A Year On A
Lead Strip Line.
One gauge, four lanes, 157,420 scans. Part of the year ran under closed-loop control and part ran under manual operation, which makes this the comparison you almost never get: the same line, the same year, the same specification window, with and without the loop.
Then, four years on, we went back to the same gauge on the same line and pulled the last twelve weeks. The second half of this article is what the first half predicted, checked against what actually happened.
One Scan In Eight,
Or One In Three.
Every scan of the year, split by control mode and plotted against the same 0.700 to 0.760 mm window. Both distributions are shown together first. Isolate either one and note what does not move: the target. Both modes aimed at the same place. What changes is the spread.
Every bar is a count, not a fitted curve: all 157,420 scans of the year, binned at 4 microns, drawn the same way as the second site's distribution further down. 0.9% of manual scans and 0.1% of closed-loop scans fall outside the plotted range and are not drawn.
The means land 1.6 microns apart, 0.7488 mm against 0.7472 mm. The standard deviations land 0.0132 mm against 0.0290 mm, a spread 2.2 times tighter with the loop closed. Against a 60 micron window, a 29 micron standard deviation simply does not fit, and a third of manual production fell outside as a result.
The counts also show something the two published curves could not. With the loop closed, 19% of the year's scans land in a single 4 micron bin. A normal curve of the same standard deviation would put about 12% there. The standard deviation is carrying a thin tail of excursions, and the ordinary production behind it is tighter than that one number suggests.
Tighter Control Buys You Room To Move.
Both modes aimed high, well above the product nominal. Under manual control that made sense as insurance against running thin. With the spread cut in half, that insurance is no longer needed, and the target can come down. Drag it and see.
The line was running hard against the upper limit, not the lower one, so coming down off the target does not cost specification compliance, it buys it. Running 16 microns thinner, every coil carries 2.1% less lead and the same scans read 96.2% in specification instead of 88.2%. The second half of this article is what happened when they did it.
The shifted shape is the measured one, slid down: the same 115,719 scans, counted again against the same window at each target. The assumption is that the line holds the same spread at a lower target, which is the thing to test before acting on it. The shape itself is not assumed.
The Operators Voted.
Nobody was told to use the loop. Scan volume by month shows manual operation carrying the line through the spring, then the balance flipping. By the second half of the year closed loop was running almost everything.
Monthly values read from the published figure. Annual totals are exact.
All Heavy. All Manual.
The year had three major out-of-spec events. Each ran heavy, each peaked more than 100 microns above the 0.760 mm limit, and each happened under manual control. The chart is not the peak, it is the whole window: every scan in the event, counted into 8 micron bins against the specification band. Pick one.
Measured from the daily results files, every scan in each window. The dotted line is the target, in the middle of the band. The blue line is what the strip that left the window averaged, and the cost is that tonnage at that thickness: remelt at $0.15 per kg plus the lead carried above target at $2.40 per kg, on the line basis set out in the next section. Auto Mode reads off for every scan in all three windows.
The Advantage Of Records
The gauge measured through all three events regardless of mode. 752 scans across the three windows, logged continuously, means every excursion is profiled from ramp to recovery whether or not the loop was on. It is also the difference between knowing a coil peaked high and knowing how much of the window sat over the limit, which is the number that decides what comes back.
One Excursion, Priced.
This is the late shift of 14 February, scan by scan, straight out of the results file. 253 scans, one every nine seconds, 37 minutes of production. The gauge was recording in manual mode throughout, so nothing corrected the drift.
Two things cost money here, and they are different things. Strip above the target line is lead bought and not sold. Strip outside the specification window has to come back and be remelted. Drag through the window and both add up.
And Then The Part Nobody Invoices
The figures above are lead and remelt charges only. On top of them sits the furnace time to put three tonnes back through, the handling to pull and stage the coil, the line availability lost while it happens, and the scheduling knock-on for everything queued behind it. Those are real and they are not small, but they are not in the results file, so no number is put on them here.
