We teach shop-floor AI around one real-time thread: capture from PLCs, sensors and cameras, feature windows, model scoring, dashboards and alerts. A person owns safety limits, change control and go-live. Complements Industrial Automation & Industry 4.0.
Domain Knowledge We Teach
- Industry 4.0, ISA-95 and smart-factory concepts
- PLC, sensor, MES and shop-floor signal chains
- Quality, yield, scrap and OEE metrics
- Maintenance strategies and downtime drivers
- Production constraints, functional safety and OT cyber basics
AI Skills You Will Learn
- OPC UA / MQTT capture with timestamps and quality flags
- Time-series ML and SPC baselines without leakage
- Computer vision for defect detection with cost-based thresholds
- Predictive maintenance health indices tied to CMMS actions
- Edge deployment, drift monitoring and management of change
Hands-on Outcomes
- Build a working real-time capture and scoring pipeline
- Run a live inspection demo inside a latency budget
- Design a maintenance trigger with a cost-benefit note
- Deploy and roll back an edge model with a runbook
- Defend a capstone with quantified OEE, scrap, downtime or energy results
Programme shape: 12 weeks, about 96 hours (live sessions plus labs and project work), optional week-0 pre-work. Subject to batch and faculty planning. The cohort and any certificate are confirmed on enquiry.
Real-time thread: sources feed a broker; the stream is windowed into features; a model scores each window or frame; results drive a live dashboard and alerts. Latency budgets set in week 2 are checked in every lab.
Week 1 — Smart manufacturing, OEE and use-case framing
Where AI fits in the automation pyramid (ISA-95). Calculate OEE and the six big losses from event data. Score and select a use case with an accountable owner. Deliverable: one-page use-case brief with a baseline KPI.
Week 2 — Industrial data capture
PLC scan cycles, tags and sensor chains. Build an OPC UA to MQTT to time-series pipeline. Detect dropouts, clock drift and stuck values. Deliverable: working capture pipeline and a data-quality report.
Week 3 — Python and machine learning for time-series
Rolling-window features, FFT and statistical features. Anomaly detection compared with Shewhart limits. Time-based splits — never random. Soft sensors with lab-lag alignment. Deliverable: notebook plus a live anomaly-detection demo.
Week 4 — Computer vision for inspection I
Lighting, lens and camera before the network. Transfer-learning classifiers and detectors. Label audit and inspector agreement. Deliverable: model card with data, metrics, limits and intended use.
Week 5 — Computer vision for inspection II (real time)
Live-frame inference inside the latency budget. Thresholds from the cost of false rejects versus escapes. Reject logic with human review and golden-sample checks. Deliverable: live inspection demo and latency report.
Week 6 — Predictive maintenance
Link failure modes to measurable signals. Health index, alert design and CMMS integration. Cost comparison with fixed-interval maintenance. Deliverable: maintenance trigger design and cost-benefit analysis.
Week 7 — Process optimisation and digital twins
Designed experiments and response surfaces. A simple digital twin for what-if scenarios under validated process limits. Energy and scrap before/after on the same shift pattern. Deliverable: KPI before-and-after report.
Week 8 — Robotics and AI perception
Frames, transforms and kinematics. Hand-eye calibration and a vision-guided pick. Measure success rate and cycle time over repeated trials. Deliverable: pick success report with failure analysis.
Week 9 — Edge AI deployment
Edge versus cloud trade-offs. Quantisation, container deploy, latency/throughput tests and a rollback drill. Plan for network loss so the line stays safe when AI is down. Deliverable: deployment runbook and latency test results.
Week 10 — MLOps, monitoring and change control
Data drift versus concept drift. Retraining triggers, model registry and audit trail. Management of change and re-validation. Deliverable: monitoring plan and a sample change record.
Week 11 — GenAI copilots, safety and cybersecurity
Retrieval over SOPs and manuals with citations. Audit for accuracy and unsupported claims. Functional safety, zones and conduits, and a hazard assessment of the AI-enabled cell. Deliverable: audit report and a safety case.
Week 12 — Capstone defense and business case
Integrate pipeline, model, deployment and safety work. Present measured results and a business case in OEE, scrap, downtime or energy terms. Plan handover and operator adoption. Deliverable: capstone package and defense.