Closed-Loop Universal Multivariable Optimizer with Artificial Intelligence (AI) for Model Predictive Control (MPC)

Colombo easily connects to any DCS or PLC using OPC-DA server. Connectivity is quite easy and there is no need for installation of complicated external MPC software that requires a lot of configuration, testing and training for both engineers and operators.

Colombo can connect to any commercial DCS or PLC, including but not limited to Allen Bradley, Siemens, Honeywell, ABB, Yokogawa, Emerson, Foxboro, or any PLC/SCADA system that is equipped with an OPC-DA server. Unlike other MPC competitors that require a complex watchdog for communication between the DCS to the MPC system, Colombo approach is far simpler, intuitive and requires much less time; typically, one-third of the time required by other commonly known MPC technologies.

Colombo uses a powerful nonlinear artificial intelligence-based optimizer that requires less than one-fifth of the data required by conventional MPCs for step testing. While all other known conventional MPCs from all MPC vendors worldwide require step test data comprising of conventional step tests, Colombo can analyze closed-loop data with the slave PIDs not only in auto mode but even in cascade mode. Columbo can even use oscillatory closed-loop data and identify open-loop dynamic models needed by the MPC algorithm. Columbo can also successfully use data where setpoints were changed by the operator for the normal plant operation, including data when the operator needed to ramp set points. Based on the ramping of normal plant operational data, Colombo can identify open-loop dynamic models from complete closed-loop data and use the models in the MPC algorithm.

Internally, Colombo has a proprietary AI (artificial intelligence) based system that integrates AI with MPC optimization action that helps the plant to maximize production rates while minimizing utilities and other variables and providing full plant automation to help the operators in reducing their work-stress in their job and allowing them to focus on other important errands and job duties. The AI-MPC integrated component generates smart messages and detects abnormal events and automatically turns off the MPC rather than causing a plant upset. In most common MPCs, operator mistakes- e.g., entering a wrong setpoint, putting one or more slave PIDs in auto or manual for calibration reasons may result in the MPC solution to flip and cause unwanted, undesirable or even dangerous control action. Just by the nature of Columbo, this cannot happen – Columbo will turn itself off, not produce harmful MV (manipulated variable) trajectory and generate AI-based advisory and warning messages for the operator.

Columbo is superior to all conventional MPCs in the way it handles nonlinearities and unmeasured disturbances. Conventional MPCs suffer from having difficulties while tackling nonlinearities and unmeasured disturbances that are common in all industrial processes. If the conventional MPC models are not accurate, or because of nonlinearities, or if there are unmeasured disturbances, experienced MPC users and control room operators get frustrated with conventional MPC’s poor feedback control action. This is evidenced by noticing a persistent offset (persistent error between the CV- controlled variable and its target) and often reminds one of a PID control analogy with weak integral action. Also, operators of conventional MPCs are frustrated and harassed by seeing MPC future predictions, predicting that the variable that is violating a high limit is nicely going to come down, but in reality, it stays high and the predictions continue to predict that everything is going to be fine but does not.

In contrast from the above problems and frustrations in conventional MPCs, Colombo provides AI-based aggressive, fast disturbance rejecting action thus improving the controller quality much better than conventional MPCs. Columbo control action is fast, aggressive, appropriately impatient upon encountering persistent deviations like the case of ramp disturbances. The AI-based algorithm in Columbo will make the MV control action temporarily “more impatient” resulting in faster and larger moves in the MV, resulting in the violating CV to return to its target limits faster than a conventional MPC.

In a conventional MPC, there is a global switch that turns on everything or turns off everything. With this switch, all configured preset MVs (manipulated variables) and all configured and preset CVs (controlled variables) are all turned on or off (active or inactive). In case one or more MVs are in manual or auto (not in cascade, not ready to accept a setpoint from the conventional MPC), then the MPC solution can go completely crazy in case the scenario encountered is kind of rare or was not simulated during the MPC testing and development. Absence of one or more MVs or CVs can cause the controller solution to go wild and cause MVs to flip from their high limits to low limits or vice versa. The operator may encounter this on a weekend or night and may be completely confused as to what may be happening. Or worse still, after turning the global switch to on (active), he may be too busy to wait and watch what the MPC did or is doing and only after half-one hour, after off spec alarms come or some critical limits are violated, he will be alerted that the MPC did bad things. So many plants worldwide have been struck by this problem caused by conventional MPCs where one global switch turns everything on and active. There are just too many scenarios based on the large number of the permutation and combination of how many and which MVs and CVs are active and inactive. All this can cause confusion and stress for the operator and a major upset or shutdown impacting the plant operation and efficiency. Six months of increase in profits by 3% can all be washed out and lost in one bad event caused by confusion as described above.

The Columbo project execution plan is much faster, easier and more convenient compared to a typical conventional competitor MPC. PiControl will work with the onsite control room engineers completely remotely or onsite and first discuss the process, how the operator runs the plant. With this information, PiControl will develop the Columbo MPC control matrix. Then PiControl engineers jointly with control room engineers will collect the old data from the plant historian or conduct a few new step tests as needed. The time needed for step testing with Columbo is typically one-fifth of what is needed by competitor MPCs. Then, PiControl will use Colombo-Pitops software to determine the open-loop models for the entire Columbo MPC controller. The next step is to build, configure and program the controller, all tasks done completely by PiControl engineers. During this step, PiControl will develop a plant simulation simulating the MV-CV model interactions. This helps to test the controller and make sure everything is correct and ready for field commissioning.

At this stage, PiControl will build a simulation with dummy tags for testing the whole Columbo MPC controller. With the simulation, PiControl is able to do a Hazop and “what-if scenario” study of the controller action. This action is demonstrated to the customer and then with the help of an experienced operator, many normal, abnormal and safety-related scenarios can be tested for production rate maximization, utility minimization, steam minimization etc. PiControl will also simulate scenarios like sudden drop in the flow or temperature signal in a slave PID controller because of signal malfunction or signal failure. Or another possible scenario included in the Hazop testing is when an operator erroneously enters a wrong set point – e.g., the operator enters a value of 30 instead of an intended value of 300. When such an abrupt change or abnormal condition like signal failure or operator error are noticed, then PiControl Columbo AI module will take over and will automatically turn itself off without causing any harmful action in the controller.

PiControl engineers then will come to site or work remotely sharing screens using remote meeting or webcam and commission the Columbo MPC controller and perform fine tuning and training at the plant site.

The whole process of Colombo is quite novel, attractive, convenient and cheaper. The project execution is faster, easier, convenient and overall, there is less time needed for maintenance on an annual basis overall. Another benefit of Columbo is that there is no new complex hardware servers needed, no new software, no need of a powerful and expensive dedicated MPC server, so there is less work for systems engineering, IT personnel and there are no tasks in Columbo related to constant Windows upgrades, software upgrades, and server refresh. All these hassles, headaches and costs associated with typical MPC systems can be avoided by using the new Colombo AI-MPC technology. Contact PiControl Solutions today and apply Colombo MPC software to maximize production rates, minimize utilities at your plant and maximize profits in your plant. Website:www.picontrolsolutions.com, email: info@picontrolsolutions.com.