Every figure is measured. Target 0.730 mm with a 0.700 to 0.760 mm window, taken from the product record in the file. 253 scans at a nine second interval, 218 of them outside the window, peak 0.877 mm. Line basis 333.8 mm of strip across four lanes at 30.48 m/min with lead at 11.34 g/cm³, which is 5,053 kg an hour. Lead is held at $2.40 per kg and remelt at $0.15 per kg throughout. One caveat: there is a 21 minute gap in the middle of this window. It is treated here as the line being stopped, which is the conservative reading. If the line was running and the gauge was not recording, every figure below is roughly 60% higher.
A thousand dollars in thirty-seven minutes is not what sinks a plant. What matters is that this happened 123 times in February and March alone, every one of them under manual control, and that nobody knew the cost of any of them until the gauge was logging every nine seconds.
The Same Line,
Four Years Later.
Same plant · Same gauge · Same four strands · Same window
Everything above ends on a hypothesis. The line was aiming at 0.7488 mm against a 0.700 to 0.760 mm window, a long way above the 0.730 mm the product calls for, and we wrote that with the loop closed the target could come down and the lead be saved. We called it hypothetical, because it was.
Here is the same line four years later. Nothing about the comparison has been arranged. Same gauge, same product, same four strands, same specification window, every scan from the first of July to the twenty third of September.
Seventeen Microns Down.
Eight Points Up.
Three distributions, one thickness axis, one specification window. The manual year, the closed-loop year, and the last twelve weeks. Every bar is a count of real scans.
All three binned at 4 microns against the same 0.700 to 0.760 mm window. The 2022 figures are the ones published above. The 2026 figures are every scan with the loop engaged across 74 production days.
The mean moved from 0.7488 mm to 0.7318 mm, 17 microns down, onto the product nominal. In-specification production went from 88.2% to 96.6%, which is 8.4 points.
Almost all of that is the move, not the machine. The standard deviation came in from 13.2 microns to 11.2, which is real but modest, and the section below shows the split: sliding the 2022 scans down to today's target on its own accounts for 8.0 of the 8.4 points. The tighter spread is worth the last 0.4. The plant did not get a better gauge. It stopped paying for headroom it no longer needed.
One number does show the tightening plainly. The tallest bar in 2022 held 19% of that year's closed-loop scans. Today the tallest bar holds 27%.
Manual operation has not changed either. The 3,263 scans in this window that ran with the loop off sit at 26.3 microns of spread, which is about where manual control sat in 2022 as well at 29.0. Against 11.2 microns with the loop on, that is 2.3 times wider, and 87.6% in specification against 96.6%.
We Wrote Down A Number.
Here Is What Happened.
The chart in the first half of this article lets you drag the 2022 target downward and counts the same 2022 scans again at each position. Drag it 17 microns, to where this line is programmed today, and it predicts 96.2% in specification.
We were slightly conservative, and the reason is the part worth taking away. Moving a target is not free if your process is loose, because everything you gain at the top you give back at the bottom. It is close to free if your process is tight enough that the bottom limit never comes into play. That is what the loop is for. It does not move your target for you. It is what makes moving it survivable.
One caveat we would rather raise than have raised for us. The match is closer than it deserves to be, because two days in this window went badly wrong and pulled the measured figure down by about three points. Take those two days out and the same scans read 99.33%. The prediction was conservative and the process is better than it. Those two days are the next section.
Two Days In Twelve Weeks
Here is every production day in the window, loop engaged, against the programmed target and the two limits. All but two of them sit on the line.
Daily means, loop engaged, days with at least 50 scans. The target in the results file is 0.730 mm on every one of the 74 days.
On the second and third of July this line ran 46 microns thick against a target that had not moved, with Auto Mode reading engaged the whole time. The tell is in the controller column: on those two days the output sat between 2.07 and 2.14 volts and barely moved, where an ordinary week of closed-loop running swings between 1.93 and 2.68. The gauge was measuring correctly, the strands agree with each other, and the loop was asking for a correction that was not arriving.
Those two days are 1,419 of the 48,065 closed-loop scans in the window and they cost about three points of the headline. Across the other 69 days that ran with the loop engaged, this line was 99.33% in specification at 7.7 microns of spread. Both numbers are true and we would rather print both than pick one.
