A financially successful advanced process control (APC) project requires a project manager with skills to adequately qualify the project, negotiate contracts, and shape project schedules. In most APC efforts, the volume of information is not large, but a precise understanding of process characteristics and performance objectives is critical from the beginning, since it is the basis for application design.
With large system projects, procedures for accurately and efficiently gathering and using large amounts of configuration information are critical to overall success. A large portion of the project manager's efforts must be devoted to overseeing this data flow, while information related to process response characteristics is relatively unimportant until commissioning.
How to qualify a project
You have to pick your battles. The most important step in achieving a successful APC project is picking the right project in the first place. A good candidate will have four essential characteristics.
1. Significant economic opportunity. Previous installments in this series have presented techniques for evaluating the economic benefits from an APC project. The same techniques can provide estimates to justify the cost of a project.
The contracts section below presents additional information on setting performance objectives. But typically, increasing production will provide the largest benefit, by far. An APC effort that can recover unused unit capacity by operating against variable constraints without requiring any process modifications will almost always provide a very significant return on investment (ROI) and easily justify itself in a very short time, often a few weeks.
In estimating the constraints for these benefits, it is critical to consider all the operating ranges for material and energy flows that can limit the process production rate. First, the designer must consider the capacities of the pumps, valves, fans, and conveyors that move material through the process. Next, consider the measurement range of process instrumentation. Third, consider unmeasured flows, such as heat transfer to the environment, as well as other physical limits, such as equipment operating pressures and temperatures, along with required residence times for process reactions, which can decrease throughput.
It's also necessary to consider other non-physical bounds, such as regulatory limits and permits for emissions. In one case, a significant study was conducted to justify the installation of an advanced control system for kiln control, only to find out at the last minute that the plant did not have a permit for firing the higher fuel rates that would have been required.
2. Sufficient information. Justifying and designing an advanced control system requires some very specific physical and economic information. Before beginning such a project, ensure the following information is available or can be obtained.
Accurate descriptions of process flows, operating conditions, procedures, and limits;
Historical operating data that is adequate in amount and resolution to reveal the variation in process operating conditions and dynamic characteristics;
Accurate documentation of the existing control system, including P&IDs (piping and instrumentation drawing), instrument ranges, current DCS (distributed control system) configurations, sequence documentation, etc.; and
Relevant cost factors for materials, energy, and product values.
3. Motivated personnel. To be successful, an APC project requires strong technical cooperation between vendor and user. Operators must be willing to accept new controls and techniques for running the plant. Management support at plant and corporate levels will probably be required, as well as engineering support at plant level. Unless a technically qualified person from the plant is fully involved in the project and takes ownership of the system after commissioning, the project is likely to end with disappointment. Execution will be difficult for both sides, and the system is likely to fall into disuse after commissioning due to lack of understanding, maintenance, operator training, or support.
4. Adequate field instrumentation and a capable control infrastructure. An advanced control system relies on a solid foundation of field equipment and regulatory controls. Instruments and actuators must be calibrated accurately and in good operating condition. Regulatory control loops must be properly tuned and operate normally in automatic control. The DCS must be capable of performing all necessary advanced functions that may be required, such as external integral feedback and supervisory set point control. It must be open to networking if the advanced control software resides in another platform.
Advanced control strategies often use many more inputs than basic regulatory controls, so additional instrumentation to measure these new constraints may be needed. More difficult and expensive measurements of economic variables from analyzers may also be needed. A high quality data historian and information network will be required if the project scope includes proving performance of the system.
A capable network will also support remote monitoring of the system after commissioning. Advanced control applications need support to maintain their performance. Given the increasing cost of onsite engineering, the budget for post-commissioning support from the vendor is likely to be limited. Remote monitoring is a cost-effective way to minimize this cost and keep the stream of economic benefits flowing.
Sign a good contract
The prior article on proving system performance presented basic techniques for measuring control system performance. The difficult part is setting reasonable goals for these metrics. Economic justification is calculated from expected changes in operating conditions, but how much change can be reasonably expected
The most common approach is based on the normal distributions discussed in the earlier article on optimization and control. The mean and standard deviation are calculated for a representative set of historical data. The assumption is made that an advanced control application can reduce standard deviation by half. This allows moving a set point one standard deviation closer to specification or operating limits without increasing the amount of off-specification product.
