Quality and yield
Scrap is found at the end of the line, and nobody knows which setting caused it.
- Kinds of AI
- Models
- Who decides
- Process engineer
- Where it runs
- Your servers or your cloud account
Model · what moves scrap risk
| Driver | Strength | Proposed |
|---|---|---|
| Mold temperature, zone 2Risk rises at the top of the range; now a wide window | Narrower window, test by DoE | |
| Resin lot moistureWet lots raise risk; not checked now | Add an incoming check | |
| Injection speed | No change |
Model · shadow run
High-risk batches flagged before inspection matched most scrap batches.
- Model
Written limitThe model never writes setpoints.
DecidesProcess engineer
1. The desk
Today
- MeasureFirst-pass yield swings week to week, and batch, lot, and inspection data sit in different systems.
- ExplainScrap is found at the end of the line, and nobody can say which setting caused it.
- AdjustThe process engineer changes three settings at once and never learns which one helped.
With the system
- MeasureBatch, lot, and inspection records are joined into one row per batch, and the defect-risk model gives each batch a scrap probability.Model
- ExplainDriver attribution shows which parameters move the risk, in which direction and range, and a candidate window per parameter is proposed for testing.Model
- AdjustThe process engineer tests the proposed window in a shadow run and a designed experiment (DoE), and decides, as before; the quality manager signs off any window change.Person decides
What lands on the desk
The process engineer sees which settings move scrap, in which direction, and how today's window compares with a narrower one to test. Every proposal goes to a designed experiment first; the model never writes a setpoint.
The process engineer decides; the quality manager signs off any window change; the model never writes setpoints.
Model · what moves scrap risk
| Driver | Strength | Proposed |
|---|---|---|
| Mold temperature, zone 2Risk rises at the top of the range; now a wide window | Narrower window, test by DoE | |
| Resin lot moistureWet lots raise risk; not checked now | Add an incoming check | |
| Injection speed | No change |
Model · shadow run
High-risk batches flagged before inspection matched most scrap batches.
- Model
Written limitThe model never writes setpoints.
DecidesProcess engineer
2. The system
For your technical team — skip it if you like.
How it works
In
- Batch records, execution system (MES)
- Process parameters
- Raw-material lot certificates
- Ambient, inspection, and scrap codes
Code · calculates
Batch data join
Batch, lot, and inspection records joined in code
Reads
- Everything under In
Out
- One row per batch
Model · estimates
Defect-risk model
GBM against a logistic or linear baseline, time-based checks
Reads
- The batch table
Out
- A scrap probability per batch
Model · estimates
Driver attribution (SHAP)
SHAP: how much each parameter moves the predicted risk
Reads
- The model
Out
- The parameters that move risk
- Their direction and range
Model · estimates
Process-window proposal
Candidate windows from the drivers, confounding noted
Reads
- The drivers
Out
- A candidate window per parameter, to test
Decision
Process engineer, after a DoE
A shadow run, then a designed experiment (DoE) by the plant
Reads
- The proposed window
Out
- The process engineer decides
- The quality manager signs off any window change
None, and no language model: tabular data goes in, risk and drivers come out; a model standing on its own.
How the models are trained and tested
Your expert writes the test first. A step the test does not need is skipped, and we say why.
Show the training steps
Defect-risk model
Target
Scrap or defect per batch (or first-pass yield).
Data
6–24 months of batches; the joins are the real work.
Method and challenger
GBM against simpler models with a time-based split; no method is presumed best.
Validation evidence
Out-of-time accuracy, stability of the drivers across periods, and shadow-run agreement with actual scrap.
Monitoring
Product changes and new materials trigger a refit check.
Causal caution
Historical data is confounded by operator adjustments; a DoE confirms before any setpoint changes.
3. Value and conditions
Where the time goes, and what your institution will measure.
Where the value comes from
- Less scrap and rework.
- Higher yield.
- Faster root cause.
What your institution measures
- Scrap rate
- First-pass yield
- Cost of poor quality
- Time to root cause
Measured in the Decision Blueprint
Baseline scrap and yield per line set the starting point; the plant sets the target.
Conditions
The rules, the data, and the known error, with its guard.
- Data joins and product changes; the DoE must be run by you.
- Correlation is not cause.
Start here
We join one line's batch, lot, and inspection records, measure today's scrap and yield, and check whether the data can name a driver before a model is built.
Similar desks
- ModelsSLMsAgentic AI
Maintenance on the bottleneck line
TodayThe press fails on the night shift; its vibration had changed weeks before.
With the systemRising stop risk flagged, with a draft work order and a proposed slot.
Maintenance planner decides
- ModelsSLMsAgentic AI
TodayInspectors check every part under a lamp; night shift misses more, and each nonconformance report takes half a day.
With the systemA draft nonconformance report with its evidence, ready for a hold-or-release call.
Quality engineer decides
Equalize the gap.
Bring one decision. Leave knowing the first step.
A short description of the decision is enough; no confidential records.
https://eqlx.ai/use-cases/quality-yield · Printed copy; the website is the current version.