Smart manufacturing for excellence in process plants: Part 1

Greg: Now, let’s get serious. In this month’s column, Houston, Texas-based Umesh Mathur, P.E., offers insightful practical guidance and explains how to achieve excellence in process plant automation.

What are the main ingredients?

Umesh: Thanks, Greg, for initiating this discussion. My list of the “clean dozen” is:

1. Sensor validation/smart sensors are essential tools for screening online data, and using it to compute critical performance parameters that make sense to operators and engineers alike.
2. Sample time recording is important because there are instances where lab sampling time isn’t recorded, so results in data historians are from the time the results were returned by the lab—possibly many hours later—rather than at the time of sampling. It’s then impossible to build reliable online inferential calculations, for example, for product quality.
3. Inferential calculations are built using validated online data and synchronized lab results. They can be valuable for minimizing product quality giveaway.
4. Regulatory control configuration minimizes poorly designed, or illogical, control loops in the basic (regulatory) control system (BCS). They pose a challenge to ensuring reliable plant operations, and are often found in manual mode.
5. Online calculations (DCS) include other calculations, such as heat exchanger duty or fouling resistance, which can help optimize exchanger cleaning cycles. Furnace efficiencies can pinpoint fuel-firing problems with burners or excessive air leaks.
6. Regulatory control tuning is important because I continue to find many instances of poorly tuned BCS loops, which come from not understanding process dynamics and/or PID control logic. This is inexcusable.
7. Equipment health monitoring accompanies online calculations, and can be useful for keeping track of equipment reliability, such as mean time between failures (MTBF).
8. Intelligent alarming comes into play because industrial accidents can be aggravated by alarm floods, in which low-level alarms cause multiple higher-level alarms to go off, hindering operator response to underlying problems. Intelligent alarm management systems are designed to eliminate alarm floods by focusing on the true, underlying problems. Emergency shutdown (ESD) systems must have redundant instrumentation (not used for process control) and an independent control logic execution capability, such as PLCs. Instruments and DCS control loops used for normal operations shouldn’t be used for ESDs.
9. Multivariable, model-predictive control (MPC) is field-proven in refinery and petrochemical applications. It enables effective decoupling of multivariable interactions inherent in most chemical processes (e.g., distillation columns). It also enables implementing of local economic optimization in these applications to ensure the unit is driven, minute by minute, to the most profitable constraint limits.
10. Rigorous reactor modeling. Economically optimal operation of chemical reactors represents a huge opportunity to maximize operating profit. First-principles, reaction kinetic models calibrated against plant data can then run under a nonlinear economic optimizer (with constraints) to ensure proper set-points for the BCS (e.g., feed rate, reactor pressure and temperature).
11. Closed-loop, real-time optimization (CLRTO) involves steady-state, plantwide optimizers designed to maximize profit, while respecting safety, environmental, process and equipment constraints. Dozens of such applications have been implemented successfully since the early 1990s.
12. Planning and scheduling optimization focuses on complex-wide, economic optimization (e.g., for entire refineries or petrochemical complexes). Mixed-integer, non-linear programming (MILNP) software is used most often for these purposes. Optimization questions such as feedstock selection, responses to seasonal marketing demands or feedstock availability, and decisions about using alternate processing plants can be made using simplified models.

Greg:What are the primary economic benefits?

Umesh: The strategies outlined above improve profitability, while ensuring that plant operations run safely and observe environmental, process and equipment constraints. Left to their own devices, operators often depart from the economically optimal conditions. We also see significant performance variations from one shift to the next. Ensuring the best results requires a coordinated campaign to assemble the proper teams from the technology, engineering and operations groups that work in concert with equipment and software vendors. Payback is usually within two to three years.

Greg: What are the value and methods of sensor validation?

Umesh: They add economic value by maximizing unit availability and operational reliability with remarkably high return on investment (ROI).

Greg: What are the benefits of smart sensors?

Umesh: Properly implemented sensor validation techniques can prevent needless shutdown of BCS loops and higher-level applications such as MPC and CLRTO. They can be valuable to refinery applications as diverse as fuel blending and storage tank emissions control.

Greg: What’s some key guidance on practices and uses of process samples?

Umesh: Online analyzers are always preferred over laboratory sampling for critical applications. For properties not measurable online (e.g., kerosine or jet fuel smoke point, or diesel cetane number) inferential calculations or soft sensors can be developed that provide good estimates of instantaneous online product quality. Data from laboratory samples is used to provide feedback updates for ensuring the reliability of inferential calculations.

Get your subscription to Control's tri-weekly newsletter.

Greg: How do we achieve the best inferential measurements of product quality?

Umesh: Using all relevant, historical, online information (F, T, P, x), with time-stamped laboratory results from the data historian (e.g., PI or PHD), techniques such as partial-least squares, neural networks, equation-based/dynamic simulation models and other proprietary techniques have been deployed successfully for many years.

Greg: What are some examples of inferential measurement calculations?

Umesh: There are innumerable examples, including distillation product purity, petroleum fraction D-86 specifications such as the 5% or 95% temperature, gasoline octane number, jet fuel color, kerosine smoke point, asphalt penetration index and polymer viscosity.

Greg: What are some improvements in regulatory control configuration?

Umesh: There are countless situations where dynamic simulation can help devise more effective control strategies. In distillation columns with a high reflux ratio (ethylene or propylene superfractionators, deisobutanizers or styrene fractionators), the reboil ratio is high. 

I’ve witnessed instances where column bottom level was controlled using the bottom flow. By changing the logic to place the bottoms product on flow control, and managing level using the reboiler duty, I enabled these applications to reduce product quality fluctuations dramatically, while also improving the disturbance rejection capability (changes in feed rate or composition, effects of rainstorm deluges).

In fired heaters, instead of adjusting fuel firing to maintain a heater outlet temperature, I switched to a strategy of maximum fuel firing (within firebox draft constraints), and adjusting feed rate to manage heater outlet temperature. This improved fuel burner management, while eliminating issues with rapid changes in furnace fuel firing and damper positioner controls. Making such changes requires deep process knowledge, coupled with experience in basic and advanced (model-predictive) control project implementation.

Greg: In future columns, Umesh will offer his knowledge on important aspects in regulatory control, equipment health monitoring, safety systems, multivariable control and real-time optimization.