Unplanned downtime and missed defects are expensive. We build AI that watches your machines and your line — predicting failures before they happen and catching quality issues the eye misses — so you run more, scrap less, and ship better.
Last updated: 11 June 2026
Manufacturing already generates the data AI needs — sensor readings, machine logs, camera feeds. The opportunity is turning that stream into action: spotting the early signs of a failure, flagging a defect before it ships, and tuning the process for less waste and energy.
We build for the factory floor, not the slide deck — robust to real conditions, deployable on the edge when the cloud isn't practical, and wired into the systems your operators already use.
Predict equipment failures from sensor data before they cause unplanned downtime.
Computer vision spots defects and deviations on the line — faster and more consistently than the eye.
Forecast demand and plan production and inventory with fewer shortages and less overstock.
Tune parameters to reduce waste, energy use and cycle time across the process.
Catch unusual machine behaviour early, before it becomes a failure or a quality issue.
Turn machine and line data into clear OEE insight your team can act on.
We map the costliest problems, your machines and the data already available.
By the second call you get a working prototype on your use case — not a spec.
We deploy to edge or on-prem and wire alerts into your MES, ERP and maintenance flow.
We track accuracy in real conditions and keep the model performing as the line changes.
Most plant managers can quote their machine hour rate but underestimate what a stoppage truly drains. The visible number is idle labour and lost output. The hidden number is the cascade: a packaging line that halts upstream of a filler, the rush order pushed to overtime, the late-pallet penalty, the quality drift in the first parts after a cold restart. Across Dutch discrete and process manufacturing, credible industry benchmarks put unplanned downtime at roughly 5 to 20 percent of available production time — in many plants the single largest controllable loss on the books.
The reason it persists is not a lack of data. A modern line already streams vibration, temperature, current draw, pressure, cycle counts and PLC fault codes. The gap is that this data sits in historians and SCADA screens nobody watches in real time, and the patterns that precede a failure — a bearing that runs a few degrees warmer each shift, a motor whose current signature shifts before it seizes — are too subtle and too constant for an operator to track by eye. That is precisely the work AI does well, and it is why AI for manufacturing in the Netherlands starts with the losses you can already see in your OEE report rather than with a technology shopping list.
Predictive maintenance is often sold as magic; the mechanics are concrete. A model learns the normal operating signature of an asset from its own historical sensor data — the vibration spectrum of a healthy pump, the thermal profile of a gearbox under load, the current draw of a conveyor motor at a given speed. When live readings begin to deviate from that learned baseline in a way that historically preceded a breakdown, the system raises a flag with enough lead time for planned intervention.
The business shift is from two bad extremes to one good middle. Run-to-failure means catastrophic, unscheduled stops and collateral damage to surrounding components. Fixed-interval servicing means replacing healthy parts on a calendar and still missing the failures that do not respect the schedule. A condition-based approach lets maintenance happen when the asset actually needs it — fewer emergency call-outs, less premature part replacement, and crucially, work scheduled into a planned window instead of erupting mid-shift.
What changes operationally:
For a worked example of failure-signature learning on industrial equipment, see our predictive maintenance case study. The same approach extends to compressors, CNC spindles, injection-moulding presses and any rotating or cycling asset that logs its own behaviour.
Manual visual inspection has a hard ceiling. Attention fades across a shift, the standard drifts between inspectors, and at line speed a human simply cannot examine every unit. Computer vision quality control removes that ceiling: a camera over the line, a model trained on your defect classes, and a pass or fail decision in milliseconds on every single part, consistently, shift after shift.
The value is not only the defects you catch but where you catch them. A defect found at the camera is a few cents of scrap. The same defect found by your customer is a return, a credit note, a quality complaint and a dent in a hard-won relationship. Pushing detection upstream — to the moulding station, the weld, the print, the fill — is the difference between a contained internal scrap number and an external failure that reaches the field.
Defect detection generalises across surprisingly different products: surface flaws on metal and plastic, missing or misaligned components on an assembly, print and label errors, fill-level and seal integrity on packaging, contamination on food and pharma lines, and structural flaws in heavy materials. Our concrete crack detection case study shows the method applied to precast segments — spotting cracks, spalling, chipping and voids automatically, right on the production line — and the same vision pipeline retargets to your own parts and your own defect taxonomy. The engineering behind it lives in our computer vision service.
One practical point that shapes every deployment: on a fast line you cannot wait for a round trip to the cloud. We run models on the edge — an industrial PC or device next to the machine — so the inspection decision is instant and the line keeps moving even if the network drops. It also keeps sensitive production imagery inside your own walls, which matters for both IP and compliance.
OEE — Overall Equipment Effectiveness — is the cleanest scoreboard manufacturing has, because it forces you to confront all three of its components honestly: availability (is the machine running when it should be), performance (is it running at its rated speed), and quality (how much of what it makes is good first time). A line that looks busy can still post a poor OEE if micro-stops sap performance or scrap eats into quality. Most plants we meet already track OEE; few use it to direct AI investment.
