HAH Software blog

Hybrid modelling & digital twins: pharma & chemical scale-up success

gavin.helinski@hahsoftware.com (Gavin Helinski) — Thu, 11 Jun 2026 14:52:20 GMT

A science‑driven path to process excellence across the lifecycle of chemicals and pharmaceuticals

Pharmaceutical, biopharma, and chemical manufacturers are under unprecedented pressure. Processes are becoming more complex, product portfolios more diverse, and regulatory expectations more stringent. Yet many organisations still rely on empirical optimisation, with incremental tweaks, trial‑and‑error experiments, and historical intuition, even when their plants generate terabytes of data.

Data is abundant, but actionable process understanding is scarce. Mechanistic models alone cannot capture the full behaviour of modern processes, while purely data‑driven models struggle with limited datasets, extrapolation, and regulatory acceptance. Hybrid modelling offers a way out of this trap.

Why traditional approaches fall short

Mechanistic models
Mechanistic or first‑principles models provide scientific interpretability with data on—mass balances, energy balances, reaction kinetics, fluid dynamics. But they require simplifying assumptions, and building them for complex biological or multiphase systems is time‑consuming and often incomplete.

Data‑driven models
Machine learning models excel at pattern recognition and nonlinear relationships. However, they depend on large, representative datasets and often fail when conditions shift, such as the introduction of new equipment, new raw materials, new scales. They also lack the mechanistic traceability regulators expect.

The result
Manufacturers face recurring pain points:

⚠️ Unreliable scale‑up and scale‑down
⚠️ Tech transfer failures between sites
⚠️ High batch‑to‑batch variability
⚠️ Limited visibility into CQAs and “unmeasurable” states
⚠️ Long development timelines and heavy experimental burden

Hybrid modelling: The best of both worlds

Hybrid modelling integrates first‑principles science with machine learning, creating models that are:

✔️ Mechanistically grounded: Every prediction is tied to physical reality
✔️ Highly predictive: ML captures nonlinearities and interactions
✔️ Robust to change: Models extrapolate more reliably across scales and equipment
✔️ Regulator‑friendly: Mechanistic traceability supports QbD and PAT

When deployed as digital twins, these hybrid models become real‑time decision engines. They reconstruct critical parameters that cannot be measured directly, such as mixing intensity, shear stress, mass transfer, local gradients, and provide early warnings when processes drift out of their ideal operating envelope.

Case studies across the value chain

1. Biopharma: Mammalian cell culture tech transfer

A biopharma manufacturer struggled to transfer a mammalian cell culture process between sites using different commercial bioreactors. Variations in vessel geometry and agitation systems caused inconsistent mixing and oxygen transfer.

A physics‑based digital twin was used to characterise hydrodynamics, shear, and mass transfer across vessels. By integrating plant data with mechanistic models, engineers:

✔️ Identified the root causes of variability
✔️ Matched mixing environments across scales
✔️ Reduced the number of physical optimisation runs
✔️ Improved viable cell density and titer consistency

This approach de‑risked tech transfer and accelerated time to performance.

2. Drug product: Protein dilution and shear control

During final formulation, a pharmaceutical company observed protein aggregation caused by high shear during dilution.

Using CFD‑based digital twins, engineers mapped shear stress and mixing profiles across operating conditions. The model revealed:

✔️ Agitation speeds that minimised shear
✔️ Addition strategies that maintained blend uniformity
✔️ Operating windows that protected sensitive proteins

The optimised process reduced protein damage and delivered savings of ~$75,000 per batch.

3. Biopharma: Bioreactor process health monitoring

Process drift, caused by subtle, cumulative deviations in bioreactor conditions, is a major cause of performance loss. It often goes undetected because key biological states cannot be measured directly.

