedures;
• monitor logs and unusual behaviour;
• control removable media and supplier access;
• rehearse incident response without endangering production or safety.

Cybersecurity and functional safety are related but different. Cybersecurity protects against intentional or accidental interference; functional safety addresses hazardous failures and required risk reduction. A cyber event can create a functional-safety hazard, so the disciplines must cooperate.

Maintenance systems may contain worker names, location records, photographs, health-related exposure information, productivity data and contractor details. Privacy governance should define lawful purpose, data minimisation, access rights, retention periods, transparency and secure disposal. Monitoring a worker’s activity through a mobile device should not be expanded beyond the legitimate safety or operational purpose without consultation and appropriate authorisation.

AI governance should document the model owner, training data, intended use, prohibited use, performance limits, review frequency, human authority and escalation path. Algorithmic bias may arise when one asset type, shift, geography or operating condition is overrepresented in training data. A model that performs well on large, modern assets may perform poorly on older equipment, even though the latter may have greater safety significance.

8. Sustainability, resilience and human capability

Digital maintenance can reduce waste by preventing catastrophic failures, optimising lubrication, reducing unnecessary replacement and improving energy performance. It can also increase environmental impact through additional sensors, batteries, network equipment, data centres, electronic waste and premature component replacement. A sustainability assessment should consider the whole lifecycle rather than assuming that more data are automatically greener.

Maintenance supports resilience by improving detection, redundancy planning, spare-parts visibility, cross-training and recovery from disruption. However, excessive dependence on a single cloud vendor, proprietary data format or unavailable communications channel can create new fragility. Critical operations require degraded-mode procedures: local control, paper or offline work instructions, manual inspection and tested restoration from backup.

Emerging skills include condition-monitoring interpretation, reliability engineering, data literacy, systems integration, cybersecurity awareness, digital ethics, visual inspection, communication and change leadership. Technicians do not need to become software developers, but they must understand sensor limitations, data quality and the meaning of model outputs. Data specialists need operational knowledge; otherwise they may optimise a metric that has little relation to safe production.

A mature organisation combines technology with a learning culture. Failure reports should examine system conditions rather than blame individuals. Communities of practice, competence matrices, simulation and joint operator–technician problem solving help embed digital tools into real work.

9. Implementation framework

A staged implementation reduces technical and organisational risk.

Stage 1: Define value and risk. Identify critical assets, failure consequences, regulatory obligations and decisions that currently suffer from poor information. Establish baseline measures such as emergency-work percentage, schedule compliance, repeat failures, downtime and safety exposure.

Stage 2: Establish information foundations. Rationalise asset hierarchies, standardise failure codes, verify preventive tasks, assign data ownership and synchronise time and units.

Stage 3: Select a focused pilot. Choose an asset family with a meaningful failure problem, measurable benefit and sufficient technical access. Do not select only the easiest asset if it cannot demonstrate operational value.

Stage 4: Integrate work execution. Ensure alerts create appropriate work requests, planners can assess priority, technicians can act safely and completed work returns useful feedback to the model.

Stage 5: Validate and scale. Compare results with a baseline or control population where practical. Review false alarms, missed failures, user acceptance, cybersecurity events and total cost. Scale only after the process—not merely the software—has been proven.

A business case should include acquisition, integration, training, calibration, connectivity, support, cybersecurity, model maintenance and eventual retirement costs. Benefits may include avoided downtime, reduced exposure, lower spare consumption, better energy performance and improved regulatory evidence. Claimed benefits should be measured against credible counterfactuals; a reduction in downtime may result from production changes rather than the digital intervention.

10. Illustrative case: connected cooling-water pumps

Background. An illustrative medium-sized process plant operated twelve cooling-water pumps. Repeated bearing failures caused production interruptions and required technicians to work near hot equipment. The existing strategy combined monthly operator rounds with time-based bearing replacement.

Problem. Failures occurred between rounds, while many scheduled replacements found healthy bearings. The CMMS contained duplicate equipment records and inconsistent failure codes. Managers considered installing wireless vibration sensors and an AI service.

Analysis. A criticality review showed that four pumps supported the most consequential process trains. A failure-mode analysis identified lubrication contamination, misalignment and bearing damage as important modes. Engineers determined that vibration and temperature trends could support detection, but a single threshold could not distinguish all operating states. The team also found that pump speed varied with production demand, requiring operating context in the analysis.

