Beware of Model Predictive Control (MPC) and Distributed Control System (DCS) companies trying to sell you expensive software that you do not need.
We live in a world full of products and services. Many products and services are good and fall in the win-win category where the customer needs them and the supplier provides them. Both the customer and the supplier win and benefit fairly from each other.
But in order to further maximize their sales, suppliers often resort to unscrupulous tactics, selling products and services knowing fully well that these are not needed by their young, inexperienced and ignorant customers.
Selling ice to Eskimos, selling perfume to passersby at an airport, using beautiful looks to seduce, tempt and surreptitiously sell products unwanted by male customers, taking advantage of ignorance and vulnerabilities in potential customers to make more sales, we can go on and on when it comes to unscrupulous salesmen and saleswomen.
Do these unscrupulous salesmen and women only fleece the large-scale consumer industry? No, not at all.
We were surprised to see how the large DCS and MPC vendors in the industrial process control industry fleece the engineers and managers in chemical plants and factories.
MPC is abbreviation for Model Predictive Control. Let us briefly look at the history of MPC evolution. In the 1980s, Dr. Charles Cutler, a Shell employee (Royal Dutch Shell, the mega oil, gas and chemical company) during his doctorate PhD thesis developed the DMC algorithm (dynamic matrix control algorithm). It was an instant enormous success in the oil refining industry where many of the process units had many interactive characteristics where several Manipulated Variables (MVs) impacted several Controlled Variables (CVs). An excellent example in the Crude Distillation Unit which has many feed streams and many side draws. A change in any of the feeds or any of the side draws impacted all the temperatures on the body of the column. This is a perfect, classical, textbook multivariable control problem where Dr. Charles’s Cutlers DMC algorithm was perfect.
With DMC's outstanding success in the 1980s, many non-expert and non-elite process control engineers backed by their ignorant and non-technical admin managers rushed to apply DMC to all conceivable processes. Some worked and some did not depending on the nature and the characteristics of the process.
Ignorant and inexperienced engineers applied to totally misfit processes like exothermic polymer reactions which were dominated by strong and frequent unmeasured disturbances and nonlinearities.
When MPC projects failed when applied on the wrong type of processes, the control engineers and the control managers did not and could not afford to tell their bosses that the projects were a disaster since someone would have to admit a loss of several 100s of thousands of dollars of money allocated on the project. A lot of hush hush activity concealed the MPC project failure. This was a boon for the MPC salesmen as they were not held accountable. Control engineers and managers tried to sing praise and success even though the benefits were minimal or even zero. No one was held accountable. The company lost several 100s of 1000s of dollars on the project and the MPC and DCS companies got away Scot free.
Our research shows almost 30-50% of the MPC projects fall in this above category. Many MPCs are implemented on processes where MPC is not the right strategy to begin with. To understand when to apply MPC and when not to apply MPC needs a lot of skill and experience. In today's world, many experienced control engineers have retired and a lot of new engineers do not know when to apply MPC and when not to apply MPC. Couple this situation with unscrupulous MPC salesmen and saleswomen and we have a perfect storm combining MPC salesman's greed, control engineer's inexperience, process control manager's ignorance and result in a flushing money down the toilet.
MPC from Aspen (DMC - dynamic matrix control), Honeywell (RMPCT), Yokogawa (SMOC and PACE), Emerson (PredictPro), Foxboro (Connoisseur) should not be applied to processes that have the following characteristics:
1) Process is in series where the fluids flow serially from upstream to downstream with little or minimal interaction and multivariability. A good simple example is 5 distillation columns in series, each column has a single feed, single reflux flow and a single reboiler. In this example, there is no multivariability and the APC can be designed well and implemented all inside the DCS or PLC with no need for OPC, no need for MPC, no need for new servers, no need for new hardware and software.
The control matrix is diagonal and the control matrix density is very low. Often MPC vendors push and successfully sell MPC software, servers, hardware, training, 5000-page MPC manual and create hours and hours of endless work for system engineers when all MPC could have been developed and implemented all inside the existing plant's DCS.
2) Disturbance dominated processes - processes like polymer processes and LNG (liquified natural gas) are disturbance dominated. In a polymer process, catalyst is coming in bags and can have a varying quality impacting the chemical reaction. In an LNG plant, changes in wind direction, ambient temperature impacts the production and refrigeration a lot. MPCs like DMC, RMPCT, SMOC, PACE, PredictPro and Connoisseur do not provide the strong kick necessary to quickly reject unmeasured disturbances like what can be done inside DCS-resident APC.
Only 20% of all industrial processes need a full blown MPC like DMC, RMPCT, SMOC, PACE, PredictPro or Connoisseur. The remaining 80% of all industrial processes can be well controlled by DCS-based APC and the performance and benefits can outperform and put to shame an MPC.
If your process does not need MPC and can be controlled using DCS-based APC, you will enjoy the following benefits:
- Runs inside existing plant DCS
- Based on standard and custom DCS blocks and some code/calculations
- No need to buy new hardware
- No need to buy new servers
- No need for expensive and cumbersome software
- No need for special software training nor maintenance, remote help easy
- Fewer step tests or even closed-loop historic data can be used to build models
- Low initial cost, low initial effort and low long-term maintenance
- No mandatory annual fees – there is no software and no server
- Cheaper than MPC by 2-3 times
Cheaper than MPC by 2-3 times
If you choose MPC instead of DCS-APC, then you will suffer from the following:
- Need new hardware and servers
- Developmental effort much higher
- Constant software patches, hardware refresh programs add up to be very expensive both at start and ongoing
- Model development needs week-long step testing
- Maintenance is tedious and needs experienced folks at site
- Model updates and tuning often need site trips by MPC vendor
- More expensive than DCS-based APC by 2-3 times
We are not saying MPC is bad and not needed. No! MPC is wonderful and irreplaceable in the case of oil refining units like FCC - fluidized catalytic cracker, Hydrocracker, Crude Distillation Unit (CDU), Primary Fractionator. For such processes, MPC is great and applying DCS-based APC would be clumsy, convoluted and not recommended.
However, in many processes like LNG, polyethylene, styrene, hydrogen, industrial gases, specialty chemicals, distillation columns in series etc., MPC is not needed and a DCS-based APC project will be not only cheaper, faster but will actually provide more benefits than an MPC project.
Beware of salesmen from DCS and MPC companies - they may tempt you into spending US$300K to US$1 million for MPC software and hardware that you do not need and you could do the same inside your existing DCS at one-fifth the cost. Contact PiControl today to evaluate your APC or MPC opportunity, PiControl is vendor neutral and will give you an honest answer - [email protected], www.PiControlSolutions.Com.
Excellent blog. Thanks