'Fuzzy logic' makes controls think like you do.

Though still a bit controversial, fuzzy logic is now controlling temperature and pressure on a lot of plastics machinery. Even its advocates say fuzzy logic is no panacea, but when it's properly applied, it can offer real benefits in control stability.

Although "fuzzy logic" may sound exotic, it's used every day to control many conveniences of modem life - everything from elevators to dishwashers. Over the past decade, fuzzy logic has also made inroads into industrial process control, including plastics processing. To its advocates, fuzzy logic outshines conventional controls because it completely avoids overshooting process limits and dramatically improves the speed of response to process upsets. Fuzzy-logic controllers are said to accomplish both these goals at the same time, rather than trading one off for the other as might be the case with PID control. Suppliers often liken fuzzy control, which seeks to emulate the human thought process, to having an "expert operator in a box."

Despite such claims, fuzzy logic is not a cure-all. Its value is contested by control vendors who have evaluated but not adopted it. And it also appears that not all fuzzy-logic controllers are created equal - no more than are all PID controllers. "We all make cake out of the same flour, but we implement fuzzy logic differently," notes Frank Dyke, v.p. of marketing & sales at Syscon-RKC.

Even some fuzzy-logic proponents concede that it is not needed or even desirable in all applications. In fact, all the fuzzy-logic controllers used in plastics processing allow the fuzzy logic to be switched off so that traditional PID control takes over. Ed Kulawiak, marketing manager at Omron, another fuzzy-logic vendor, puts it this way: "Many plastics processing applications simply don't need fuzzy logic. But the ones that do can really benefit a great deal." Areas where fuzzy logic has carved out a niche include temperature control for extruders and auxiliary equipment. Fuzzy-logic pressure controllers and PLCs have also recently come onto the market.

FUZZY TEMPERATURE CONTROL

Thousands of fuzzy-logic temperature controllers already in the field - from vendors such as Fuji, Omron, Syscon-RKC, and Yokogawa - make this by far the most common industrial application. Suppliers say this success stems from the inherent shortcomings of standard PID temperature control. "The chief value of fuzzy logic in temperature control is its ability to deal with events that disturb the stability of the control loop," says Syscon's Dyke. For example, he says much - or sometimes all - of the heat in extrusion comes from shear energy, which can change drastically under the influence of variables such as extruder rpm or material consistency. Such process upsets require retuning of PID terms. "PID constants are not really constant," Dyke points out. And without proper tuning, PID control will not perform optimally.

Yokogawa marketing manager Steve Crotty observes that PID control typically involves a tradeoff: Set the tuning parameters too loose in order to control overshoot, and the system will be slow to react to an upset. Opt for a tighter PID tuning, and the system will respond quickly - with more considerable overshoot.

Fuzzy logic addresses this tradeoff - not with a new mathematical model but with rules that describe how the process behaves in human-language terms such as "too warm" or "slow down."

Fuzzy-logic techniques then turn that seemingly subjective understanding of the process into a quantitative heater-output control signal (see sidebar).

Fuzzy logic does not do away with PID control altogether. The approach taken by Syscon-RKC, Fuji, and Omron uses fuzzy logic to decide when to re-tune and to calculate the PID terms. "Fuzzy logic is just another way of coming up with the tuning constants," says Dyke. "The fuzzy algorithm can respond to a broad range of phenomena. Our rule base is sensitive to rate of change, rate of oscillation, amplitude of oscillation, and oscillation above and below setpoint," he explains.

Taking a different approach, Yokogawa's fuzzy-logic algorithm directly manipulates the controller's output, says Steve Crotty. "Our Super Control fuzzy logic does not calculate the PID terms," Crotty emphasizes. Instead it generates temporary "artificial" setpoints designed to "trick" an otherwise conventional PID loop into maintaining the original "true" setpoint. For example, to eliminate overshoot from the true setpoint, the Super Control might feed a lower artificial setpoint into the controller's autotuning PID algorithms. The Super Control's knowledge base plays a key role in determining those artificial set-points. "This method is close to what an expert human operator would do based on his knowledge of how the process performs," Crotty says.

The quality of the fuzzy-logic rule base, or how well fuzzy control mimics the behavior of an expert operator, marks another difference between systems. Crotty argues that some suppliers were too eager to jump on the fuzzy-logic bandwagon and rushed to market with stripped-down, ineffectual rule bases. "We use nearly 50 different rules to control temperature while others may have one, three, or four added to a PID algorithm just so they can say they have fuzzy logic," he asserts. "We've seen tests where good PID control has outperformed poor fuzzy-logic control," Crotty claims. So don't be seduced by the high-tech sound of fuzzy logic. At the end of the day, good control is all that matters.

