Scrap-Slag Classification AI for Steel Mills

USE CASEANWENDUNGSFALLUSE CASEUSE CASEUSE CASEUSE CASE · Scrap-slag classification · LiDAR + camera fusionSchrott-Schlacke-Klassifikation · Fusion aus LiDAR und KameraScrap-slag classification · LiDAR + camera fusionScrap-slag classification · LiDAR + camera fusionScrap-slag classification · LiDAR + camera fusionScrap-slag classification · LiDAR + camera fusion

Scrap-vs-slag, sorted at the portal, before it hits the charge bucket. Schrott und Schlacke am Portal trennen, bevor sie im Chargenkorb landen. Scrap-vs-slag, sorted at the portal, before it hits the charge bucket. Scrap-vs-slag, sorted at the portal, before it hits the charge bucket. Scrap-vs-slag, sorted at the portal, before it hits the charge bucket. Scrap-vs-slag, sorted at the portal, before it hits the charge bucket.

Scrap-slag classification AI is a portal-mounted perception pipeline that distinguishes ferrous scrap from cooled slag in real time on a steel-mill yard, using fused LiDAR and camera data to feed the EAF charge-mix planner.Schrott-Klassifikation KI ist eine portal-montierte Wahrnehmungs-Pipeline, die Eisenschrott in Echtzeit auf dem Stahlwerks-Hof von erkalteter Schlacke trennt — mittels Fusion von LiDAR- und Kameradaten — und das Ergebnis direkt in die EAF-Chargenmischung speist.Scrap-slag classification AI is a portal-mounted perception pipeline that distinguishes ferrous scrap from cooled slag in real time on a steel-mill yard, using fused LiDAR and camera data to feed the EAF charge-mix planner.Scrap-slag classification AI is a portal-mounted perception pipeline that distinguishes ferrous scrap from cooled slag in real time on a steel-mill yard, using fused LiDAR and camera data to feed the EAF charge-mix planner.Scrap-slag classification AI is a portal-mounted perception pipeline that distinguishes ferrous scrap from cooled slag in real time on a steel-mill yard, using fused LiDAR and camera data to feed the EAF charge-mix planner.Scrap-slag classification AI is a portal-mounted perception pipeline that distinguishes ferrous scrap from cooled slag in real time on a steel-mill yard, using fused LiDAR and camera data to feed the EAF charge-mix planner.

The pain is well-known to every EAF operator: scrap deliveries arrive contaminated with cooled slag, refractory debris, oxidised fines and the occasional non-ferrous lump. When this mix gets graded the wrong way, the charge bucket carries the wrong mass of actual iron — the heat ends up under- or over-charged, the energy balance drifts, fluxes have to be adjusted on the fly, and in the worst case the tap goes off-spec. The cost shows up as wasted electricity per tonne, extra refractory wear, and slag pits that fill faster than the schedule predicted.Den Schmerz kennt jeder EAF-Betreiber: Schrott-Anlieferungen kommen kontaminiert mit erkalteter Schlacke, Feuerfest-Bruch, oxidierten Feinanteilen und gelegentlich nicht-eisenhaltigen Stücken. Wenn diese Mischung falsch eingestuft wird, trägt der Chargenkorb die falsche Eisenmasse — das Heat wird unter- oder überchargiert, die Energiebilanz driftet, Schlackebildner müssen während des Schmelzens nachjustiert werden, und im schlimmsten Fall geht der Abstich Off-Spec. Die Kosten zeigen sich als Mehr-Strom pro Tonne, zusätzlicher Feuerfest-Verschleiß und Schlackegruben, die schneller voll werden als geplant.The pain is well-known to every EAF operator: scrap deliveries arrive contaminated with cooled slag, refractory debris, oxidised fines and the occasional non-ferrous lump. When this mix gets graded the wrong way, the charge bucket carries the wrong mass of actual iron — the heat ends up under- or over-charged, the energy balance drifts, fluxes have to be adjusted on the fly, and in the worst case the tap goes off-spec. The cost shows up as wasted electricity per tonne, extra refractory wear, and slag pits that fill faster than the schedule predicted.The pain is well-known to every EAF operator: scrap deliveries arrive contaminated with cooled slag, refractory debris, oxidised fines and the occasional non-ferrous lump. When this mix gets graded the wrong way, the charge bucket carries the wrong mass of actual iron — the heat ends up under- or over-charged, the energy balance drifts, fluxes have to be adjusted on the fly, and in the worst case the tap goes off-spec. The cost shows up as wasted electricity per tonne, extra refractory wear, and slag pits that fill faster than the schedule predicted.The pain is well-known to every EAF operator: scrap deliveries arrive contaminated with cooled slag, refractory debris, oxidised fines and the occasional non-ferrous lump. When this mix gets graded the wrong way, the charge bucket carries the wrong mass of actual iron — the heat ends up under- or over-charged, the energy balance drifts, fluxes have to be adjusted on the fly, and in the worst case the tap goes off-spec. The cost shows up as wasted electricity per tonne, extra refractory wear, and slag pits that fill faster than the schedule predicted.The pain is well-known to every EAF operator: scrap deliveries arrive contaminated with cooled slag, refractory debris, oxidised fines and the occasional non-ferrous lump. When this mix gets graded the wrong way, the charge bucket carries the wrong mass of actual iron — the heat ends up under- or over-charged, the energy balance drifts, fluxes have to be adjusted on the fly, and in the worst case the tap goes off-spec. The cost shows up as wasted electricity per tonne, extra refractory wear, and slag pits that fill faster than the schedule predicted.

Our approach fuses two sensors at the yard entrance. A 3D-LiDAR scans the load geometry as the truck rolls under a portal — bulk density signature, surface texture, void structure. A high-resolution colour camera adds material colour, oxidation tint, and visible refractory or slag signatures. A custom classifier — trained on your scrap categories (shred, bushelling, plate-and-structural, heavy melt, turnings) and your slag types — returns a per-truck breakdown: tonnage estimate per category, slag fraction, contamination flags. The result writes directly into the charge-planner or MES over OPC UA, REST or MQTT.Unser Ansatz fusioniert zwei Sensoren am Hof-Eingang. Ein 3D-LiDAR scannt die Ladegeometrie, während der LKW durchs Portal rollt — Schüttdichte-Signatur, Oberflächentextur, Hohlraum-Struktur. Eine hochauflösende Farbkamera ergänzt Materialfarbe, Oxidations-Ton und sichtbare Schlacke- oder Feuerfest-Signaturen. Ein eigener Klassifikator — trainiert auf Ihren Schrott-Kategorien (Schredder, Bushelling, Schwerschrott, Späne, Plattenschrott) und Ihren Schlacke-Typen — liefert pro LKW eine Aufschlüsselung: geschätzte Tonnage je Kategorie, Schlacke-Anteil, Kontaminations-Flags. Das Ergebnis schreibt direkt in die Chargenplanung oder ins MES — OPC UA, REST oder MQTT.Our approach fuses two sensors at the yard entrance. A 3D-LiDAR scans the load geometry as the truck rolls under a portal — bulk density signature, surface texture, void structure. A high-resolution colour camera adds material colour, oxidation tint, and visible refractory or slag signatures. A custom classifier — trained on your scrap categories (shred, bushelling, plate-and-structural, heavy melt, turnings) and your slag types — returns a per-truck breakdown: tonnage estimate per category, slag fraction, contamination flags. The result writes directly into the charge-planner or MES over OPC UA, REST or MQTT.Our approach fuses two sensors at the yard entrance. A 3D-LiDAR scans the load geometry as the truck rolls under a portal — bulk density signature, surface texture, void structure. A high-resolution colour camera adds material colour, oxidation tint, and visible refractory or slag signatures. A custom classifier — trained on your scrap categories (shred, bushelling, plate-and-structural, heavy melt, turnings) and your slag types — returns a per-truck breakdown: tonnage estimate per category, slag fraction, contamination flags. The result writes directly into the charge-planner or MES over OPC UA, REST or MQTT.Our approach fuses two sensors at the yard entrance. A 3D-LiDAR scans the load geometry as the truck rolls under a portal — bulk density signature, surface texture, void structure. A high-resolution colour camera adds material colour, oxidation tint, and visible refractory or slag signatures. A custom classifier — trained on your scrap categories (shred, bushelling, plate-and-structural, heavy melt, turnings) and your slag types — returns a per-truck breakdown: tonnage estimate per category, slag fraction, contamination flags. The result writes directly into the charge-planner or MES over OPC UA, REST or MQTT.Our approach fuses two sensors at the yard entrance. A 3D-LiDAR scans the load geometry as the truck rolls under a portal — bulk density signature, surface texture, void structure. A high-resolution colour camera adds material colour, oxidation tint, and visible refractory or slag signatures. A custom classifier — trained on your scrap categories (shred, bushelling, plate-and-structural, heavy melt, turnings) and your slag types — returns a per-truck breakdown: tonnage estimate per category, slag fraction, contamination flags. The result writes directly into the charge-planner or MES over OPC UA, REST or MQTT.

