Food Product Design: QAQC - February 2004 - Determining Sampling Frequency

February 2004

Determining Sampling Frequency

By Bruce FloydContributing Editor

In a production-planning meeting, "How often do you want me to take samples?" often arises as the last topic of discussion. With proper planning, the company has already given some thought to that subject, along with where to sample and how large a sample to take. It does not matter what system is used -- TQM, six sigma, or just common-sense statistical QC -- sampling is basic to all of them.

Defining our world Determining sample frequency depends on the situation. Manufacturers might use a static population, where the product has been made and is sitting somewhere; or a dynamic population, where sampling occurs during the manufacturing process.

The following determine sample rate and are part of both of the above:

· acceptability limits; · sample preparation; · analytical repeatability; · raw-material variability; · process stability; · storage conditions; · container size or definition of the sample unit; and · number of containers and/or line speed.

Also, an intangible is at work here -- the capacity of the testing facility to process the samples.

If you make or buy something, it will eventually find its way into some type of container. But remember that someone might sample that product in the warehouse sometime in the future. Even if a company conducts its quality-control (QC) program during processing, the customer or regulatory authority -- or both -- will sample the product in a static situation.

Crafting a plan To establish a sampling plan, a company needs to know the acceptable quality level for each product attribute. For physical attributes, this is defined as the acceptable quality level (AQL). For objective data, the QC department can define this as an average value with a plus-or-minus range, or by maximum or minimum acceptable values.

Some defects are more important than others. Commonly, the company will define classes of defects, such as critical, major and minor, and assign an AQL for each class. Usually, critical defects have no acceptable level. If more than one type of defect exists in a classification, the sum of all the defects in that class would equal the AQL.

This difficult process -- to determine what is acceptable to the customer -- should not be assigned to the QC department. Sales, and even the customer, should have a say. Focus groups can help determine retail-product acceptability, and the manufacturer should also define safety issues when developing the product's HACCP program. The company must address regulatory issues, such as the percentage of key ingredients in products with a standard of identity. For ingredients, someone should produce product with raw materials that fall at the specification's extremes.

How would they know? A sampling plan may create as many questions as it answers. Will the analytical method and sample preparation create a problem? How repeatable is the analytical method? Can other laboratories run the method? Are the validated methods approved by AOAC International, the American Chemical Society, the International Dairy Federation or similar independent groups? These are just some of the concerns that may arise.

The quantity of the sample used and preparation affect the results as much as the method. Preparing minute quantities for analysis is the current trend. However, most food is a mixture, and it may not be a homogeneous one. In meat processing, for example, fat is not uniformly distributed in muscle; the case is the same for ground beef. So, when sampling, start with a quantity large enough to know the fat in a standard quantity. This applies to other processes, as well. For many products, knowing that a certain measured amount has identical composition doesn't matter. But it is important that every serving is similar; this is defined when acceptability limits are established. Small samples may indicate an irrelevant variation to the final user.

Analytical variation can be a part of attribute sampling. For example, how crooked must a label be before it is "officially" crooked? The company must establish clear standards and definitions in terms that all constituencies can understand.

Gathering and stability Manufacturing requires raw materials, including packaging materials. All materials vary, so it is important to know the flexibility of the system. This is a chicken and the egg problem: it takes in-depth knowledge of the raw materials available to design a process, but once the system is built, its flexibility determines the uniformity requirements of the raw materials.

Lack of confidence in a supplier increases the difficulty of this task. John Flaig, Ph.D., president of Applied Technology, San Jose, CA, recommends that the supplier send control charts for key factors along with the certificate of analysis. He also mentions that a review of process stability is an important part of a vendor audit.

The stability of the process determines the sampling rate: The more uniform the process, the fewer samples required to verify it. To learn a process's stability, pull a series of samples until the average and range do not change and use this data to calculate the normal mean and normal deviation. The mean and standard deviation will change as more samples are collected. If the sample averages do not converge and the system is bimodal or trimodal, the system is not under statistical control and other factors require examination. This is a big problem if the variation in the individual samples is outside of the specification limits for the product.

Definition and storage What is the sample unit? A truck may carry 44,000 lbs. of product, but what is the discrete sample unit involved? Bulk trailers present problems in how to take a representative sample. Stratification and bulk-container cleanliness issues also require resolution. Some people pull samples during unloading to make sure all sections have an equal opportunity for sampling. Manufacturers can use special equipment for taking samples from bulk trailers, and for sampling bulk bags, as well. Smaller units are easier to define and sample.

What happens to the sample after it is taken is important to consider during the sampling process. If the samples are not handled correctly, it can affect the validity of the analytical results. This is not a question of aseptic sampling technique. Will the transport and storage of lab samples differ from the actual product? As an example, if one receives raw-material samples on ice overnight, while actual raw materials are in-transit in an insulated trailer for three days from the West Coast, is the analysis valid?

