Background: As some of you might be aware, one of the oldest and best known Pittsburgh products is the Clark candy bar. The original company has changed hands several times over the last few years - to be honest, I don't even know who currently owns the company (it might be Hershey...). Some years ago, I had the opportunity to work with a graduate student on a project at Clark - at that time the company was owned by a corporation in Europe (some place in Scandinavia if I recall correctly) called LEAF, Inc. We were asked to come out to their manufacturing facility in RIDC Park in Harmarville to help them with some quality problems that they were having. At this facility they made two brands of candy - Clark and Zagnut, and while they were having trouble with both, we were asked to focus on the Clark line initially.
The candy bars are manufactured using an interesting combination of methods - some, very high-tech and others, distinctly low-tech! It's been a while, but from my recollection the process is roughly as follows: at the first stage the ingredients for the "core" of the bar (the crunchy part) are mixed together in batches - the primary ingredient is peanut butter and the batch also has molasses, flour and some other stuff that I don't exactly recall now. The mixing is done in a mixing tank with the temperature, pressure, concentrations, etc. being controlled by PLC's (programmable logic controllers), sensors, limit switches and other electrical/electronic gizmos. When the mixing process is complete the product has the approximate consistency of pizza-dough (i.e., the stuff that is used to make the crust of a pizza). At this stage things get low-tech for a while. A guy on the line breaks off a chunk of the dough (much like a pizza-maker does, except that these chunks are much larger) and then he tosses it around, adds some flour and beats it over and over against a metal table until he's satisfied that it has the right consistency!
At the second stage the dough is fed through a set of rollers and comes out at the other end as a long snaky strip about an inch wide and about half an inch high. As it moves along a conveyer, a knife-blade comes down and cuts up the strip into pieces that are about 4 inches long - the speed of the conveyer as well as the times at which the blade comes down are electronically controlled so that each piece is the same length. The "train" of pieces then enters a baking tunnel where the temperature is once again closely monitored and exits the tunnel as little bars baked to a "crunchy" consistency. The bars then wander around the facility along a set of roller conveyers through a number of loops to cool down to room temperature.
At the third stage the bars go through a coating chamber where the chocolate coating is added on - by the way, it's not really chocolate but some brown goop that looks and tastes (sort of...) like chocolate! The coating basically rains down on the bars from an overhead chamber with the excess being drained off and recycled, and when the bars come off the end of the chamber they look like the finished candy bars that you eat. At this point inspectors manually scan the pieces coming out and remove defectives such as broken or chipped bars, bars that are not fully coated, bars that look too skinny, etc., etc. (when we were at the plant we always got to take a bunch of these home - they may not have looked attractive but they tasted just as good!). The finished bars then go to a finishing area where the wrappers are added, and the wrapped bars are then boxed together.
Problem: The main reason we were called in was because the bars were very inconsistent with repect to their final weights. These were 1.75 oz. candy bars and legally, they had to weigh at least this much (since this was what the label stated) - otherwise the company could be sued for false advertising. In fact, the plant manager claimed that there was this guy in New Jersey whose only occupation was weighing food products so that he could sue the manufacturer whenever he found a package that was below the advertised weight!
Ideally, one would like each bar to weigh exactly 1.75 oz., but this of course is statistically impossible given the randomness inherent in the manufacturing process - the actual weights were sometimes lower and sometimes higher. To be on the safe side, the company had set all the controllable manufacturing parameters so that they essentially made bars that were bigger and weighed more than they had to; that way the fallout at the lower end would be small. Unfortunately, this meant they were "giving away" candy for free and it was having a significant negative impact on the bottom line of the business. To make matters worse, despite their economically short-sighted strategy of "playing it safe" they were still producing underweight bars much more often than they would have liked. Our job was to analyze the entire process to see what could be done to improve it. The company was considering several courses of action such as buying more modern equipment, modifying process parameters, using better feedback control systems, instituting statistical process control methods, and even possibly changing the process itself. However, it needed more direction on what would be appropriate.
Data: Nominally, the specifications
that Clark had set were as follows:
| Target | Range | ||
| Center weight: 35.7
to 38.0 grams
Finished weight: 49.6 to 53.0 grams Coating weight: 28% of total weight |
length
width height |
12.95 cm
2.63 cm 1.55 cm |
12.60 -
13.30 cm
2.50 - 2.82 cm 1.45 - 1.70 cm |
As the first step we decided that it was important to get a handle on the degree of randomness and variability in the process as it was currently run, and to see how close they were to meeting these specifications. A large volume of data was collected through random sampling of candy bars coming off the end of the line prior to packaging; more characteristics of the data will be detailed as we go along.
A subset of this data is contained
in the file clark1.xls. This is a Microsoft Excel
spreadsheet containing a total of 220 data points - each datum is the weight
of a bar. The file clark1.mtw contains the
same data as a MINITAB data file.
Back to the ENGR
0020 Home Page