Maintenance, Automation and Smart Manufacturing

Introduction and Context

Modern workshops are no longer collections of manually operated machines. They are interconnected production systems in which mechanical equipment, sensors, software, operators and maintenance practices jointly determine safety, quality, cost and delivery performance. A lathe with excessive spindle vibration, a CNC machine with contaminated coolant, or a robot with deteriorating gearbox lubrication may produce defective work long before complete failure occurs. Contemporary maintenance therefore seeks not merely to repair breakdowns, but to anticipate degradation, preserve capability and use data responsibly.

This chapter develops the practical foundations of preventive and predictive maintenance, lubrication, alignment and condition monitoring. It then connects these practices with sensors, programmable logic controllers (PLCs), industrial robots, industrial Internet of Things (IIoT), digital twins and responsible artificial intelligence (AI). The emphasis is on workshop and laboratory operations suitable for mechanical engineering students.

Learning Objectives

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

1. Distinguish corrective, preventive, predictive and reliability-centred maintenance.
2. Prepare a safe maintenance plan involving inspection, lubrication and alignment.
3. Interpret basic vibration, temperature, ultrasound and lubricant-condition data.
4. Explain sensor selection, signal paths and PLC-based control.
5. Describe the roles and limitations of robots, IIoT systems and digital twins.
6. Evaluate AI applications using criteria of safety, validity, cybersecurity, explainability and human responsibility.
7. Apply maintenance concepts to a realistic workshop case.

1. Maintenance as a Systematic Engineering Function

Maintenance is the combination of technical, administrative and managerial actions intended to retain or restore equipment so that it can perform its required function. The phrase “required function” is important: maintenance is not automatically successful because a machine runs. It must run safely, accurately, at the required capacity and within acceptable environmental and economic limits.

1.1 Maintenance strategies

Corrective maintenance restores equipment after a fault or failure. It may be appropriate for inexpensive, non-critical items whose failure has little safety or production consequence. It is hazardous when a failure can injure people, damage expensive components or interrupt a critical process.

Preventive maintenance is performed at predetermined intervals or according to equipment usage. Examples include replacing a filter every 1,000 operating hours, checking belt tension monthly and changing gearbox oil according to the manufacturer’s schedule. Its weakness is that fixed intervals may replace components too early or too late.

Predictive maintenance uses measurements of actual condition to determine when intervention is needed. Vibration analysis, thermography, oil analysis and motor-current analysis are common methods. Predictive maintenance does not mean predicting an exact failure time with certainty; it means identifying degradation and supporting a risk-based decision.

Proactive maintenance seeks to remove root causes, such as poor foundations, contamination, misalignment, incorrect assembly or unsuitable operating conditions.

Reliability-centred maintenance (RCM) selects maintenance tasks according to functions, functional failures, failure modes, effects and consequences. A safety-critical failure may require rigorous condition monitoring even when the component is expensive; a low-consequence failure may be allowed to run to failure.

A useful maintenance programme combines these approaches rather than treating them as competitors.

1.2 Reliability and availability measures

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

where R(t) is reliability at time t, λ is the constant failure rate in failures per unit time, and t is operating time. This model assumes statistically independent failures and a constant hazard rate; it is not generally valid during early-life or wear-out periods.

For repairable equipment, mean time between failures (MTBF) and mean time to repair (MTTR) are commonly used. A simple steady-state availability estimate is:

MTBF and MTTR must use the same time units. The expression excludes planned downtime, logistics delays and other practical effects unless they are incorporated into the definition of the times.

Worked example. A machine operates, on average, 480 hours between failures and requires 8 hours to restore. Its inherent availability estimate is A = 480 ÷ (480 + 8) = 0.9836, or approximately 98.36%. This does not mean that production availability is 98.36% if setup, inspection, material shortages or planned maintenance also stop the machine.

