Closed-loop model identification
Identifies open-loop dynamic models from complete closed-loop data, no step tests required.
COLUMBO - the Closed-Loop Universal Multivariable Optimizer - is closed-loop MPC model identification software from PiControl Solutions that also runs as a compact, modular, PLC-based multivariable predictive controller. COLUMBO identifies and refreshes the dynamic models behind an MPC directly from historical closed-loop data, without conventional step tests, and it can be deployed as a new multivariable controller running wholly inside an existing PLC or DCS, with no separate host server required.
Dynamic models are the heart of every MPC. As feed composition, catalyst activity, and operating conditions drift, those models go stale and MPC control quality deteriorates, often leading to controllers that get switched off with the benefits they were installed for. Re-identifying models the conventional way means one to two weeks of disruptive open-loop step testing. COLUMBO removes that requirement by identifying models from data the plant already has.
COLUMBO identifies and refreshes multivariable MPC dynamic models directly from closed-loop plant data, and it can deploy a compact, modular MPC that runs inside your existing PLC or DCS. Because it works from data the plant already collects, COLUMBO closes the gap between how conventional advanced process control is maintained and how plants actually operate.
Identifies open-loop dynamic models from complete closed-loop data, no step tests required.
Deploys wholly inside an existing PLC or DCS, with no new Level 3 server.
Each manipulated variable can be switched in or out without affecting other MVs.
Separates steps, pulses, and ramps from the true process dynamics during identification.
Blends identified dynamic models with AI to strengthen disturbance rejection and robustness.
Improves models inside Aspen DMC, Honeywell RMPCT, and other commercial MPC systems.
Plants use COLUMBO to deploy a compact, modular multivariable predictive controller directly inside their existing PLC or DCS logic, rather than standing up a new host computer, OPC layer, and standalone operator interface. Because each manipulated variable is independently enabled or disabled, operators can pull an MV for maintenance without the cross-MV upset risk that conventional MPCs carry.
Where an existing MPC's control quality has degraded, COLUMBO analyzes closed-loop data with the controller still active, slave PIDs in cascade, and identifies improved dynamic models without taking the MPC offline for step tests. It refreshes the models behind FCC, hydrotreating, and other advanced controllers as feed and catalyst conditions drift.
For the engineering story behind COLUMBO's identification method, see how COLUMBO identifies MPC models from closed-loop data.
COLUMBO's feature set is built for one outcome: multivariable models that stay accurate, without disrupting production to get there.
Identifies open-loop dynamic models from complete closed-loop historian data, with slave PIDs in auto or cascade.
No conventional one-to-two-week step-test campaign, and no production risk to get an accurate model.
Separates steps, pulses, and long-running ramps from true process dynamics and displays the residual trend.
Disturbance-contaminated data is the usual root cause of bad MPC models; isolating it protects model accuracy.
No external Level 3 host server, no additional OPC layer, no new operator interface to install.
Removes the systems-integration work that stretches out most MPC projects and support burden after go-live.
Each MV moves to cascade or auto/manual independently, with no global on/off or prediction/standby switch.
Operators can pull an MV for maintenance without the cross-MV upset risk conventional MPCs carry.
Uses AI in a limited, targeted way alongside identified models to strengthen disturbance rejection and robustness.
Avoids the extrapolation and repeatability failures full black-box AI controllers hit on real plant disturbances.
Improves transfer-function and step-response-coefficient models inside DMC, RMPCT, PredictPro, PACE, SMOC, and Connoisseur.
Fixes the MPC you already have without a rip-and-replace project or a new vendor relationship.
Every MPC's control quality depends on the accuracy of its dynamic models, and those models do not stay accurate on their own. Feed composition, catalyst activity, and operating conditions drift continuously, and the conventional way of fixing a stale model, turning the MPC off and re-running open-loop step tests, is exactly the disruption plants installed the MPC to avoid.
Conventional model identification means one to two weeks of small, deliberate moves on each MV in open-loop mode, one at a time, while the plant waits out the settling time. Many plants defer this work indefinitely rather than accept the disruption, and the MPC keeps running on a stale model in the meantime.
Conventional identification methods, including ARMAX and Box-Jenkins approaches, work poorly on data with unmeasured disturbances mixed in, whether fast noise, medium-frequency drift, or slow ramps. Contaminated data produces a wrong-shaped model, and a wrong-shaped model produces poor MPC predictions.
Most conventional MPCs use a single global on/off or prediction/standby switch. Disabling one MV for maintenance can leave the controller trying to manage a CV with the wrong MV, and many plant upsets trace back to an operator who did not fully understand how the multivariable system was coupled.
COLUMBO removes the structural problem: model identification from closed-loop data the plant already has, unmeasured-disturbance isolation built into the identification method, and per-MV control that will not cross-couple an operator's maintenance action into another loop. For MPCs that need fixing today, PiControl's MPC maintenance and improvement service builds a structured engagement around COLUMBO.
COLUMBO handles multivariable MPC model identification and modular PLC-based MPC. It works alongside the rest of PiControl's process control software suite to cover a plant's PID and multivariable loops end to end.
Where COLUMBO identifies and maintains multivariable MPC models, PITOPS handles closed-loop PID tuning and single-loop system identification, the foundation those MPCs are built on top of.
The slave PID loops that COLUMBO's MVs cascade through stay tuned with SUPERTUNE online PID auto-tuning, so the base regulatory layer keeps pace with the multivariable layer above it.
APROMON real-time loop monitoring watches the PID and MPC loops COLUMBO maintains, flagging oscillation and model drift early enough for a COLUMBO model refresh before control quality erodes further.
Plants across every sector PiControl serves use COLUMBO to keep multivariable MPC models accurate, or to deploy new modular, PLC-based MPC without new host hardware.
COLUMBO is part of the engineering practice behind PITOPS, SUPERTUNE, and APROMON: 50+ MPCs re-stabilized across 200+ industrial facilities since 2005, on-site and remote, worldwide. Read more in our customer success stories.
Request the COLUMBO overview brochure and project details for your plant's MPC.
See COLUMBO identify a multivariable model from your own closed-loop data. A PiControl engineer will demonstrate the software on loop dynamics relevant to your plant and map out a model-refresh or new-deployment path for your MPC.