# Digital Twins in Metals Manufacturing: A Complete Guide



> Digital twins cut steel scrap from 5-8% to under 2.5%. Learn the standards, data foundations and pilot plan that make one work on your line.
> 
> **URL:** https://www.gosmarter.ai/blog/digital-twins-steel-production-guide/

**Date:** 2026-07-20
**Author:** BlogSmarter AI

**Categories:** blog

**Tags:** artificial-intelligence, automation, data-strategy, digital-transformation, inventory, manufacturing, quality

## 



**Your mill already has the data. It just cannot use it in time.** A digital twin turns live plant data into a working model, so you can spot drift, scrap risk and line trouble before it burns cash.

The pain is not steelmaking or rolling aluminium. The pain is **messy certs, lagging reports, mismatched timestamps and spreadsheet guesswork**. You do not lose margin because the model is weak. You lose it because the records feeding the model are a mess.

I see the fix like this. Start small. Clean the material data first. Then run one twin on one painful process. [GoSmarter](/), built by [Nightingale HQ](https://nightingalehq.ai/), helps metals manufacturers do that by pulling data from mill certs, stock records, scrap figures, sensor tags and schedules into one usable flow.

What you get fast:

-   **[Lower scrap](/hubs/scrap-waste-yield-optimisation/)**. Cutting plans can push scrap from **5-8%** down to **below 2.5%**
-   **Less admin grind**. [Mill cert automation](/blog/how-to-automate-mill-certificate-management-in-5-steps/) can save **120+ hours a year**
-   **Earlier warnings** on quality drift, thermal risk, bottlenecks and bad stock matches
-   **Faster shift decisions** with live inputs, not yesterday's report
-   **Less [Carbon Border Adjustment Mechanism (CBAM)](/blog/lifecycle-carbon-tools-cbam-compliance/) reporting pain** because the records already line up
-   **A pilot you can run in 1-2 days**, not some bloated software circus

If you want a twin that helps during the shift, not after the damage, here is how to sort it: what a twin actually models, where the money sits, the standards that stop it becoming another dashboard nobody trusts, and the pilot plan that proves it works.

## How Tata Steel and Jindal Steel Increased Profits with Digital Twins

{{< youtube width="480" height="270" layout="responsive" id="kfncq-dN9fc" >}}

## What a Digital Twin Actually Models

On a metals line, the first thing to sort out is scope. Are you modelling one asset, one route, or the whole plant? That matters because a **static simulation** and a **live twin** do different jobs.

A static simulation answers one project question, then stops. A live twin keeps updating with production data and helps you make day-to-day calls. On a furnace, for example, a live twin can recalibrate from current line data and predict discharge temperature, energy use and breakout risk before the shift changes.

|     | Static Simulation | Live Digital Twin |
| --- | --- | --- |
| **Data source** | Historical or assumed values | Live sensor, supervisory control and data acquisition (SCADA), and quality data |
| **Update frequency** | Run once for a study | Continuous - updates with live production data |
| **Typical decisions** | Equipment design, feasibility studies | Daily scheduling, quality control, energy tuning |

### Machine Twin, Process Twin and Full Plant Twin

Each level answers a different kind of question, and each is a bigger commitment than the last.

A **machine twin** mirrors one asset. On a reheating furnace, it models slab heating, radiation and convection, combined with live burner and temperature data, to predict discharge temperature, skid-mark risk and energy use per tonne. On a continuous caster, it models solidification and strand temperature along the caster length, using casting speed and secondary-cooling set-points to forecast shell thickness and breakout risk. Because a machine twin stays focused on one asset, it is usually the first step. You can put it in place without ripping apart the whole plant setup.

A **process twin** covers several machines across one route. In a hot-strip mill, it might link the reheating furnace, roughing and finishing stands, run-out table and coilers in one continuous simulation of coil behaviour. This is where you can ask joined-up questions: if furnace throughput goes up by 5%, what happens to rolling forces, strip crown and final mechanical properties? A process twin shows you before you touch a set-point.

A **full plant twin** mirrors the whole site, including production, utilities and logistics. It layers an asset registry, maintenance history and energy models on top of the machine and process layers. It pulls live data from programmable logic controllers (PLCs), supervisory control and data acquisition (SCADA), laboratory systems and the manufacturing execution system (MES), then simulates how one change, such as altering caster speed, ripples through capacity, quality, energy and emissions across the site.

