Many MPCs (model predictive control) systems fail to deliver the promised and expected benefits:
Many MPC (model predictive control) systems start with a lot of excitement, hope and expectations. But after installation and project commissioning, many fail to deliver the expected monetary, tangible and intangible benefits. There are many reasons why MPCs do not deliver the expected benefits. These are:
- Difficulty in identifying models. In some cases, large step tests cannot be made as the process may go off-spec or running the plant perilously close to serious equipment, process and operating constraints. The small steps on the MVs (manipulated variables) are washed by the impact of unmeasured disturbances resulting in poor or outright bogus dynamic process models. Subsequently, the MPC control action is poor.
- Applying MPC to processes with frequent, large and unmeasured disturbances. MPCs do not work well compared to APC (advanced process controllers) and EPC (expert predictive controllers) and it is an inherently bad choice to use MPC like DMC (dynamic matrix control) from Aspen Tech, RMPCT from Honeywell (robust model predictive controller technology), PredictPro from Emerson, Connoisseur from Foxboro Schneider Electric, PACE or SMOC from Yokogawa and Shell Global Solutions or any other known MPC system. MPC feedback control action is rather weak compared to APC, EPC and other multivariable controllers. Subsequently, the MPC predictions show nice and comforting steady state trends down in the future, but as time goes by, the CVs (controlled variables) are moving very embarrassingly in the wrong direction much to the annoyance and ridicule of the control room operators.
- Ignoring base-level PID control tuning monitoring and optimization. After an MPC project is awarded, it is considered like magic, Utopia or something totally amazing. Some believe with MPC, it is not important to check and tune primary base-level PID control loops. Also control valve problems like control valve stiction, hysteresis is ignored or not considered worth checking. These are basic, fundamental and critical faults and will always result in a poor MPC.
- Relying on dynamic transfer function models or dynamic step response coefficient models based only on data analysis and open-loop model identification. This is a serious limitation with MPC design as done by inexperienced control engineers. Using chemical engineering knowledge, knowledge from an experienced operator and then using this information to improve the dynamic model is an advanced and a more modern development compared to old-fashioned MPC as typically implemented by Aspen Tech, Honeywell, Emerson, Foxboro, Schneider Electric and Yokogawa Shell MPCs. With EPC, chemical engineering knowledge, geometric dimensions, operator and process engineer knowledge are all used to improve the dynamic process models and the control performance with EPC is better than MPC.
EPC -Expert Predictive Control system – new APC/MPC technology from PiControl Solutions:
PiControl Solutions LLC (www.PiControlSolutions.Com) has been implementing APC (advanced process control system) and EPC (expert predictive control system) with control action and benefits superior to traditional MPC (model predictive control system). Benefits of PiControl EPC:
- The process knowledge and other specific and useful knowledge from an experienced operator or process engineer are directly used in the generation and finalization of dynamic process models.
- Both closed-loop data analysis and open-loop data analysis are used to determining dynamic process models used in EPC. PiControl Solutions LLC is the only company in the world that can take data with slave PID control loops in not only automatic mode but also in cascade mode. PiControl software is both multivariable and can use completely closed-loop data for determining process models. The ability to use closed-loop data with multivariable changing inputs, allows PiControl to determine dynamic process models more accurate than current MPC model identification software from Aspen Tech, Honeywell, Yokogawa, Emerson, Foxboro Schneider Electric and other MPCs. With current MPCs, small step tests are made with the base-level slave PIDs in auto mode, and the step size is only 1-2% around the prevailing setpoints. If bigger moves are made then they can upset the process or will not allow holding the step for 0.33 to 1.5 times the TTSS (time to steady state) and then the gain of the transfer function model or step response coefficient model may be inaccurate. PiControl technology is unique and a one-of-a-kind in the world because it can identify multiple open-loop process models with large moves, simultaneous moves and models identified in this manner are more accurate than the models identified using small steps.
- PiControl technology is able to even control online analysis on processes that have just lab samples and do not have online analyzers nor inferentials. PiControl uses unique proprietary technology and many years of experience to allow such functionality.
- While the feedback action in MPC is weak in response to unmeasured disturbances, PiControl’s EPC (expert predictive control) can produce a strong initial kick, mimicking lead factor in a feedforward or a large proportional kick in a slave PID control loop or a cascade PID control loop and this helps to provide superior control than a conventional MPC.
Many MPCs do not work well for complex and nonlinear distillation super-fractionators due to bad or missing models. PiControl EPC works well in these cases. PiControl EPC has been successfully applied on styrene monomer plants, olefins plants, MTBE plants, electric power plants, LNG plants, polymers – polyethylene, polypropylene and elastomers, cyclohexane, para-xylene, polyester and many others.
Migrate from conventional old MPCs into the new world of EPC for 2022 and beyond. Expert Predictive Control is the new upgrade and new version for the old conventional MPC – Model Predictive Control from the 1980s.
Contact PiControl Solutions LLC today – [email protected], www.PiControlSolutions.Com