This is not an off-the-shelf SaaS. Every steel mill has its own scrap sources, its own slag pit, its own EAF charge model. We treat every engagement as a discovery + fixed-scope build, descending from our hub service Industrial Perception AI. Typically four to twelve weeks from contract to live tool, with the code and the model handed over to you at the end.Das ist kein SaaS von der Stange. Jedes Stahlwerk hat seine eigenen Schrott-Quellen, seine eigene Schlackegrube, sein eigenes EAF-Chargenmodell. Wir behandeln jedes Projekt als Discovery + Festscope-Build, abgeleitet aus unserem Hub-Service Industrielle Objekterkennung. Typisch vier bis zwölf Wochen vom Vertrag bis zum produktiven Tool — Code und Modell gehören Ihnen.This is not an off-the-shelf SaaS. Every steel mill has its own scrap sources, its own slag pit, its own EAF charge model. We treat every engagement as a discovery + fixed-scope build, descending from our hub service Industrial Perception AI. Typically four to twelve weeks from contract to live tool, with the code and the model handed over to you at the end.This is not an off-the-shelf SaaS. Every steel mill has its own scrap sources, its own slag pit, its own EAF charge model. We treat every engagement as a discovery + fixed-scope build, descending from our hub service Industrial Perception AI. Typically four to twelve weeks from contract to live tool, with the code and the model handed over to you at the end.This is not an off-the-shelf SaaS. Every steel mill has its own scrap sources, its own slag pit, its own EAF charge model. We treat every engagement as a discovery + fixed-scope build, descending from our hub service Industrial Perception AI. Typically four to twelve weeks from contract to live tool, with the code and the model handed over to you at the end.This is not an off-the-shelf SaaS. Every steel mill has its own scrap sources, its own slag pit, its own EAF charge model. We treat every engagement as a discovery + fixed-scope build, descending from our hub service Industrial Perception AI. Typically four to twelve weeks from contract to live tool, with the code and the model handed over to you at the end.

„A bad charge mix is the most expensive way to learn what was actually in the truck." „Eine falsche Chargenmischung ist der teuerste Weg, hinterher herauszufinden, was wirklich im LKW war." „A bad charge mix is the most expensive way to learn what was actually in the truck." „A bad charge mix is the most expensive way to learn what was actually in the truck." „A bad charge mix is the most expensive way to learn what was actually in the truck." „A bad charge mix is the most expensive way to learn what was actually in the truck."

Built by the same team that delivers our LiDAR wagon-detection pipeline — six lid centroids, sub-5 cm error, two-second response. Gebaut von dem Team, das auch unsere LiDAR-Waggon-Erkennung liefert — sechs Deckel-Mittelpunkte, Fehler unter 5 cm, Antwort in zwei Sekunden. Built by the same team that delivers our LiDAR wagon-detection pipeline — six lid centroids, sub-5 cm error, two-second response. Built by the same team that delivers our LiDAR wagon-detection pipeline — six lid centroids, sub-5 cm error, two-second response. Built by the same team that delivers our LiDAR wagon-detection pipeline — six lid centroids, sub-5 cm error, two-second response. Built by the same team that delivers our LiDAR wagon-detection pipeline — six lid centroids, sub-5 cm error, two-second response.

Three stagesDrei StufenThree stagesThree stagesThree stagesThree stages

From truck to charge bucket — three stages of the pipeline. Vom LKW bis zum Chargenkorb — drei Stufen der Pipeline. From truck to charge bucket — three stages of the pipeline. From truck to charge bucket — three stages of the pipeline. From truck to charge bucket — three stages of the pipeline. From truck to charge bucket — three stages of the pipeline.

1 · Sensor portal at the yard entrance1 · Sensor-Portal am Hof-Eingang1 · Sensor portal at the yard entrance1 · Sensor portal at the yard entrance1 · Sensor portal at the yard entrance1 · Sensor portal at the yard entrance

One LiDAR plus one calibrated colour camera mounted on a steel portal at the scrap-yard entrance or weighbridge. Trucks drive through at normal speed — the system captures load geometry and surface imagery on the fly, no stopping required. IP65+ housings, dust-tolerant optics, runs in fog, snow and night light. Ein LiDAR plus eine kalibrierte Farbkamera, montiert auf einem Stahl-Portal am Schrottplatz-Eingang oder über der Waage. LKW fahren mit normaler Geschwindigkeit durch — das System erfasst Ladegeometrie und Oberflächen-Bild im Vorbeifahren, ohne Anhalten. IP65+-Gehäuse, staubtolerante Optik, läuft in Nebel, Schnee und Nacht. One LiDAR plus one calibrated colour camera mounted on a steel portal at the scrap-yard entrance or weighbridge. Trucks drive through at normal speed — the system captures load geometry and surface imagery on the fly, no stopping required. IP65+ housings, dust-tolerant optics, runs in fog, snow and night light. One LiDAR plus one calibrated colour camera mounted on a steel portal at the scrap-yard entrance or weighbridge. Trucks drive through at normal speed — the system captures load geometry and surface imagery on the fly, no stopping required. IP65+ housings, dust-tolerant optics, runs in fog, snow and night light. One LiDAR plus one calibrated colour camera mounted on a steel portal at the scrap-yard entrance or weighbridge. Trucks drive through at normal speed — the system captures load geometry and surface imagery on the fly, no stopping required. IP65+ housings, dust-tolerant optics, runs in fog, snow and night light. One LiDAR plus one calibrated colour camera mounted on a steel portal at the scrap-yard entrance or weighbridge. Trucks drive through at normal speed — the system captures load geometry and surface imagery on the fly, no stopping required. IP65+ housings, dust-tolerant optics, runs in fog, snow and night light.

2 · Classification + segmentation2 · Klassifikation + Segmentierung2 · Classification + segmentation2 · Classification + segmentation2 · Classification + segmentation2 · Classification + segmentation

The fusion classifier separates the load into your scrap categories — shred, bushelling, plate-and-structural, heavy melt, turnings — plus cooled slag, refractory debris and unknown contamination. Per-truck tonnage estimate per category, segmentation mask for visual audit. Models retrained on your labelled data, not on a public benchmark. Der Fusions-Klassifikator trennt die Ladung in Ihre Schrott-Kategorien — Schredder, Bushelling, Schwerschrott, Späne, Plattenschrott — plus erkaltete Schlacke, Feuerfest-Bruch und unbekannte Kontamination. Pro LKW geschätzte Tonnage je Kategorie, Segmentierungs-Maske für visuelle Prüfung. Modelle trainiert auf Ihren gelabelten Daten — nicht auf einem öffentlichen Benchmark. The fusion classifier separates the load into your scrap categories — shred, bushelling, plate-and-structural, heavy melt, turnings — plus cooled slag, refractory debris and unknown contamination. Per-truck tonnage estimate per category, segmentation mask for visual audit. Models retrained on your labelled data, not on a public benchmark. The fusion classifier separates the load into your scrap categories — shred, bushelling, plate-and-structural, heavy melt, turnings — plus cooled slag, refractory debris and unknown contamination. Per-truck tonnage estimate per category, segmentation mask for visual audit. Models retrained on your labelled data, not on a public benchmark. The fusion classifier separates the load into your scrap categories — shred, bushelling, plate-and-structural, heavy melt, turnings — plus cooled slag, refractory debris and unknown contamination. Per-truck tonnage estimate per category, segmentation mask for visual audit. Models retrained on your labelled data, not on a public benchmark. The fusion classifier separates the load into your scrap categories — shred, bushelling, plate-and-structural, heavy melt, turnings — plus cooled slag, refractory debris and unknown contamination. Per-truck tonnage estimate per category, segmentation mask for visual audit. Models retrained on your labelled data, not on a public benchmark.