### Why Spreadsheets and Gut Feel Break Down

Metals production is tightly linked and messy by nature. A production manager juggling heat sequences, chemistry windows, casting speeds, rolling schedules, coil routing, maintenance windows and shipping commitments cannot hold all of that in a spreadsheet forever. Something slips.

Spreadsheets are **static snapshots**. They do not pull live data from PLCs or SCADA, so heat numbers, slab locations and coil statuses go stale fast. You end up deciding on yesterday's picture. Gut feel makes that worse: it works until product mixes grow, new grades arrive, or the mill runs close to capacity with frequent changeovers. That is when "we've always done it this way" turns into expensive guesswork.

|     | The Manual Way | The Digital Twin Way |
| --- | --- | --- |
| **Scenario testing** | A handful of options, recalculated by hand; complex "what ifs" avoided | Many scenarios simulated rapidly, showing predicted impact on throughput, scrap and energy |
| **Data accuracy** | Manual entry and ad-hoc imports; mismatched heat numbers and inconsistent scrap tallies | Live, calibrated data from SCADA, sensors and quality systems |
| **Response speed** | Slower to re-plan after a disruption | Recalculates quickly when a rush order arrives or a machine goes down |
| **Traceability** | Fragmented across PDFs, paper and memory | Automatic link from heat number to finished piece, updated continuously |

## Where Digital Twins Save Real Money in Steel and Aluminium

Digital twins save money where variation turns into **scrap, rework or downtime**: the furnace, caster and rolling mill. Some gains sit at asset level. Some hit the whole line. Others show up across the plant, mainly in scheduling and traceability.

### Furnace and Caster: Catching Drift Before It Becomes a Breakout

Blast furnace and caster twins help operators cut fuel rate, yield loss and breakout risk without breaking throughput, chemistry, temperature or safety limits. Continuous casting twins combine solidification models with live data from casting speed, mould level and secondary cooling water flow, so you can run virtual heats at different speeds and see where defect risk, such as surface cracks and hot tears, starts to climb.

In Electric Arc Furnace (EAF) steelmaking, energy usually makes up 20-30% of total production cost [\[9\]](https://gosmarter.ai/hubs/metals-manufacturing-glossary/), so even a small gain in furnace efficiency protects margin.

| The Manual Way | The Automated Way |
| --- | --- |
| Decisions depend on experience and lagging readings | Live process data supports earlier intervention |
| Breakout risk is identified only after thermal readings lag | Predictive models flag thermal issues before breakout occurs |
| Traceability relies on paper records and spreadsheets | The system creates a digital audit trail on receipt |

### Rolling Mills and Aluminium Lines: Catching the Coil Before It Is Scrap

The same pattern shows up downstream. In the rolling mill, process drift turns into scrap fast. For a rolling-mill-focused twin, the most useful inputs are [heat number](/docs/what-is-a-heat-number/), chemical composition, tensile strength, elongation and dimensions [\[5\]](/docs/what-is-a-mill-test-certificate/). Grade and chemistry must stay as hard constraints, so matching dimensions never hide a material mix-up [\[1\]](/hubs/cutting-optimiser/).

Aluminium plants face the same coupling problem. Furnace temperature, casting speed and cooling intensity are tightly linked, so you cannot tune one in isolation. In hot-strip rolling, a digital-twin-based shape-control model let plants test process settings virtually before making production changes [\[10\]](https://pubmed.ncbi.nlm.nih.gov/38257707/). One simulation of aluminium direct-chill casting matched melting time and fuel use within ±10% of actual plant data [\[11\]](https://www.programmaster.org/PM/PM.nsf/ApprovedAbstracts/5A2C2AAA93290CBB85258CAE001C23AF?OpenDocument) — close enough to test a scenario before you touch the plant.

