Leading the Intelligent Factory:  Chapter 8

Where it costs least

Your history shows where you have run. It does not show where you should run. Statistics make the plant repeatable. A cost study makes it cheapest.

Spread 1

Many recipes hit the target. One costs least.

The decision. Once the rate is set, hundreds of other settings remain: excess air, refining load, moisture target, chemical doses, steam to the boxes. Many combinations make good product. They do not cost the same.

What goes wrong today. Most targets come from habit or from history. A centerline built from past data records where the plant has run, good periods and bad. That makes the process repeatable, which is a real step. But if the operating point drifted, the statistics drift with it. And history cannot show a better point the plant has never tried.

What AI changes. The work happens in two steps. First, targets and limits come from your own run history, so every crew runs the same way. A compliance score then shows where you run and which tags are off. Second, a cost study compares a year of runs within each grade, tests every link, and finds the lowest-cost point inside the specification. The targets move there, and the compliance score holds them.

What still needs people. Specification calls, customer limits and the decision to run a trial. A good study separates changes you have run before from trials on new ground.

Smart targetsset, moved, held1. Targets from historyEvery crew runs the same wayREPEATABLE2. A cost study moves themTo the lowest cost inside the specOPTIMIZED3. Compliance holds themWhere you run, which tags are offOPTIMIZED4. Review and study againMonthly review, quarterly studyADAPTIVE
Figure 8.1. Targets are set from history, moved by a cost study, and held by compliance.

Spread 2

A real example: the lime kiln

A lime kiln ran its stack O2 at 7.1% through the fall. In January it moved up to 7.6% and stayed there. Nothing broke. A centerline built from the last six months now says 7.6% is normal.

The cost study says otherwise. Above about 7.2% O2, residual carbonate runs half a point higher. Lime inside the 2.0 to 3.5% carbonate band fell from 68% of samples to 49%. Purchased makeup lime rose from 60.8 to 70.7 tons per 1,000 tons of lime. You pay twice for the extra air: heat up the stack, and lime bought back.

Savings potential

About $141,000 a year

New target: 7.0 to 7.2% stack O2, taken out with the ID fan. No capital, no outage.

A paper machine shows the second limit of history. It ran its reel moisture target at 7.5% on 96% of hours. The specification allows up to 8.6% before an advisory. Statistics could never suggest 8.0%, because the machine never ran there. A trial on one grade, with a warp check, tests it.

6.87.07.27.47.67.8Lowest cost7.0 to 7.2%Where historysays to runWhere you ranJanuary stepAug 25Nov 25Feb 26May 26Aug 26Monthly average stack O2, %
Figure 8.2. Lime kiln stack O2 by month, Aug 2025 to Aug 2026. The dashed line is a trailing six-month average: a centerline built from history follows the drift up.

Questions for Monday morning

  1. Where did our current targets come from, and when were they last checked against cost?
  2. Which targets have not moved in two years?
  3. For our three largest cost lines, do we know the lowest-cost point inside the specification?
  4. Which specification limits do we run well inside of, and why?
  5. When a tag runs off target all month, do we run to the target or reset it? Who decides?

In practice. Smart Centerlines works in this order: targets from your history first, then a quarterly cost study moves them, then a monthly compliance score holds them.