3 · Charge planner + MES integration3 · Chargenplanung + MES-Anbindung3 · Charge planner + MES integration3 · Charge planner + MES integration3 · Charge planner + MES integration3 · Charge planner + MES integration

Classification results write into the charge-planner or MES over OPC UA, REST or MQTT — same shift, before the bucket is filled. The charge model sees real category masses instead of operator estimates. Off-spec deliveries are flagged before they enter the mix; aging slag-contaminated trucks get routed back to the supplier with the evidence attached. Klassifikations-Ergebnisse schreiben in die Chargenplanung oder ins MES per OPC UA, REST oder MQTT — gleiche Schicht, bevor der Korb gefüllt wird. Das Chargenmodell sieht echte Kategorie-Massen statt Bediener-Schätzungen. Off-Spec-Anlieferungen werden vor der Mischung markiert; schlackehaltige LKW gehen mit Beweis-Anhang an den Lieferanten zurück. Classification results write into the charge-planner or MES over OPC UA, REST or MQTT — same shift, before the bucket is filled. The charge model sees real category masses instead of operator estimates. Off-spec deliveries are flagged before they enter the mix; aging slag-contaminated trucks get routed back to the supplier with the evidence attached. Classification results write into the charge-planner or MES over OPC UA, REST or MQTT — same shift, before the bucket is filled. The charge model sees real category masses instead of operator estimates. Off-spec deliveries are flagged before they enter the mix; aging slag-contaminated trucks get routed back to the supplier with the evidence attached. Classification results write into the charge-planner or MES over OPC UA, REST or MQTT — same shift, before the bucket is filled. The charge model sees real category masses instead of operator estimates. Off-spec deliveries are flagged before they enter the mix; aging slag-contaminated trucks get routed back to the supplier with the evidence attached. Classification results write into the charge-planner or MES over OPC UA, REST or MQTT — same shift, before the bucket is filled. The charge model sees real category masses instead of operator estimates. Off-spec deliveries are flagged before they enter the mix; aging slag-contaminated trucks get routed back to the supplier with the evidence attached.

Pipeline architecture · #how-it-worksPipeline-Architektur · #how-it-worksPipeline architecture · #how-it-worksPipeline architecture · #how-it-worksPipeline architecture · #how-it-worksPipeline architecture · #how-it-works

How the pipeline works — concretely. Wie die Pipeline arbeitet — konkret. How the pipeline works — concretely. How the pipeline works — concretely. How the pipeline works — concretely. How the pipeline works — concretely.

We keep the architecture boring on purpose. Three loosely coupled stages, each one independently testable, each one swappable when the sensor stack changes. Built on the same stack we use across all our perception work: PCL, Open3D, OpenCV, PyTorch. Wir halten die Architektur absichtlich langweilig. Drei lose gekoppelte Stufen, jede einzeln testbar, jede einzeln tauschbar, wenn sich der Sensor-Stack ändert. Gebaut auf demselben Stack, den wir für unsere gesamte Wahrnehmungs-Arbeit nutzen: PCL, Open3D, OpenCV, PyTorch. We keep the architecture boring on purpose. Three loosely coupled stages, each one independently testable, each one swappable when the sensor stack changes. Built on the same stack we use across all our perception work: PCL, Open3D, OpenCV, PyTorch. We keep the architecture boring on purpose. Three loosely coupled stages, each one independently testable, each one swappable when the sensor stack changes. Built on the same stack we use across all our perception work: PCL, Open3D, OpenCV, PyTorch. We keep the architecture boring on purpose. Three loosely coupled stages, each one independently testable, each one swappable when the sensor stack changes. Built on the same stack we use across all our perception work: PCL, Open3D, OpenCV, PyTorch. We keep the architecture boring on purpose. Three loosely coupled stages, each one independently testable, each one swappable when the sensor stack changes. Built on the same stack we use across all our perception work: PCL, Open3D, OpenCV, PyTorch.

Portal scan capturePortal-Scan-AufnahmePortal scan capturePortal scan capturePortal scan capturePortal scan capture

The LiDAR delivers a 3D point cloud of the full truck load — typically a million-plus points per pass at portal speed. The camera adds a synchronised colour image, lens-calibrated and time-stamped. Raw data writes to local storage with a per-truck UUID for audit trail and re-labelling. Der LiDAR liefert eine 3D-Punktwolke der gesamten LKW-Ladung — typisch über eine Million Punkte pro Durchfahrt bei Portal-Geschwindigkeit. Die Kamera ergänzt ein synchronisiertes Farbbild, objektiv-kalibriert und zeit-gestempelt. Rohdaten schreiben auf lokalen Speicher mit pro-LKW-UUID für Audit-Trail und Re-Labelling. The LiDAR delivers a 3D point cloud of the full truck load — typically a million-plus points per pass at portal speed. The camera adds a synchronised colour image, lens-calibrated and time-stamped. Raw data writes to local storage with a per-truck UUID for audit trail and re-labelling. The LiDAR delivers a 3D point cloud of the full truck load — typically a million-plus points per pass at portal speed. The camera adds a synchronised colour image, lens-calibrated and time-stamped. Raw data writes to local storage with a per-truck UUID for audit trail and re-labelling. The LiDAR delivers a 3D point cloud of the full truck load — typically a million-plus points per pass at portal speed. The camera adds a synchronised colour image, lens-calibrated and time-stamped. Raw data writes to local storage with a per-truck UUID for audit trail and re-labelling. The LiDAR delivers a 3D point cloud of the full truck load — typically a million-plus points per pass at portal speed. The camera adds a synchronised colour image, lens-calibrated and time-stamped. Raw data writes to local storage with a per-truck UUID for audit trail and re-labelling.

Fusion classifierFusions-KlassifikatorFusion classifierFusion classifierFusion classifierFusion classifier

The point cloud goes through ground removal, voxel down-sampling and a PointNet-style classifier trained on your scrap categories. The colour image goes through a CNN trained on the same labels. The two heads are fused into a single per-voxel prediction — typically 85–95% classification accuracy on production data, depending on how clean your scrap categories actually are. Die Punktwolke läuft durch Boden-Entfernung, Voxel-Downsampling und einen PointNet-artigen Klassifikator, trainiert auf Ihren Schrott-Kategorien. Das Farbbild läuft durch ein CNN, trainiert auf denselben Labels. Beide Köpfe werden zu einer voxel-weisen Vorhersage fusioniert — typisch 85–95 % Klassifikations-Genauigkeit auf Produktionsdaten, abhängig davon, wie sauber Ihre Schrott-Kategorien tatsächlich abgegrenzt sind. The point cloud goes through ground removal, voxel down-sampling and a PointNet-style classifier trained on your scrap categories. The colour image goes through a CNN trained on the same labels. The two heads are fused into a single per-voxel prediction — typically 85–95% classification accuracy on production data, depending on how clean your scrap categories actually are. The point cloud goes through ground removal, voxel down-sampling and a PointNet-style classifier trained on your scrap categories. The colour image goes through a CNN trained on the same labels. The two heads are fused into a single per-voxel prediction — typically 85–95% classification accuracy on production data, depending on how clean your scrap categories actually are. The point cloud goes through ground removal, voxel down-sampling and a PointNet-style classifier trained on your scrap categories. The colour image goes through a CNN trained on the same labels. The two heads are fused into a single per-voxel prediction — typically 85–95% classification accuracy on production data, depending on how clean your scrap categories actually are. The point cloud goes through ground removal, voxel down-sampling and a PointNet-style classifier trained on your scrap categories. The colour image goes through a CNN trained on the same labels. The two heads are fused into a single per-voxel prediction — typically 85–95% classification accuracy on production data, depending on how clean your scrap categories actually are.