2. Preventive Maintenance Planning

A preventive-maintenance programme begins with an asset register. Each asset should have an identification code, manufacturer information, capacity, drawings, manuals, safety precautions, criticality and maintenance history. Criticality assessment considers safety, environmental impact, quality, production loss, repair cost and availability of alternatives.

A task specification should state:

• what is to be inspected or serviced;
• the acceptance criterion;
• required tools, instruments and consumables;
• isolation and personal protective equipment requirements;
• the responsible person and estimated duration;
• the record to be completed;
• the action required if the result is unsatisfactory.

A generic inspection such as “check machine” is weak. A better instruction is “inspect spindle-belt condition; replace if cracked, frayed or contaminated; record belt tension according to the manufacturer’s specification.”

2.1 Safe maintenance workflow

The practical sequence is: review the work order; identify hazards; isolate electrical, hydraulic, pneumatic, gravitational and stored mechanical energy; apply lockout/tagout; verify zero energy; perform the task; replace guards; test under controlled conditions; document findings; and close the work order. A functional test is not permission to bypass guards or operate an unsafe machine.

Maintenance records should distinguish observations from interpretations. “Outer-race frequency sidebands increased by 6 dB” is an observation; “bearing will fail tomorrow” is an unsupported interpretation unless validated by evidence.

3. Lubrication Engineering

Lubrication reduces friction, wear, heat generation and corrosion, while also helping seal contaminants and remove heat. The lubricant must be selected for load, speed, temperature, environment, material compatibility and relubrication method.

The principal lubrication regimes are:

• Boundary lubrication: surfaces are partly separated by chemically active films; asperity contact remains significant.
• Mixed lubrication: some load is carried by the lubricant film and some by asperities.
• Hydrodynamic lubrication: relative motion and lubricant viscosity generate a full fluid film separating surfaces.
• Elastohydrodynamic lubrication: elastic deformation and pressure-dependent viscosity are important, especially in rolling contacts.

Oil viscosity is central but is not the only selection factor. Grease consists broadly of base oil, thickener and additives. Grease compatibility cannot be assumed merely because two products have similar colours or nominal grades. Over-greasing can cause churning, heat and seal damage; under-greasing causes starvation and wear.

A lubrication schedule should identify lubricant type, quantity, point, method, interval, cleanliness controls and disposal requirements. The correct quantity is normally obtained from the manufacturer or a validated engineering calculation rather than guessed. Lubricants should be stored sealed, labelled and protected from water, dust and extreme temperature.

Contamination control is often more valuable than simply increasing lubricant quantity. Clean transfer containers, dedicated dispensing equipment, breathers, seals and filtration reduce particles and water. Used-oil analysis can reveal viscosity change, oxidation, water, particle contamination and wear metals, but laboratory results must be interpreted with operating history and trend data.

4. Alignment and Mechanical Integrity

Misalignment between coupled shafts produces radial and axial forces, elevated vibration, coupling wear, seal leakage and premature bearing failure. Soft foot—an uneven condition in which a machine foot does not sit properly on its base—can distort the housing when bolts are tightened and make alignment unstable.

Alignment should be performed after checking foundation condition, pipe strain, shaft runout, coupling condition and soft foot. Common methods include straightedge and feeler gauge, dial indicators and laser alignment systems. Laser instruments improve measurement efficiency, but they do not eliminate the need for correct brackets, clean reference surfaces, thermal-growth compensation and competent interpretation.

For two parallel shafts, angular and offset errors are distinct. A coupling may tolerate limited error, but its rating is not a justification for poor alignment. Thermal growth may require the cold position of a machine to be deliberately offset so that shafts become aligned at operating temperature. The manufacturer’s values and the machine’s operating conditions govern this decision.

5. Condition Monitoring

Condition monitoring measures variables whose change is associated with degradation. It is most effective when a baseline is established after commissioning or a known good repair, measurements are repeatable, and trends are examined rather than isolated readings.