That is the most useful thing AI does with your OEE data — it tells you which loss to attack first. Decompose the score and the biggest contributor becomes obvious: if availability is the drag, predictive maintenance is the lever; if quality is the drag, vision inspection and root-cause analysis are; if performance is the drag, process optimisation on speeds, feeds and parameters earns its keep. The point is to spend on the loss that actually moves your number.
The arithmetic is unforgiving in your favour. Because OEE multiplies its three factors, a modest gain in each compounds. Lift availability, performance and quality by a few points apiece and the combined effect on good output is larger than any single factor suggests — often equivalent to capacity you would otherwise pay for in new equipment or extra shifts. That is why downtime reduction and scrap reduction are not separate projects; they are two routes to the same throughput, costed differently.
We connect these signals into the systems your team already lives in — MES, ERP and your maintenance flow — so a prediction becomes a work order and a vision reject becomes a logged, traceable event. Wiring that integration cleanly is core data engineering work, and it is what separates a dashboard nobody opens from a system that changes behaviour on the floor.
Crux Digits is a boutique, senior-led AI consultancy in Nieuwegein, in the province of Utrecht, serving the Utrecht region and manufacturers across the Netherlands and Europe. We are a deliberate alternative to two things the maakindustrie knows well: the large enterprise consultancies whose day rates and ramp-up rarely fit a mid-sized plant, and the web agencies that have recently rebranded around AI without ever shipping a model into production. We are the engineering partner — not a marketing shop — and on our projects senior people stay on the work from audit to deployment.
Compliance is not an afterthought bolted on at the end. We design with the EU AI Act and GDPR/AVG in view from day one, which matters more in manufacturing than people expect. Vision systems on a line capture imagery; quality and maintenance models touch operational and sometimes personal data; and as the AI Act phases in, knowing how each system is classified and documented protects you from rework and exposure later. Building that in early is far cheaper than retrofitting it.
Across 13 delivered case studies — spanning computer vision, predictive maintenance, forecasting, NLP and more — the pattern that earns trust with Dutch SMEs is the same: prove value on a real use case quickly, keep the scope honest, and hand over a solution the client genuinely owns. Pricing is transparent and fixed-step, excluding VAT: an AI Audit and Strategy at EUR 2,500 to map the costliest losses and the data you already have, a Proof of Concept at EUR 20,000 to prove the model on your line, and a production launch from EUR 50,000 to deploy and integrate it. You always know the next step and its cost before you commit.
The fastest route to value is rarely the flashiest use case — it is the loss with the biggest, clearest number against it. If unplanned downtime dominates your OEE report, start there. If quality complaints or scrap are the recurring pain, start with vision inspection. If demand swings leave you short or overstocked, forecasting earns its place first. The audit exists precisely to make that call on evidence rather than enthusiasm.
A typical engagement moves from an AI audit and strategy session — where we map your machines, your data and the costliest problems — to a focused proof of concept on a single line, and only then to a production deployment wired into MES, ERP and maintenance. You see a working prototype on your own use case early, not a specification document, so the decision to scale rests on something you can watch run.
If you have a downtime, scrap or planning problem with a real cost attached, that is the conversation worth having. Tell us where the line hurts and we will map a concrete path to value — uptime regained, defect rate lowered, OEE moved — in a free consultation. Reach Tom Joseph and the team at info@cruxdigits.nl or +31 6 44384676, and see how we work on our about page.
In the Brainport region? See AI for manufacturers in Eindhoven for the high-tech and maakindustrie cluster.
Usually, yes. We design around your current equipment, sensors and cameras wherever possible, rather than asking you to rip and replace.
Not necessarily. For speed and reliability on the line, we can run models on the edge or on-premise, close to the machines.
Yes — we connect predictions and alerts into your MES, ERP and maintenance systems so they drive real action, not just dashboards.
We'll tell you early whether what you log today is enough — and if not, the simplest way to capture what the model needs.
Yes. Crux Digits is a boutique applied-AI firm based in the Utrecht region, working with manufacturers across the Netherlands and the EU, on-site or remotely as needed. We have built vision-based quality inspection, predictive maintenance and production-planning tools for shop-floor use. Senior engineers do the work directly, and you own the source code, so you are never locked in afterwards.
We start with a short paid discovery to scope one real use case, then deliver at a fixed price. We design for the EU AI Act from day one, keeping quality and maintenance models transparent and human-checked rather than fully automated. Where AI touches machine control we respect your CE and machinery-safety obligations, and any worker or personnel data stays GDPR-compliant.
Learn the signatures of failure and flag them early — so maintenance happens on schedule, not in a crisis.
Catch cracks, spalling, chipping and voids on precast segments automatically, right on the production line.
Spot the prized two-leaves-and-a-bud growth, catch pests and disease early, and put crews where they matter most — across 3,200 ha.
Tell us where the line hurts — failures, scrap or planning — and we'll map a path to value in a free consultation.
Book a free consultation →