Charge-planner write-backRück-Schreibung in die ChargenplanungCharge-planner write-backCharge-planner write-backCharge-planner write-backCharge-planner write-back

Per-truck output: estimated tonnage per category, slag fraction, contamination flags, audit images. Writes into the EAF charge planner over OPC UA — or into the MES, the ERP, the yard-management system, whatever the plant already runs. The charge model now reads real category masses instead of operator estimates. Pro LKW: geschätzte Tonnage je Kategorie, Schlacke-Anteil, Kontaminations-Flags, Audit-Bilder. Schreibt in die EAF-Chargenplanung per OPC UA — oder ins MES, ERP, Hof-Management-System, was die Anlage eben hat. Das Chargenmodell liest jetzt echte Kategorie-Massen statt Bediener-Schätzungen. Per-truck output: estimated tonnage per category, slag fraction, contamination flags, audit images. Writes into the EAF charge planner over OPC UA — or into the MES, the ERP, the yard-management system, whatever the plant already runs. The charge model now reads real category masses instead of operator estimates. Per-truck output: estimated tonnage per category, slag fraction, contamination flags, audit images. Writes into the EAF charge planner over OPC UA — or into the MES, the ERP, the yard-management system, whatever the plant already runs. The charge model now reads real category masses instead of operator estimates. Per-truck output: estimated tonnage per category, slag fraction, contamination flags, audit images. Writes into the EAF charge planner over OPC UA — or into the MES, the ERP, the yard-management system, whatever the plant already runs. The charge model now reads real category masses instead of operator estimates. Per-truck output: estimated tonnage per category, slag fraction, contamination flags, audit images. Writes into the EAF charge planner over OPC UA — or into the MES, the ERP, the yard-management system, whatever the plant already runs. The charge model now reads real category masses instead of operator estimates.

All three stages run on an industrial PC at the portal cabinet. No cloud dependency, no external API, no licence dial-home. The code is yours at handover. Alle drei Stufen laufen auf einem Industrie-PC im Portal-Schaltschrank. Keine Cloud-Abhängigkeit, keine externe API, kein Lizenz-Heimruf. Der Code gehört Ihnen bei der Übergabe. All three stages run on an industrial PC at the portal cabinet. No cloud dependency, no external API, no licence dial-home. The code is yours at handover. All three stages run on an industrial PC at the portal cabinet. No cloud dependency, no external API, no licence dial-home. The code is yours at handover. All three stages run on an industrial PC at the portal cabinet. No cloud dependency, no external API, no licence dial-home. The code is yours at handover. All three stages run on an industrial PC at the portal cabinet. No cloud dependency, no external API, no licence dial-home. The code is yours at handover.

What you getWas Sie bekommenWhat you getWhat you getWhat you getWhat you get

Three deliverables — from the same pipeline. Drei Liefer-Ergebnisse — aus derselben Pipeline. Three deliverables — from the same pipeline. Three deliverables — from the same pipeline. Three deliverables — from the same pipeline. Three deliverables — from the same pipeline.

Per-truck classification reportKlassifikations-Report pro LKWPer-truck classification reportPer-truck classification reportPer-truck classification reportPer-truck classification report

Every delivery gets a structured record: scrap-category mass breakdown, slag fraction, contamination flags, audit images, supplier ID, weighbridge ticket cross-link. Drops into the supplier-quality dashboard, into ERP, and into the yard-acceptance workflow. Jede Anlieferung erhält einen strukturierten Datensatz: Massen-Aufschlüsselung je Schrott-Kategorie, Schlacke-Anteil, Kontaminations-Flags, Audit-Bilder, Lieferanten-ID, Querverweis auf den Waage-Ticket. Geht ins Lieferanten-Qualitäts-Dashboard, ins ERP und in den Hof-Annahme-Workflow. Every delivery gets a structured record: scrap-category mass breakdown, slag fraction, contamination flags, audit images, supplier ID, weighbridge ticket cross-link. Drops into the supplier-quality dashboard, into ERP, and into the yard-acceptance workflow. Every delivery gets a structured record: scrap-category mass breakdown, slag fraction, contamination flags, audit images, supplier ID, weighbridge ticket cross-link. Drops into the supplier-quality dashboard, into ERP, and into the yard-acceptance workflow. Every delivery gets a structured record: scrap-category mass breakdown, slag fraction, contamination flags, audit images, supplier ID, weighbridge ticket cross-link. Drops into the supplier-quality dashboard, into ERP, and into the yard-acceptance workflow. Every delivery gets a structured record: scrap-category mass breakdown, slag fraction, contamination flags, audit images, supplier ID, weighbridge ticket cross-link. Drops into the supplier-quality dashboard, into ERP, and into the yard-acceptance workflow.

Yard-level inventory heatmapHof-Inventur als HeatmapYard-level inventory heatmapYard-level inventory heatmapYard-level inventory heatmapYard-level inventory heatmap

Across many trucks, the yard accumulates an aging map: which bay holds which scrap mix, how long it has sat, how much slag contamination is sitting in each pile. Same temporal model we use on stockpile monitoring — applied to scrap categories. Über viele LKW hinweg entsteht auf dem Hof eine Alterungs-Karte: welche Box welchen Schrott-Mix hält, wie lange er liegt, wie viel Schlacke-Kontamination in welcher Halde sitzt. Dasselbe zeitliche Modell, das wir auf Halden-Monitoring einsetzen — übertragen auf Schrott-Kategorien. Across many trucks, the yard accumulates an aging map: which bay holds which scrap mix, how long it has sat, how much slag contamination is sitting in each pile. Same temporal model we use on stockpile monitoring — applied to scrap categories. Across many trucks, the yard accumulates an aging map: which bay holds which scrap mix, how long it has sat, how much slag contamination is sitting in each pile. Same temporal model we use on stockpile monitoring — applied to scrap categories. Across many trucks, the yard accumulates an aging map: which bay holds which scrap mix, how long it has sat, how much slag contamination is sitting in each pile. Same temporal model we use on stockpile monitoring — applied to scrap categories. Across many trucks, the yard accumulates an aging map: which bay holds which scrap mix, how long it has sat, how much slag contamination is sitting in each pile. Same temporal model we use on stockpile monitoring — applied to scrap categories.

Charge-mix recommendation per heatChargen-Mix-Vorschlag pro HeatCharge-mix recommendation per heatCharge-mix recommendation per heatCharge-mix recommendation per heatCharge-mix recommendation per heat

For every planned heat, the system proposes a charge mix that hits the target chemistry within typical fines and slag tolerances — drawing from real yard inventory instead of assumed averages. The metallurgist still approves; the planner does the bookkeeping. Für jedes geplante Heat schlägt das System eine Mischung vor, die die Ziel-Chemie innerhalb typischer Feinanteil- und Schlacke-Toleranzen trifft — gezogen aus dem realen Hof-Bestand statt aus angenommenen Mittelwerten. Der Metallurg gibt frei; der Planer macht die Buchhaltung. For every planned heat, the system proposes a charge mix that hits the target chemistry within typical fines and slag tolerances — drawing from real yard inventory instead of assumed averages. The metallurgist still approves; the planner does the bookkeeping. For every planned heat, the system proposes a charge mix that hits the target chemistry within typical fines and slag tolerances — drawing from real yard inventory instead of assumed averages. The metallurgist still approves; the planner does the bookkeeping. For every planned heat, the system proposes a charge mix that hits the target chemistry within typical fines and slag tolerances — drawing from real yard inventory instead of assumed averages. The metallurgist still approves; the planner does the bookkeeping. For every planned heat, the system proposes a charge mix that hits the target chemistry within typical fines and slag tolerances — drawing from real yard inventory instead of assumed averages. The metallurgist still approves; the planner does the bookkeeping.

Why customWarum CustomWhy customWhy customWhy customWhy custom

Why a custom build — not an off-the-shelf product. Warum nach Maß — und nicht von der Stange. Why a custom build — not an off-the-shelf product. Why a custom build — not an off-the-shelf product. Why a custom build — not an off-the-shelf product. Why a custom build — not an off-the-shelf product.