The main use cases in steel and aluminium look like this:

| Sector | Example scenario in the twin | Main target metric | Typical operational goal |
| --- | --- | --- | --- |
| Steel | Continuous casting speed vs secondary cooling settings | Defect rate (cracks, surface defects) | Maximise slab throughput without increasing scrap or downgrades |
| Steel | Hot strip mill speed and entry temperature profile | Tonnes/hour and flatness quality | Increase output while keeping flatness and gauge within tolerance |
| Aluminium | Furnace temperature ramp and soak time changes | kWh/tonne and melt loss | Reduce energy use and oxidation while maintaining melt quality |
| Aluminium | Rolling speed and coolant flow settings | Surface defect occurrence and strip flatness | Push line speed safely while maintaining first-time-through quality |

In rebar and structural steel, AI-optimised cutting plans can cut scrap rates from an industry average of **5-8%** to **under 2.5%** [\[1\]](/hubs/cutting-optimiser/). In a two-week production trial, [Midland Steel](https://midlandsteelreinforcement.com/) ran GoSmarter's Rebar Optimiser across **193 jobs** and **734 tonnes** of steel, cutting scrap by **50%** [\[1\]](/hubs/cutting-optimiser/). Read the full [Midland Steel case study](/casestudies/midland-steel/) for the wider engagement. If a machine fails or a rush order lands, AI can replan at once, which is what makes a one-line pilot practical during a live shift.

### When a Plant-Wide Twin Beats Another Spreadsheet

A plant-wide view helps once the problem crosses team lines, not just one asset. The gain comes from linking inventory, heat codes and certificates across teams. A digital audit trail created on receipt links straight to heat codes [\[3\]](/), giving each team one reliable source instead of three versions of the same story.

The same data flow cuts admin time. [AI-powered mill test report processing](/blog/ai-mill-test-report-traceability/) digitises PDFs in seconds, removes manual data entry errors and saves production teams more than **120 hours** of admin time a year [\[2\]](/hubs/gosmarter-for-metals-operations/)[\[3\]](/). Those gains only last if you feed the twin clean signals and connect it without disrupting the kit that already runs the line.

## Connecting a Twin Without Breaking What Already Works

{{< image src="6a7525dcd642d19a979268e2-1786069363773.jpg" alt="Digital Twin Stack vs. Plant Setup: Which Approach Fits Your Metals Plant?" >}}

Most UK metals manufacturers need an **overlay**, not a rip-and-replace job. You have got mixed PLC generations, messy historians and an MES set up for product mixes that made sense years ago. That is normal. The job is not to build a perfect digital twin in a vacuum. The job is simpler, and harder: **add intelligence without upsetting the line**.

The twin sits next to your current control setup, SCADA, MES and quality systems, and acts as an integration and intelligence layer. It reads live signals without changing the control logic. Legacy PLCs keep the safety-critical work. The twin adds context, links plant data and gives operators actions through the dashboards or Human-Machine Interfaces (HMIs) they already know. Start read-only. Check the data. Test offline. Move to advisory mode. Then, if it proves itself, go closed loop.

### The Technology Stack That Makes a Twin Useful

A twin only earns its keep if the stack makes sense. At the bottom, sensors pick up what the process is doing: temperature, power draw, flow rate, position and speed. Above that sits the edge layer, where [Open Platform Communications Unified Architecture (OPC UA)](https://opcfoundation.org/about/opc-technologies/opc-ua/) servers, Modbus TCP gateways and similar industrial middleware turn proprietary machine signals into data you can use. Low-cost edge devices can expose OPC UA data on the shop floor, which matters in older plants where budgets do not stretch to full hardware replacement [\[12\]](http://scielo.senescyt.gob.ec/scielo.php?script=sci_abstract&pid=S1390-65422017000100287&lng=en&nrm=iso).

The storage layer, usually a historian or time-series database, holds time-series data with context such as heat numbers, grades and quality results, letting the twin trace an outcome back to a specific heat or coil. Above that, the analytics layer runs the twin logic: a physics-based furnace model, a machine-learning anomaly detector, or a discrete-event simulation of the rolling schedule. The application layer then puts the result in front of operators through dashboards, alarms and what-if tools. **Standards come first. Fancy optimisation comes later.**

These three twin setups fit different plant problems:

| Setup | Primary Strength | Best Fit |
| --- | --- | --- |
| **[Reference Architecture Model Industrie 4.0 (RAMI 4.0)](https://www.plattform-i40.de/IP/Redaktion/EN/Downloads/Publikation/rami40-an-introduction.html)-based** | Architectural standardisation across a heterogeneous plant | Large, multi-line plants with many asset types and vendors |
| **Asset Administration Shell (AAS)-led** | Asset interoperability and data portability across systems | Plants where machines from different suppliers need a common data language |
| **AI-led** | Prediction, optimisation and anomaly detection | Plants with decent data quality that need better decision support |

No single setup fits every plant. (See the [metals manufacturing glossary](/hubs/metals-manufacturing-glossary/) for definitions of OPC UA, AAS and other terms used here.)