This shift can provide a reasonable estimate of a potential improvement in performance metrics. However, one important point is often overlooked. These concepts, along with concepts of process capability, are only rigorously accurate for normal data distributions.
A distribution that is “normal” is a symmetrical bell-shaped curve with a specific mathematical form. Not all bell-shaped curves are normal. Two additional statistical indicators, skew and kurtosis, can measure deviations from normality. Skew measures lack of symmetry in the data distribution. A distribution that is extended in the negative direction has a negative skew, and vice versa. For normal distributions, skew = 0. As the first two figures indicate, positive and negative values can indicate process non-linearities and operator intervention.
Kurtosis is a measure of whether a distribution is more peaked or flat than a normal distribution. For normal distributions, kurtosis = 0. A taller, more narrow distribution has a positive kurtosis value, while a flatter distribution has negative kurtosis. In the context of control, a high kurtosis can indicate tight control performance while negative kurtosis can indicate loose control, perhaps from poorly tuned controllers. Still, high kurtosis can also indicate excessive signal filtering, transmitter insensitivity because of poor location, or deposits that isolate the sensor from the process.
Process data is almost never normally distributed. Human behavior, process characteristics, and mechanical influences always distort the distribution. So, characterizing these distributions with simple averages and standard deviations can be quite misleading. The graphics illustrate.
“Effect of operator actions,” shows 10 minute samples for the fuel flow to a lime kiln over a two-week period. At irregular intervals, the fuel flow is either fixed or suddenly drops. These are periods when the operator put the fuel flow controller into manual and either left the output constant or dropped it abruptly in response to current operating conditions.
The data distribution resulting from this sample will have a negative skew and the average will be drawn down by the sharp negative dips.
A out-of-control process is an operator's worst fear, so it's only human nature for operators to take actions that push a process in the safe direction more drastically than they push it towards higher performance. Often, operators wait too long to deal with a developing condition, hoping that it will correct itself or remain minor. Then when action is finally unavoidable, a correction has to be large and abrupt. Conversely, moves toward less stable or difficult operating conditions are usually done gradually, and often reluctantly.
Data in “Effect of process non-linearities” shows hourly samples for residual carbonate in the product from a lime kiln over a two-week period. The normal value is relatively low, but upsets cause positive excursions that are quite significant. The resulting behavior of this variable is highly non-linear. The amount of residual carbonate cannot go negative, so the lowest possible value is zero, but strong disturbances can drive it relatively high.
In this case, non-linearity will create a distribution with a positive skew, and the average will be higher than the normal operating value. The high peaks also increase standard deviation, relative to variation under normal conditions.
The third graphic charts 10-minute samples of the feed flow to a lime kiln over a 2-week period. This is a difficult measurement and the process flow is often interrupted for mechanical reasons. Consequently, the graph shows a great many drops in the feed rate reading, some of which are real while others are only transmitter failures.
In either case, the result is that the mean will be lower than the normal value, while the standard deviation indicates more variation in the process than actually exists.
Historical data has to be carefully examined and processed to remove as much misleading information as possible. Shutdown and startup periods have to be filtered out, along with anything corrupted by transmitter or other equipment failures, if it can be identified. The data examined must represent normal operations. Avoid periods using alternative or supplemental fuels, unusual process configurations, off-specification feeds, and other abnormal conditions. Those must be identified and eliminated from the dataset. Engineering judgment and process understanding are the key to this step.
After data processing, statistical analysis of the remaining data provides the best basis for predicting potential benefits, subject to engineering interpretation. Random linear effects create normal data distributions. So, if a distribution is not normal, there is always a reason that needs to be taken into account while evaluating the opportunity and setting goals for economic benefits. Failure to recognize these influences can easily lead to unreasonable goals that set up a project for economic failure, even if it is a control success.
Today's economic environment for advanced control projects is very difficult. Control system hardware is becoming a commodity item. At the same time, increasing standardization in communications protocols and open operating systems means that users can shop around to “mix and match” components of control systems from many vendors.
