Measurement systems variability for sand compactibility.

This article explains how this valuable statistical analysis is used to measure a foundry process' consistency.

In Part 1 of this two-part series, Measurement Systems Variability (MSV) was introduced through an AFS committee report. It showed that pairing the sample or part variation with the Gage Repeatability and Reproducibility (GR&R) calculation enabled foundrymen to use the analysis for a single gage--or, more TABULAR DATA OMITTED importantly--for destructive tests and/or processes.

This article introduces the compactibility sand test to illustrate how MSV can be used in a practical way.

The GR&R analysis determines how well a gage can reproduce its accuracy by repeated measurements of the same pieces under controlled conditions. MSV determines the capability of a gage or test unit to measure process equipment's ability to reproduce its measurements.

In a GR&R test, repeatability also is termed equipment variation, such as with a vernier caliper or micrometer. In MSV analysis, all of the equipment incorporated in the process is included together. Reproducibility also is termed appraiser or operator variation. With MSV, this includes all operators' influence on the process.

The GR&R analysis includes part variation as a factor of the repeatability in the reproducibility calculation. In a process, the parts or samples have their own variation and, therefore, should be treated as a separate entity. This is accomplished in the MSV analysis.

The compactibility sand test was chosen to illustrate some of the practical advantages of the MSV analysis. It is a simple, well-known test that is actually a process because it requires a sand sample, a compacting device and a measurement to determine the compactibility value. It is a destructive test because once the sand is compacted it can not be used for repeat measurement.

In-Foundry Study

One of the committee's first projects was to have a member foundry do a GR&R study of its compactibility test using its production sand. For a true study of the test, it was decided the sample variation should be minimized by added mulling for better homogeneity of the sand.

After thorough mulling, sand was placed in covered pails numbered one through 10. For a trial, an appraiser rammed one compactibility sample from each pail and measured its height with a caliper. This measurement was then converted to percent compactibility.

A copy of the MSV work sheet is shown in Table 1. The compactibility to the nearest 0.5% was recorded under the correct sample number in the row for each appraiser's trial. Each appraiser followed the same procedure, testing the pails in a random sequence.

After data for all the tests are entered, various calculations of range and average are made. These calculations must be verified to assure that they are statistically correct.

The first check of the data should be determining that the range of the three trials of any sample for each appraiser does not exceed the Upper Control Limit of Range (UCLR). If an individual range value exceeds the UCLR, it is best to delete the three values in question, or, if one individual number is a "flyer," to delete it and then recalculate the average range.

The second check is to determine whether there is adequate discrimination in the range values. If the range numbers show only one, two or three values, or four values with one-fourth of TABULAR DATA OMITTED TABULAR DATA OMITTED the range reading zero, there is inadequate discrimination of the measurements and the test results are questionable. This occurs when the part-to-part variation is less than the smallest unit of measurement and the differences between parts will be lost.

To correct this situation, the discrimination of the measurements must be increased, such as reading to the nearest hundredth rather than the nearest tenth.

Technically, it is incorrect to show calculated numbers to more decimal points than are used in the data. Because of the need to show trends of some of these values, we have shown one decimal point more than the data for all calculations except Cpk, which is generally reported to two decimals.

Compactibility Studies

Table 1 is a copy of a computer work sheet programmed to do the necessary calculations from the data of the original GR&R study of compactibility. For the variability calculations, the specification limits or tolerances must be entered. Both the GR&R and the MSV calculations are included to illustrate the main difference in accounting for the sample or part variation.

In the GR&R calculation, the sample or part variation is included as a correction factor to the reproducibility. The MSV calculation treats the sample or part variation as a separate entity. In this case, there is little difference between the results. Also shown are the standard deviation and the Cpk values for the data.

Ideally, the GR&R value should be no greater than 10%, although up to 30% is acceptable if improvement is being shown over earlier tests. In this illustration, the 20% would be tentatively acceptable and the Cpk of 4.56 is more than double the nominally accepted range. The committee found that if a MSV calculation is not over 100%, the process is viable and acceptable. Many checks have shown that an MSV of about 60% will have a Cpk of about 1.33. For continuous improvement, every effort should be made to reduce the MSV to as low a number as possible.

Table 2 compares the MSV calculations for the preceding GR&R test procedure to the suggested MSV "R&R" procedure for a process. Test samples for the process were collected from the discharge belt of a continuous muller in separate gallon containers. Again, each of the three operators checked compactibility for each trial, from each container, in random order. However, in the latter case, the sand samples were collected over a period of time and compactibility was read as a whole number directly from the percent compactibility scale mounted on the lab squeezer.

The equipment variability for the MSV analysis is greater because of the increased number of pieces of equipment making up the process. The biggest difference between the two was in the sample variation. This difference was expected because the samples were taken over a period of time from a continuous production muller rather than remulling in an attempt to achieve a homogeneous batch of sand.

Production Examples

Shown in Table 3 are actual comparisons of two sand systems with continuous mullers that produce sand for similar molding lines. They differ in that one system is automated and the other is not, but has a full-time muller operator. Compactibility was taken at the muller discharge.

It would be expected that the nonautomated mulling system would have better control because the full-time operator is continuously checking and correcting for compactibility variation. As the comparison shows, the operator variability is substantially lower on the nonautomated unit. However, the equipment and sample variations caused the MSV to be greater. In this TABULAR DATA OMITTED TABULAR DATA OMITTED TABULAR DATA OMITTED example, the MSV is greater than 100% and, therefore, is "out of control" based on the committee's recommendation.

To correct this situation, the muller operator was instructed to make and record compactibility results every four minutes instead of random tests and hand checks. Corrections were to be made when necessary, as in the past. Table 4 shows the effect of the change on the unit. The MSV dropped from 137% to 93%.

For further improvement, there must be some changes to the equipment involved. These two examples show the value of the MSV analysis--the first by comparing equipment and the second through process changes. Sometimes a relatively simple process change can improve an operation that too often is put off because perception indicates excessive equipment changes would be required.

Another example of the effects of a process change is shown in Table 5. Compactibility test samples are taken at the molding machines on this unit. Control of the sand is difficult because of wide fluctuations in return sand temperature. To maintain any consistency at the molding machine, it is necessary to vary the muller discharge compactibility according to temperature; the hotter it is, the wetter the sand.

Some years ago, a tempering tank was installed in the system between the muller and the machines. This tank was being used as a cushion with the level varying at times from full to empty and various stages between. As part of the process change, instructions were to keep the tank full at all times.

If the sand department fell behind, it was instructed to stop the discharge belt from the tank until it caught up again. The MSV dropped from 92% to 81%. While this change showed improvement, there is still a need to further reduce the equipment variability.

In many foundries, sand technicians check compactibility at the muller for on-line control but take samples to the sand lab for reported process control. It was decided to record the compactibility of the production sand check sample taken at the muller deck and also tested in the lab for perceived differences between the two locations. The lab samples were also measured with a vernier caliper and compactibility calculated from the measurement to check discrimination of the test.

The results were surprising. There was almost no difference in the MSV values between the muller deck and the lab in spite of rather large differences in the operator and sample variation. The slight drop in average value was anticipated. The measured sample analysis actually showed a higher MSV value that was not expected. Further testing of these results is now being done.

MSV: A Valuable Tool

The use of the MSV analysis has proven to be a valuable statistical tool. It not only shows how well the process is under control, but also indicates which of the variables (equipment, operator or part) should be concentrated on for continuous improvement.

Being able to analyze production test results on processes--and destructive and nondestructive tests--is beneficial.