-   If measurements are missing or unreliable, start with sensor retrofits and edge collection.
-   If data sits in silos, start with an AAS-led asset layer.
-   If the data already exists but nobody acts on it well, start with a focused AI-led twin for one use case.

### Why Standards Matter on Mixed Old and New Kit

The day-to-day problem in most metals plants is simple. Assets speak different languages. Tag names differ. Time references differ. Data formats differ. Without a common standard, connecting them to a twin means writing custom integrations for every asset, which gets expensive fast and breaks often.

RAMI 4.0 gives you the plant-wide blueprint. The [Asset Administration Shell](https://industrialdigitaltwin.org/en/content-hub/aasspecifications) gives each asset a standard digital identity: what it is, what data it exposes and how it behaves, in a format any compliant system can read, no matter who built the kit [\[6\]](https://ieomsociety.org/proceedings/brazil2025/53.pdf)[\[7\]](https://link.springer.com/chapter/10.1007/978-3-030-98636-0_9)[\[8\]](https://industrialdigitaltwin.org/wp-content/uploads/2023/04/Diskussionspapier-Zielbild-und-Handlungsempfehlungen-fuer-industrielle-Interoperabilitaet-5.3.pdf). The International Electrotechnical Commission (IEC) is standardising AAS as [IEC 63278](https://industrialdigitaltwin.org/wp-content/uploads/2023/06/IDTA-01001-3-0_SpecificationAssetAdministrationShell_Part1_Metamodel.pdf).

**OPC UA** is the communication backbone that moves data between assets and the twin, supporting secure, reliable data exchange across mixed-age industrial equipment [\[8\]](https://industrialdigitaltwin.org/wp-content/uploads/2023/04/Diskussionspapier-Zielbild-und-Handlungsempfehlungen-fuer-industrielle-Interoperabilitaet-5.3.pdf)[\[9\]](https://opcconnect.opcfoundation.org/2015/06/opc-ua-in-the-reference-architecture-model-rami-4-0/). Fraunhofer research shows that teams can integrate legacy robots and machines with digital twins through OPC UA without full replacement [\[5\]](https://publica.fraunhofer.de/entities/publication/664e09ba-1e84-4c13-bdc6-bf8b9d41a46d).

AAS and OPC UA do different jobs and do not compete. AAS defines what the data means. OPC UA moves it. Together, they let older furnace systems and newer analytics platforms share the same operating picture, which is what traceability by heat number actually needs. The practical move is to start with OPC UA connectivity and a basic AAS layer for the assets that matter most: a furnace, a caster, or a key rolling stand. Standardise enough to make one pilot trustworthy, then extend it line by line.

## How to Build a Twin Without Drowning in Data

{{< image src="6a5d67feb4101d23b971ce7e-1784512343948.jpg" alt="Digital Twin Model Types for Steel Production: A Quick Comparison Guide" >}}

A lot of digital twin projects stall for one boring reason: **the data is a mess**. Heat numbers sit in one system. Mill certificates hide in PDFs. Sensor tags do not match historian records. That is what blocks progress, not a lack of clever modelling.

### Why Most Twin Projects Fail on Data Before They Fail on Models

Once you connect a twin, **data quality** becomes the problem, not the model. Most twin failures start with three old headaches:

-   **Drifting sensors**. Worn thermocouples and badly scaled PLC tags push biased inputs into the model [\[13\]](https://www.informatica.com/content/dam/informatica-com/en/collateral/ebook/5-signs-your-data-architecture-is-not-ready-for-digital-twins_ebook_5302en.pdf).
-   **Unsynchronised timestamps**. If clocks do not match across PLC, SCADA, MES and enterprise resource planning (ERP), the twin cannot link a furnace reading to the right heat number [\[13\]](https://www.informatica.com/content/dam/informatica-com/en/collateral/ebook/5-signs-your-data-architecture-is-not-ready-for-digital-twins_ebook_5302en.pdf).
-   **Broken traceability**. If people reuse or mistype heat numbers between the melt shop and finishing, the twin cannot piece together what happened to a given steel product [\[19\]](https://www.testcert.co/guides).