It is also the argument for the record itself. Two bad days inside a quarter are invisible in a monthly average and invisible on a coil weight. They are obvious the moment somebody plots the file.
What It Does
When The Strip Drifts.
One ordinary correction, scan by scan, nothing modelled. The top panel is thickness at each of the four strands with the scan mean in bold. The bottom panel is the voltage the gauge sends to the caster speed drive. Press play and watch the order of events.
Every point is a logged scan, one every 17 seconds. The shaded band is the controller deadband, 8 microns either side of target, from the controller settings on file. No operator action is involved.
The strip drifts thick, the gauge sees it, the voltage steps within 33 seconds, and the strip is back inside the deadband 1.4 minutes later and sits there. The whole thing happens 17 microns from target, well inside a window that runs to 30. That is the part worth noticing: nothing here was ever going to be scrap. The loop is working on drift, not on rejects.
Across the twelve weeks this happened 92 times with the loop engaged and recovered every one of them, median 1.1 minutes, slowest 3.5 minutes at the ninetieth percentile. A larger disturbance than this one comes back with some overshoot before it settles, which is a tuning choice rather than a fault, and the settings that decide it are in the file.
2.27 Percent,
On Every Metre.
Lead is bought by weight, and thickness is the part of that weight a caster controls. 2.27% off the thickness is 2.27% off the lead, on every metre of every strand, for as long as the line runs. That is not a quality number, it is a materials number. Set the slider to your own tonnage.
Lead held at $2.20 per kilogram throughout. The percentage is measured against where this line ran in 2022; the tonnage and the price are yours. Nothing here counts scrap, remelt, furnace time or the line availability lost to an excursion, all of which are real and none of which are in a results file.
Our own savings estimator assumes a plant can hand back 1.5%. This one handed back 2.27% and is running at 98.9% in specification in September while it does it.
Two Percent Off Every Coil.
What Would Yours Be?
How far you can move your own target depends on your spread, your specification window and your line speed, not on ours. The estimator shows the arithmetic it used, so you can check every number in it.
A Year On A
Lead Strip Line.
One gauge, four lanes, 157,420 scans. Part of the year ran under closed-loop control and part ran under manual operation, which makes this the comparison you almost never get: the same line, the same year, the same specification window, with and without the loop.
Then, four years on, we went back to the same gauge on the same line and pulled the last twelve weeks. The second half of this article is what the first half predicted, checked against what actually happened.
One Scan In Eight,
Or One In Three.
Every scan of the year, split by control mode and plotted against the same 0.700 to 0.760 mm window. Both distributions are shown together first. Isolate either one and note what does not move: the target. Both modes aimed at the same place. What changes is the spread.
Every bar is a count, not a fitted curve: all 157,420 scans of the year, binned at 4 microns, drawn the same way as the second site's distribution further down. 0.9% of manual scans and 0.1% of closed-loop scans fall outside the plotted range and are not drawn.
The means land 1.6 microns apart, 0.7488 mm against 0.7472 mm. The standard deviations land 0.0132 mm against 0.0290 mm, a spread 2.2 times tighter with the loop closed. Against a 60 micron window, a 29 micron standard deviation simply does not fit, and a third of manual production fell outside as a result.
The counts also show something the two published curves could not. With the loop closed, 19% of the year's scans land in a single 4 micron bin. A normal curve of the same standard deviation would put about 12% there. The standard deviation is carrying a thin tail of excursions, and the ordinary production behind it is tighter than that one number suggests.
Tighter Control Buys You Room To Move.
Both modes aimed high, well above the product nominal. Under manual control that made sense as insurance against running thin. With the spread cut in half, that insurance is no longer needed, and the target can come down. Drag it and see.
The line was running hard against the upper limit, not the lower one, so coming down off the target does not cost specification compliance, it buys it. Running 16 microns thinner, every coil carries 2.1% less lead and the same scans read 96.2% in specification instead of 88.2%. The second half of this article is what happened when they did it.
The shifted shape is the measured one, slid down: the same 115,719 scans, counted again against the same window at each target. The assumption is that the line holds the same spread at a lower target, which is the thing to test before acting on it. The shape itself is not assumed.