5.1 Vibration

Accelerometers measure vibration acceleration, commonly in m/s² or g. Velocity, often expressed in mm/s RMS, is widely used for general machine-severity assessment, while displacement can be useful at lower frequencies. Frequency analysis helps distinguish unbalance, misalignment, looseness, gear-mesh problems and bearing-related patterns. Diagnosis must consider speed, load, sensor location, structural resonance and measurement direction.

RMS for N sampled values xᵢ is:

For a sinusoidal signal with zero mean, RMS equals the peak amplitude divided by √2. The formula is not a diagnosis by itself; it describes signal magnitude.

5.2 Temperature and infrared thermography

Temperature trends can indicate friction, overload, poor cooling, electrical resistance or lubrication problems. Infrared cameras measure emitted radiation and infer surface temperature. Emissivity, reflected radiation, viewing angle, distance and atmospheric effects influence accuracy. Shiny metal surfaces are particularly difficult to measure reliably; a high-emissivity reference patch may improve comparison.

5.3 Ultrasound, electrical and lubricant monitoring

Ultrasound can detect compressed-air leaks, arcing and early bearing friction in noisy environments, especially when measurements are trended. Motor-current signature analysis may identify some electrical and mechanical abnormalities, but it is affected by load and power quality. Oil particle counts, ferrography and elemental analysis provide complementary evidence rather than a complete diagnosis.

5.4 Data interpretation

A practical alarm system uses baseline, alert and action levels. Thresholds should be based on manufacturer guidance, standards where applicable, machine history and risk. A single threshold cannot suit every machine. False alarms create alarm fatigue, while missed alarms create risk. Every alert should have a defined response: verify measurement, inspect, reduce load, schedule repair or stop immediately when safety demands it.

6. Sensors, PLCs and Automation

A sensor converts a physical quantity into a usable signal. Examples include inductive proximity sensors for metal detection, photoelectric sensors for presence, thermocouples and resistance temperature detectors for temperature, pressure transducers, encoders for position and accelerometers for vibration.

Sensor selection requires consideration of range, accuracy, resolution, response time, repeatability, environmental protection, mounting, calibration, wiring, electromagnetic interference and failure behaviour. A sensor should be selected from the measurement need, not from novelty.

A typical automated cell contains sensors, signal conditioning, a controller, actuators, a human–machine interface and safety functions. A PLC cyclically reads inputs, executes a program and updates outputs. Ladder logic is common because it resembles relay-control schematics; structured text and function-block programming are also used.

A PLC sequence must separate ordinary control from safety control. Emergency stops, guard interlocks and safety-rated devices require appropriate architecture and validation; a normal PLC input used as an emergency-stop substitute is unacceptable where regulations require a safety system. Interlocks should be designed so that loss of power, broken wiring or controller fault leads to a safe state where practicable.

A basic sequence might be: confirm guard closed; confirm workpiece present; clamp; verify clamp pressure; start spindle; perform operation; stop spindle; unclamp; confirm part removed. Each step needs timeout and fault handling. “Start motor” without feedback that the motor actually started is incomplete control design.

7. Industrial Robots

Industrial robots provide repeatability, speed and ergonomic benefits in welding, handling, machine tending, inspection and assembly. A robot system includes the manipulator, controller, end effector, fixtures, sensors, guarding and safety-rated access control.

Robot programming requires understanding coordinate frames, tool-centre points, payload, reach, singularities, collision risks and cycle-time constraints. Teaching a path is not equivalent to validating a safe cell. Risk assessment must address unexpected restart, stored energy, pinch points, dropped loads, collaborative operation and maintenance access.

Collaborative robots may limit speed, force or power under specified conditions, but “collaborative” does not mean inherently safe. The complete application—including tool, workpiece, speed, posture and foreseeable misuse—must be assessed.

8. Industrial Internet of Things and Digital Twins

IIoT connects industrial assets through sensors, networks, edge devices, databases and analytical applications. A useful architecture separates:

1. Physical layer: machines, sensors and actuators.
2. Control layer: PLCs and local safety/control systems.
3. Edge layer: protocol conversion, filtering, buffering and rapid analysis.
4. Platform layer: storage, asset models and analytics.
5. Application layer: dashboards, work orders, quality and maintenance decisions.