Before any serious twin deployment, a UK metals plant needs five basics in place: calibrated sensors, one timestamp standard across all systems, stable heat and coil identifiers in every system, reliable links between PLC/SCADA, MES, the laboratory information management system (LIMS) and ERP, plus mill certificates in structured, machine-readable form [\[14\]](https://sustainableatlas.org/post/case-study-digital-twins-for-infrastructure-industry-a-pilot-that-failed-and-what-it-taught-us-839)[\[17\]](https://www.sciencedirect.com/science/article/pii/S2452414X24000219).

### Start with the Signals You Already Have

Before you model anything, map the signals you already have. Start with PLC, SCADA, MES, ERP and lab signals. Then link each tag to a heat number, certificate and grade. Put material data first: chemistry, heat numbers, grade certificates. This is where most plants trip up. PDF mill certificates come in unstructured. Legacy records sit all over the place. Someone re-keys the lot into spreadsheets and hopes nothing breaks. It usually does. Clean this layer first. Everything else sits on top of it [\[2\]](/hubs/gosmarter-for-metals-operations/)[\[4\]](/hubs/mill-cert-automation/).

### Choosing the Right Model Type for the Job

Once you can see the data properly, model choice gets a lot less painful. Different jobs need different model types. Pick the wrong one and you can lose months to software theatre.

| Model Type | Primary Inputs | Computational Effort | Best-Fit Application |
| --- | --- | --- | --- |
| **Physics-Based** | Dimensions, density, thermal constants | Low | Weight, area and heat-transfer calculations |
| **Data-Driven (AI)** | PDF certificates, sensor logs, legacy records | Medium | Certificate extraction and pattern recognition |
| **Optimisation** | Inventory lengths/grades, open orders | Moderate | Cut planning and scrap reduction |
| **Hybrid** | Cleaned mill test record data + real-time sensor signals | Medium-High | Plant-wide quality prediction |

For most metals plants, the hybrid approach works best [\[2\]](/hubs/gosmarter-for-metals-operations/)[\[4\]](/hubs/mill-cert-automation/). It uses AI to clean and digitise material data first — extracting values from PDFs with tools trained on metals terms, so "Rp0.2" gets read correctly as yield strength — then feeds that cleaned data into physics-based or optimisation models. On the modelling side, the best metals twins mix physics-based models (thermal, solidification, material-flow) with data-driven models that patch the gaps left by incomplete records [\[15\]](https://pmc.ncbi.nlm.nih.gov/articles/PMC12787485/).

Use [cutting optimisation](/docs/what-is-cutting-optimisation/) for 1D cutting stock, where spreadsheets cannot test enough combinations [\[1\]](/hubs/cutting-optimiser/). Manual cut planning usually produces **5-8% scrap**. AI-optimised plans cut that to **below 2.5%** [\[1\]](/hubs/cutting-optimiser/).

### Calibrating the Twin Until Operators Trust It

If operators do not trust the twin, they will not use it. Calibration closes the gap between what the model predicts and what the plant actually does. Offline calibration starts with old process and quality data. Online calibration runs beside the live process, comparing predictions with incoming sensor readings and adjusting as it goes.

The material data layer matters just as much as the sensor layer. If a heat number in the twin does not match the physical stock on the floor, quality predictions fall apart. Link every mill certificate to a specific inventory record when stock arrives. That gives you one source of truth the twin and the operators can both trust [\[2\]](/hubs/gosmarter-for-metals-operations/)[\[3\]](/). Start with one process that is easy to check against physical measurements. Prove the output. Then expand once operators accept the results.

## Running a Digital Twin Pilot That Proves Itself

{{< image src="6a704a082f6aeb480bfec7e6-1785751676277.jpg" alt="Digital Twin Pilot Roadmap for Metals Plants: 4 Phases to Live Deployment" >}}

Do not start with a plant-wide model. That is how you end up with a big, expensive diagram nobody trusts. Clean the data first, then run a tight pilot: one line, one question, one result you can measure.

