/01
Advanced Control
- PID design & auto-tuning
- Model Predictive Control
- Adaptive & fuzzy logic
- State-space methods
- Cascaded & feedforward loops
We unify the rigour of classical control theory with applied machine learning — one team handling everything from PID tuning to neural model predictive control.
Engagements across project management, AI & data science, industrial automation, and software development — spanning food & beverage, lime production, logistics, and process manufacturing.
Bayesian optimisation layer over an ISA-88 batch sequence, learning recipe set-points from historian data to lift yield without process redesign.
Ensemble model (Isolation Forest + autoencoder) catching incipient faults on rotating equipment 4–6 hours before alarm thresholds trip.
Random Forest classifier and Isolation Forest hybrid driving a predictive maintenance schedule from vibration, temperature, and current signatures.
Retrieval-augmented assistant grounding LLM responses in P&IDs, FDS documents, and decades of plant manuals — with citation traceability.
YOLOv8 line-side defect detector for fill levels, cap orientation, and label registration on a high-speed beverage line; runs on an industrial edge GPU.
LSTM-based day-ahead electricity load forecast for a high-voltage manufacturing site, integrated with the operator's market scheduling tool.
Gradient-boosted model estimating tonnes-per-hour from drive currents, vibration, and feed dynamics where the physical weightometer was unreliable.
Multivariate SPC and autoencoder system flagging slow product-quality drift on a petrochemical reactor hours before any spec breach.
TIA Portal redesign of a 14-tank fermentation profile with cascaded jacket cooling, recipe-driven set-points, and clean termination curves.
Migration from a legacy Wonderware estate to AVEVA System Platform for a regional water treatment site, with full EEMUA 191 alarm rationalisation.
Studio 5000 merge and divert logic for an e-commerce fulfilment centre, including jam-recovery sequences and live throughput balancing across twelve merges.
Multivariable O₂ and NOₓ trim controller layered over a precalciner kiln, sequenced against fuel mix and clinker quality targets.
Beckhoff TwinCAT 3 coordination of three FANUC robots in a press-tending cell, with EtherCAT-based safety and a deterministic 1 ms cycle.
Pressure-cascade and recipe-driven HVAC control for a sterile fill-finish line, validated to GAMP 5 Category 4 with full 21 CFR Part 11 audit trail.
OPC UA radar-level integration across a 24-tank fuel farm, with custody-transfer reconciliation logic and operator-facing volume dashboards.
Bay-level GOOSE messaging and MMS reporting across four substation bays at a private generation site, brought into a central control HMI.
Beckhoff TwinCAT 3 recipe handling for a high-speed FFS bagging line, with cross-coupled checkweigher and metal-detector handshakes and a full reject-lane audit trail.
Cascade temperature loop on an HTST pasteuriser with flow-diversion-valve sequencing per regulatory hold-tube logic, plus full divert-event reporting to the historian.
Studio 5000 master-axis coordination across an infeed metering belt, case erector, top-load packer, and label verifier — including dynamic gap control under product mix changes.
Coordinated obsolescence audit of eleven production sites for a Tier-1 automotive supplier; consolidated risk register and a five-year replacement roadmap.
End-to-end FAT and SAT delivery for a new reactor line: 340 test cases, six-week site programme, and zero punch-list items carried into production.
Principal designer interface for a brownfield process upgrade under CDM 2015: designer risk assessments, hazard log, and stage-gate sign-off package.
ISA-88 batch standard rollout across four production cells, covering the master/control/site recipe model, exception library, and operator training.
Site-level IACS cybersecurity assessment per IEC 62443-3-2: zones-and-conduits diagram, SL-T determination, and a prioritised remediation backlog.
FMEA-driven migration strategy for legacy-to-modern PLC replacement on a 24/7 line, including dry-run protocol, rollback criteria, and validation matrix.
Engineering brief and concept-to-FEED stage-gate submission for a £1.8M control system replacement, including TCO model and vendor evaluation matrix.