Approach. The plant first corrected asset identities, standardised units and recorded pump speed with each measurement. Sensors were installed on the four critical pumps, with local buffering for network interruptions. A rule-based baseline model was used before machine learning. Alerts were reviewed weekly by a reliability engineer and planner. An alert could recommend inspection, but it could not automatically stop a pump. Technicians used mobile work orders and recorded cause, action and evidence using controlled codes.

Cybersecurity staff segmented the sensor gateway, restricted supplier access and tested recovery. Workers were informed about the purpose and limits of monitoring. The team tracked false alarms, missed events, inspection findings, emergency work and bearing consumption.

Outcome. After the pilot period, the plant reported earlier identification of two lubrication problems and avoided one unplanned pump outage. It also discovered that several alerts were caused by loose sensor mounting, demonstrating the importance of installation quality. The financial result was positive only after the team included avoided exposure and reduced unnecessary bearing replacement; the sensors alone would not have delivered the benefit.

Learning points. The case illustrates that data governance precedes analytics, operating context improves interpretation, human review remains essential, and a pilot should test the complete sense-to-action process. It also shows why technology suppliers should not define success solely as the number of alerts generated.

Challenges and limitations

Smart maintenance has significant limitations. Sensors fail and require calibration. Wireless systems may suffer interference. Historical failure data are often sparse and biased toward severe events. Models may produce false confidence, especially when rare failures are represented by few examples. Cloud services may be unavailable during network disruption. Integration projects can exceed budgets because asset hierarchies and interfaces are more complex than expected.

There is also a risk of automation bias: users may accept a computer recommendation despite contradictory physical evidence. Conversely, alert fatigue may cause users to ignore valid warnings. These risks require workload design, alarm rationalisation, explanation of evidence, competency development and periodic independent review.

No digital system removes the need for safe isolation, permits, competent inspection, engineering standards or statutory examinations. A prediction is not proof of condition, and a normal reading is not proof that an asset is safe. Digital evidence must be combined with physical verification and sound risk assessment.

Best-practice principles

1. Start with critical decisions and failure consequences, not fashionable technology.
2. Build trustworthy asset and failure data before advanced analytics.
3. Link every alert to an authorised action and a feedback mechanism.
4. Preserve human authority, explainability and safe fallback modes.
5. Measure economic, safety, environmental and reliability outcomes together.
6. Design cybersecurity, privacy and data ownership into procurement.
7. Validate models across assets, operating states and time periods.
8. Monitor model drift, sensor health and changing failure patterns.
9. Develop technicians, planners, engineers and managers as a connected capability.
10. Scale only after a controlled pilot demonstrates repeatable value.

Chapter Summary

Smart maintenance integrates condition information, connected assets, enterprise maintenance systems, analytics and human work execution. CMMS platforms primarily organise maintenance work, while EAM systems address the broader asset lifecycle. Sensors and IIoT create value only when measurements are relevant, contextualised, secure and linked to decisions. Digital twins provide purpose-specific representations rather than perfect copies of physical assets.

Mobile systems, drones and augmented reality can improve access, inspection and work quality, but they require competent procedures and fallback arrangements. AI-enabled predictive maintenance can support diagnosis, prognosis and prescription; however, model performance depends on representative data, appropriate validation, drift monitoring and human oversight. Precision and recall describe different aspects of alert performance and must be interpreted in relation to the consequences of false and missed alarms.

Successful transformation requires accurate master data, cybersecurity, privacy protection, bias assessment, sustainability analysis and resilience planning. The most durable capability is socio-technical: it combines technology with reliability engineering, safe work, organisational learning and developing workforce skills.

Key Terms

Augmented reality: Digital information overlaid on a user’s view of the physical environment.

CMMS: Computerised maintenance management system for organising assets, work, labour, materials and history.

Condition monitoring: Repeated measurement and interpretation of equipment condition.

Digital twin: A purpose-specific digital representation linked to a physical asset or process.

EAM: Enterprise asset management across the asset lifecycle.

Edge computing: Processing data near the source rather than sending all data to a central system.

IIoT: Industrial Internet of Things connecting industrial assets, systems and users.

Model drift: Decline in model suitability as data, equipment or operating conditions change.

MTBF: Mean operating time between relevant failures.

MTTR: Mean time required to restore an asset after failure, according to a defined boundary.

Predictive maintenance: Maintenance supported by condition information and forecasts of developing failure.

Precision: Proportion of issued positive alerts that are true positives.

Prognosis: Estimation of how a diagnosed condition may develop.

Recall: Proportion of actual positive events detected by a model.