### Pick One High-Value Line and One Question

Look at the last 12-24 months of production data. Find the line with the biggest mix of scrap, energy use and unplanned downtime. In UK steel and aluminium plants, that often means a hot strip mill, a continuous caster, a reheating furnace, or an aluminium rolling line.

Put a pound figure on the pain. If a 2-3% scrap cut, or a 1-2% energy saving per tonne, is worth at least **£250,000-£500,000 a year** on that line, you have your pilot [\[1\]](/hubs/cutting-optimiser/). Then cut the scope again. Ask one question, not five:

> "Can we cut edge-trim scrap on the hot strip mill by 10% without breaching customer tolerances?"

> "Can we shift rolling schedules to off-peak tariffs to reduce kWh per tonne without creating downstream bottlenecks?"

Set your key metrics before you build anything: pounds saved per month, tonnes of scrap cut, kWh per tonne, and on-time delivery percentage. If you cannot measure the result inside 12 weeks, the question is still too broad.

### Validate Against Real Runs, Then Use It in Daily Planning

Feed the twin 6-12 months of old runs, using the same product mix, schedules and set-points from that period. Compare its scrap, throughput and energy-per-tonne predictions with what actually happened on the line. If the caster ran at 2.5% scrap and 450 kWh per tonne, the twin should land near those figures, and match the swings around them within an agreed tolerance.

When that checks out, move to controlled trials on the line. Change one high-impact setting the twin points to, then run matched coils or billets under the old settings and the twin-recommended settings. Measure the gap in scrap tonnes, defect counts, energy use and compliance metrics linked to customer or regulatory rules. Get your quality team to sign it off. Savings mean nothing if you create spec breaches or traceability holes. Only then put the twin into routine planning.

| Phase | Key Activities | Indicative Duration | Primary Owner |
| --- | --- | --- | --- |
| **Foundation** | Define scope, agree key metrics (£, tonnes, kWh/tonne), map data sources | 2-4 weeks | Digital/Innovation Lead + Plant Manager |
| **Data Setup** | Connect SCADA/PLC data, ingest ERP/MES exports and mill certs, clean and standardise inputs | 4-8 weeks | Data Engineer + Process Engineer |
| **Model Validation** | Test against historical runs, run controlled trials, confirm outputs within agreed tolerance | 4-6 weeks | Lead Process Engineer + Data Scientist |
| **Operational Rollout** | Embed twin into daily planning, train planners and engineers, set governance for model updates | 4-8 weeks | Operations Manager + Planning Lead |

In day-to-day use, planners run the twin against the next 7-14 days of rolling or casting sequences before they commit. Process engineers run what-if tests before touching reheating curves or coolant flow settings. Operations leads use a stripped-back dashboard in daily huddles to spot bottlenecks and energy-heavy runs before they hit the floor.

## Run a Pilot with GoSmarter on One Painful Process

{{< image src="1a2522ef04568f801ea774469c6af8d8.jpg" alt="GoSmarter's MillCert Reader dashboard showing extracted mill certificate data" >}}

Start your pilot where bad data does the most damage. For most metals firms, that means **material certificates**. It ties the pilot to a job that affects output, not just another dashboard.

### Clean Material Data First with MillCert Reader

Your digital twin is only as good as the data you feed it. If heat numbers sit in scanned PDFs, grades get typed by hand and certificates hide in someone's inbox, the model is guessing. GoSmarter, built by Nightingale HQ, fixes that first. [GoSmarter's MillCert Reader](/features/mill-certificate-reader/) reads PDFs, scans and photos of mill certificates on its own. It pulls out heat numbers, grades, and chemical and mechanical properties in seconds, then links each record to the right stock item [\[3\]](/)[\[4\]](/hubs/mill-cert-automation/).