A hybrid digital twin combined historical data with biological and mechanistic models to reconstruct these “unmeasurable” states in real time. This enabled:

✔️ Early detection of abnormal behaviour
✔️ Improved upstream robustness
✔️ Higher cell viability
✔️ Reduced batch variability

Regulatory impact: A stronger scientific foundation**

Regulators increasingly expect mechanistic understanding, data‑driven justification, and transparent control strategies. Hybrid modelling directly supports:

Quality by Design (QbD): Hybrid models quantify how CPPs influence CQAs, enabling robust design spaces.
Process Analytical Technology (PAT): Digital twins act as soft sensors, providing real‑time visibility into states that cannot be measured directly.
Model‑informed regulatory submissions: Mechanistic traceability ensures that predictions are scientifically defensible, which is a key requirement for global regulatory agencies.

By combining scientific rigour with data‑driven adaptability, hybrid modelling reduces regulatory risk and strengthens process understanding across the global supply chain.

Why this matters now

Hybrid modelling and digital twins deliver:

✔️ Faster development
✔️ More reliable scale‑up
✔️ Fewer failed batches
✔️ Stronger tech transfer
✔️ Lower cost of experimentation
✔️ Greater confidence in regulatory submissions

This is not just a modelling upgrade — it is a shift toward science‑driven, data‑empowered manufacturing. Hybrid modelling is becoming the backbone of modern process development and commercial manufacturing. For organisations seeking to reduce variability, accelerate innovation, and build resilient global supply chains, hybrid digital twins offer a clear, proven path forward.

Download your free guide

" src="https://hubspot-no-cache-eu1-prod.s3.amazonaws.com/cta/default/147368144/interactive-419403712752.png" style="height: 100%; width: 100%; object-fit: fill; margin: 0 auto; display: block; margin-top: 20px; margin-bottom: 20px" align="center">

Are your defence vehicles safe from chemical attacks?

gavin.helinski@hahsoftware.com (Gavin Helinski) — Thu, 11 Jun 2026 14:26:09 GMT

Why digital leakage simulation is redefining crew protection in the defence sector

For ISTAR (Intelligence, Surveillance, Target Acquisition and Reconnaissance) and other protected mobility platforms, armour and active protection systems tend to dominate the conversation. Yet a quieter, less visible vulnerability is rapidly becoming a frontline design priority: enclosure integrity. In an era of CNBR (Chemical, Nuclear, Biological and Radiological) threats, the question isn’t simply whether a vehicle is “airtight,” but whether it is sealed against uncontrolled air ingress.

Airtight doesn’t mean airless
Modern armoured vehicles are not sealed like submarines. Crews still need breathable air, and that air must come from outside. The difference is that it must enter through a single, intentional pathway: the vehicle’s CBRN filtration and overpressure system. This unit draws in external air, filters out contaminants, and pushes clean air into the crew compartment. By maintaining positive pressure, the system ensures that if any micro‑gaps exist, air flows outward, not inward.

But this only works if the hull is tight enough for the overpressure system to do its job. Even a 1mm gap in the wrong place can break the pressure balance, overwhelm filtration, and expose the crew to chemical or biological agents. That’s why “airtightness” in defence engineering really means no unfiltered air entering the vehicle at any point.

The hidden vulnerability: structural leaks
Traditionally, leak detection has been a reactive, late‑stage process. Engineers build a physical prototype, fill it with smoke or tracer gas, and manually search for leaks. This “blower door” approach is slow, expensive, and often too late to influence the design. If a leak is discovered after welding, the options are limited: add sealant, add weight, or accept a compromise.

In a world where NBC threats are evolving, “close enough” is no longer acceptable.

From reactive testing to ‘secure by design’
HAH Software is bringing Elsyca LeakageMaster to the UK and Ireland defence sector to change this paradigm. Instead of waiting for a physical prototype, designers can now simulate leakage digitally during the CAD phase — long before steel is cut or plates are welded.

With advanced digital leakage simulation, defence teams can:

Automate detection: Identify microscopic gaps, imperfect seals, and geometric inconsistencies in complex CAD files.
Verify integrity: Ensure the hull supports stable overpressure and that all airflow passes through the filtration system.
Build a digital thread: Create a repeatable, quantitative reference for future vehicle programmes, reducing reliance on manual checks and human interpretation.

Precision that protects lives
Elsyca’s simulations have shown remarkable correlation with physical test results. This isn’t just about reducing prototype costs or avoiding unnecessary sealant weight. It’s about confidence — the confidence that when a vehicle enters a high‑threat environment, the crew is protected by a hull that performs exactly as intended.

