Reference curve in cobalt; a fictional batch in ember. Two in eight tiles from it showed cracks at sorting.
Batch history
Every tile carries a history. Most plants keep it in five places.
Material lots live with stores, firing curves with the kiln controller, inspection results on the sorting line. KilnMetric joins them per batch, so a rejected batch can be traced back through every stage it passed.
Raw material lot
Clay L-77 · moisture 5.7%
Pressing
Body density, moisture
Glazing
Glaze G-12 · weight
Firing
Peak 1,151 °C · cooling 36 °C/min
Inspection
11.8% rejected · cracks 41
Defect-risk prediction
Know which batches deserve a closer look — before the sorting line.
KilnMetric estimates each batch’s defect risk from its material lot, pressed-body moisture and firing record, and shows exactly what pushed the estimate up. Engineers decide whether to hold a batch for extra inspection.
KB-309 · clay L-77 · glaze G-12 · fired Fri 8
Why the estimate is high
Clay lot L-77+0.70
Faster cooling than usual+0.55
Pressed-body moisture above 5.5%+0.48
Peak temperature off target+0.15
Contributions to the model’s log-odds. Not yet inspected — this is a prediction, not a result.
Pattern analysis
Conditions that keep turning up beside rejected batches.
Clay lot L-77
9.4%with (3)
4.5%without (5)
Cooling faster than 33 °C/min
9.8%with (2)
5.2%without (6)
Pressed-body moisture above 6%
9.4%with (1)
5.9%without (7)
Peak more than 10 °C off target
5.2%with (1)
6.5%without (7)
Average rejection rate of inspected sample batches, with and without each condition. Small samples, and association is not chemistry — which is why the next step is a trial.
Trial tracking
Trials, not hunches.
When a pattern looks real, an engineer approves a controlled trial inside the approved process window — slower cooling on half the next batches, say — and KilnMetric compares the results with the control.
Engineer states the hypothesis
Setting stays inside the approved window
Inspectors record control and trial results
The finding is kept with the batch history
8.6%Controlcooling 34 °C/min
vs
4.1%Trialcooling 28 °C/min
Illustrative trial result, for a proposed product.
How KilnMetric works with your engineers.
I
Engineers decide
KilnMetric estimates and explains. Holding a batch, opening an investigation or changing a process is a person’s call.
II
No automatic kiln control
It never changes setpoints or curves. Trials are approved by engineers and run by your team.
III
One product family first
Models start narrow, on your own historical records, and widen only when they hold up.
IV
Patterns get tested
Associations are labelled as associations until a controlled trial supports them.
A proposed design. Nothing is connected to this website; the firing ledger runs in your browser on fictional data.
Amazon SageMaker AIDefect-risk models
Amazon BedrockClassifying inspection notes
Amazon S3Production datasets
Amazon DynamoDBBatch histories
AWS LambdaProcessing imported logs
Founders
The people behind KilnMetric.
OF
Founder & CEO
Olumide Fashola
Olumide has spent his career around kilns and sorting lines, and founded KilnMetric so that every rejected batch teaches the plant something. He leads product, customer partnerships and every pilot line.
AB
Co-founder
Amina Bello
Amina leads the analytics: connecting firing curves, material lots and inspection results, and building defect-risk models that engineers can question and test.
FAQ
Frequently asked questions.
What data does KilnMetric need?
Historical firing curves or controller logs, raw-material batch records and inspection results for one product family. Inspection notes can be free text.
Does it control the kiln?
No. It estimates risk and surfaces patterns. Engineers approve any trial or process change, and your team runs it.
How reliable are the risk estimates?
They start from your own records and are checked against inspection results. Every estimate shows what drove it, so engineers can question it.
Is KilnMetric live?
Not yet. It is a proposed product; the firing ledger is a prototype running in your browser on fictional data.
Pilot
Start with one product family and its history.
A pilot uses your historical firing logs, material batch records and inspection results for one product family. Your engineers check every pattern and risk estimate against what they already know before relying on anything.