The Operators Voted.
Nobody was told to use the loop. Scan volume by month shows manual operation carrying the line through the spring, then the balance flipping. By the second half of the year closed loop was running almost everything.
Monthly values read from the published figure. Annual totals are exact.
All Heavy. All Manual.
The year had three major out-of-spec events. Each ran heavy, each peaked more than 100 microns above the 0.760 mm limit, and each happened under manual control. The chart is not the peak, it is the whole window: every scan in the event, counted into 8 micron bins against the specification band. Pick one.
Measured from the daily results files, every scan in each window. The dotted line is the target, in the middle of the band. The blue line is what the strip that left the window averaged, and the cost is that tonnage at that thickness: remelt at $0.15 per kg plus the lead carried above target at $2.40 per kg, on the line basis set out in the next section. Auto Mode reads off for every scan in all three windows.
The Advantage Of Records
The gauge measured through all three events regardless of mode. 752 scans across the three windows, logged continuously, means every excursion is profiled from ramp to recovery whether or not the loop was on. It is also the difference between knowing a coil peaked high and knowing how much of the window sat over the limit, which is the number that decides what comes back.
One Excursion, Priced.
This is the late shift of 14 February, scan by scan, straight out of the results file. 253 scans, one every nine seconds, 37 minutes of production. The gauge was recording in manual mode throughout, so nothing corrected the drift.
Two things cost money here, and they are different things. Strip above the target line is lead bought and not sold. Strip outside the specification window has to come back and be remelted. Drag through the window and both add up.
And Then The Part Nobody Invoices
The figures above are lead and remelt charges only. On top of them sits the furnace time to put three tonnes back through, the handling to pull and stage the coil, the line availability lost while it happens, and the scheduling knock-on for everything queued behind it. Those are real and they are not small, but they are not in the results file, so no number is put on them here.
Every figure is measured. Target 0.730 mm with a 0.700 to 0.760 mm window, taken from the product record in the file. 253 scans at a nine second interval, 218 of them outside the window, peak 0.877 mm. Line basis 333.8 mm of strip across four lanes at 30.48 m/min with lead at 11.34 g/cm³, which is 5,053 kg an hour. Lead is held at $2.40 per kg and remelt at $0.15 per kg throughout. One caveat: there is a 21 minute gap in the middle of this window. It is treated here as the line being stopped, which is the conservative reading. If the line was running and the gauge was not recording, every figure below is roughly 60% higher.
A thousand dollars in thirty-seven minutes is not what sinks a plant. What matters is that this happened 123 times in February and March alone, every one of them under manual control, and that nobody knew the cost of any of them until the gauge was logging every nine seconds.
The Same Line,
Four Years Later.
Same plant · Same gauge · Same four strands · Same window
Everything above ends on a hypothesis. The line was aiming at 0.7488 mm against a 0.700 to 0.760 mm window, a long way above the 0.730 mm the product calls for, and we wrote that with the loop closed the target could come down and the lead be saved. We called it hypothetical, because it was.
Here is the same line four years later. Nothing about the comparison has been arranged. Same gauge, same product, same four strands, same specification window, every scan from the first of July to the twenty third of September.
Seventeen Microns Down.
Eight Points Up.
Three distributions, one thickness axis, one specification window. The manual year, the closed-loop year, and the last twelve weeks. Every bar is a count of real scans.
All three binned at 4 microns against the same 0.700 to 0.760 mm window. The 2022 figures are the ones published above. The 2026 figures are every scan with the loop engaged across 74 production days.
The mean moved from 0.7488 mm to 0.7318 mm, 17 microns down, onto the product nominal. In-specification production went from 88.2% to 96.6%, which is 8.4 points.
Almost all of that is the move, not the machine. The standard deviation came in from 13.2 microns to 11.2, which is real but modest, and the section below shows the split: sliding the 2022 scans down to today's target on its own accounts for 8.0 of the 8.4 points. The tighter spread is worth the last 0.4. The plant did not get a better gauge. It stopped paying for headroom it no longer needed.