Get that sample Manufacturers have used the same sampling plans for many years. One of the older plans, U.S. Department of Defense Military Standard 105D, was first implemented in 1963 and, with revisions, is still in use. The interesting thing about MIL STD 105D is that knowledge of the suppliers' manufacturing capability is basic to the system, since the standard calls for inspecting product from processes known to be capable of meeting the specification. The simplicity of the standard is that after finding out the AQL and lot size, the sample size is given. A random-number table determines which units are selected.

Harold F. Dodge devised one way to sample product from an unknown source: randomly choose a quantity of samples equivalent to the square root of N, plus one, where N is equal to the total number units in the lot. Many companies use this method; if the noncomplying product is not uniformly distributed, a good chance exists that a company will accept a noncomplying lot.

The word random is also important. When palletized product comes in, how random are the actual samples taken? People don't like to tear down pallets; however, dishonest suppliers have been known to bury defects inside of pallets.

Remember that with large sample units, such as a bulk bag, the formula devised by Dodge does not apply unless there is a way to take a top-to-bottom sample.

Sometimes, all a company wants to know is the general average for the incoming lot. The process can be adjusted to the new average. In this case, the variation from the average causes the problem.

Be skeptical of composite samples. Do they hide unacceptable product variation? If a company receives truckloads of raw materials, but only uses a part of the truckload in each batch, it makes sense to run samples at levels that represent actual usage.

This applies to manufacturers as well as users. If a company makes 40,000 lbs. per shift and tests a sample every hour for composition, what will a user who uses 1,000 lbs. per batch of finished product and tests every batch find? This depends on the stability of the process. W.E. Deming, Ph.D., called this the economics of sampling: the cost of looking versus the cost of not looking. Companies can determine this by a breakeven analysis of the process. Flaig refers to this as the "average time to signal." This measure says that the sampling frequency should be directly proportional to the frequency of out-of-control signals of the process. The less-stable the process is, the more frequently someone will have to take samples to know what is going on.

If a company has experience with a particular type of defect, it can design a sampling program to detect it. One example is cluster sampling. According to Flaig, defects are not "randomly distributed" when they occur consistently in a specific location. It is possible to design a specific sampling plan that will detect this cluster of nonconforming material with better accuracy than with a random sampling plan.

Keeping it consistent Many companies take samples in-process. If they understand the product and process, in-process sampling will reduce the cost associated with rejected materials. Companies first must determine what they can actually test during the run, including what physical restrictions exist to obtaining the required analyses results. Taking a sample removes a very thin slice out of the process. Does this slice represent what is happening in general? The company must determine this if it wants to rely on the in-process data.

It is possible to do 100% machine inspection of various quality factors. Some examples include label placement, label verification, fill weight, vacuum seal "duds," soluble solids, pH, color and so forth. If automatic testing is accurate and practicable, this option warrants consideration.

The point of it all Why is the company taking samples? Some processes have built-in adjustment steps. The sample is used to determine the current composition and calculate any adjustments that are necessary. This requires taking a series of samples from several batches and determining when the mean value becomes constant. This mixing time is used on all subsequent batches. One would repeat the analysis after the adjustment until they are confident that the process works as designed. This method works well when the raw materials and/or upstream processes vary greatly, and the company wants to maintain tight finished-product specifications.

In-process sampling also can determine if the process is still in control. However, Jonathan Cryer, Ph.D., professor emeritus, department of statistics and actuarial science, University of Iowa, Iowa City, notes that an in-control process doesn't necessarily mean manufacturing is making good product. Being in statistical control and meeting specification are different issues. The machine may work as well as it can, but that might not be good enough. A company's ability to process samples usually limits sampling frequency; if that's a problem, take samples often enough to detect changes that fall outside of normal variation.

This assumes that samples are taken at the right point in production. Flaig goes further, saying that improper sampling gives misleading results. Take care that a random sample detects nonrandom behavior.

No magic number of samples applies to every situation. Direct knowledge of the process and product are critical if one truly wants information that will lead to better quality products and more efficient processes.

Bruce Floyd established Process Systems Consulting, Iowa City, IA, after working for more than 30 years in the food-processing industry. He has had extensive experience in sanitation, quality control, regulatory relations, and product and process development (both domestic and international), and specializes in integrating ingredient and manufacturing specifications into total process systems. A graduate of Georgia State University, Atlanta, he has successfully completed all areas of the Better Process Control School at the University of Minnesota, Minneapolis, and has been qualified by the International HACCP Alliance as an instructor. He can be reached via e-mail at [email protected].

Back to top

3400 Dundee Rd. Suite #100Northbrook, IL 60062Phone: 847/559-0385Fax: 847/559-0389E-mail: [email protected]Website: www.foodproductdesign.com