volume measurement Archives - CRATUS Technology

AI Inventory Management Needs Measured Inputs, Not Book Values

Cratus — Mon, 19 Oct 2026 17:38:18 +0000

AI inventory management promises better forecasts, leaner stock and fewer stockouts. The models are capable. The question is what they are fed.

For discrete goods, the data is good: barcodes, RFID and transactions give an accurate count. For bulk materials, the count is a calculation. Tons received are logged from scale tickets, tons consumed are estimated from production, and the difference is assumed to be on the pile. Moisture, spillage, compaction, theft and scale drift all widen the gap between the book and the ground.

A forecasting model trained on that data learns the error. It recommends reorder points that are too high or too low, and planners stop trusting it. The project is labeled an AI problem when it is a measurement problem.

Give the model a measured number. CRATUS VMS measures bulk inventory directly with 3D LiDAR, at intervals from minutes to hours. Each reading is a time-stamped volume and tonnage, available by API to the planning platform, the data lake or the model itself. The model sees actual consumption rates, actual receipts and actual on-hand.

In customer installations, volume accuracy moved from 70% to 80% with visual and manual methods to 99%+ with LiDAR. Readings flow into Ignition™ and other SCADA platforms, where they are stored time-stamped and available to the planning model.

Results consultants and planners can show:

Forecasts trained on measured inventory, not reconstructed book values.
Anomaly detection on receipts and consumption, catching short loads and losses.
Reorder points set from real variance.
Audit-ready history for every pile and silo.

AI inventory management works when the inventory is measured. CRATUS supplies that layer as a managed service, so the AI team does not have to become a sensor team.

FAQ: AI Inventory Management for Bulk Materials

What is AI inventory management?

AI inventory management uses machine learning to forecast demand, set reorder points and flag anomalies. Its results depend on the accuracy and frequency of the inventory data it is trained and run on.

Why do AI inventory models underperform on bulk materials?

Bulk on-hand figures are calculated from scale tickets and production estimates, not measured. Moisture, spillage, compaction and scale drift create errors the model learns and repeats.

How accurate is LiDAR bulk inventory data for AI models?

CRATUS VMS delivers 99%+ volumetric accuracy in engineered and calibrated installations, compared to 70% to 80% for visual and manual methods. Readings are time-stamped, minutes to hours apart.

How does measured inventory data reach an AI or planning platform?

Through open APIs to the data lake, ERP or model, or through SCADA such as Ignition™ by Inductive Automation. Data is processed on site at the edge, so cloud is optional.

Can consulting firms use CRATUS VMS in AI and digital twin engagements?

Yes. CRATUS provides the measured physical data layer as a managed service, so the consulting team focuses on models, planning and process change.

The post AI Inventory Management Needs Measured Inputs, Not Book Values appeared first on CRATUS Technology.

You Can’t Manage What You Can’t Measure: Why Supply Chain Autonomy Starts at Receiving

Cratus — Wed, 07 Oct 2026 11:20:19 +0000

Supply chain autonomy is the goal of most operations leaders right now: plans that adjust themselves, replenishment that triggers on its own, and fewer people reconciling spreadsheets. Most of these programs start in the ERP and the planning layer. Few start where the material enters the plant.

That is the gap. For bulk materials such as ore, aggregate, grain, feedstock, salt, pulp and biomass, the inventory number in the ERP is usually a book value. It is corrected once a month, sometimes once a quarter, by a drone flight, a surveyor, or someone with a tape and a calculator. Between surveys the system assumes receipts minus consumption equals what is on the ground. It rarely does.

When the input is wrong, every downstream decision inherits the error. Production schedules run against material that is not there. Safety stock grows to cover the uncertainty. Month-end write-offs absorb the difference. An autonomous system acting on a wrong number makes the wrong decision faster.

Measure first, then automate. CRATUS VMS uses LiDAR to measure stockpiles, silos, bunkers and bins continuously. Volume is converted to tonnage and published through open APIs to SAP, Oracle, Azure, AWS or any historian. The inventory figure becomes a measured value, refreshed on a schedule you set, not an estimate refreshed when someone has time.

