Many mechanical designs are on the mature end of their design life cycle. While there are certainly innovations in the details, the changes are often evolutionary than revolutionary. The evolution comes from field-testing and refinement, emulating successful approaches used elsewhere and, most importantly, the application of sound engineering and science. Even with such refinements, the types of guides we use now; unwind, winder, displacement and steering, are the same types as used decades ago. The same is true of other machines such as winders. We have center winds, centerwinds with layon rollers and center-surface winders, just as in the past. We have reels, turrets and two-drums, just as in the past.

In stark contrast, however, most process automation is still on the "adolescent" end of its design life cycle. Process automation can be likened to a teenager, who outgrows clothes or tires of the style in less than a year. Like a teenager, process controls are quite capable of doing useful work. Like a teenager, however, they sometimes lack direction and motivation, and can even be obstinate and troublesome. It is not uncommon for them to be self-centered rather than benevolent.

For better or worse, however, our future is in the hands of automation—and teenagers. They give us the certainty of change, if not the promise of better ways. It is our responsibility to understand these changes and guide them to their maximum potential.

To know where we are going, we must first know where we came from. The evolution of process automation begins with sensors and actuators. A sensor might be a load cell for tension, or a beta gage for basis weight or caliper, or a camera for defect detection. An actuator might be a drive motor, or a CD profile thermal jack on a die lip, or a web-marking system.

Even though sensors and actuators are generally more mature than their supervisory controls, they are still evolving. Onboard computer chips can do local data manipulation and simple control, as well as manage two-way communication with the outside world. Local digital control is generally faster than supervisory control but, surprisingly, still not always as fast as analog circuits are. Local intelligence also allows sensors and actuators to be more precise because they can compensate for certain complications of the real world such as ambient temperature changes, profile imperfections and so on.

The digital revolution has on occasion radically changed mechanical design. For example, the new servo drives on printers, packaging and other machinery give us an 'electronic line shaft' option with many advantages.

• First, we can exchange inflexible mechanical complexity for flexible control complexity.
• Second, we can do things that are difficult to do mechanically, such as change registration and repeat length on the fly.
• Third, the drives can report on their status and health in much more detail than can a mechanical drive.

Another example is the 'smart' layon roller, dancer or other pivoting device. The control system 'knows' where in the stroke the device is and can compensate for varying gravity, mechanical advantage, friction, interaction with other variables and so on. The result is better precision and more mechanical and control flexibility.

However, the biggest trend may not be the mere digitization of sensors and actuators. Rather, it is in the number of sensors. Machines built only just a couple of decades ago may have been equipped with not much more than a speed meter and a few motor ammeters. Even web makers may have had only a few pressure gages and temperature readouts to guide the process developer and the machine operator. As recently as three years ago, the most abundantly sensored paper mill had 10,000 sensors. These days, many paper mills have more than 20,000 sensors that report to data acquisition systems. Obviously, most of us do not have that many sensors, much less log them all into a computer, but the trend is still the same.

As computer automation began to have an impact on American manufacturing, it was assumed that knowledge is power and that data was the key to knowledge. Thus, we bought sensors and data acquisition technology with the expectation that our processes, and thus our products, would somehow improve. However, the results have been variable. In some cases, the sensors allowed closed-loop control that made the machine, and thus the product, much more consistent. In not a few cases, however, either the sensor, the actuator, or the control algorithm was not trustworthy enough to be allowed to control a process. In other words, the process was more stable when the automatic controls were turned off. Managers assumed that technology would make a better product, but the potential benefits are not always realized due to oversight of a few key details.

The most serious limitation we are now running into is the lack of process knowledge. For example, a new winder can program 'curves' for each of the tension, nip and torque functions. Yet few operators or managers know how to select a curve to minimize a single defect, much less minimize the sum of several defects.

The result is an infinite number of potential curves without guidance on how to make use of the capabilities. In fact, the computer controls sometimes get in the way of learning about our processes because one can't just go in and 'twist a knob' because of the many interlocks and other control restrictions.