Every steel mill has its own scrap sources, its own EAF charge model and its own existing yard sensors. A generic classifier solves the generic case; your case is not the generic case. Jedes Stahlwerk hat eigene Schrott-Quellen, ein eigenes EAF-Chargenmodell und eigene bestehende Hof-Sensorik. Ein generischer Klassifikator löst den generischen Fall; Ihr Fall ist nicht der generische Fall. Every steel mill has its own scrap sources, its own EAF charge model and its own existing yard sensors. A generic classifier solves the generic case; your case is not the generic case. Every steel mill has its own scrap sources, its own EAF charge model and its own existing yard sensors. A generic classifier solves the generic case; your case is not the generic case. Every steel mill has its own scrap sources, its own EAF charge model and its own existing yard sensors. A generic classifier solves the generic case; your case is not the generic case. Every steel mill has its own scrap sources, its own EAF charge model and its own existing yard sensors. A generic classifier solves the generic case; your case is not the generic case.

Material-specific calibrationMaterial-spezifische KalibrierungMaterial-specific calibrationMaterial-specific calibrationMaterial-specific calibrationMaterial-specific calibration

A shred pile in one mill looks different from a shred pile in another — different shredder, different feedstock, different oxidation history. We calibrate the model on your material under your lighting and dust conditions, and we re-calibrate when the supplier mix changes. Eine Schredder-Halde sieht in einem Werk anders aus als im nächsten — anderer Schredder, anderes Vormaterial, andere Oxidations-Geschichte. Wir kalibrieren das Modell auf Ihrem Material unter Ihrem Licht und Staub — und wir re-kalibrieren, wenn sich der Lieferanten-Mix ändert. A shred pile in one mill looks different from a shred pile in another — different shredder, different feedstock, different oxidation history. We calibrate the model on your material under your lighting and dust conditions, and we re-calibrate when the supplier mix changes. A shred pile in one mill looks different from a shred pile in another — different shredder, different feedstock, different oxidation history. We calibrate the model on your material under your lighting and dust conditions, and we re-calibrate when the supplier mix changes. A shred pile in one mill looks different from a shred pile in another — different shredder, different feedstock, different oxidation history. We calibrate the model on your material under your lighting and dust conditions, and we re-calibrate when the supplier mix changes. A shred pile in one mill looks different from a shred pile in another — different shredder, different feedstock, different oxidation history. We calibrate the model on your material under your lighting and dust conditions, and we re-calibrate when the supplier mix changes.

Plant-specific labelling + training dataWerks-spezifisches Labelling + TrainingsdatenPlant-specific labelling + training dataPlant-specific labelling + training dataPlant-specific labelling + training dataPlant-specific labelling + training data

We bring the labelling tooling and the engineering hours. You bring the operator who actually knows what's in each truck. The result is a labelled dataset that belongs to you and that you can keep using as the model evolves — no vendor lock-in on the training data. Wir bringen das Labelling-Tooling und die Engineering-Stunden. Sie stellen den Bediener, der tatsächlich weiß, was in jedem LKW war. Das Ergebnis ist ein gelabelter Datensatz, der Ihnen gehört und den Sie weiterverwenden, wenn sich das Modell weiterentwickelt — kein Vendor-Lock-in auf Trainingsdaten. We bring the labelling tooling and the engineering hours. You bring the operator who actually knows what's in each truck. The result is a labelled dataset that belongs to you and that you can keep using as the model evolves — no vendor lock-in on the training data. We bring the labelling tooling and the engineering hours. You bring the operator who actually knows what's in each truck. The result is a labelled dataset that belongs to you and that you can keep using as the model evolves — no vendor lock-in on the training data. We bring the labelling tooling and the engineering hours. You bring the operator who actually knows what's in each truck. The result is a labelled dataset that belongs to you and that you can keep using as the model evolves — no vendor lock-in on the training data. We bring the labelling tooling and the engineering hours. You bring the operator who actually knows what's in each truck. The result is a labelled dataset that belongs to you and that you can keep using as the model evolves — no vendor lock-in on the training data.

Integration with existing yard sensors + cranesAnbindung an bestehende Hof-Sensorik + KraneIntegration with existing yard sensors + cranesIntegration with existing yard sensors + cranesIntegration with existing yard sensors + cranesIntegration with existing yard sensors + cranes

Most steel-mill yards already have weighbridges, RFID gates, ANPR cameras and overhead cranes. We fuse the classification output into what's already there — we do not replace your yard-management system. OPC UA, REST, MQTT and analog outputs all supported. Die meisten Stahlwerks-Höfe haben bereits Waagen, RFID-Tore, ANPR-Kameras und Hallenkrane. Wir fusionieren die Klassifikation in das, was schon da ist — wir ersetzen Ihr Hof-Management nicht. OPC UA, REST, MQTT und Analog-Ausgänge — alles unterstützt. Most steel-mill yards already have weighbridges, RFID gates, ANPR cameras and overhead cranes. We fuse the classification output into what's already there — we do not replace your yard-management system. OPC UA, REST, MQTT and analog outputs all supported. Most steel-mill yards already have weighbridges, RFID gates, ANPR cameras and overhead cranes. We fuse the classification output into what's already there — we do not replace your yard-management system. OPC UA, REST, MQTT and analog outputs all supported. Most steel-mill yards already have weighbridges, RFID gates, ANPR cameras and overhead cranes. We fuse the classification output into what's already there — we do not replace your yard-management system. OPC UA, REST, MQTT and analog outputs all supported. Most steel-mill yards already have weighbridges, RFID gates, ANPR cameras and overhead cranes. We fuse the classification output into what's already there — we do not replace your yard-management system. OPC UA, REST, MQTT and analog outputs all supported.

IP ownership + clean handoverIP-Eigentum + saubere ÜbergabeIP ownership + clean handoverIP ownership + clean handoverIP ownership + clean handoverIP ownership + clean handover

You own the source code, the model weights and the labelled dataset at handover. We document the system, train your team, and walk away clean. No black box, no monthly per-truck licence, no service contract you can't exit. See our FAQs on IP and engagement model for the standard terms. Sie besitzen Quellcode, Modell-Gewichte und gelabelten Datensatz nach der Übergabe. Wir dokumentieren das System, schulen Ihr Team und gehen sauber raus. Keine Black Box, keine monatliche Pro-LKW-Lizenz, kein Servicevertrag, aus dem Sie nicht rauskommen. Standard-Bedingungen siehe unsere FAQs zu IP und Zusammenarbeits-Modell. You own the source code, the model weights and the labelled dataset at handover. We document the system, train your team, and walk away clean. No black box, no monthly per-truck licence, no service contract you can't exit. See our FAQs on IP and engagement model for the standard terms. You own the source code, the model weights and the labelled dataset at handover. We document the system, train your team, and walk away clean. No black box, no monthly per-truck licence, no service contract you can't exit. See our FAQs on IP and engagement model for the standard terms. You own the source code, the model weights and the labelled dataset at handover. We document the system, train your team, and walk away clean. No black box, no monthly per-truck licence, no service contract you can't exit. See our FAQs on IP and engagement model for the standard terms. You own the source code, the model weights and the labelled dataset at handover. We document the system, train your team, and walk away clean. No black box, no monthly per-truck licence, no service contract you can't exit. See our FAQs on IP and engagement model for the standard terms.

FAQ

Questions about scrap-slag classification. Fragen zur Schrott-Schlacke-Klassifikation. Questions about scrap-slag classification. Questions about scrap-slag classification. Questions about scrap-slag classification. Questions about scrap-slag classification.

The engagement-model questions we hear from every steel-mill customer considering a custom perception build. Need something more specific to your yard? Ask us. Die Fragen zum Zusammenarbeits-Modell, die wir von jedem Stahlwerks-Kunden hören, der einen Custom-Perception-Build erwägt. Brauchen Sie etwas Spezifischeres zu Ihrem Hof? Sprechen Sie uns an. The engagement-model questions we hear from every steel-mill customer considering a custom perception build. Need something more specific to your yard? Ask us. The engagement-model questions we hear from every steel-mill customer considering a custom perception build. Need something more specific to your yard? Ask us. The engagement-model questions we hear from every steel-mill customer considering a custom perception build. Need something more specific to your yard? Ask us. The engagement-model questions we hear from every steel-mill customer considering a custom perception build. Need something more specific to your yard? Ask us.