LOPA-supported HAZOP facilitation and SIL determination across twelve nodes on a thermal oxidiser, with a full safety requirements specification.
Multi-protocol (OPC UA / Modbus TCP / MQTT) industrial monitoring stack with a FastAPI back-end, time-series storage, and a React control-room dashboard.
End-to-end Python repository for manufacturing defect prediction — data pipeline, model training, evaluation harness, and a thin FastAPI inference service.
Containerised Python service computing real-time OEE from PLC tag streams across six production lines, pushing to InfluxDB and Grafana.
Static-analysis CLI that enforces ISA-style tag naming conventions across exported Studio 5000 .L5X files; integrates straight into a CI pipeline.
Lightweight Python bridge replicating OPC UA subscriptions into TimescaleDB, with deterministic backfill on disconnect and tag-level back-pressure.
FastAPI + React tool capturing shift-end events, alarm clusters, and operator commentary — with an LLM-generated summary of the last twelve hours.
Edge gateway streaming Sparkplug-B payloads into a central Kafka cluster across thirty-six sites, with exactly-once semantics and store-and-forward.
Open-source-style Python library for industrial document RAG — bespoke chunkers for P&IDs, FDS, and alarm rationalisation tables.
Three engagement models, from a focused diagnostic to a full intelligent control system build. Most engagements run hybrid — remote design with on-site commissioning support.
Diagnostic engagement, 2–3 weeks
Design → commissioning, 8–16 weeks
Ongoing capacity, 6-month minimum
A pragmatic stack — vendor platforms where they earn it, open-source where it gives leverage. Selected from a decade of production deployments.
Working papers and applied research at the intersection of industrial control engineering and machine intelligence — written from the perspective of someone who has to defend it in front of an operator, not just a reviewer.
A hybrid MPC architecture that learns the residual between a first-principles process model and observed plant behaviour, retaining the stability guarantees of classical MPC while improving tracking on highly nonlinear regimes.
A reference framework for layering predictive safety analysis on top of a process digital twin — covering scenario generation, simulation-in-the-loop testing, and the boundary between safety analytics and the certified SIS.
How Bayesian optimisation and contextual bandits can sit cleanly above an ISA-88 recipe engine to tune set-points across master, control, and site recipe layers without breaking traceability or batch records.
A practitioner-focused architecture spanning sensor through historian through feature store through model serving — with explicit treatment of OPC UA semantics, time-alignment, and back-pressure under plant disturbances.
An argument and worked examples for control architectures that combine physics-based models with learned components — with an explicit framing of where in the stack each component is allowed to act.
A practical taxonomy of explainability techniques for industrial ML — SHAP, counterfactuals, attention maps — tied to specific operator-facing artefacts and regulatory audit requirements.
Selected drafts available on request. Several of these underpin the SmartOpsTech engineering approach — specifically how machine intelligence is layered on top of certified control without ever replacing it.
Selected outcomes from production engagements — every figure verified by the client's own historians, OEE systems, or financial reporting.
Cascade PID with an ML-driven feedforward layer eliminated 30-year-old quality issues caused by feed-stock variability.
Model Predictive Control with an embedded energy objective reduced steam draw across four pasteurisation lines.
AI-powered failure prediction replaced calendar-based PM on critical rotating assets across three shifts.
Closed-loop control of three chemical dosing skids using real-time turbidity and pH feedback — replacing manual operator adjustments.
“The best control system is the one your operators forget is there. That’s the bar we hold ourselves to.”
SmartOpsTech is a specialist control systems and AI engineering consultancy serving industrial clients across the UK and Europe. The practice is led by certified automation engineers with postgraduate research in control systems and applied AI.
From PID optimisation to neural network-assisted model predictive control, our work bridges classical engineering and modern AI — with quantifiable improvements in efficiency, product quality, and uptime.
Send a short note describing the plant, the symptom, and what success looks like. We’ll reply within two working days.
uzoeshijuliet@gmail.com