That gives you a traceable material record your twin can trust. Most teams go live in one to two days, at £350 a month in the Workshop tier, and most sites recover that inside the first quarter from admin hours saved alone. GoSmarter sits on top of your current [ERP in metals manufacturing](/docs/what-is-erp-metals-manufacturing/) system, whether that's [Infor](https://www.infor.com/solutions/erp), [Epicor](https://www.epicor.com/en/products/enterprise-resource-planning-erp/), [Dynamics](https://www.microsoft.com/en-us/dynamics-365) or [Sage](https://www.sage.com/en-gb/sage-business-cloud/sage-x3/). It connects through a Representational State Transfer (REST) API, authenticated with OAuth 2.0 or Microsoft Entra single sign-on (SSO), with CSV sync as a fallback. Your data stays on UK Azure infrastructure and is never shared to train Microsoft's underlying models. You do not need a rip-and-replace project to get started [\[2\]](/hubs/gosmarter-for-metals-operations/)[\[3\]](/).

### Add Product Lineage, Scrap and Scheduling Data

This is where the twin starts doing useful work. **Product Lineage** builds and keeps an end-to-end chain from melts to casts, slabs, coils, cuts and finished parts, joined to process data and quality outcomes. Instead of an engineer stitching together spreadsheet exports and historian queries to answer an audit question, the twin can pull a ready-made record of where each product went.

GoSmarter's scrap calculation and [production scheduling tools](/docs/optimised-production-plans/) connect real material flow back to certified stock. The same heat-number spine feeds the Scrap Calculator, the [Smart Production Scheduler](/solutions/production/), Product Lineage and the MillCert Reader — one record, every tool, covering offcuts, remnants and cut plans. So the twin reflects what the floor is producing, not what the plan predicted [\[1\]](/hubs/cutting-optimiser/)[\[4\]](/hubs/mill-cert-automation/).

Run 2-4 weeks of data from your pilot line through GoSmarter: mill cert PDFs, scrap records with weight, time and type, and a production export with coil identifier, heat code, grade and shift time. GoSmarter's MillCert Reader pulls out fields including heat code, grade, chemistry, mechanical properties and [carbon equivalence (CEQ)](/hubs/metals-manufacturing-glossary/#carbon-equivalence-ceq), then puts them into one consistent schema even when certificate formats vary. It links each heat code to the matching coil or batch identifiers in your production records, then links those to scrap entries using the same identifiers.

Before you build the full twin, run three checks on your cleaned batch:

-   Can you trace **more than 80% of coils** from mill cert to finished product and scrap record? If not, the gaps will skew the result.
-   Do derived metrics like yield percentage, [scrap rate](/blog/how-to-calculate-scrap-rate/) and throughput match your engineers' spreadsheet figures within about **±2-3%**?
-   If you run the pipeline again, do you get the same figures?

If those three checks pass, you have an input layer you can trust for scenario testing.

### What the Board Cares About: Margin, Energy and CBAM

These data fixes matter because they turn twin output into margin, energy and compliance gains. **Scrap** is the clearest link: clean, consistent scrap labels let the twin learn the connection between process conditions and yield loss. **Energy decisions** only work with time-aligned data. If power draw, gas use, throughput and product mix sit in different systems with different timestamps, the twin cannot tell you when to shift energy-heavy work into lower-tariff periods.

**CBAM** is the compliance issue pushing this up the board agenda now. Auditable, structured data turns [Carbon Border Adjustment Mechanism (CBAM)](https://taxation-customs.ec.europa.eu/carbon-border-adjustment-mechanism_en) reporting into a routine twin output instead of a last-minute scramble [\[14\]](https://sustainableatlas.org/post/case-study-digital-twins-for-infrastructure-industry-a-pilot-that-failed-and-what-it-taught-us-839)[\[16\]](https://www.informatica.com/blogs/why-digital-twins-fail-without-the-right-data-foundation.html). A 5% scrap rate linked to defects you cannot trace can cost a mid-sized mill more than **$50,000 a month** in lost material and rework [\[22\]](https://ifactory.jrsinnovation.com/blog/steel-quality-traceability-system-manufacturing). That is money leaking out because the paperwork still runs the show.

You do not need a multi-year transformation programme for any of this. You need the right first process and clean data. Prove the pilot on one painful process, then add the next layer with lineage, scrap and scheduling data.

## Key Takeaways for Metals Plants Starting with Digital Twins

A [digital twin](/blog/digital-twins-and-ai-for-manufacturers/) falls apart if your data is a mess. Clean that up first. No shortcuts.