Reliability: Ability of an item to perform a required function for a specified time and conditions.

Smart maintenance: Integrated, data-enabled maintenance involving connected assets, analytics and human decisions.

Total cost of ownership: All relevant acquisition, operation, support, risk and disposal costs.

Work-order close-out: Recording the completed work, findings, causes, resources and evidence in the maintenance system.

Review Questions

Short Answer Questions

1. Distinguish corrective, preventive, predictive and proactive maintenance.
2. What is the principal difference between a CMMS and an EAM platform?
3. State four important dimensions of maintenance-data quality.
4. Why is human oversight important in AI-enabled maintenance?
5. Identify two cybersecurity controls appropriate for connected industrial assets.

Long Answer Questions

1. Explain the six-layer architecture of a smart-maintenance system and show how information should move from sensing to verified work.
2. Discuss the criteria that should be used to evaluate and procure a CMMS or EAM platform.
3. Explain how digital twins, mobile systems, drones and augmented reality can improve maintenance while introducing new risks.
4. Discuss the ethical and governance issues associated with worker data, algorithmic bias and automated maintenance recommendations.
5. Develop a staged implementation plan for introducing predictive maintenance in a production plant.

Analytical / Application Questions

1. A plant has frequent pump failures but incomplete failure records and inconsistent asset identifiers. Recommend the first three actions before purchasing an AI prediction platform, and justify each action.
2. A model produces 80 alerts, of which 60 are confirmed genuine; 20 actual failures occurred, including 15 detected by the model. Calculate precision and recall, and explain which operational risks the two measures reveal.
3. Select one critical asset from a manufacturing plant and design a pilot covering failure mode, sensor choice, data governance, work execution, cybersecurity, human review and success measures.

Multiple Choice Questions with Answers

Question 1

Which statement best describes an EAM platform?

A. It is limited to automatic vibration measurement.

B. It manages the wider asset lifecycle, including operation, maintenance and renewal.

C. It replaces all safety procedures with artificial intelligence.

D. It is only a paper-based spare-parts register.

Correct Answer: B. It manages the wider asset lifecycle, including operation, maintenance and renewal.

Question 2

In the simplified availability expression A = MTBF ÷ (MTBF + MTTR), what does MTTR represent?

A. Mean time to repair

B. Maximum temperature tracking range

C. Monthly total reliability

D. Mean time between readings

Correct Answer: A. Mean time to repair

Question 3

Which data-quality dimension concerns whether information arrives when it is needed?

A. Uniqueness

B. Validity

C. Timeliness

D. Accuracy

Correct Answer: C. Timeliness

Question 4

A model issues 80 alerts and 60 are confirmed true positives. What is its precision?

A. 25%

B. 60%

C. 75%

D. 80%

Correct Answer: C. 75%

Question 5

Which activity is an example of proactive maintenance?

A. Replacing every bearing at a fixed calendar date

B. Repairing a motor only after it stops

C. Removing the cause of repeated misalignment

D. Ignoring low-priority equipment failures

Correct Answer: C. Removing the cause of repeated misalignment

Question 6

What is a major purpose of edge computing in an IIoT system?

A. To process selected data near the asset

B. To eliminate the need for asset identification

C. To guarantee that sensors never fail

D. To prevent all human review

Correct Answer: A. To process selected data near the asset

Question 7

Which practice best reduces automation bias?

A. Allowing the model to stop all equipment automatically

B. Removing explanations from alerts

C. Requiring competent review and allowing recommendations to be challenged

D. Ignoring physical inspection results

Correct Answer: C. Requiring competent review and allowing recommendations to be challenged

Question 8

Why should model validation use time-separated or asset-separated data where appropriate?

A. To increase the number of duplicate records

B. To reduce the risk of overly optimistic performance caused by information leakage

C. To remove the need for labels

D. To guarantee perfect predictions

Correct Answer: B. To reduce the risk of overly optimistic performance caused by information leakage

Question 9

Which cybersecurity measure reflects least privilege?

A. Giving every supplier administrator access

B. Allowing only necessary users and services to access required functions

C. Connecting all control systems directly to the public internet

D. Disabling audit logs

Correct Answer: B. Allowing only necessary users and services to access required functions

Question 10

What is the most appropriate first step in a smart-maintenance implementation?

A. Purchase the most complex AI platform available

B. Define business value, asset risk and the decisions requiring better information

C. Install sensors on every item immediately

D. Replace the CMMS before reviewing its data quality

Correct Answer: B. Define business value, asset risk and the decisions requiring better information