Connectivity does not automatically create value. Data must have a defined purpose, adequate quality, ownership, cybersecurity protection and a response process. Time synchronisation, naming conventions, units and calibration are essential for comparing records.

A digital twin is a digital representation linked to a physical asset or process, maintained with data and used for monitoring, simulation, prediction or optimisation. A static CAD model is not necessarily a digital twin. A useful twin states its boundary, assumptions, update frequency, data sources, validation evidence and intended decision.

Digital twins may range from a condition dashboard to a physics-based model coupled with live measurements. Model predictions are uncertain; a twin should display confidence and data quality rather than presenting estimates as facts.

9. Responsible AI in Workshop Operations

AI can support anomaly detection, remaining-useful-life estimation, visual inspection, scheduling and energy optimisation. Supervised learning requires labelled examples; anomaly detection can learn normal behaviour and flag deviations. Neither method replaces engineering judgement.

A responsible AI deployment should address:

• Validity: Is the model tested on representative machines and operating conditions?
• Data quality: Are labels reliable, sensors calibrated and missing values handled?
• Robustness: Does performance remain acceptable under load, speed, temperature and tooling changes?
• Explainability: Can personnel understand the evidence behind an alert?
• Human authority: Who approves a shutdown or maintenance action?
• Cybersecurity and privacy: Are access, updates and industrial data protected?
• Fairness and workforce impact: Are workers evaluated by unreliable automated measures, and are they trained for changed roles?
• Lifecycle control: Are models monitored for drift, versioned and periodically revalidated?

An AI recommendation should normally be treated as decision support unless it has undergone formal safety validation. A model trained on one bearing type may fail on another. Data leakage, class imbalance and changing production conditions can create impressive laboratory accuracy but poor field performance. Safety-critical interlocks should not depend solely on an unvalidated statistical model.

10. Integrated Maintenance and Smart-Workshop Workflow

A mature workflow links physical work with information:

1. Define the asset’s required function and criticality.
2. Establish baseline condition after commissioning or repair.
3. Select failure-relevant measurements.
4. Collect repeatable, time-stamped data.
5. Screen trends and validate abnormal readings.
6. Diagnose using multiple evidence sources.
7. Rank risk using consequence and likelihood.
8. Plan the intervention, parts, permits and skills.
9. Execute safely and verify restoration.
10. Record the cause, action and post-repair condition.
11. Review the result and improve the maintenance task.

Useful performance indicators include planned-maintenance percentage, schedule compliance, repeat failures, mean time to repair, maintenance-related downtime, spare-parts stockouts and percentage of alerts receiving a documented response. Indicators should encourage reliable work rather than manipulation of numbers.

Applied Case: Illustrative CNC Cell Improvement

Background

A university workshop operates a CNC turning cell with a spindle motor, gearbox, coolant system and robot loader. Operators report increasing surface-finish variation and occasional spindle alarms. Maintenance has been mainly corrective.

Problem

The workshop must determine whether the problem is caused by tooling, workholding, spindle condition, coolant or automation without making unsafe assumptions or replacing expensive components unnecessarily.

Analysis

The team first reviews work orders, cutting conditions and rejected parts. A baseline is collected with repeatable accelerometer locations and operating speeds. Vibration velocity increases mainly at spindle speed, while temperature near the drive-end bearing rises gradually. Thermography shows a persistent hot region, but the shiny housing surface creates measurement uncertainty. A contact temperature sensor confirms the trend. Inspection finds contaminated grease and slight coupling misalignment. The robot also presents parts with a small but repeatable seating variation.

