THERE ARE a multitude of advanced process control (APC) and artificial intelligence (AI) solutions available for process applications. These solutions can provide process optimization, predict failures, and simulate process variables that are difficult or impossible to measure.
Solutions not only are available, theyre also field-proven. Thats the good news. The bad news is that APC and AI wont work unless applied and maintained by automation professionals that know the process, understand the basic principles of APC/AI, and possess computer skills.
Personnel involved in advanced process control projects need a minimum level of familiarity with computers and process control fundamentals, says Fred Woolfrey, a productivity solutions consultant with Yokogawa Corp. of America.
Paul Kesseler, the manager for North America APC Solutions at Invensys Process Systems, seconds this opinion. While the software tools are relatively easy to work with, applying them to the right objectives and operating and maintaining them effectively is challenging. We work closely with customers to target APC solutions, configure the software, and train control engineers. But, no matter how well defined the application or well trained the operator, making the most of APC requires thinking about processes in an entirely different way.
Vendors typically and habitually understate the difficulty of implementation, so their warnings should be heeded. APC solutions require highly skilled personnel, and there are three ways for end users to secure and use these experts.
The first is to staff up at the plant level. This is impractical for all but the very largest plants. The second is to provide a central engineering staff that can service multiple plants. This makes more sense for most. The third alternative is to rely on vendor or independent system integrators services. This solution can be effective for plants and organizations with limited internal resources.
If the right people are in place, many benefits can be derived from APC and AI implementations. In terms of artificial intelligence, we use fuzzy logic to assimilate and evaluate outputs from various early event detection models, reports Randy Wagler, the profit suite product manager at Honeywell Process Solutions.
Fuzzy logic early event detection reduces the frequency and scale of process upsets, and it also reduces costs associated with process upsets. Fuzzy logic can minimize false alerts without compromising detection and notifcation of actual events, but there is always a tradeoff between sensitivity and increasing likelihood of false notification, adds Wagler.
Invensys provides a multivariable, model-predictive control software package with a full array of tools, including support for neural networks and fuzzy logic. Invensys reports that APC/AIs bottom-line benefits come much more from optimization than from control. Control only addresses response to upsets, which occur in only a small fraction of operational time, while optimization works all the time, says Lew Gordon, a principal application engineer with Invensys.
National Instruments provides a range of APC and AI solutions, including Model Free Adaptive control from partner Cybosoft, fuzzy logic capabilities, and toolkits that support other types of advanced control such as linear quadradic control. These solutions are used to tackle tricky control problems with multiple inputs and outputs and noisy disturbances, or to deal with systems in which not all of the process variables can be measured, according to Brian MacCleery, the industrial product manager at National Instruments.
In some cases our customers use these tools to estimate the value of a process variable that cant be measured in real-time. This is done by using other signals that can be measured. For example, the amount of a chemical reagent remaining in a tank can be estimated by measuring the pressure and temperature in the tank, adds MacCleery.
Yokogawa has taken a slightly different approach to APC. Rather than developing solutions in-house and refining these solutions in client applications, it instead uses technology originally developed and proven by Shell Global Solutions. Yokogawas multivariable optimizing controller is a linear multivariable controller that has undergone continuous improvement since it was originally developed by Shell in the early 1980s. The controller periodically adjusts the level of several manipulated variables to bring and keep controlled variables on given targets, taking into account steady-state dynamic interactions between variables.
Yokogawa also provides a toolkit that allows users to build real-time models of process quality variables, which are difficult or expensive to measure. The toolkit provides an environment for linear and non-linear models, including principal component analysis, partial least squares, and neural network models. The kits updating procedures include data validation, SPC rules, and Kalman filtering.