Industrial Perception AIIndustrielle ObjekterkennungIndustrial Perception AIIndustrial Perception AIIndustrial Perception AIIndustrial Perception AI

What does custom perception AI software cost?Was kostet maßgeschneiderte Wahrnehmungs-Software?What does custom perception AI software cost?What does custom perception AI software cost?What does custom perception AI software cost?What does custom perception AI software cost? ＋

 We don't publish list prices because every project is scoped against your data and your decision logic — but we work to three predictable tiers.

Discovery & Assessment — a 1–3 day workshop, on-site or remote, fixed price in the low-four-figure range. You receive a written feasibility note, a recommended next step and (if applicable) a fixed-price quote for the follow-on project. Useful even if you don't then go ahead — many customers use the note to evaluate two or three vendors.

Custom Pipeline / Tool — a 4–12 week project delivering a working perception pipeline (parser + algorithm + dashboard or OPC UA integration). Fixed scope, fixed price, you own the code at handover. Typical project size is in the mid-five to low-six-figure range depending on data volume, sensor count and integration depth.

Long-term Partnership — a monthly retainer for ongoing development and on-call support. Sized to the team capacity you need (typically 0.5–2 FTE equivalent). Quarterly road-mapping included.

All three tiers include source code, written documentation and a clean handover — no licence dependence on us. See our Industrial Perception AI service for the full engagement model.
Wir veröffentlichen keine Listenpreise, weil jedes Projekt auf Ihre Daten und Ihre Entscheidungs-Logik zugeschnitten wird — aber wir arbeiten in drei berechenbaren Tiers.

Discovery & Assessment — ein 1–3-tägiger Workshop, vor Ort oder remote, Festpreis im unteren vierstelligen Bereich. Sie erhalten eine schriftliche Machbarkeits-Note, einen empfohlenen nächsten Schritt und ggf. ein Festpreis-Angebot für das Folge-Projekt. Lohnt sich auch ohne Folge-Auftrag — viele Kunden nutzen die Note, um zwei oder drei Anbieter zu vergleichen.

Custom Pipeline / Tool — ein 4–12-Wochen-Projekt, das eine fertige Wahrnehmungs-Pipeline liefert (Parser + Algorithmus + Dashboard oder OPC-UA-Anbindung). Fester Scope, Festpreis, Code-Übergabe am Ende. Übliche Projektgröße liegt im mittleren fünf- bis unteren sechsstelligen Bereich, je nach Datenvolumen, Sensor-Anzahl und Integrations-Tiefe.

Langfristige Partnerschaft — monatlicher Retainer für laufende Entwicklung und On-Call-Support. Dimensioniert nach benötigter Team-Kapazität (typisch 0,5–2 FTE-Äquivalente). Quartalsweise Roadmap inklusive.

Alle drei Tiers enthalten Quellcode, schriftliche Dokumentation und saubere Übergabe — keine Lizenz-Abhängigkeit von uns. Vollständiges Zusammenarbeits-Modell siehe unsere Industrielle Objekterkennung.
We don't publish list prices because every project is scoped against your data and your decision logic — but we work to three predictable tiers.

Discovery & Assessment — a 1–3 day workshop, on-site or remote, fixed price in the low-four-figure range. You receive a written feasibility note, a recommended next step and (if applicable) a fixed-price quote for the follow-on project. Useful even if you don't then go ahead — many customers use the note to evaluate two or three vendors.

Custom Pipeline / Tool — a 4–12 week project delivering a working perception pipeline (parser + algorithm + dashboard or OPC UA integration). Fixed scope, fixed price, you own the code at handover. Typical project size is in the mid-five to low-six-figure range depending on data volume, sensor count and integration depth.

Long-term Partnership — a monthly retainer for ongoing development and on-call support. Sized to the team capacity you need (typically 0.5–2 FTE equivalent). Quarterly road-mapping included.

All three tiers include source code, written documentation and a clean handover — no licence dependence on us. See our Industrial Perception AI service for the full engagement model.
We don't publish list prices because every project is scoped against your data and your decision logic — but we work to three predictable tiers.

Discovery & Assessment — a 1–3 day workshop, on-site or remote, fixed price in the low-four-figure range. You receive a written feasibility note, a recommended next step and (if applicable) a fixed-price quote for the follow-on project. Useful even if you don't then go ahead — many customers use the note to evaluate two or three vendors.

Custom Pipeline / Tool — a 4–12 week project delivering a working perception pipeline (parser + algorithm + dashboard or OPC UA integration). Fixed scope, fixed price, you own the code at handover. Typical project size is in the mid-five to low-six-figure range depending on data volume, sensor count and integration depth.

Long-term Partnership — a monthly retainer for ongoing development and on-call support. Sized to the team capacity you need (typically 0.5–2 FTE equivalent). Quarterly road-mapping included.

All three tiers include source code, written documentation and a clean handover — no licence dependence on us. See our Industrial Perception AI service for the full engagement model.
We don't publish list prices because every project is scoped against your data and your decision logic — but we work to three predictable tiers.

Discovery & Assessment — a 1–3 day workshop, on-site or remote, fixed price in the low-four-figure range. You receive a written feasibility note, a recommended next step and (if applicable) a fixed-price quote for the follow-on project. Useful even if you don't then go ahead — many customers use the note to evaluate two or three vendors.

Custom Pipeline / Tool — a 4–12 week project delivering a working perception pipeline (parser + algorithm + dashboard or OPC UA integration). Fixed scope, fixed price, you own the code at handover. Typical project size is in the mid-five to low-six-figure range depending on data volume, sensor count and integration depth.

Long-term Partnership — a monthly retainer for ongoing development and on-call support. Sized to the team capacity you need (typically 0.5–2 FTE equivalent). Quarterly road-mapping included.

All three tiers include source code, written documentation and a clean handover — no licence dependence on us. See our Industrial Perception AI service for the full engagement model.
We don't publish list prices because every project is scoped against your data and your decision logic — but we work to three predictable tiers.

Discovery & Assessment — a 1–3 day workshop, on-site or remote, fixed price in the low-four-figure range. You receive a written feasibility note, a recommended next step and (if applicable) a fixed-price quote for the follow-on project. Useful even if you don't then go ahead — many customers use the note to evaluate two or three vendors.

Custom Pipeline / Tool — a 4–12 week project delivering a working perception pipeline (parser + algorithm + dashboard or OPC UA integration). Fixed scope, fixed price, you own the code at handover. Typical project size is in the mid-five to low-six-figure range depending on data volume, sensor count and integration depth.

Long-term Partnership — a monthly retainer for ongoing development and on-call support. Sized to the team capacity you need (typically 0.5–2 FTE equivalent). Quarterly road-mapping included.

All three tiers include source code, written documentation and a clean handover — no licence dependence on us. See our Industrial Perception AI service for the full engagement model.

How long does it take to build a custom perception pipeline?Wie lange dauert eine maßgeschneiderte Wahrnehmungs-Pipeline?How long does it take to build a custom perception pipeline?How long does it take to build a custom perception pipeline?How long does it take to build a custom perception pipeline?How long does it take to build a custom perception pipeline? ＋

 From signed contract to a working tool in production, expect 4–12 weeks for a standard custom-pipeline engagement. The spread is driven by three factors.

Data readiness. If you already have labelled data or representative recordings, we start training in week 1. If we need to set up data collection, add a sensor, or label from scratch, the front end adds 1–3 weeks.

Integration depth. A standalone dashboard with REST output is the fastest path — typically 4–6 weeks. OPC UA integration into a running control system, with interlocks, factory-acceptance tests and a customer change-management process, is closer to 8–12 weeks.

Sensor count and site complexity. One LiDAR + one camera in a clean indoor environment is faster than a six-sensor outdoor portal with dust, snow and dynamic lighting. We scope the difference up-front in the Discovery workshop.

For a faster first signal we recommend the Discovery & Assessment tier — 1–3 days, fixed price, written feasibility note within a week. See Industrial Perception AI for the full model.
Vom unterzeichneten Vertrag bis zum produktiven Tool rechnen Sie mit 4–12 Wochen für ein Standard-Custom-Pipeline-Projekt. Die Spanne wird von drei Faktoren bestimmt.