CRATUS VMS is already integrated in customer plants through SCADA, including Ignition™ by Inductive Automation and other SCADA and historian platforms. The plant queries the edge unit, stores time-stamped volume readings in its own database and uses them at the frequency it needs.

What changes for the operation:

Receiving and consumption are verified against measured change, not assumed.
Planning and replenishment run on current inventory.
Survey costs and month-end adjustments drop.
Safety stock can be set from measured variance instead of fear.

Customers report volume accuracy rising from 70% to 80% to 99%+, about 10 hours of survey and post-processing removed per cycle, and data available in seconds instead of 1 to 2 days. Typical payback is 6 to 9 months.

Supply chain autonomy is a data problem before it is an AI problem. You can’t manage what you can’t measure. Start at the gate.

FAQ: Supply Chain Autonomy and Bulk Inventory

What is supply chain autonomy?

Supply chain autonomy is the ability of planning and execution systems to sense conditions and act, such as triggering replenishment, without manual intervention. It depends on accurate, current inventory data at every stage, including raw material receiving.

Why do supply chain autonomy programs stall at bulk materials?

Bulk inventory is usually a book value corrected by monthly drone or manual surveys, at 70% to 80% accuracy. Autonomous systems acting on that number need human override, so the program stops at a dashboard.

How does LiDAR inventory measurement support digital supply chain transformation?

Fixed 3D LiDAR measures stockpiles, silos and bunkers continuously at 99%+ volumetric accuracy and publishes volume and tonnage to ERP, SCADA and cloud platforms. Planning and replenishment then run on measured inventory.

Can CRATUS VMS integrate with SAP, Oracle or SCADA systems?

Yes. Data is delivered through open APIs and through SCADA. Customer installations run on Ignition™ by Inductive Automation and other SCADA and historian platforms.

What ROI can a supply chain autonomy program expect from automated bulk measurement?

Customers report payback in 6 to 9 months, about 10 hours of survey work removed per cycle, and seasonal inventory mismatches reduced from hundreds of thousands of dollars to hundreds.

The post You Can’t Manage What You Can’t Measure: Why Supply Chain Autonomy Starts at Receiving appeared first on CRATUS Technology.

Accurate Volume Measurement for Silos

Cratus — Wed, 16 Sep 2026 15:29:58 +0000

Go Beyond Single-Point Sensors to stop your silo lying to you

Here is the first one. The silo is full, the feeder is running, and nothing is coming out. Somewhere above the outlet the material has formed a stable arch across the opening, strong enough to carry its own weight, and the entire column above it is now standing on that arch instead of feeding the discharge. The second is worse, because it lies in the opposite direction. The material bores a narrow channel straight down the middle and empties through it, leaving the rest standing against the walls where it slowly hardens. Your level reading falls toward empty. Most of what you are storing never moved.

Neither of these shows up on anything you already have installed, and the reason deserves precision, because it is not a quality problem with your instruments. A radar or ultrasonic gauge measures the distance to whatever surface sits under its beam. During a rathole that surface is the floor of the chimney, so the gauge reports a silo running empty while the perimeter is still packed. Load cells look like they ought to catch it, and they do not. A load cell gives you total mass and says nothing whatsoever about where that mass is sitting, so a ratholed silo and a properly draining one holding the same tonnage read identically. Every instrument in that list returns a single number, and what has gone wrong is a shape. One number cannot describe a shape.

Which means the first you hear about it is when production stops, or when somebody climbs up and looks. That is the point where the cost stops being about material. Clearing an arch, or breaking down material caked on the wall, means going in with a pole, and it gets scheduled blind because nobody can see whether it is necessary. Meanwhile the structure itself is paying. Jenike and Johanson, who have worked on this problem since the 1960s, are blunt about it: eccentric discharge develops non-uniform pressures around the circumference, “leading to circumferential bending and axial compression on the cylinder shell.” In thin metal silos that produces high membrane stresses which “may deform the shell and in some cases cause catastrophic failure.” In concrete it produces bending moments and serious cracking. Then they add the line that should keep an engineer awake, that the non-catastrophic failures “may be the prelude to a much larger event in the near future.”