The goal is simple: eliminate vulnerabilities in the digital phase so they never reach the battlefield.

If you’re involved in the design or procurement of protected mobility platforms, we’d welcome a conversation. Digital leakage simulation is reshaping how defence teams think about integrity — and redefining what “airtight” truly means.

Unlock engineering value with open data and models

gavin.helinski@hahsoftware.com (Gavin Helinski) — Thu, 11 Jun 2026 13:38:32 GMT

The shift from documents to open and simulation models is here and it's reshaping how complex products get built.

For decades, engineering teams have wrestled with the same friction: disconnected documents, siloed CAD and simulation files, late-stage physical tests that catch problems too late, and tribal knowledge that walks out the door when an engineer changes roles.

Digital engineering is finally breaking that cycle and the organisations that move first are pulling away from the pack.

From documents to a connected model

Digital engineering replaces the document-centric workflow with a unified, model-centric environment. Simulation, data, and decisions stay connected across the entire product lifecycle. That means faster iteration, earlier flaw detection, and engineering choices backed by traceable, high-quality data instead of gut feel.

The proof is already on the road, in the air, and on the motor racing grid. Renault, Toyota, Boeing, and BMW have all demonstrated major gains in throughput, quality, and program predictability after adopting digital engineering and simulation-driven workflows. Across automotive, aerospace, energy, defence, and construction, the pattern is consistent: shorter development cycles, less rework, and significant cost savings.

The authoritative source of truth

The destination most engineering leaders are aiming for is what we call an Authoritative Source of Truth (ASoT). This is a single, validated foundation for engineering information that satisfies the FAIR principles: Findable, Accessible, Interoperable, and Re-usable & Reproducible.

Layer advanced modelling, simulation, and AI-assisted analysis on top of that foundation, and slow sequential processes give way to continuous development. The payoff is a connected engineering ecosystem that accelerates innovation, strengthens assurance, and critically, prepares teams for the next wave of AI-enabled product development.

The open answer: openDEMS and openSPDM

Most enterprise simulation data management platforms are bolted onto heavyweight PLM systems. They're expensive, rigid, and slow to deploy. For many organisations, that's overkill.

openDEMS (Digital Engineering for Modelling & Simulation) and openSPDM (Simulation Process and Data Management) offer an open, lightweight foundation for organisations that need scalable simulation without the cost or rigidity of traditional PLM-based systems.

openDEMS centralises simulation knowledge so engineers can quickly find, reuse, and trust past work.

openSPDM automates workflows, captures provenance, and ensures repeatability across complex engineering programs.

Together, they remove one of the biggest barriers to digital engineering adoption: slow, bespoke, fragile integrations. A neutral parameter-exchange layer lets simulation tools connect without relying on proprietary APIs, enabling reusable connectors for:

✔️ CFD solvers and FEA tools
✔️ Multibody dynamics codes
✔️ In-house and legacy code
✔️ Python, MATLAB, and other scripting environments

The result is a modern engineering environment with traceable workflows, automated processes, integrated toolchains, and AI-ready data.

Compute with a conscience: Qarnot

Simulation is only as fast as the compute behind it and traditional HPC carries a heavy environmental footprint. Pairing openSPDM with Qarnot's sustainable HPC platform gives organisations high-performance simulation with dramatically lower environmental impact. Qarnot recovers heat from compute workloads to warm buildings, turning what would be wasted energy into something useful, while delivering predictable, scalable capacity.

Why this matters now

The companies winning on cycle time and quality aren't the ones with the biggest CAD licences. They're the ones who have stitched their simulation, data, and decision-making into a single connected thread and who can feed AI the clean, traceable data it needs to be useful.

If you're still moving spreadsheets between teams and re-running simulations because nobody can find last quarter's results, the gap is only going to widen.

Download our free eBook

" src="https://hubspot-no-cache-eu1-prod.s3.amazonaws.com/cta/default/147368144/interactive-419401327832.png" style="height: 100%; width: 100%; object-fit: fill; margin: 0 auto; display: block; margin-top: 20px; margin-bottom: 20px" align="center">