Page 1 – Foundations of Predictive Maintenance

1.1 Maintenance Paradigms

Engineering maintenance strategies have evolved across four key paradigms:

• Reactive Maintenance – repair after failure; high downtime and risk.
• Preventive Maintenance – scheduled service intervals; assumes uniform degradation.
• Condition-Based Maintenance (CBM) – maintenance triggered by sensor conditions.
• Predictive Maintenance (PdM) – AI predicts failures before they occur.

1.2 Purpose of AI-Driven Predictive Maintenance

• Identify degradation patterns invisible to human inspection.
• Predict Remaining Useful Life (RUL).
• Reduce unexpected downtime and catastrophic failures.
• Optimise maintenance schedules and spare-part inventory.
• Support reliability-centred engineering decisions.

1.3 Engineering Assets Suitable for PdM

• Rotating machinery – bearings, shafts, gearboxes, turbines.
• Civil structures – bridges, pipelines, tunnels.
• Electrical systems – transformers, switchgear, motors.
• HVAC and industrial systems – compressors, pumps, chillers.

1.4 Data Requirements

• High-frequency time-series (vibration, current, torque).
• Intermittent metrics (temperature, pressure).
• Discrete operational states (load, speed, duty cycle).
• Failure and maintenance logs.

1.5 Reliability Considerations

• Hazard rates and lifetime distributions (Weibull, lognormal).
• Failure modes and mechanisms.
• Environmental effects (humidity, thermal cycling).
• Machine–operator interaction variance.

Page 2 – Sensors, Signal Processing & Feature Engineering

2.1 Common Engineering Sensors

• Accelerometers: vibration signatures, bearing fault harmonics.
• Thermal sensors: overheating, lubrication loss, resistive heating.
• Acoustic emission (AE): crack initiation, frictional impacts.
• Electrical sensors: current, voltage, phase imbalance.
• Strain gauges: structural deformation and fatigue cycles.
• Ultrasonic transducers: subsurface crack detection.

2.2 Signal Preprocessing

• Band-pass filtering for noise suppression.
• Resampling and windowing for time-series alignment.
• Normalisation and trending removal.
• Fourier analysis for spectral signatures.
• Wavelet transforms for multiscale transient detection.

2.3 Diagnostic Feature Engineering

Features extracted from sensor signals feed AI models:

• Time-domain features: RMS, kurtosis, peak-to-peak, crest factor.
• Frequency features: spectral peaks, harmonic patterns.
• Time–frequency features: wavelet coefficients, spectrograms.
• Statistical descriptors: skewness, variance evolution.
• Operational context features: load, speed, temperature.

2.4 Feature Targets for Common Faults

• Bearing outer-race defects → high-frequency harmonics.
• Gear tooth breakage → sideband modulation frequencies.
• Thermal runaway → PDE-driven thermal gradients.
• Electrical faults → current unbalance & harmonics.

2.5 Edge vs Cloud Processing Pipelines

• Edge AI: low latency, local inference, real-time detection.
• Cloud AI: complex models, long-term analysis, remote aggregation.
• Hybrid architectures: tiered analytics across plant-level infrastructure.

Page 3 – ML for Failure Prediction & Remaining Useful Life (RUL)

3.1 Classification Models for Fault Detection

• Random Forests and Gradient Boosting (XGBoost, LightGBM).
• SVMs for high-dimensional vibration features.
• CNNs for spectrogram-based fault signatures.
• Graph neural networks (GNNs) for structural mesh health.

3.2 RUL Prediction via Regression Models

• LSTM & GRU networks: capture sequential degradation patterns.
• Temporal Convolutional Networks (TCN): efficient long-range modelling.
• Gaussian Process Regression (GPR): uncertainty-rich prognostics.
• Ensemble degradation models: combine physical & statistical signatures.

3.3 Hybrid Physics–ML Prognostics

Essential for engineering domains where physics cannot be ignored:

• Integrating crack growth models (Paris law) into ML-based RUL.
• Thermal ageing models combined with neural networks.
• CFD-informed degradation for rotating machinery cooling systems.

3.4 Model Evaluation Metrics

• Accuracy, precision/recall for fault classification.
• MAE, RMSE, MAPE for RUL prediction.
• Prognostic Horizon (PH).
• α–λ metrics for evaluating prediction envelopes.
• Confidence intervals for risk-sensitive engineering decisions.

3.5 Practical Challenges

• Class imbalance – rare failures.
• Non-stationary degradation behaviour.
• Sensor drift and calibration inconsistencies.
• Transferability across machines and environments.

Page 4 – Reliability Engineering, FMEA & Digital Twins

4.1 Reliability Frameworks

AI complements classical reliability tools:

• Failure Rate (λ) and Mean Time To Failure (MTTF).
• Reliability Function R(t).
• Weibull analysis for lifetime distributions.
• Fault Tree Analysis (FTA).

4.2 Digital Twins for Predictive Maintenance

Digital twins combine simulation + real-time data + AI:

• Dynamic state estimation.
• Predicting crack propagation under real load cycles.
• Thermal modelling with adaptive ML correction.
• Structural deformation tracking from strain gauges.
• Process optimisation in manufacturing lines.

4.3 FMEA Enhanced by AI

• Automatic failure mode detection from sensor trends.
• Real-time calculation of RPN (Risk Priority Number).
• Data-driven prioritisation of maintenance tasks.
• Automated mapping of failure causes in complex systems.

4.4 AI for Structural Health Monitoring (SHM)

• Civil bridges – vibration-based modal analysis.
• Pipelines – corrosion prediction & ultrasonic inspection.
• Aircraft structures – fatigue crack early detection.
• Turbines – blade imbalance and tip clearance detection.

4.5 System-Level Prognostics

• Model-based + data-driven hybrid SHM frameworks.
• Operating envelope modelling for safe limits.
• Component–system interactions in degradation.

Page 5 – Maintenance Optimisation & Integration into Engineering Workflows

5.1 Optimisation of Maintenance Strategy

• Optimal scheduling using stochastic models.
• Cost–risk trade-off optimisation.
• Inventory optimisation for spare parts.
• Downtime minimisation via predictive triggers.

5.2 Real-Time Maintenance Decision Systems

• Continuous monitoring dashboards.
• Realtime anomaly alerts with confidence scores.
• Integration with SCADA and MES systems.

5.3 Human–Machine Collaboration

AI enhances—not replaces—engineering judgement:

• AI flags risks; engineers validate actions.
• Explainability for maintenance recommendations.
• Feedback loops to refine ML models.

5.4 Ethical & Safety Considerations

• Ensuring safe operating limits under AI guidance.
• Preventing false negatives in high-risk systems.
• Transparent reporting and audit trails.
• Compliance with engineering codes of ethics.

5.5 Summary

Module 2 explored predictive maintenance, condition monitoring and reliability analytics, covering sensing, diagnostics, RUL modelling, digital twins, and maintenance optimisation. These skills underpin modern asset management in mechanical, civil, electrical and industrial engineering.

Module 3 will examine Digital Twins, Simulation Intelligence & Engineering Optimisation.