In response, vendors have been forced to drive down prices for hardware as much as possible and differentiate themselves from one another through project services and the benefits that their designs provide. Naturally, promising larger benefits makes a vendor's proposal more attractive. But while promising benefits is easy, delivering them can be quite difficult.
This creates a strong relationship between risk and price. At one extreme is a contract that only requires the vendor to deliver a functioning system, rather than specific economic benefits. In this case there is almost no risk to the vendor because the bill gets paid no matter how well the system actually performs. The vendor can offer a rock-bottom price when the user assumes all performance risk.
This may seem like a good deal for the vendor, but it is a lose-lose proposition. Vendors lose because to be competitive, they often have to sacrifice margin just to get the job, even supplying the system below cost. The user loses because the a lowest-cost system will have only minimum functionality and scarce engineering support during and after the project.
The other extreme is a contract that requires the system to deliver a specific level of economic benefits or there will be no payment at all to the vendor. This shifts almost all performance risk to the vendor, since there will be no compensation for the considerable costs of engineering, commissioning, testing, and removing the control system if all promised benefits aren't actually achieved.
This may seem like a good deal for the user, but it is also lose-lose proposition. To be competitive, vendors are forced into making risky promises while taking chance that another vendor will take a higher risk by quoting a lower price or even bigger promises. For users, there is considerable risk that time and opportunity will be lost when an under-funded or over-promised system doesn't quite meet expectations. Prices for even simple applications must be higher to compensate for others that are removed when they don't make the cut. Easy applications end up subsidizing the risky ones.
A pay-for-economic performance contract is a compromise that is good for both parties. Unless the system is poorly engineered, total failure is unlikely. If targets are only partly met, the user pays less while the vendor can still cover basic costs. When performance exceeds its targets, benefits to the user are greater and the margin for the vendor is higher. These mutual incentives change the user-vendor relationship from a zero sum game with a winner and a loser into a win-win form. User and vendor maximize gain and share pain. This is the best arrangement for a long-term relationship between user and vendor.
Another advantage of a pay-for-performance contract is that the testing it requires demonstrates cash flow gained from application of the system. With this information, it becomes reasonable to fund a project out of the benefit stream it creates, instead of having to fund it from capital budgets for which there is so much competition.
Every project needs a schedule to coordinate and prioritize necessary tasks and to command necessary resources of people, time, and money. But scheduling an APC project has some unique characteristics that need to be taken into account.
Every APC project should begin with a control study to qualify the project, using the perspectives outlined earlier. A study helps the user avoid committing resources to poor projects, provides a sound basis for critical early decisions, identifies obstacles, reduces costs, and provides reference for evaluation. (See “Make better decisions; do a control study.”)
APC projects require a high degree of iteration and review, so a realistic schedule must make appropriate allowance. At the beginning of a project, the necessary functionality and proper design will not be clear. Successful and efficient completion of a project always requires periodic reviews and approval of iterative adjustments to the design as additional considerations and requirements become clear.
Applying the effort necessary to anticipate problems that will come up in a project is not as obvious as it sounds. During the initial contact and sales phase, both user and vendor have a mutual and often unrecognized interest in ignoring difficult issues. For the user, it is tempting to leave many details of the requirements and problems unrecognized to shift the burden and expense of solving problems that come up later onto the vendor. At the same time, leaving significant issues unaddressed can allow a vendor to offer a more competitive price initially to get the job, expecting to gain additional revenue through change orders later on.
But this kind of thinking leads to conflicts over price, schedule, and scope late in the project when progress is most critical. Apart from creating unpleasantness, it consumes time and energy (better directed toward testing, tuning, training, and documentation), making errors and overruns more likely. A proper control study helps highlight these issues and avoid this kind of disruption.
Finally, since an APC project will most often be applied to an existing operation, the schedule must allow for commissioning in a way that eliminates, or at least minimizes, interruptions to current production.
In short, a project schedule is an ideal thing, while engineering an advanced control system is a real thing. As has been said of love and marriage, confusion between an ideal thing and a real thing never goes unpunished.