The same is true of sensors. So what if the sensor shows a process variation? The real question is whether this variation does (as opposed to can) contribute significantly to product variability. Next, whether that specific variability of the product causes a real or a perceptual issue with the customer. All variations are not equally important, yet the control system makes no distinction. We human operators must make that distinction based on our knowledge of our process, product and customer.

Sometimes technology is not used the way we think it is. I recall a recent visit to a client whose engineer insisted that they were using taper on the winder. I discovered that the operators had not used taper for years. The engineer knew what should be done but did not know what was being done. The operator knew what was being done, but did not know what should be done. The sad thing here is that the client wanted to buy a new winder just because it had more knobs and thus might give them a better product. They did not effectively use the one knob they already had. In order to use new technology, we must go to school—both in the classroom and in our plant.

Sometimes technology forces us to learn more about our plant systems. Consider scheduling software that takes customer orders as inputs and then outputs a schedule for a machine. The simplest scheme and most attractive for just-in-time (JIT) is first-in-first-out (FIFO). However, some customers don't need as quick a turnover as others. Also, to run this type of schedule raises waste and delay because it requires machine changes from one grade to another and then back again.

An improvement might be to weight scheduling so that it groups similar chemistries together, so personnel don't have to clean the machine and vats as often. In contrast, changing thickness or speed can be done more gracefully and is thus weighted less for scheduling priority. On top of all of this, it gets really complicated when one tries to nest various individual smaller wound-roll widths to fill out the width of a machine with minimum trim waste. Scheduling software forces us to decide what is important. For example, is a day quicker turnaround on an order worth more than a 5 percent loss of throughput and a 2 percent increase in waste?

Every investment must provide an economic payback. The easiest payback for process automation may be reduced delay. Control systems can report the time cycle or heartbeat of the machine. If it skips a beat during the night shift, you will know right away when you sign on. You can look at the control data before the event so see if there were any precursors for the trouble. If a new crew is struggling, it will show up. Obviously, this will not replace the need to go down to the floor and talk with the operator, only that it will make the dialog more meaningful. Process automation can also reduce the time to make grade changes and other setups, especially if the recipe has been preloaded.

Though not new, defect mapping is becoming much more common and useful. One example is a high-speed camera that inspects the entire web for variations in brightness and passes the data to a computer. The computer is taught how to classify different patterns of brightness variations as an indication of a certain type of defect such as, for example, a streak, thin spot or hole. These defects are color-coded by type and displayed on a plot for the width and length of a roll.

In some cases, the mapping is used merely for troubleshooting. In more sophisticated installations, the defect map is downloaded to the rewinder, which automatically stops at each location for visual inspection by the operator. Manual inspection is often required, as the system may generate false positives and inappropriate classifications.

Other important returns for process automation are reduced waste and customer complaints. However, a subtle limitation on data systems used in plants can cripple its utility here. The difficulty begins because process information is acquired on time-based intervals while lab test, waste and customer complaints are acquired on lot-based intervals such as wound rolls or boxes. Also, the time the data is acquired is different. Process data is available immediately, test data is delayed by minutes to hours, and customer feedback is delayed by weeks or months. To stitch the data together from the various sources requires a lot of effort.

• First, you must get the roll number from the customer and then look up what time interval that roll was produced. Hopefully, both you and the customer are absolutely consistent in what you mean when referring to a defective roll.
• Then you must go to the process system and download the data for the interval when the roll was produced.
• Finally, you may need to attach the lab data for that roll or the closest roll to it that was tested. While this is doable for a single roll, a single roll rarely gives you insight into a problem. Rather, you must statistically compare sets of rolls that are good with sets of rolls that are bad for a particular reason on the same grade.

In the real world, the data is obtainable, but is almost unusable for problem solving because few of us want to take the time for data synthesis. More often, we make use of only what is convenient.

It is imperative that the next generation of software allows us to simply and artfully sieve the wealth of data just sitting in the computers. We don't need more data and more graphs. What we need is more insight—the kind of insight that can truly help us improve the process and the product.

If you found this article helpful, ENTER 240 or Inquire Online.