Courses for tuning mold level control

Has anyone taken a course that goes deep on mold level control and breakout detection, ideally AIST or OEM-led? I’m on a two-strand slab caster (180 cpm, 3.5 mm stroke) and want training with hands-on PID tuning and signal filtering using historian data to cut sticker alarms and tighten surface quality.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠‌‌⁠⁠‌⁠‌​‌‍⁠⁠‌⁠​​‌‍‍‌‌‍​⁠​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠​‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​‍⁠‌‌⁠‌‌‌‌‌‌‌​‌​‌⁠‍​‌‌​‍‌​​‍‌‍‌⁠‌‌​⁠‌​‌​‌​‍‌​⁠​​‌‍‍‍‌​‍⁠​⁠‌​‌‌​‍​‍​‍‌⁠⁠‌

Are you on Primetals, SMS, or Danieli? Check AIST’s technology training and the OEM-led classes — Primetals Mold Expert and SMS Concast both offer solid MLCS tuning; start here: 404 - Page Not Found. While you’re waiting, pull a 30–60 min steady-state historian slice and do small valve bumps to ID the loop; at 180 cpm (about 3 Hz) add a notch or tight low-pass at the oscillation (earplugs for the sensor) and retune the PID in velocity form with casting-speed feedforward.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‌‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌⁠‍​‌‍​‍​⁠‌​‌‌‍‍‌‍‍‍‌⁠‌‍​⁠​⁠‌‌‍​‌‍​‍‌‌‍‌​⁠‌‍‌‍‍​‌​⁠​‌‍‍‌‌​‍‌‌‌‍‌​‍​‍‌⁠⁠‌​​

And good option: AIST’s Continuous Casting training usually includes a PID/IMC lab and real case work — check the calendar: 404 - Page Not Found. Ahead of that, export a shift of PV/CO and trial a narrow notch around 3 Hz (your 180 cpm) before retuning with IMC to put a leash on the wobble; OEM classes are solid but tool‑specific — @lucy_wilson02 which OEM are you on?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌⁠‌‍‌⁠​​‌​‍‌​⁠‌‍‌‍‌⁠‌‌​‌‌‍​‍‌‍‌‍‌‍‍​‌​‍⁠‌‌‍‌‌‌‌‍‌​‍‌‌​‍‌‌​‍‌​⁠​‌​‍​‍‌⁠⁠‌​​

Big win for us: run a clean 10–15% speed bump, fit a FOPDT from the historian, set IMC-PI with lambda about 2*tau, then insert a tight notch at the oscillation frequency (about 3 Hz) before the PID — false stickers dropped about 30% and surface tightened. EMBr on or off, and what level sensor are you running, @lucy_wilson02?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​​​⁠​‍​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‌​​‌​⁠‍‌‍​‍‌‌‌‌‌‌‌​‌‍⁠‌‌​⁠‌‌​‌​‌​⁠‌‌​‌⁠‌‍‌⁠​⁠​‍‌‍​‍​⁠​‍​⁠​⁠​⁠‌⁠​‍​‍‌⁠⁠‌​​

One trick that helped us was moving to two‑DOF PI so setpoint steps don’t kick the meniscus; we kept our IMC gains but set ‘beta’ around 0.4 and added a 1–1.5 s low‑pass on the stopper feed‑forward, which cut nuisance stickers without dulling rejection. If you can, log a couple of small setpoint steps and compare error vs. control effort before/after to confirm you’re not starving the actuator. Does your controller expose setpoint weighting and feed‑forward filtering?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​​​⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‍‍‌​⁠‌‍​⁠​⁠‌​‍‌‌⁠‌‍‌⁠‌‌‌‍‍‌‌⁠‍‍‌⁠​⁠‌⁠‌⁠‌⁠‍​‌‌‌⁠‌‍‌​‌‌​‍‌‍‍​‌⁠‌⁠​‍​‍‌⁠⁠‌​​

