by Juha Karisola
Universities, research centers and corporate product development units
continuously look for new ways to improve the reliability, quality and repeatability of
production processes to meet the industrys ever-rising quality requirements. Fuzzy
control is a new, exciting way to improve process performance in a number of different
Fuzzy logic is a new form of computer logic that provides a systematic way to handle qualitative information. It enables an automated control system to reason and make judgements in ways similar to the way human beings make judgments. Simply put, fuzzy logic converts vague concepts or values into a mathematical format, which is then used by the computer to precisely control the process.
Our company has developed a system that uses this technique to control the windshield bending process. The main benefit is the ability to repeat a process consistently and accurately, but the new system also makes it possible to mix different windshields freely in the same production run. This article gives general information about fuzzy controls and explains why it has been applied to the serial windshield bending furnace.
Fuzzy logic is based on the theory of fuzzy sets. Normal binary logic states that something either belongs to a group or it does not, but fuzzy logic accepts that something can belong partially to a group. For instance, binary logic says that when it is 30 C, water is warm and also when it is 20 C, but, abruptly, when it is 19 C it is not. But, warm water is an unclear, fuzzy concept, something that binary logic cannot understand.
By using membership functions (relationships of degree), fuzzy logic assigns quantitative values to something that partially belongs to a group. Water at 20 C belongs entirely to the group warm, while water at 19 C only belongs to this group to a certain degree.
The membership functions in each group may be bell-shaped, triangular or trapezoidal and usually there are three, five or seven groups.
Fuzzy control consists of the following steps (see figure 1):
Fuzzification; Converting the measured process values into membership functions;
Processing; Using a set of rules;
Defuzzification; To produce precise control values.
To explain how a fuzzy logic control procedure works, lets look at a simplified example of target temperature and actual temperature. The aim is to define controls that keep the actual temperature steadily on the target curve by adjusting the heating power. The first step is to define the membership functions in the fuzzy sets for the input and output values (see figure 2).
Figure 2 shows the the fuzzy set for the temperature difference between the actual and the target temperatures (DT). Corresponding sets also have to be defined for other input values (in this example for the derivate of DT, which indicates whether the difference is increasing or decreasing).
The fuzzy input values are then processed through the set of rules. The rules in fuzzy control consist of a condition, IF, followed by a control action, THEN. In normal language a rule would state something like: if the actual temperature is slightly lower than the target temperature and the difference is slightly increasing, then increase the heating power considerably. As a logic rule it states: if DT is NS (negative small) and DT/dt is NS (negative small), then P is PL (positive large). (A matrix of the processing rules is shown in figure 3). Each rule processes the information using different parameters; the output of each rule is different. The correct heating power required is obtained by combining the results from all the rules and finding its center of mass (see figure 4).
Fuzzy Control Benefits
Fuzzy control is simple for the user. It provides precise control values from non-precise input data, a situation in which a conventional controller fails. Since fuzzy control systems process rules in parallel they are very fast. Since each rule operates in parallel, the effect of individual errors is minimal. The whole system is resistant to errors and more reliable than conventional controls that process data in series. Fuzzy control constantly adapts automatically to changing conditions, something that normal PID (Proportional Integrated Derivated) controls, for example, are unable to do.
Why Apply Fuzzy Control?
The serial bending furnace (TFA) is a tunnel furnace consisting of two conveying tracks, one above the other. The upper track is for pre-heating and bending, the lower for cooling (edge compression and annealing). The furnace has a compact construction and has very low energy consumption. It originally was developed for producing replacement windshields, where glass is bent in small or medium runs. The capability to process different windshields at the same time was an essential requirement, as each wagon could possibly contain a different windshield.
Thanks to its versatility, the furnace was adopted rapidly by the industrynot only for replacement windshields, but also for OEM glass. The construction limits the cycle time to approximately 60 seconds, so whenever the capacity requirement is lower, the furnace can also be used for OEM runs. Glass companies using serial furnaces supply low volume car manufacturers in Europe (e.g. truck manufacturers and manufacturers of sport cars and special vehicles) as well as in Latin America and the Far East.