What actually reads both failures is the shape of the surface, and it turns out you need only two values off it to get the whole story: how much material is in there, and how tall the highest point is. Watch those two together through a drawdown and each failure signs its own name. Healthy discharge, both fall. An arch, both go flat while the feeder keeps running. A rathole, volume drops away while peak height sits exactly where it was, because the walls are holding it up. Two numbers moving against each other, and you know which problem you have well before anything downstream starves.

The CRATUS VMS-HS128 publishes both of them. LiDAR sensors mounted in the roof space scan the full surface on a schedule, and the compute unit puts volume in cubic meters and peak pile height in meters straight into your SCADA over Modbus TCP, as two ordinary registers. That detail matters more than it sounds, because the diagnostic above is not something you have to buy from us or wait for a release to get. Your own integrator can historize those two tags and write the divergence alarm in an afternoon, inside the control system you already run. Everything else the system does, inventory that is current rather than remembered and a reason to stop opening the roof, is worth having on its own. But the first time it tells you an arch is forming while the silo still reads full, it has paid for itself.

The post Accurate Volume Measurement for Silos appeared first on CRATUS Technology.

A Belt Scale Cannot Tell a Centered Load From One That Is Destroying Your Belt

Cratus — Mon, 07 Sep 2026 13:04:40 +0000

Somebody sweeps up under your conveyor. Maybe every shift, maybe every week, and the pile is always in roughly the same stretch of the run. Meanwhile the belt scale reports tonnage that looks fine, because from where that instrument sits, the tonnage is fine. A belt scale measures force on a weighbridge and integrates it over belt travel. Take the same load and push it a hand’s width toward one edge and the force is identical. The scale is doing exactly what it was built to do. What it hands you has nothing at all to say about where the material is sitting.

That position is the thing doing the damage. Martin Engineering puts the mechanism plainly: the load’s center of gravity seeks the lowest point of the troughing idlers, so a belt that is not center loaded gets pushed toward its more lightly loaded side. From there you get edge wear, material coming off along the run, and belt damage that costs more than everything you spilled. Off center loading sits among the primary causes of mistracking, next to a misaligned tail pulley and a worn splice. The difference is that the other two announce themselves. This one hides behind a tonnage number that reads correct all the way to the point where you are replacing belt.

None of this is a complaint about belt scales, which are good at the job they were built for. NIST Handbook 44 defines two accuracy classes for legal for trade belt conveyor scales, 0.25 and 0.1, and manufacturers rate their best multi idler systems to around 0.125 percent. Thayer Scale’s own engineering guide is refreshingly honest about what holds that up: belt tension variation, idler misalignment, material building on the weighbridge, belt stretch with temperature. Their phrase for it is that accuracy is not static. Even running perfectly, the instrument is telling you about mass, and most of what stops a line is geometric.

So think about what you would actually want on the screen instead. Whether the belt is running well under its design cross section, which is motor load spent moving rubber rather than product. Where the load sits across the belt, trending over weeks, so you see the drift before the belt starts wandering. A surge building ahead of a transfer chute that is about to plug. A tramp item or a size excursion heading into a crusher. A load cell cannot answer any of these, because none of them are questions about weight.

The CRATUS VMS-FL180C was built to answer them. A solid state flash LiDAR looks down at the belt over a fixed idler section and returns the full cross section of the load at up to 10 Hz, with color imaging from the same field of view. Fill level warnings and material flags come out of the box, and because the whole profile is there every frame, where the load sits across the belt is arithmetic on data the sensor already has. Volume integrates from measured belt displacement rather than assumed belt speed, and if the encoder stalls or reports something implausible the totalizers hold rather than drift. The digital outputs already cover belt run permissive, alarm, and watchdog, so this closes a loop instead of adding one more dashboard for somebody to ignore. Your belt scale goes on owning the tonnage. This owns the shape of what is riding on the belt, which is where the maintenance budget actually goes.

The post A Belt Scale Cannot Tell a Centered Load From One That Is Destroying Your Belt appeared first on CRATUS Technology.