One number does show the tightening plainly. The tallest bar in 2022 held 19% of that year's closed-loop scans. Today the tallest bar holds 27%.
Manual operation has not changed either. The 3,263 scans in this window that ran with the loop off sit at 26.3 microns of spread, which is about where manual control sat in 2022 as well at 29.0. Against 11.2 microns with the loop on, that is 2.3 times wider, and 87.6% in specification against 96.6%.
We Wrote Down A Number.
Here Is What Happened.
The chart in the first half of this article lets you drag the 2022 target downward and counts the same 2022 scans again at each position. Drag it 17 microns, to where this line is programmed today, and it predicts 96.2% in specification.
We were slightly conservative, and the reason is the part worth taking away. Moving a target is not free if your process is loose, because everything you gain at the top you give back at the bottom. It is close to free if your process is tight enough that the bottom limit never comes into play. That is what the loop is for. It does not move your target for you. It is what makes moving it survivable.
One caveat we would rather raise than have raised for us. The match is closer than it deserves to be, because two days in this window went badly wrong and pulled the measured figure down by about three points. Take those two days out and the same scans read 99.33%. The prediction was conservative and the process is better than it. Those two days are the next section.
Two Days In Twelve Weeks
Here is every production day in the window, loop engaged, against the programmed target and the two limits. All but two of them sit on the line.
Daily means, loop engaged, days with at least 50 scans. The target in the results file is 0.730 mm on every one of the 74 days.
On the second and third of July this line ran 46 microns thick against a target that had not moved, with Auto Mode reading engaged the whole time. The tell is in the controller column: on those two days the output sat between 2.07 and 2.14 volts and barely moved, where an ordinary week of closed-loop running swings between 1.93 and 2.68. The gauge was measuring correctly, the strands agree with each other, and the loop was asking for a correction that was not arriving.
Those two days are 1,419 of the 48,065 closed-loop scans in the window and they cost about three points of the headline. Across the other 69 days that ran with the loop engaged, this line was 99.33% in specification at 7.7 microns of spread. Both numbers are true and we would rather print both than pick one.
It is also the argument for the record itself. Two bad days inside a quarter are invisible in a monthly average and invisible on a coil weight. They are obvious the moment somebody plots the file.
What It Does
When The Strip Drifts.
One ordinary correction, scan by scan, nothing modelled. The top panel is thickness at each of the four strands with the scan mean in bold. The bottom panel is the voltage the gauge sends to the caster speed drive. Press play and watch the order of events.
Every point is a logged scan, one every 17 seconds. The shaded band is the controller deadband, 8 microns either side of target, from the controller settings on file. No operator action is involved.
The strip drifts thick, the gauge sees it, the voltage steps within 33 seconds, and the strip is back inside the deadband 1.4 minutes later and sits there. The whole thing happens 17 microns from target, well inside a window that runs to 30. That is the part worth noticing: nothing here was ever going to be scrap. The loop is working on drift, not on rejects.
Across the twelve weeks this happened 92 times with the loop engaged and recovered every one of them, median 1.1 minutes, slowest 3.5 minutes at the ninetieth percentile. A larger disturbance than this one comes back with some overshoot before it settles, which is a tuning choice rather than a fault, and the settings that decide it are in the file.
2.27 Percent,
On Every Metre.
Lead is bought by weight, and thickness is the part of that weight a caster controls. 2.27% off the thickness is 2.27% off the lead, on every metre of every strand, for as long as the line runs. That is not a quality number, it is a materials number. Set the slider to your own tonnage.
Lead held at $2.20 per kilogram throughout. The percentage is measured against where this line ran in 2022; the tonnage and the price are yours. Nothing here counts scrap, remelt, furnace time or the line availability lost to an excursion, all of which are real and none of which are in a results file.
Our own savings estimator assumes a plant can hand back 1.5%. This one handed back 2.27% and is running at 98.9% in specification in September while it does it.
Two Percent Off Every Coil.
What Would Yours Be?
How far you can move your own target depends on your spread, your specification window and your line speed, not on ours. The estimator shows the arithmetic it used, so you can check every number in it.