Most fuzzy-logic temperature-control use in plastics has been on extruders and related equipment, especially in cast-film lines. Molding-machine control has been less popular. "Injection molding machines tend to have dedicated multi-loop, custom-designed controllers," says Omron's Kulawiak. However, all fuzzy-control suppliers cite auxiliary equipment applications such as dryers and chillers for molding machines. "Hot-runner control has become important for us," Crotty adds.

Other plastics applications for fuzzy-logic control have started to appear, in part because all the companies mentioned here offer "universal" fuzzy logic controllers with input/output capabilities for pressure and other control applications. One pressure-control supplier, CMC Inc., uses a fuzzy algorithm to control the switchover from boost to hold pressure in injection molding, says president Rami Racin. And in Yokogawa's own injection molding operations in Newnan, Ga., fuzzy logic controls barrel temperature and injection pressure. Omron even offers general-purpose fuzzy-logic PLCs.

TURNING OFF FUZZY

As noted above, all fuzzy temperature controllers come with an autotune PID control option. Fuzzy logic isn't always needed, and sometimes you get better results with it switched off. "Our experience has been that most customers use it when they need very tight control - within one degree of set-point," says Kulawiak. "Many plastics applications don't require that kind of control. In those cases, standard PID control works just fine."

"In any process that is steady-state, fuzzy logic might not hurt but would be unnecessary," says Syscon's Dyke. "It seems to work best when process conditions are changing rapidly." Kulawiak adds that it works best with systems that have fast response times - less than a minute is his rule of thumb. "Fuzzy logic is ideal in quick-responding systems," he says. When applied in the wrong setting, these vendors concede, the result could be control instability in the form of oscillations around setpoint.

Yokogawa takes a different view. Because that firm's fuzzy logic manipulates setpoints rather than calculates PID terms, it loses its benefit in processes that respond very quickly - those that have little so-called dead time. "As the dead time approaches zero, fuzzy logic's benefit does too. If the loop is very fast, fuzzy logic could even make the loop unstable," says Crotty.

WHAT FUZZY COSTS

Once thought to be too expensive to use for temperature control, fuzzy logic today doesn't necessarily carry a price premium over any other high-end controller. Omron and Yokogawa controllers all offer fuzzy logic as standard whether you choose to turn it on or not.

But according to Dyke, fuzzy logic appears only in high-end controllers that cost 10-50% more than the simplest single-loop temperature controllers. He says that premium partly reflects fuzzy-logic software's high development costs but is mostly a consequence of fuzzy logic's need for substantial data-processing power and memory.

As computer memory and processing power become cheaper all the time, fuzzy logic could become available at less and less cost. Still, in cases where there's no real benefit, Dyke says, "We would recommend a simpler controller."

RELATED ARTICLE: How a Controller Thinks 'Fuzzy'

Although fuzzy-logic controllers are not all the same, they all rely on a "fuzzy inference engine." This element of fuzzy-logic software seeks to imitate the human thought process by delivering precise control outputs based on the kind of linguistic, rule-based process understanding a human operator would have.

Taking Syscon's version of temperature control as an example, the fuzzy inference engine would break the problem down into three steps, which are combined graphically in the figure above.

* Rule-base evaluation: Fuzzy logic starts by defining the process dynamics with seemingly loose linguistic terms instead of a strict mathematical model. According to Syscon's Frank Dyke, the rule base embodies the methodology of the so-called "expert operator" by relating process values to one another, assigning a degree of importance to each value, and deciding when to initiate a control response. A typical temperature-control rule might take the form: If temperature is "Too Hot," then set heat to "Low."

* Fuzzification: Linguistic labels, as defined by the rule base, relate to actual temperature inputs by way of so-called "fuzzy sets." These sets are simply collections of mathematical values - in this case temperature readings from the sensors. But fuzzy logic permits a value to belong simultaneously in one or more contradictory sets - for example the set of temperatures that are "Too Hot" and the set that are "Just Right." A particular process value has a degree of "membership" in each set that is expressed mathematically by a number between zero and one. The rule base is what defines the various sets and determines the degree to which a given process value belongs in each set.

In the figure, an extruder temperature can belong to a set called "Just Right" and "Too Cool" at the same time but to differing degrees. In conventional control logic, that temperature could not be "Too Cool" and "Just Right."

* Defuzzification: This final step translates the logical sum of the membership information into a precise heater output. According to Dyke, defuzzification typically involves finding a weighted average of the relevant membership functions; this single number is the output value fed to the heater.