Approach and solution

The machine is isolated under a documented procedure. The coupling is inspected, soft foot is corrected and alignment is performed using a laser tool with the manufacturer’s tolerance and thermal condition considered. The bearing lubricant is replaced with the specified compatible product using clean equipment. The robot fixture is modified with a positive locating feature, and a proximity sensor verifies seating before machining. The PLC sequence is changed so that machining cannot start without clamp and seating confirmation.

Data are recorded through an edge device, but the control system retains deterministic interlocks independent of the analytics platform. A simple trend dashboard provides alert and action levels. AI is not used to command a shutdown; it is trialled only to rank anomalous vibration records for technician review.

Outcome

After repair, vibration and temperature return close to baseline, seating variation is reduced, and rejected parts decline. The team validates the result over several production cycles rather than declaring success after one measurement. The maintenance plan is updated to include alignment verification, lubricant contamination control, robot-fixture inspection and periodic sensor verification.

Learning points

The case demonstrates that a production symptom may have multiple interacting causes. Condition monitoring narrows the investigation but does not replace mechanical inspection. Automation is valuable when it confirms process conditions, while safety and essential control remain independent of uncertain analytics. The most effective intervention combines maintenance engineering, metrology, controls, operator knowledge and disciplined records.

Challenges, Limitations and Best Practices

Smart maintenance requires investment in instruments, training, data infrastructure and time for analysis. Sensors can drift, networks can fail and dashboards can encourage attention to easily measured variables rather than important failure modes. Predictive models may produce false positives and false negatives. Legacy machines may lack interfaces, and proprietary protocols can create vendor dependence.

Best practice is to begin with critical assets and a clearly defined decision. Use calibrated instruments and repeatable measurement procedures. Keep local control functional during network loss. Apply least-privilege cybersecurity, secure backups and controlled software updates. Involve operators and maintenance technicians in system design. Validate every important alarm against physical evidence. Document uncertainty, protect personal and production data, and review whether the technology actually reduces risk, waste or downtime.

Chapter Summary

Maintenance engineering preserves the required function of equipment through a combination of corrective, preventive, predictive and proactive methods. Reliability and availability measures help describe performance, but their assumptions and boundaries must be recognised. Effective preventive maintenance depends on asset criticality, precise task instructions, safe isolation and useful records.

Lubrication requires correct lubricant selection, contamination control and disciplined application. Alignment, soft-foot correction and thermal considerations protect bearings, couplings and seals. Condition monitoring uses vibration, temperature, ultrasound, electrical signatures and oil evidence; trends and validated thresholds are more useful than isolated readings.

Sensors and PLCs connect physical events to control decisions, while robot cells require comprehensive risk assessment. IIoT architecture links assets, edge processing, platforms and applications, but connectivity alone creates no benefit. Digital twins must be defined, data-linked and validated. AI can assist anomaly detection and planning, but responsible use requires representative data, robustness, cybersecurity, explainability, human accountability and lifecycle monitoring. Smart manufacturing is therefore an integrated socio-technical practice, not simply the installation of more software.

Key Terms

Maintenance: Actions that retain or restore required equipment function.

Corrective maintenance: Maintenance performed after a detected fault or failure.

Preventive maintenance: Planned maintenance based on time, usage or predetermined intervals.

Predictive maintenance: Condition-based maintenance using measured degradation evidence.

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

Availability: Proportion of time equipment is capable of performing its function.

MTBF: Mean operating time between failures of repairable equipment.

MTTR: Mean time required to restore equipment after failure.

Lubrication regime: The type of load-carrying contact between surfaces and lubricant film.

Soft foot: A condition in which a machine foot does not sit correctly on its base.

Condition monitoring: Measurement and interpretation of variables associated with equipment health.

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

PLC: Programmable logic controller used for industrial control.

Interlock: A control condition that prevents an unsafe or incorrect sequence.

Industrial robot: Programmable manipulator used for automated industrial work.

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

Digital twin: A data-linked digital representation of a physical asset or process.

Anomaly detection: Identification of behaviour that differs from a learned or defined normal condition.

Model drift: Reduction in model performance as data or operating conditions change.