Datenlage. Wenn Sie bereits gelabelte Daten oder repräsentative Aufnahmen haben, starten wir in Woche 1 mit dem Training. Müssen wir Datenerfassung aufsetzen, einen Sensor ergänzen oder von Null labeln, kommt vorne 1–3 Wochen dazu.

Integrations-Tiefe. Ein eigenständiges Dashboard mit REST-Schnittstelle ist der schnellste Weg — typisch 4–6 Wochen. OPC-UA-Anbindung in eine laufende Leittechnik, mit Verriegelungen, Werks-Abnahme und kundenseitigem Change-Management, eher 8–12 Wochen.

Sensor-Anzahl und Anlagen-Komplexität. Ein LiDAR + eine Kamera in einer sauberen Innen-Umgebung geht schneller als ein 6-Sensor-Portal im Freien mit Staub, Schnee und wechselndem Licht. Den Unterschied schneiden wir vorab im Discovery-Workshop sauber zu.

Für ein schnelleres erstes Signal empfehlen wir das Discovery & Assessment-Tier — 1–3 Tage, Festpreis, schriftliche Machbarkeits-Note innerhalb einer Woche. Vollständiges Modell siehe Industrielle Objekterkennung.
From signed contract to a working tool in production, expect 4–12 weeks for a standard custom-pipeline engagement. The spread is driven by three factors.

Data readiness. If you already have labelled data or representative recordings, we start training in week 1. If we need to set up data collection, add a sensor, or label from scratch, the front end adds 1–3 weeks.

Integration depth. A standalone dashboard with REST output is the fastest path — typically 4–6 weeks. OPC UA integration into a running control system, with interlocks, factory-acceptance tests and a customer change-management process, is closer to 8–12 weeks.

Sensor count and site complexity. One LiDAR + one camera in a clean indoor environment is faster than a six-sensor outdoor portal with dust, snow and dynamic lighting. We scope the difference up-front in the Discovery workshop.

For a faster first signal we recommend the Discovery & Assessment tier — 1–3 days, fixed price, written feasibility note within a week. See Industrial Perception AI for the full model.
From signed contract to a working tool in production, expect 4–12 weeks for a standard custom-pipeline engagement. The spread is driven by three factors.

Data readiness. If you already have labelled data or representative recordings, we start training in week 1. If we need to set up data collection, add a sensor, or label from scratch, the front end adds 1–3 weeks.

Integration depth. A standalone dashboard with REST output is the fastest path — typically 4–6 weeks. OPC UA integration into a running control system, with interlocks, factory-acceptance tests and a customer change-management process, is closer to 8–12 weeks.

Sensor count and site complexity. One LiDAR + one camera in a clean indoor environment is faster than a six-sensor outdoor portal with dust, snow and dynamic lighting. We scope the difference up-front in the Discovery workshop.

For a faster first signal we recommend the Discovery & Assessment tier — 1–3 days, fixed price, written feasibility note within a week. See Industrial Perception AI for the full model.
From signed contract to a working tool in production, expect 4–12 weeks for a standard custom-pipeline engagement. The spread is driven by three factors.

Data readiness. If you already have labelled data or representative recordings, we start training in week 1. If we need to set up data collection, add a sensor, or label from scratch, the front end adds 1–3 weeks.

Integration depth. A standalone dashboard with REST output is the fastest path — typically 4–6 weeks. OPC UA integration into a running control system, with interlocks, factory-acceptance tests and a customer change-management process, is closer to 8–12 weeks.

Sensor count and site complexity. One LiDAR + one camera in a clean indoor environment is faster than a six-sensor outdoor portal with dust, snow and dynamic lighting. We scope the difference up-front in the Discovery workshop.

For a faster first signal we recommend the Discovery & Assessment tier — 1–3 days, fixed price, written feasibility note within a week. See Industrial Perception AI for the full model.
From signed contract to a working tool in production, expect 4–12 weeks for a standard custom-pipeline engagement. The spread is driven by three factors.

Data readiness. If you already have labelled data or representative recordings, we start training in week 1. If we need to set up data collection, add a sensor, or label from scratch, the front end adds 1–3 weeks.

Integration depth. A standalone dashboard with REST output is the fastest path — typically 4–6 weeks. OPC UA integration into a running control system, with interlocks, factory-acceptance tests and a customer change-management process, is closer to 8–12 weeks.

Sensor count and site complexity. One LiDAR + one camera in a clean indoor environment is faster than a six-sensor outdoor portal with dust, snow and dynamic lighting. We scope the difference up-front in the Discovery workshop.

For a faster first signal we recommend the Discovery & Assessment tier — 1–3 days, fixed price, written feasibility note within a week. See Industrial Perception AI for the full model.

Who owns the code and IP we pay you to build?Wem gehören Code und IP, die wir bei Ihnen beauftragen?Who owns the code and IP we pay you to build?Who owns the code and IP we pay you to build?Who owns the code and IP we pay you to build?Who owns the code and IP we pay you to build? ＋

 You do. Our standard contract transfers full ownership of the project-specific code, model weights, training data (where the data is yours) and documentation to you on final acceptance. We don't keep a back-door licence and we don't lock you into a maintenance contract.

What we retain is our pre-existing toolbox — reusable libraries, calibration utilities, generic data parsers, the OWL EYE® core — which we license to you, royalty-free, for use on the delivered tool. This is the same boundary that any reputable engineering firm draws: we bring the toolbox, you keep the deliverable.

If you want a different IP arrangement — for example a joint patent on a novel algorithm, or a co-developed model that we both reuse with consent — we structure that explicitly in the project contract. Tell us up-front in the Discovery workshop.

Full engagement details on our Industrial Perception AI page.
Ihnen. Unser Standard-Vertrag überträgt das volle Eigentum am projektspezifischen Code, an den Modell-Gewichten, an Trainingsdaten (soweit die Daten Ihnen gehören) und an der Dokumentation bei der Abnahme an Sie. Wir behalten keine Hintertür-Lizenz und binden Sie nicht an einen Wartungsvertrag.

Was bei uns bleibt, ist unser vorhandener Werkzeugkasten — wiederverwendbare Bibliotheken, Kalibrier-Utilities, generische Daten-Parser, der OWL EYE®-Kern — den wir Ihnen lizenzgebührenfrei für das gelieferte Tool lizenzieren. Das ist dieselbe Grenze, die jedes seriöse Ingenieurbüro zieht: Wir bringen den Werkzeugkasten, Sie behalten das Ergebnis.

Wenn Sie eine andere IP-Vereinbarung wollen — etwa ein gemeinsames Patent auf einen neuartigen Algorithmus oder ein gemeinsam entwickeltes Modell, das wir beide mit Einwilligung wiederverwenden — regeln wir das explizit im Projektvertrag. Sagen Sie es vorab im Discovery-Workshop.

Vollständige Details auf unserer Seite zur Industriellen Objekterkennung.
You do. Our standard contract transfers full ownership of the project-specific code, model weights, training data (where the data is yours) and documentation to you on final acceptance. We don't keep a back-door licence and we don't lock you into a maintenance contract.

What we retain is our pre-existing toolbox — reusable libraries, calibration utilities, generic data parsers, the OWL EYE® core — which we license to you, royalty-free, for use on the delivered tool. This is the same boundary that any reputable engineering firm draws: we bring the toolbox, you keep the deliverable.

If you want a different IP arrangement — for example a joint patent on a novel algorithm, or a co-developed model that we both reuse with consent — we structure that explicitly in the project contract. Tell us up-front in the Discovery workshop.

Full engagement details on our Industrial Perception AI page.
You do. Our standard contract transfers full ownership of the project-specific code, model weights, training data (where the data is yours) and documentation to you on final acceptance. We don't keep a back-door licence and we don't lock you into a maintenance contract.

What we retain is our pre-existing toolbox — reusable libraries, calibration utilities, generic data parsers, the OWL EYE® core — which we license to you, royalty-free, for use on the delivered tool. This is the same boundary that any reputable engineering firm draws: we bring the toolbox, you keep the deliverable.

If you want a different IP arrangement — for example a joint patent on a novel algorithm, or a co-developed model that we both reuse with consent — we structure that explicitly in the project contract. Tell us up-front in the Discovery workshop.

