Article:
Optimizing Product Fill Weights
Using statistical process control (SPC) analysis in a program of continuous monitoring and improvement can help manufacturers optimize product fill weights, allowing them to minimize overfill and still comply with regulatory requirements. This paper describes the application of NWA Quality Analyst to fill weight situations in various industries. The examples used here are composites of several NWA Quality Analyst users.
Reducing variation in product fill weights has a large impact on a manufacturer's bottom line. Take, for example, a product for which a government agency requires that the amount of product in the container, on average, be at least as much as the label on the package states, a common requirement with consumer products. Since under-filling the container puts the producer at risk with government agencies (not to mention customers), producers tend to avoid problems by over-filling. Yet, over-filling can create significant cost problems. Using statistical process control (SPC) analysis in a continuous monitoring and improvement program, a company can reduce over-fill situations and still meet the regulatory requirements. The resulting savings can be huge.
Another potential problem with product fill-weight results when manufacturers use materials pre-measured by their supplier. Often buyers use the container as it comes from the supplier as a unit of measure. If the supplier has under- or over-filled the container, the buyer may not be using the correct amount of raw material. The result can be an end product that does not meet specifications. Suppliers that provide unpredictable product may be removed from the qualified purchasing list. Again, using SPC in a continuous monitoring and improvement program, a company can optimize its product fill to meet its customer requirements predictably and, potentially, enhance its market position.
The following examples illustrate how SPC with NWA Quality Analyst can be used to optimize product fill weights.
Example 1. Balancing Cost Concerns with Regulatory Requirements
A large commercial bakery uses three dough-mixing machines for its dinner roll line. When each mixer has prepared its 1,200 pounds of dough, it dumps its batch into a pouring machine, which then pours dough onto baking sheets. Dough for six rolls is poured onto a baking sheet lane with each sheet holding 4 lanes of rolls. The entire sheet is then baked after which the rolls are cut and packaged in twelve-roll packs.
The pouring machine keeps dispensing dough onto baking sheets until it is empty. It then receives another 1,200 pounds of dough from the next mixing machine in a fixed rotation and begins the pouring process again.
Since the bakery labels the package as weighing 454 grams (the bakery uses grams rather than ounces for its internal reporting), the bakery requires that 100% of the packages weigh at least 454 grams in order to avoid any problems with the Food and Drug Administration (FDA), the governing regulatory agency. If the FDA finds the bakery is not meeting the weight requirement (i.e., it is under-filling the packages), the bakery can face product recalls or serious fines. On the other hand, excessive over-filling can be expensive.
In order to determine if it could reduce costs in its dinner roll line and still meet the FDA weight requirement, the bakery's process engineers conducted a study using SPC with NWA Quality Analyst. The first step was to determine if the dinner roll production process was predictable (this is called "in control"). To do this, samples of baked dinner rolls from three lanes were taken every hour on both shifts and weighed. This information was then analyzed using an X-Bar/Range chart. As shown in Figure 1, the process was not in control (note all of the points above and below the control limits). These results indicated that external factors (i.e. factors that were not due to the inherent variation in the pouring/filling process itself) were influencing the process.
Additional investigation revealed that there was little variation between individual rolls in a lane or between lanes on a baking sheet. The engineers also found little variation between shifts or operators. To understand the problem more clearly, the engineers took samples every minute. This analysis revealed a pattern - the excessive variation was occurring approximately every 30 minutes, the time between dough batches from one of the three mixing machines.
Since each dough batch came from a different mixing machine, the engineers analyzed the dough coming from each of the three machines. One machine, Mixer #3, exhibited significantly higher variation in the texture of the dough it produced. Further examination of the texture revealed that the machine was not receiving flour and water in the proportions the bakery's formula required. Adjustments were made to this machine and samples were then taken. As shown in Figure 2, the X-Bar/Range chart for the pouring/filling process indicates that the process was still not in perfect control but exhibited much less variability.
At this point, the engineers examined the "capability" of the process—i.e., they measured how predictably the process is producing within specifications. Normally capability should not be examined before a process is in-control because the analysis is unreliable. However, in this case the engineers did the analysis to look at the capability of the process relative to the specification because the variability was reduced so significantly.
The most commonly used statistic to evaluate the relationship between the variation in the process and the specifications is Cpk, which looks at the ratio of the spread of the specifications and the spread of the process (the distance between the +/- 3 Standard Deviation lines). A Cpk of 1.3 is usually considered adequate. This process, as shown in Figure 3, shows a Cpk of 1.058, indicating that it will not be capable of consistently producing within these specifications. However, this chart also shows that the process is capable of producing rolls that meet the FDA weight requirement predictably as long as the overfill averages 15 grams. (Note that virtually all of the distribution is above the specification of 454 grams. Also, note that the mean for this process is 468.74 grams).
