Industrial Process NMPC/APC — Model-Based Control for Energy, Quality, and Throughput

Public reference use case (by others): This page summarizes publicly available reference implementations and papers. Not client results of Dr. Rafal Noga.

Why This Matters (Executive Summary)

• Industrial processes lose 5—15% of energy and yield to suboptimal manual control and fixed-schedule operation; NMPC/APC can recover a significant fraction of that.
• First-principles or hybrid models inside the controller handle thermal inertia, reaction kinetics, and multi-variable coupling that PID cascades cannot coordinate.
• Reference implementations across steel, cement, pharma, chemicals, and HVAC report measurable improvements: temperature accuracy from 13 K mean error down to 0.9 K, precalciner variability reduced by over 50%, and HVAC energy use cut by approximately 17%.
• Staged commissioning (shadow mode, advisory, closed-loop) de-risks deployment and builds operator trust before full automation.

The Design Pattern Explained

Industrial process NMPC uses physics-based models — heat transfer, reaction kinetics, mass balances — inside an optimization loop that runs every seconds-to-minutes. The controller computes optimal manipulated-variable trajectories (temperatures, feed rates, valve positions) over a prediction horizon, subject to hard constraints on equipment limits, product quality, and safety.

Unlike PID cascades that handle one loop at a time, NMPC coordinates multiple interacting variables simultaneously. For batch processes, this means time-varying trajectory optimization (temperature profiles, feed schedules). For continuous processes, it means setpoint tracking with disturbance rejection while minimizing energy per ton.

The architecture follows: online estimation (soft sensors, PAT, inferential models) provides observability; the optimizer computes constrained trajectories; safety layers enforce equipment limits; and operator interfaces allow seamless handoff between advisory and closed-loop modes.

Applications & Reference Implementations

Slab Reheating Furnace NMPC — Steel Industry (Dillinger, Germany)

A first-principles nonlinear MPC was deployed at Dillinger Huettenwerke in Dillingen/Saar to control slab exit temperatures in a continuous reheating furnace for heavy-plate rolling. The controller computes local furnace temperature targets so that slabs reach desired final temperatures even during non-steady-state operation caused by changing product mix and throughput. Mean slab temperature error dropped from 13.0 K to 0.9 K, and out-of-range slabs fell from 59% to 12%. The controller was commissioned in February 2011 and later applied to additional furnaces.¹

Coal-Free Precalciner MPC — Cement Industry (Holcim Laegerdorf, Germany)

ABB implemented its Expert Optimizer (MPC + Mixed Logical Dynamic control) at Holcim’s Laegerdorf plant to stabilize precalciner temperature while handling multiple alternative fuels with high variability. The MPC explicitly accounts for combustion delays, thermal inertia, and air supply disturbances. Calciner temperature variation narrowed from -45/+80 degrees C (manual) to -30/+50 degrees C (MPC), with overall precalciner temperature variability reduced by over 50%. This enabled coal-free precalciner operation starting June 2007, with coal on standby only for fast recovery.²³

Raw Mix Quality Stabilization — Cement Industry (Holcim Untervaz, Switzerland)

At Holcim Untervaz, ABB extended its Expert Optimizer with a Raw Mix Preparation module to control seven feeders with variable chemistry and time-delayed measurements (up to 30 minutes). The MPC+MLD scheme with adaptive modeling and explicit delay handling reduced AR and SR variability by approximately 20% without any hardware modifications. Blending-bed operation was simplified, reducing the need for multiple blending programs.³

Continuous Wet Granulation MPC — Pharma (Novartis, Switzerland)

Novartis implemented MPC on a pharmaceutical continuous wet granulation line, controlling API content and loss-on-drying (LOD) from solid feeders through to the dryer. The system was validated on two drug products (Diclofenac and Paracetamol), demonstrating accurate control of critical quality attributes to produce consistent tablet quality. This represents a step toward real-time release in continuous pharmaceutical manufacturing.⁴

Online MPC for Batch Processes — Chemicals (BASF RECOBA, Germany)

The EU-funded RECOBA project (EUR 6 million, 3-year term), coordinated by BASF, developed online MPC for complex batch processes — specifically emulsion co-polymerization. The project combined new sensor technologies, process models, and automation tools to move from fixed-schedule batch operation toward model-based online trajectory tracking. Target outcomes included narrower quality specifications, higher productivity, and energy savings across polymers, steel, and silicon batch processes.⁵

Economic NMPC for Superfluid Helium Cryogenics — Research Infrastructure (CERN LHC)

Output-feedback economic NMPC was applied to the superfluid helium cryogenic circuit of the Large Hadron Collider, targeting constrained temperature recovery after disturbances. The architecture pairs a first-principles thermo-hydraulic model with a Luenberger observer and Moving Horizon Estimator (MHE). Set-point recovery was achieved in approximately 1 hour after perturbation, with computation times of approximately 7 seconds per optimization cycle. The strict constraint at 2.1 K (magnet powering limit) and 2.16 K (superfluidity loss) makes this a compelling demonstration of economic NMPC under hard physical constraints.⁶

