Model predictive control (MPC)
Predicts the process several steps ahead and optimises several variables at once within process constraints.
Cortergy Industrial
Industrial Solutions
Industrial equipment and solutions for power plants, substations, and production lines — hardware, instrumentation, and delivered projects.
Review the industrial equipment and delivered projects of this division.
// engineering solutions
What we do in smart industry
Three specialist areas where we combine process know-how, engineering simulation and artificial intelligence to improve how industrial equipment actually performs — from component design to real-time control and maintenance.
AI-Driven Process Control
Intelligent process control with AI and machine learning
We upgrade classic control loops with data-driven models so industrial equipment — from boilers and chillers to compressors, pumps and steam networks — runs more stably, more precisely and on less energy. What we deliver is a controller that sits on top of your existing PLC and control system, not a report on paper.
In many plants, PID controllers run on factory or rule-of-thumb settings and do not respond optimally to changes in load, fuel quality, ambient temperature or equipment wear. The result is oscillating process variables, wasted energy and early wear of actuators and control valves.
Our approach combines process knowledge with machine-learning models. We first build a dynamic model of the process from historical SCADA and sensor data, then design model predictive control (MPC), automatic and adaptive PID tuning, or reinforcement-learning control policies on top of it. Before anything touches real equipment, the controller is tested in simulation and then in an advisory mode alongside the operators.
Safety and reliability always come first: the intelligent controller only acts inside the limits and safety interlocks defined in the PLC, and if data is faulty or communication drops it automatically falls back to classic control.
Solution architecture · The intelligent control loop
- Sensors & SCADAtemperature, pressure, flow, load
- ML process modelpredicts process behaviour
- Predictive controllerMPC / adaptive PID
- PLC & actuatorsvalves, VFDs, burners
- Processboiler, chiller, compressor
Live feedback from the process back into the model
Performance charts
Outlet temperature response to a setpoint change and a load disturbance
Classic PID vs intelligent predictive control (simulated example)
- Intelligent predictive control
- Classic PID
- Setpoint
Show the data as a table
| Time (s) | Intelligent predictive control | Classic PID | Setpoint |
|---|---|---|---|
| 0 | 60 | 60 | 60 |
| 20 | 70.1 | 69.3 | 80 |
| 40 | 79.7 | 84.1 | 80 |
| 60 | 80 | 79.7 | 80 |
| 80 | 80 | 79.6 | 80 |
| 100 | 79.2 | 74 | 80 |
| 120 | 80 | 81.7 | 80 |
| 140 | 80 | 79.7 | 80 |
| 160 | 80 | 79.9 | 80 |
Control-valve command in the same test
Smoother actuator movement means less wear and steadier energy use
- Intelligent predictive control
- Classic PID
Show the data as a table
| Time (s) | Intelligent predictive control | Classic PID |
|---|---|---|
| 0 | 40 | 40 |
| 20 | 61 | 74 |
| 40 | 65 | 58 |
| 60 | 65 | 67 |
| 80 | 65 | 65 |
| 100 | 74 | 80 |
| 120 | 77 | 73 |
| 140 | 77 | 78 |
| 160 | 77 | 77 |
The charts are simulated examples that show typical behaviour, not data from a specific project.
What we deliver
Automatic & adaptive PID tuning
System identification from operating data and controller gains that follow changing conditions.
Soft sensors
Real-time estimates of variables that are costly or slow to measure, such as combustion efficiency or product quality.
Reinforcement learning for control
Control policies trained in simulation and transferred to the real process in a controlled, safe way.
Setpoint optimisation
The best operating points for load, energy tariff and ambient conditions — automatically and continuously.
PLC & SCADA integration
Connected over OPC UA, Modbus and MQTT without replacing existing hardware.
Technical specifications
- Control cycle
- 100 ms to 5 s, depending on process dynamics
- Protocols
- OPC UA, Modbus TCP/RTU, MQTT; Profinet via gateway
- Runs on
- Industrial edge PC next to the PLC or an on-site server; no internet dependency
- Data needed
- Historical SCADA / historian data, ideally 3 to 6 months
- Safety
- Acts inside PLC limits and interlocks; automatic fallback to PID on any fault
- Model updates
- Periodic, or triggered by monitoring of model accuracy
- Operator interface
- HMI screens and web dashboard with advisory, automatic and manual modes
Typical outcomes
5–15%lower energy use in heating and cooling loops
30–60%less variability in key process variables
20–40%shorter settling time after load changes
10–30%fewer unnecessary actuator and valve movements
Typical ranges reported in industry reports and the scientific literature for this kind of project; the result of each project depends on the process and its data.