Lockout/tagout: Procedure for isolating and controlling hazardous energy during work.

Review Questions

5 Short Answer Questions

1. Distinguish preventive maintenance from predictive maintenance.
2. Why is contamination control important in lubrication systems?
3. What is the purpose of establishing a condition-monitoring baseline?
4. State two functions of a PLC in an automated workshop cell.
5. Why is a digital twin more than a static CAD drawing?

5 Long Answer Questions

1. Explain how a university workshop could develop a reliability-centred maintenance programme for its machine tools.
2. Describe the causes, measurement methods and consequences of shaft misalignment.
3. Explain the architecture of an IIoT-enabled maintenance system from sensor to work order.
4. Discuss the safety requirements for integrating an industrial robot with a CNC machine.
5. Evaluate the benefits and limitations of AI-based predictive maintenance.

3 Analytical / Application Questions

1. A machine has an MTBF of 600 hours and an MTTR of 12 hours. Calculate its inherent availability and state why actual production availability may differ.
2. A bearing shows rising vibration, stable temperature and abnormal oil particles. Propose a sequence of verification and maintenance actions, including evidence you would collect.
3. Design a safe control and data strategy for a robot-loaded drilling cell. Identify which functions must remain local and deterministic, and which functions may appropriately use cloud analytics.

Multiple Choice Questions with Answers

Question 1

Which maintenance strategy uses measured equipment condition to decide when intervention is needed?

A. Corrective maintenance

B. Predictive maintenance

C. Breakdown-only maintenance

D. Unplanned maintenance

Correct Answer: B. Predictive maintenance

Question 2

For MTBF = 480 h and MTTR = 8 h, the inherent availability is approximately:

A. 80.00%

B. 96.00%

C. 98.36%

D. 99.83%

Correct Answer: C. 98.36%

Question 3

Which condition is most directly associated with incorrect machine-foot seating?

A. Soft foot

B. Cavitation

C. Thermal runaway

D. Electrical arcing

Correct Answer: A. Soft foot

Question 4

What is the main purpose of a condition-monitoring baseline?

A. To eliminate all maintenance tasks

B. To provide a reference for detecting change

C. To replace machine guards

D. To increase lubricant viscosity

Correct Answer: B. To provide a reference for detecting change

Question 5

Which device commonly executes industrial sequence-control logic?

A. PLC

B. Grease gun

C. Dial indicator

D. Thermographic lens

Correct Answer: A. PLC

Question 6

Why should an emergency stop not normally rely only on an ordinary PLC input?

A. Ordinary PLCs cannot read digital signals

B. Safety functions require suitable safety architecture and validation

C. PLCs operate only hydraulic systems

D. Emergency stops are used only during maintenance

Correct Answer: B. Safety functions require suitable safety architecture and validation

Question 7

Which statement best describes a digital twin?

A. Any three-dimensional CAD model

B. A paper maintenance instruction

C. A data-linked digital representation of a physical asset or process

D. A replacement bearing model

Correct Answer: C. A data-linked digital representation of a physical asset or process

Question 8

Which is a responsible practice when deploying AI for maintenance?

A. Allowing an unvalidated model to bypass safety interlocks

B. Ignoring sensor calibration because models correct all errors

C. Monitoring model drift and retaining human accountability

D. Training only on ideal laboratory data

Correct Answer: C. Monitoring model drift and retaining human accountability

Question 9

For N sampled values xᵢ, which expression represents RMS?

A. ∑xᵢ ÷ N

B. √[(1/N) ∑(xᵢ²)]

C. N ÷ ∑xᵢ²

D. ∑(xᵢ²) × N

Correct Answer: B. √[(1/N) ∑(xᵢ²)]

Question 10

What is a principal risk of over-greasing a bearing?

A. Reduced churning and heat

B. Increased churning, heat and possible seal damage

C. Automatic shaft alignment

D. Improved contamination removal in every case

Correct Answer: B. Increased churning, heat and possible seal damage