And quick win for us was adding speed feedforward so the stopper pre‑moves on ramps; we fit Kff from a about 6% speed step in the historian and clamp integral inside ±1.5 mm so noise at the meniscus doesn’t wind it up. With a short 3.5 mm stroke that cut sticker alarms a lot — does your controller let you add a basic feedforward term alongside the PI for that “signal filtering using historian data” push?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​​​⁠‍​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌​​⁠‌⁠​‍​⁠​​‌​‌‍‌‌‌‌‌‌‍‍‌​⁠⁠​‍⁠‌​⁠‍​‌‍​‌‌⁠‍​‌​​⁠‌​​⁠‌​​⁠​⁠‌⁠‌​​‌​‍​‍‌⁠⁠‌​​

Closest fit I’ve seen for ‘hands‑on PID tuning’ is the pre‑AISTech continuous casting workshop on automation/control, and OEM‑wise Primetals (MoldExpert) and SMS group both run onsite mold‑monitoring sessions using your historian data. Ask them to deliver a two‑speed tuning matrix and an explicit deadtime ID so you’re not tuning by flashlight at 3 a.m. What level sensor are you on — EMLI or radiometric?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​‌​⁠‌‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌‍​‌‌‍​‌‌​⁠‌‌⁠‌⁠‌‍‌⁠‌‌‌‌‌‍‍⁠​⁠‌​‌​‌‌‌‍​‍‌‍​‌​⁠‍​‌‌‌⁠‌⁠‌⁠‌​​‌​‍​‍‌⁠⁠‌​​

, at “180 cpm” (about 3 Hz) and 3.5 mm stroke, a synchronous notch tied to the oscillator encoder cleaned up our meniscus trace far better than smoothing alone, which let us push I a bit without spurious sticker alarms. For training, ISA’s Advanced Loop Tuning let us bring our own caster trends and work through filters on real data; not caster-specific, but it translated well: Training - ISA. Do you have an encoder signal you can sync to for the notch?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​‌​⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌⁠‍‌‌‌‌​‌‍​‍​⁠​‍‌‍‌‌‌‌‌​‌‍​⁠‌‌⁠⁠‌‌​​‌​‌⁠‌⁠​​‌⁠‌‍‌‍‌⁠‌‍​‍‌‍​‍​⁠‍‌​‍​‍‌⁠⁠‌​​

And building on @nathaniel_smith22, if you can swing an OEM onsite clinic, push for a two-DOF PID (setpoint weighting) plus an integral hold when meniscus velocity exceeds a small threshold; that kept our controller from ‘chasing ghosts’ and cut stickers about 25% for us. Do you log EMBr current and torch trims in the historian? Mapping those as a disturbance input during the lab moved the needle more than heavier smoothing, not a silver bullet but quick to try.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​‌​⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌​⁠‍‌​​‍​⁠​‌‌​‍‍​⁠‍‌​⁠​​‌‌‍​‌‍⁠⁠​⁠​‌‌⁠‌​‌‍‍‍‌​⁠‌‌‌‍‌‌⁠‌‍‌⁠‌⁠‌‌​⁠​‍​‍‌⁠⁠‌​​

Before you pick a course, ask if they’ll run a short ‘replay’ lab with your historian data; we did this at 180 cpm by stepping stopper position about 0.5% during a speed hold, fit a quick FOPDT in PI/Seeq, and IMC‑tuned the PI so we walked in with gains that stuck. Caveat: we had to slow the filter a touch at higher casting speeds and recheck breakout thresholds afterward; would an OEM clinic let you do that with your data, @nathaniel_smith22?

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌⁠​⁠‌⁠‌​‌‍‌‌‌‍‌‌‌‍⁠​​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠‍‌​⁠‌‍​⁠​⁠​⁠‌⁠​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‍​⁠​​​⁠​‌​⁠​‍​⁠​​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌⁠‌⁠​⁠‍​‌​‍‌‌​⁠​‌‌‍‌‌‍‌‍‌‌​⁠‌‌‍‌‌‍‍⁠‌​⁠⁠‌​⁠‌‌​‍​‌​⁠‌‌‌​‌‌​‍‌‌‌‍​​‍​‍‌⁠⁠‌​​