How a project runs
- 01Audit & data collection
Review of existing control loops and data quality; finding the biggest improvement opportunities.
- 02Process modelling
A hybrid (physics + data) dynamic model, validated against real operating data.
- 03Controller design & simulation
MPC or adaptive PID design, tested across load and disturbance scenarios.
- 04Staged deployment
Advisory mode first, then closed loop within limits and safety interlocks.
- 05Monitoring & continuous improvement
Performance dashboard, before/after comparison and periodic model updates.
Applications
- Boilers & steam networks
- Chillers & cooling towers
- HVAC of large buildings
- Air & gas compressors
- Pumping stations
- Furnaces & heat exchangers
- Production lines & batch processes
- Renewables & energy storage
Tools & technologies
- Python
- PyTorch / TensorFlow
- MPC (CasADi / do-mpc)
- System Identification
- Reinforcement Learning
- OPC UA
- Modbus TCP
- MQTT
- Siemens / Schneider PLC
- SCADA / HMI
Component Optimization
Optimising the design and performance of mechanical and thermal components
A large share of industrial energy loss comes from components that were not designed or selected optimally: a pump impeller running far from its best-efficiency point, a heat exchanger larger or smaller than needed, or piping with high pressure drop. We combine engineering simulation, AI surrogate models and optimisation algorithms to optimise components for the conditions they really run in.
Every project starts with a precise target: higher efficiency, lower pressure drop, lower weight and manufacturing cost, or longer life. We then define the design variables — blade angle and count, channel dimensions, tube layout or wall thickness — together with manufacturing, code and operating constraints.
Because each CFD or FEA run can take hours, we use machine-learning surrogate models: a limited number of well-chosen simulations train a model that predicts the component's behaviour in a fraction of a second, which makes searching thousands of candidate designs practical. Genetic algorithms, Bayesian optimisation and multi-objective methods then reveal the Pareto front between competing goals — efficiency versus cost, for example.
The result is a design confirmed by a final high-fidelity simulation, and by testing where possible, delivered with drawings, technical specifications and a before/after comparison report.
Solution architecture · The design optimisation loop
- Parametric geometrydesign variables & constraints
- CFD / FEA simulationflow, heat, stress
- AI surrogate modelpredictions in milliseconds
- Optimisergenetic, Bayesian, multi-objective
- Optimal designfinal validation & drawings
Each generation of designs feeds back into the model to sharpen the search
Performance charts
Pump efficiency curve before and after impeller optimisation
The best-efficiency point (BEP) is higher and closer to the operating range
- Optimised impeller
- Current impeller
Show the data as a table
| Flow (m³/h) | Optimised impeller | Current impeller |
|---|---|---|
| 0 | 0 | 0 |
| 40 | 43 | 44 |
| 80 | 69.6 | 68.5 |
| 120 | 79.9 | 73.4 |
| 160 | 73.7 | 58.7 |
| 200 | 51.2 | 24.5 |
Pareto front: efficiency vs manufacturing cost
140 designs evaluated with the surrogate model; designs on the front are the best trade-offs
- Evaluated designs
- Pareto front
- Current design
Show the data as a table
| Relative manufacturing cost (current = 1) | Efficiency (%) | |
|---|---|---|
| 0.82 | 70.2 | Pareto front |
| 0.84 | 72.9 | Pareto front |
| 0.89 | 74.4 | Pareto front |
| 0.92 | 75.3 | Pareto front |
| 0.93 | 75.5 | Pareto front |
| 0.94 | 75.8 | Pareto front |
| 0.95 | 76 | Pareto front |
| 0.97 | 76.2 | Pareto front |
| 1 | 74.2 | Current design |
| 1.02 | 77.3 | Pareto front |
| 1.04 | 77.7 | Pareto front |
| 1.06 | 77.9 | Pareto front |
| 1.08 | 78 | Pareto front |
| 1.1 | 78.4 | Pareto front |
| 1.16 | 78.8 | Pareto front |
| 1.17 | 79 | Pareto front |
| 1.21 | 79.1 | Pareto front |
| 1.23 | 79.4 | Pareto front |
| 1.26 | 79.6 | Pareto front |
| 1.27 | 79.6 | Pareto front |
| 1.37 | 79.9 | Pareto front |
The charts are simulated examples that show typical behaviour, not data from a specific project.