Since the process is not in control, the engineers must first find other external factors that are influencing the pouring/filling process. Once the process is in control, they can continue to make improvements in the process to reduce the over-fill. Despite the fact that the process is not in control at this point, the changes the engineers were able to make during the initial parts of the study to reduce the weight variability in the dinner roll production line saved the bakery approximately $200,000 per year.
Example 2. Multiple Head Filling
Although the use of the X-bar/Range chart is appropriate in many filling operations, it is not when there are multiple fill heads. The problem is that an X-bar/Range chart computes the average for the operation—i.e. averaging the fill for all the heads—rather than computing the fill for each head. Yet, an analysis of the average for the fill heads taken as a whole can mask the problem behaviors of each individual fill head because each fill head is actually an individual process. The proper analysis in a multiple fill head process is the Median and Individuals Chart.
In this example a milk filling line has twelve fill heads, each filling a carton in the line. FDA regulations require each carton to weigh at least 8.604 pounds. In this milk-fill line, an X-bar/Range chart was produced (as shown in Figure 4), showing that the process as a whole was in control (meaning it predictably met the regulatory requirements).
Although the dairy had detected an occasional under-fill, the problem seemed minor and did not seem to reflect a pattern. Besides, the SPC charts indicated that the overall process was ok. Therefore, the dairy was surprised when state regulators informed them that they had received several complaints about underfilled cartons. The problem was that the X-bar chart reported on the average fill weight, missing any problems with individual fill heads.
When the engineers re-examined the filling process, they used NWA Quality Analyst's Median and Individual Chart. This allowed them to treat each head as an individual process while still monitoring the median of the overall process. This analysis immediately revealed a problem with Fill Head #8. As shown in Figure 5, the variability on eleven of the fill heads was in statistical control, but Fill Head #8 exhibited excessive variation and was under-filling.
Now that the cause had been identified, Fill Head #8 could be repaired or replaced. This change brought the entire process into statistical control and the dairy was ready to determine if there were additional opportunities for process improvements that could save the dairy money by reducing over-fill (see Figure 6 showing that this process with an average overfill of .02 pounds was capable of meeting the specifications).
Example 3. Meeting Customer Requirements
A titanium dioxide (TiO2) producer supplies a paint manufacturer its product in 50 lb bags, which are used in paint batch operations as a unit of measure. Since any variation in the bags affects the paint manufacturer's final product, cans of paint, the customer specifies that variation in TiO2 bag weight be no more than +/- 1lb.
During process improvement studies using NWA Quality Analyst, the TiO2 producer discovered a +/- 7 lb variation in the 50 lb fill weight. Since this variation is well outside the buyer's accepted specifications, the manufacturer needed to investigate this situation further. The first step was to determine if the process was in statistical control. The analysis, as shown in Figure 7, showed that the process was in statistical control, suggesting that the +/- 7lb variation was a part of this process and not resulting from an external influence (such as different raw materials).
As the TiO2 manufacturer analyzed this problem, it learned that the primary source of the variation came from the bag-filling machine. After adjusting the machine and providing more training to the operators, the manufacturer reduced the fill weight variation to the necessary +/- 1lb (see the left half of Figure 7).
Although the process was now meeting the customer's specification, the TiO2 manufacturer needed to look at the processes capability, that is how reliable was the process for producing the fill weight within specification. Process capability statistics are generally based on a ratio of the product's specifications to the process variation (usually expressed as three standard deviations based on the distribution). The most common way of denoting process capability is the Cpk index, which is typically required to be at least 1.3. The higher the Cpk, the more room there is between the process and its specifications.
As the histogram in Figure 7 shows, the Cpk for this process is 0.7506. A Cpk this low does not necessarily mean the process is "incapable" of meeting the required specification. However, it is a warning to the manufacturer that it should investigate the process further because additional problems with this process may arise. For example, the manufacturer might experience new problems with its fill operations if it needs to meet tighter customer specifications (to +/- 0.5 lb, for example, from this customer or a new one) and/or it wants to make additional process improvements.
Answering Fill Weight Concerns
Companies from a variety of industries must meet fill weight specifications. The first step is getting a process in control or predictable. The next step is determining how capable the process is for meeting the required specifications. Finally comes improving the process so that it optimizes fill-weights. These are all ideal uses for NWA Quality Analyst.
 back to top
|
 |