As with any bending process, the serial bending process is subject to a number of factors that reduce the repeatability of the process. To improve the process, and the repeatability, a careful study was made of these factors. Some of these, such as differences in molds and in the float glass, are not related to the furnace itself so were not taken into account here. The main factors affecting process repeatability are:
Production Mix. Whenever the furnace is loaded for small runs (all the wagons have different windshields) or medium size runs (four different types each loaded in seven wagons; 7/7/7/7) the process involves varying bending and preheating times. When the mix is changed again, the processing times for each glass change accordingly. Adjustments in the final bending time, or temperature, can help compensate for changes in the process time but this usually causes deviation from the ideal form.
Specific Heat Balance. Even if a furnace can be taken into production one hour after preheating, it still continues to heat up for four to five hours more until it reaches the specified heat balance. Unless the furnace is being operated on a full 24-hour, three-shift basis, then the operators face the same problem every day: a piece of glass produced in the morning does not have the same shape as an identical piece of glass bent with the same program in the afternoon. Some operators compensate for this difference by using a different program once the furnace has reached the specified heat balance.
External Temperature. The external temperature around the furnace may vary considerably between morning and afternoon, summer and winter. Temperature controllers compensate for the variations automatically, but not as precisely as they could. Traditional on/off- and PID-controls usually have overstrikes, which may affect the shape of the glass.
Variations in Voltage. Good process performance usually is guaranteed for variations of less than five percent. In many countries the supply voltage varies much more than five percent. Instead of eliminating each of these factors one by one with a separate system, a new approach was taken. Standardizing the bending time and precise control of the heating and bending process provided an easy to use and reliable way of eliminating the morning factor.
FuzzyBendHow does it work?
Our FuzzyBend (patent pending) is a new fully-automated control system for the TFA and the LTFBA range of windshield bending furnaces, with clear advantages over existing automatic systems. In the new system, glass is always bent for the same period of time under uniform conditions, regardless of any changes in production mix, the heat balance in the furnace or external temperatures, or variations in the power supply. This guarantees excellent repeatability and easy use of the furnace, at the same time as it allows for optimal use of furnace capacity. There is no need to use several different bending programs for changing conditions.
The basis for controlling the bending is provided by measurements from aligned optical infrared pyrometers in the final preheating and bending sections. One of the given target process curves is chosen, depending on the thickness, size and shape of the glass. This curve then determines how the glass is heated in the pre-heating sections (see figure 5) and most variations in the glass temperature have been adjusted to the target curve before the glass enters the bending section. This balancing is controlled automatically by a fuzzy logic control that minimizes the differences between the target curve and actual temperatures and eliminates overstrikes. To ensure that the required heat is distributed over the glass during balancing, the system maintains the heating profile, while the programmable logic controller regulates the heating power for the whole profile. The glass always has the same temperature, and an even heat distribution over the glass, when it enters the bending section. The bender selects the heating profile according to the required glass shape and decides on the programs required to bend the glass.
Operators Experiences of FuzzyBend Controls
The FuzzyBend has already been installed in a number of furnaces around the world, including the large bus windshield bending furnace of Viracon Inc. Some of the furnaces are being used for OEM runs, others for replacement glass. As expected, the repeatability of the bending process has improved considerably. Good repeatability is one of the most important factors that OEM producers are looking for. Sag (cross curvature) tolerance has been typically reduced to a half or less. Sag control is especially difficult with compound shapes with a central sag exceeding 15 mm. As the trend is toward pieces with deep sagging, the need for precise controls is obvious.
Another important benefit was the fact that a furnace with FuzzyBend can be loaded with a mix that cannot be produced with traditional automatic controls without sacrificing bending tolerances. Previously, operators had to plan production on the terms of the bending furnace, dividing production into groups of more or less the same size, shape and thickness. FuzzyBend compensates for differences in size and shape by using different process curves. Operators now have almost free hands and can group production according to the order book, not in accordance with the limitations of the bending furnace.
Many people are afraid that fuzzy controls are very complicated. Even though fuzzy control in itself may be a complex technique, its user interface is very simple. For the furnace operator, it could not be easier. Just select the required process curve and switch on FuzzyBend. All the controls then work automatically in the background.
Juha Karisola is vice president of sales and marketing for Glassrobots Oy, located in Tampere, Finland.
© Copyright 1999 Key Communications, Inc. All rights reserved. No reproduction of any type without expressed written permission.