What we deliver
Computational fluid dynamics (CFD)
Flow, pressure-drop and heat-transfer analysis of pumps, fans, ducts and heat exchangers.
Stress & vibration analysis (FEA)
Strength, fatigue and natural frequencies for lighter, more reliable designs.
AI surrogate models
Fast models that replace heavy simulations so the design space can be explored widely.
Multi-objective optimisation
Genetic algorithms, NSGA-II and Bayesian optimisation to find the best trade-off between competing goals.
Equipment sizing & selection
Pumps, exchangers, valves and piping chosen for the real load profile, not a single design point.
Reverse engineering & redesign
Improving existing parts or designing local replacements for imported ones.
Technical specifications
- Simulation tools
- ANSYS Fluent / CFX, OpenFOAM, Abaqus, SolidWorks Simulation
- Design variables
- Typically 5 to 30 geometric and operating parameters
- Objectives
- Single or multi-objective: efficiency, pressure drop, weight, cost, life
- Surrogate model
- Gaussian process (Kriging), neural network, polynomial regression
- Simulation budget
- 50 to 300 targeted runs (DoE) instead of thousands
- Validation
- Final fine-mesh simulation, compared with test data where available
- Deliverables
- CAD model, manufacturing drawings, performance curves and technical report
Typical outcomes
3–8%higher hydraulic efficiency of pumps and fans
10–25%lower pressure drop in piping and exchangers
10–20%less weight or material per part
100×+faster evaluation of each design with a surrogate
Typical ranges reported in industry reports and the scientific literature for this kind of project; the result of each project depends on the process and its data.
How a project runs
- 01Targets & constraints
Performance indicators, manufacturing and code limits, and real operating conditions.
- 02Parametric modelling
A parametric geometry and a baseline simulation for validation.
- 03Design of experiments & surrogate
Targeted simulations (DoE) that train a fast, accurate model.
- 04Optimisation & selection
Searching the design space and reviewing the Pareto front with the client.
- 05Validation & delivery
Final simulation, manufacturing drawings and a before/after report.
Applications
- Pump & fan impellers
- Shell-and-tube & plate exchangers
- Turbine & compressor blades
- Piping & air ducts
- Control valves
- Coils & cooling towers
- Cast & machined parts
- Support structures & frames
Tools & technologies
- ANSYS Fluent / CFX
- OpenFOAM
- SolidWorks Simulation
- Abaqus
- Python (SciPy, pymoo)
- Bayesian Optimization
- NSGA-II
- Gaussian Processes
- Neural Surrogates
- Design of Experiments
Digital Twin · Predictive Maintenance
Digital twins and predictive maintenance
A digital twin is a live, synchronised copy of a real asset or process, fed by sensor data. It shows the health of the equipment moment by moment, lets you simulate its behaviour in different scenarios, and predicts the right time for maintenance before a failure happens.
Traditional maintenance is either reactive (after a failure) or time-based (at fixed intervals, even when the equipment is healthy). Both are costly: the first brings sudden production stops and secondary damage, the second replaces healthy parts too early. Predictive maintenance moves the repair to exactly when the data says it is needed.
For rotating and thermal equipment — turbines, compressors, pumps, motors, gearboxes and exchangers — we design condition-monitoring systems: vibration, temperature, motor current, pressure and flow are collected; anomaly-detection algorithms reveal abnormal behaviour early, and remaining-useful-life (RUL) models show how much time is left before a likely failure.
The digital twin combines this data with a physical model of the equipment, so it explains the root cause as well as raising the alert, and shows the effect of operating decisions — such as raising the load or postponing a repair — before they are made. For the maintenance team, the result is an equipment-health dashboard, timely alerts and a prioritised maintenance plan.
