Decreasing the Cost of Control
One common challenge when designing or upgrading installation is determining how to reduce costs. There are a number of factors to be considered, including development, deployment, and maintenance costs. Although many ideas may come to mind, the possibility of using advanced process control (APC) is more easily attainable due to the power of computers and new strategies that are now feasible. But what if we reduce the number of variables to be controlled? With fewer variables, less hardware is needed, the process is simpler, and it is easier to teach to others. So, how can users reduce the number of loops to control?
Often, process control systems are composed of a number of variables coupled together. Not only is effort spent on controlling variables, but also on using “fake” controllers to decouple the variables so that single-input and single-output algorithms, like the ever-present PID, can be implemented. APC now adds the strategy to control several variables simultaneously, so there is no need for decoupling. While PID is the workhorse of the control algorithms used in industry, it has limitations. First, it is a single-input, single-output system, so control engineers have to figure out how to take information from several measuring devices and combine them into a single value that makes sense for the controller. The reverse is also a challenge: how can a controller make decisions on several control variables based on the combined inputs? The most typical approach is to design several PID controllers in parallel, but this option is not without its own challenges. Since some variables are coupled, more controllers are put in place to try to isolate those variables. The challenge that comes now is tuning all those controllers that have been placed in the plant. Although there are rules to tune PID control algorithms, it is not an exact science. In the case of multivariable controllers, it is more challenging since changing one gain will affect the behavior of not only the PID loop that is being tuned, but also all the different controllers and variables that are coupled.
One example is the case of Tyco Electronics, where a coaxial cable manufacturing process was put in place. The process under consideration includes building dielectric, shield, and jacket layers on top of the center conductor wire. The dielectric layer is the most critical variable because it affects the subsequent layer diameters and their consistency. It also determines the coaxial product qualities such as impedance, capacitance, and time delay per foot. The process involves a combination of multiple input variables to control the diameter and coaxial consistency with the output variables, so a multivariable controller is necessary. The manufacturing lines run at different speeds to accommodate product requirements and equipment limitations. To make the process even more challenging, there are also variable distances between actuators and sensors. As the line speeds and distances change, time delays between inputs and outputs vary. Tyco Electronics found multivariable model predictive control (MPC) was the best choice for this application because it controls variable time delay processes while optimizing multiple output setpoints and multiple input constraints.
Dr. Kimberly Wang, principal control systems engineer at Tyco Electronics, completed modeling the process using logged data and LabVIEW modeling tools. Once the model was found, she and her colleagues used these results in a real application with further controller tuning. Dr. Wang implemented the MPC systems on more than 10 manufacturing lines with excellent Six Sigma process control levels. The control system not only controls the product quality to meet specifications, but also stops the lines automatically if the product quality is out of specification. Tyco Electronics achieved unmanned overnight production, thereby increasing productivity and reducing machine downtime. In addition, National Instruments helped set up functions that reduced coaxial final quality test time because the control system tracks the product quality and prints a label to identify whether the product passed or failed.
Vision for difficult variables
In the previous section we discussed calculations using single controllers to control multiple variables, but every control loop relies on measurements. What happens if the variable under control cannot be measured with standard sensors? Vision may be a solution.
Iggesund Workington Mill faced such a challenge in its process to analyze paperboard formation. The mill needed to provide board machine and pulp mill operators with a current formation image coupled with a formation index. The formation index is calculated from gray-level distribution of the image, because formation is a subjective science and it is important to have a visual correlation. The display of current formation images provides the operators with the most current situation and a percentage measurement of how well the board machine and pulp mill is running. Differentiation between board and pulp variations is helpful to control board quality, classify problem areas, and assist management in assigning resources to one particular area.
The source head contains the light control unit, flash system, and illumination optics. Illumination is synchronized with the detector-head camera unit through wiring in the power track between the upper and lower head. The image control unit continuously adjusts the illumination and camera parameters to maintain the best possible image quality under varying process conditions such as speed and basis weight. This makes the modulus calculation for the total board construction extremely difficult. Thus, it is necessary to employ a predictive multivariate model that uses online measured data to calculate the bending stiffness based on process variables. Dynamic model development, using the LabVIEW System Identification Toolkit, uses a real-time proportional-integral (PI) controller to control the critical polymer chemicals dosing rate.
Imaging technology-based formation measurements are already available in the market. However, the information reported is generally confined to the individual quality index figure that indicates overall formation variation. Nevertheless, it still remains difficult to use only a single figure to deduce why formation variations occur in order to optimize the process. This is because formation is an outcome of several factors, including raw materials, process equipment, and the process operating point. It is usually easier to list items that do not affect the formation. Iggesund introduced a new approach to facilitate formation optimization: divide the formation-related image information into mutually independent descriptive subcomponents. This further highlights the actual changes of the formation in question, thus making it easier to choose subsequent corrective actions. The upgraded system replaced the traditional formation analysis method and provides the operators with real-time board quality information.
Observers for many variables
Even using nontraditional sensors, such as cameras, may not be enough to support closed-loop control since not all of the variables that need to be controlled can be measured, and not all of the variables that can be measured need to be controlled. This is an area where we could think of using “virtual variables” that are composed from other variables, which are the ones that need consideration. The use of virtual variables is driven by the fact that if you can’t measure it, you can’t control it. So how should one approach problems where measurements can’t be taken using standard or advanced sensors? Academics in control theory have been working on this problem for years, and now, with the use of APC, these solutions can be taken into modern manufacturing systems. Variables that can’t be easily measured are present in a number of traditional industries, like steel manufacturing. When producing steel or any other heavy-duty material, there are often very harsh conditions that don’t work well with measurement systems, for example, measuring the thickness of steel coming off a rolling mill. To solve the challenge of virtual variables, a possible approach is to use observers. With observers, users can control variables that are either too difficult, or impossible, to measure. These “observed” variables can be inferred from other measured variables and/or a software model of the process under control. This type of controlled system allowed Iggesund to increase its throughput by 25%.
One problem with conventional pH process control is that variables are monitored individually, which indicates the state of only the individual variables and ignores the interaction of variables. Achieving optimal pH enhances paper quality and production process efficiency, so Iggesund used APC with multivariate techniques to take into account the internal interactions among variables. Therefore, it could model the process on detecting univariance with individual variables. This can then be compared with the univariance for combined variables to identify the true process performance, which can then be used to identify conditions when the process is unstable.
Iggesund designed and developed a model predictive control scheme using LabVIEW and artificial neural network (ANN) technology combined with advanced adaptive control algorithms for closed-loop adaptive refiner control. The model is bound within the normal operating conditions for the pulp mill. If conditions outside the normal operating conditions occur, an alarm sounds that not only indicates to an operator that there is a problem, but also provides details of what variable is at fault and the best action to resolve that fault.
National Instruments used LabVIEW to develop a pH model that demonstrates that it is possible to model a nonlinear complex process. It confirms the pH meter readings and provides a basic online diagnostic tool for operators. Because the model is only taking in the key variables that affect pH, if there is a change in the incoming flows of one of these, it is highlighted on the model. This change can be observed much sooner than normal due to process dwell time. The model aided in monitoring and optimizing chemicals to control the pH more smoothly by displaying real-time information. This increased process stability is due to improvement of the control of a process-critical variable.
Advancements in computer power, combined with the development of object process control (OPC), and improvements in software, enable engineers to use the processing capabilities of a desktop PC to control large plants with a combination of new technologies (sensors, algorithms, etc.) that improve throughput while reducing costs and optimizing the process.
Javier Gutierrez is senior product manager, control design and simulation, for National Instruments.