Then pick **one painful process** and **one key performance indicator (KPI)**. That focus helps operators trust the thing and gives you a result you can measure without the usual software fog. Speed matters too: if a twin cannot keep up with the shift, it is just another screen to ignore. Hybrid twins stay fast enough for shift decisions because they mix physics with live shop-floor data, and standards such as OPC UA and the Asset Administration Shell keep mixed-age plant kit talking the same language.

| What matters most | Why it counts |
| --- | --- |
| **Live data connection** | Without it, the twin drifts away from reality |
| **Trusted inputs** | Poor data leads to misleading recommendations |
| **Clear metrics upfront** | If you cannot measure it, you cannot prove the value |
| **Validation before live use** | Testing against historical and controlled runs builds trust in the output |
| **Phased deployment** | One line, one question, one result, then expand |

The best places to start are simple:

-   **Mill certificate automation**: turns unstructured PDFs into traceable material records, saving **120+ admin hours a year** [\[3\]](/)
-   **Cutting plan optimisation**: links certified stock to open orders and cuts scrap rate by **up to 50%** [\[1\]](/hubs/cutting-optimiser/)[\[3\]](/)
-   **Plant-wide scheduling**: connects scrap recovery data and multi-site stock levels to production throughput

For your first pilot, use the systems already on the floor. Do not pile on new instrumentation if you do not need it. Connect the ERP system already in use by API or CSV. Most metals manufacturers can go live with a focused pilot in **one to two days** [\[3\]](/). Prove the point on one process first. Then expand.

## FAQs

{{< faq question="What is a digital twin in metals manufacturing?" >}}
A digital twin is a live software model of a real plant asset, line or process. It sits alongside your SCADA, MES and quality systems and acts as an integration and intelligence layer, reading live sensor and production data and linking it to context such as heat numbers and grades.

You use it to test production scenarios and check performance data in a virtual setting before you change anything on the shop floor. In a cyber-physical system, the twin never replaces the control logic. Legacy PLCs keep the safety-critical work, and the twin adds context and gives operators clearer, faster decisions through the dashboards they already use.
{{< /faq >}}

{{< faq question="What data do I need to start a pilot?" >}}
To start a pilot, you need two main data sets: **your stock inventory** and **your current open order book**. That means available bar lengths, grades, mill certificates, and order details such as required lengths, quantities and material grades — the stuff older systems usually scatter across spreadsheets and inboxes.

GoSmarter, built by Nightingale HQ, takes that data and does the grunt work your software never sorted out. It pulls the key details from certificates, checks material compliance, and builds an optimised production schedule. You keep your legacy software. You just stop asking it to do jobs it was never much good at.
{{< /faq >}}

{{< faq question="Can a digital twin work with older plant systems?" >}}
Yes. You do not need to rip out legacy software or scrap hardware that still does the job. GoSmarter, built by Nightingale HQ, sits on top of your existing ERP system and shop-floor technology, connecting via a REST API with OAuth 2.0 or Microsoft Entra single sign-on, hosted on UK Azure infrastructure. Your data is never shared to train Microsoft's underlying models.

Older machines can link in through edge devices and protocols such as OPC UA, so you keep using the kit on the floor while pulling in real-time insight and automated workflows instead of patching the gaps by hand.
{{< /faq >}}

{{< faq question="How do operators learn to trust the twin?" >}}
Trust builds when the digital twin stays transparent, not when it acts like a black box. You need a system that backs up operators, not one that pretends it knows the plant better than they do.

Show people clear, useful insights they can check against what they already know on site. Let them override bad suggestions and tweak plans when the model misses something. That is how the twin earns its place as a **reliable shop-floor assistant**, not another bit of software that talks down to the people doing the work.
{{< /faq >}}

{{< faq question="How do we choose the best pilot line?" >}}
Choose the pilot line by looking at the problem that causes the most day-to-day friction. Start where the pain is worst — that is usually where you get the clearest win.

If admin is clogging up the day, start with GoSmarter's MillCert Reader. If scrap is eating your margin, start with a cutting plan pilot. Keep the pilot tight: pick one area, such as goods receivable, and track a small set of numbers from day one, such as admin minutes per certificate and rework incidents, to give yourself a clean baseline before you scale.
{{< /faq >}}