Solution architecture · The digital-twin cycle
- Physical assetturbine, pump, motor, exchanger
- Sensors & IIoTvibration, temperature, current
- Digital twinphysics + data model
- Diagnosis & predictionanomalies, faults, RUL
- Maintenance decisionalerts & repair plan
The maintenance action is applied to the real asset and monitored again
Performance charts
Bearing health index and remaining-life forecast
Monitored data up to today and the digital twin's forecast with its uncertainty band
- Monitored health index
- Digital-twin forecast
Show the data as a table
| Day | Monitored health index | Digital-twin forecast |
|---|---|---|
| 0 | 96.4 | — |
| 30 | 96.5 | — |
| 60 | 94.5 | — |
| 90 | 92.6 | — |
| 120 | 88.6 | — |
| 150 | 75.7 | 74.8 |
| 180 | — | 53.4 |
| 210 | — | 23.8 |
| 240 | — | — |
Electric-motor vibration spectrum (FFT)
Peaks at the ball-pass frequency of the outer race (BPFO) indicate bearing damage
- Vibration spectrum
- Bearing fault frequencies (BPFO)
Show the data as a table
| Frequency (Hz) | Vibration velocity (mm/s) | |
|---|---|---|
| 25 | 2.6 | 1× |
| 50 | 1.1 | 2× |
| 90 | 1.45 | BPFO |
| 175 | 0.95 | 2×BPFO |
| 265 | 0.6 | 3×BPFO |
The charts are simulated examples that show typical behaviour, not data from a specific project.
What we deliver
Condition monitoring
Vibration, temperature, current and process parameters monitored with industrial and IIoT sensors.
Anomaly detection
Models that learn the healthy behaviour of a machine and flag deviations before they become critical.
Remaining useful life (RUL)
Time-to-failure estimates for bearings, blades, seals and other critical parts.
Fault diagnosis & root cause
Identifying the fault type — unbalance, misalignment, bearing damage or fouling — from the data signature.
Physics-based digital twin
A hybrid physics-and-data model to simulate operating scenarios and weigh the effect of decisions.
Dashboards & maintenance planning
Fleet-wide equipment health, smart alerts and integration with your CMMS.
Technical specifications
- Sensors
- Vibration accelerometers, temperature (RTD / thermocouple), motor current, pressure and flow
- Sampling rate
- Up to 25.6 kHz for vibration; 1 s for process variables
- Edge processing
- FFT, RMS and health indicators computed on an industrial gateway
- Connectivity
- OPC UA, MQTT, Modbus; wired or industrial wireless network
- Storage
- Time-series database with multi-year retention
- Models
- Anomaly detection, fault classification, remaining-life estimation, physics model
- Outputs
- Health dashboard, SMS and e-mail alerts, CMMS integration
Typical outcomes
30–50%fewer unplanned shutdowns
10–40%lower maintenance costs
20–40%longer equipment life
Weeksof early warning before a failure, depending on the fault
Typical ranges reported in industry reports and the scientific literature for this kind of project; the result of each project depends on the process and its data.
How a project runs
- 01Criticality analysis
Finding the equipment whose downtime costs the most (FMEA / FMECA).
- 02Sensors & data collection
Sensor selection, measurement points and the IIoT communication layer.
- 03Building the digital twin
Physics- and data-based models of the asset, synchronised with live data.
- 04Diagnosis & prediction models
Training anomaly, fault-classification and remaining-life models.
- 05Integration with maintenance
Dashboards, alerts and links to maintenance planning and spare-parts stock.
Applications
- Steam & gas turbines
- Compressors
- Pumps & electric motors
- Gearboxes & bearings
- Fans & blowers
- Exchangers & boilers
- Wind turbines
- Switchgear & transformers
Tools & technologies
- Vibration Analysis (FFT / Envelope)
- Infrared Thermography
- IIoT Sensors & Edge Gateways
- OPC UA / MQTT
- InfluxDB / TimescaleDB
- Python (scikit-learn, PyTorch)
- LSTM / Autoencoders
- Physics-based Models
- Grafana Dashboards
- CMMS Integration
Have a project in this area?
Send us a description of your process or equipment; we will review feasibility, the data needed and a proposed approach with you.