Full engagement details on our Industrial Perception AI page.
You do. Our standard contract transfers full ownership of the project-specific code, model weights, training data (where the data is yours) and documentation to you on final acceptance. We don't keep a back-door licence and we don't lock you into a maintenance contract.

What we retain is our pre-existing toolbox — reusable libraries, calibration utilities, generic data parsers, the OWL EYE® core — which we license to you, royalty-free, for use on the delivered tool. This is the same boundary that any reputable engineering firm draws: we bring the toolbox, you keep the deliverable.

If you want a different IP arrangement — for example a joint patent on a novel algorithm, or a co-developed model that we both reuse with consent — we structure that explicitly in the project contract. Tell us up-front in the Discovery workshop.

Full engagement details on our Industrial Perception AI page.
You do. Our standard contract transfers full ownership of the project-specific code, model weights, training data (where the data is yours) and documentation to you on final acceptance. We don't keep a back-door licence and we don't lock you into a maintenance contract.

What we retain is our pre-existing toolbox — reusable libraries, calibration utilities, generic data parsers, the OWL EYE® core — which we license to you, royalty-free, for use on the delivered tool. This is the same boundary that any reputable engineering firm draws: we bring the toolbox, you keep the deliverable.

If you want a different IP arrangement — for example a joint patent on a novel algorithm, or a co-developed model that we both reuse with consent — we structure that explicitly in the project contract. Tell us up-front in the Discovery workshop.

Full engagement details on our Industrial Perception AI page.

Can you use our existing cameras and LiDAR sensors?Können Sie unsere vorhandenen Kameras und LiDAR-Sensoren weiterverwenden?Can you use our existing cameras and LiDAR sensors?Can you use our existing cameras and LiDAR sensors?Can you use our existing cameras and LiDAR sensors?Can you use our existing cameras and LiDAR sensors? ＋

 Yes — in most cases. Our toolchain is sensor-agnostic on the data layer. If your sensors deliver a standard format (PCD, LAS, E57 for point clouds; RTSP, GigE Vision, USB3 Vision for cameras) we can ingest them in a day. We've worked with Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision and IDS cameras.

We'll tell you honestly when the existing hardware isn't the right tool. Common cases where we recommend a replacement:

- The existing LiDAR is too narrow (FoV or range) to cover the area you want classified — usually solved by repositioning before buying anything new.
- The camera is auto-exposing on a high-contrast scene (sun + shadow at a gate), which breaks classification reliability — solved by switching to a manual-exposure industrial camera with HDR sensor.
- The existing sensor's vendor lock-in (proprietary stream format, no documented SDK) makes integration cost more than a Livox replacement.

We don't earn a margin on reselling hardware to you. If you do want us to source new sensors, we do that through our normal Hardware & products channel at list price. See Industrial Perception AI for how we scope it.
In den meisten Fällen ja. Unsere Toolchain ist auf der Datenebene sensor-agnostisch. Wenn Ihre Sensoren ein Standard-Format liefern (PCD, LAS, E57 für Punktwolken; RTSP, GigE Vision, USB3 Vision für Kameras), binden wir sie an einem Tag an. Wir haben mit Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision und IDS gearbeitet.

Wir sagen Ihnen ehrlich, wenn die vorhandene Hardware nicht das richtige Werkzeug ist. Übliche Fälle, in denen wir zum Tausch raten:

- Der vorhandene LiDAR ist zu schmal (FoV oder Reichweite), um den gewünschten Bereich abzudecken — meist durch Repositionierung gelöst, bevor neu gekauft wird.
- Die Kamera auto-belichtet auf einer Szene mit hohem Kontrast (Sonne + Schatten am Tor), das bricht die Klassifikations-Zuverlässigkeit — gelöst durch Wechsel auf eine manuell belichtbare Industrie-Kamera mit HDR-Sensor.
- Vendor-Lock-in des vorhandenen Sensors (proprietärer Stream, kein dokumentiertes SDK) macht die Anbindung teurer als ein Livox-Ersatz.

Wir verdienen nicht am Weiterverkauf von Hardware an Sie. Wenn Sie wollen, dass wir neue Sensoren beschaffen, läuft das über unseren regulären Kanal Handelsware & Produkte zum Listenpreis. Scoping-Modell siehe Industrielle Objekterkennung.
Yes — in most cases. Our toolchain is sensor-agnostic on the data layer. If your sensors deliver a standard format (PCD, LAS, E57 for point clouds; RTSP, GigE Vision, USB3 Vision for cameras) we can ingest them in a day. We've worked with Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision and IDS cameras.

We'll tell you honestly when the existing hardware isn't the right tool. Common cases where we recommend a replacement:

- The existing LiDAR is too narrow (FoV or range) to cover the area you want classified — usually solved by repositioning before buying anything new.
- The camera is auto-exposing on a high-contrast scene (sun + shadow at a gate), which breaks classification reliability — solved by switching to a manual-exposure industrial camera with HDR sensor.
- The existing sensor's vendor lock-in (proprietary stream format, no documented SDK) makes integration cost more than a Livox replacement.

We don't earn a margin on reselling hardware to you. If you do want us to source new sensors, we do that through our normal Hardware & products channel at list price. See Industrial Perception AI for how we scope it.
Yes — in most cases. Our toolchain is sensor-agnostic on the data layer. If your sensors deliver a standard format (PCD, LAS, E57 for point clouds; RTSP, GigE Vision, USB3 Vision for cameras) we can ingest them in a day. We've worked with Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision and IDS cameras.

We'll tell you honestly when the existing hardware isn't the right tool. Common cases where we recommend a replacement:

- The existing LiDAR is too narrow (FoV or range) to cover the area you want classified — usually solved by repositioning before buying anything new.
- The camera is auto-exposing on a high-contrast scene (sun + shadow at a gate), which breaks classification reliability — solved by switching to a manual-exposure industrial camera with HDR sensor.
- The existing sensor's vendor lock-in (proprietary stream format, no documented SDK) makes integration cost more than a Livox replacement.

We don't earn a margin on reselling hardware to you. If you do want us to source new sensors, we do that through our normal Hardware & products channel at list price. See Industrial Perception AI for how we scope it.
Yes — in most cases. Our toolchain is sensor-agnostic on the data layer. If your sensors deliver a standard format (PCD, LAS, E57 for point clouds; RTSP, GigE Vision, USB3 Vision for cameras) we can ingest them in a day. We've worked with Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision and IDS cameras.

We'll tell you honestly when the existing hardware isn't the right tool. Common cases where we recommend a replacement:

- The existing LiDAR is too narrow (FoV or range) to cover the area you want classified — usually solved by repositioning before buying anything new.
- The camera is auto-exposing on a high-contrast scene (sun + shadow at a gate), which breaks classification reliability — solved by switching to a manual-exposure industrial camera with HDR sensor.
- The existing sensor's vendor lock-in (proprietary stream format, no documented SDK) makes integration cost more than a Livox replacement.

We don't earn a margin on reselling hardware to you. If you do want us to source new sensors, we do that through our normal Hardware & products channel at list price. See Industrial Perception AI for how we scope it.
Yes — in most cases. Our toolchain is sensor-agnostic on the data layer. If your sensors deliver a standard format (PCD, LAS, E57 for point clouds; RTSP, GigE Vision, USB3 Vision for cameras) we can ingest them in a day. We've worked with Riegl, Livox, Faro, Hesai, Velodyne, Ouster, Basler, FLIR, Allied Vision and IDS cameras.

We'll tell you honestly when the existing hardware isn't the right tool. Common cases where we recommend a replacement:

- The existing LiDAR is too narrow (FoV or range) to cover the area you want classified — usually solved by repositioning before buying anything new.
- The camera is auto-exposing on a high-contrast scene (sun + shadow at a gate), which breaks classification reliability — solved by switching to a manual-exposure industrial camera with HDR sensor.
- The existing sensor's vendor lock-in (proprietary stream format, no documented SDK) makes integration cost more than a Livox replacement.

We don't earn a margin on reselling hardware to you. If you do want us to source new sensors, we do that through our normal Hardware & products channel at list price. See Industrial Perception AI for how we scope it.
 