Building Climate Control MPC — HVAC (ETH Zurich OptiControl-II, Switzerland)

ETH Zurich and Siemens Building Technologies deployed MPC on a fully occupied Swiss office building, controlling thermally activated building systems (TABS), an air handling unit (AHU), and blinds over seven months. Simulation-based comparisons showed approximately 17% reduction in non-renewable primary energy use and approximately 5,000 CHF/year net savings for one floor. Field experiments confirmed reliable MPC operation with good comfort levels in an occupied commercial setting.⁷

What This Means for Your Operations

• Start with the best-instrumented unit: process NMPC requires observability. Identify the production line where temperature, flow, or quality measurements are already available or can be added with minimal effort.
• Expect 3—6 month commissioning: shadow mode (weeks), advisory mode (weeks), and closed-loop validation (weeks) are standard milestones.
• ROI is measurable: energy per ton, quality deviation rates, and batch cycle times are directly trackable before and after deployment.
• No hardware changes required in many cases: several reference implementations (Holcim Untervaz, Novartis) achieved results purely through software-layer improvements on existing instrumentation.

How We Deliver This (Engagement Model)

• Phase 0: NDA + data request — understand your process, instrumentation, and current control baseline.
• Phase 1: Fixed-scope discovery — process modeling feasibility, control architecture concept, and expected KPI improvements.
• Phase 2: Implementation + validation + commissioning — model development, MPC tuning, shadow/advisory/closed-loop rollout.
• Phase 3: Monitoring + training + scaling — operator training, performance dashboards, and expansion to additional units.

Typical KPIs to Track

• Quality: temperature deviation from setpoint, composition variability (AR/SR), API content consistency
• Energy: energy per ton of product, non-renewable primary energy use, fuel substitution rate
• Throughput: batch cycle time, furnace throughput rate, uptime
• Operator burden: manual interventions per shift, alarm frequency, time in advisory vs. closed-loop mode

Risks & Prerequisites

• Model accuracy: first-principles models require process knowledge and validation data. Expect 2—4 weeks of step-test or historical data collection.
• Instrumentation gaps: NMPC cannot control what it cannot observe. Soft sensors or PAT may be needed for unmeasured quality attributes.
• Operator trust: staged commissioning is essential. Forcing closed-loop operation without shadow-mode validation creates resistance.
• Maintenance of models: process drift (wear, feedstock changes) requires periodic model updates or adaptive elements.

FAQ

Q: How does process NMPC differ from traditional APC (e.g., DMC)? NMPC uses a nonlinear, often first-principles model and solves a constrained optimization at each step. Traditional APC (like DMC) uses linear step-response models. NMPC handles nonlinearities, batch trajectories, and hard constraints more naturally, but requires more modeling effort.

Q: Can NMPC run on our existing DCS/PLC infrastructure? In most cases, NMPC runs on a separate compute layer (industrial PC or edge server) and sends setpoints to the existing DCS. No DCS replacement is needed — the MPC acts as a supervisory layer.

Q: What if our process changes frequently (product grades, feedstock)? NMPC handles this well because the model captures the physics of grade transitions. Multi-model or adaptive approaches can be configured for different operating regimes.

Q: What is the typical payback period? Reference implementations report payback periods of 6—18 months, depending on energy costs, quality penalty structures, and throughput value. A fixed-scope discovery engagement can estimate this for your specific case.

Book a 30-Minute Discovery Call

Ready to explore whether this pattern fits your system?

Dr. Rafal Noga — Independent APC/MPC Consultant

Email me | noga.es

Fixed-scope discovery — NDA-first — DACH on-site available

Public References

Footnotes

1. Steinboeck, Wild, Kugi, “Nonlinear model predictive control of a continuous slab reheating furnace” (Automation and Control Institute, TU Wien, 2013). PDF ↩

2. Marx et al., “Coal free Cement Plant Operation using Alternative Fuels — Modeling and Control of Pre-calciner under Alternative Fuels using Model Predictive Control” (ABB / AUCBM, 2008). PDF ↩

3. ABB / World Cement, “Expert Optimizer MPC/MLD case studies incl. Holcim Laegerdorf and Untervaz” (World Cement, March 2008). PDF ↩ ↩²

4. Novartis, “Advanced process automation of a pharmaceutical continuous wet granulation line: Model Predictive Control from solid feeders to dryer” (Powder Technology, 2023). DOI ↩

5. BASF, “BASF cooperates with partners to introduce online control of complex batch processes” (BASF News Release, 2015). Link ↩

6. “NMPC for the Superfluid Helium Cryogenic Circuit of the LHC” (IFAC PapersOnLine, September 2015). ↩

7. Sturzenegger et al., “Model Predictive Climate Control of a Swiss Office Building: Implementation, Results, and Cost-Benefit Analysis” (IEEE TCST / ETH Zurich, 2016). PDF ↩