Process control: tensile testing.

The test results didn't show what was expected. Melting knows the lab goofed. The lab knows it ran the test correctly. What happens next?

Most of the time, the lab repeats the test procedure with a purchased certified reference material, or standard. If the results come within the published limits for the standard, the lab tells melting it was right. If the results are not close to the published value for the standard, the lab investigates what happened to its testing process.

Getting a result close to the published value proves to the lab that it's in control of its process. However, sample preparation is part of the testing process and the organization providing the standard performed some type of sample preparation. Running standards excludes sample preparation and, therefore, does not check the entire testing process, but it remains a good way to see if the rest of the procedures are being performed correctly.

There are no standards, however, to monitor tensile testing. So, usually, another test bar representing the period is pulled. If the bar passes, the melting department says, "See, we told you we were right!" Usually, the reason for the questioned test results is not pursued too far.

The tensile test remains one of the most important foundry tests because it proves the metal meets customer requirements. However, while there are other procedures that can be performed, checking the tensile test process is usually limited to an annual calibration check of the tensile machine's load cell.

Unfortunately, the tensile test is sensitive to more variables than just the load cell. Head alignment of the tensile machine, rate of the test, alignment of the test bar during machining and surface finish attained on the test bar can affect the results. Yet, there is no process control on the testing because there are no standard test bars.

Foundries have produced "in-house" standards for their spectrometers for some time. Following are three steps to using "in-house" standards to monitor the process of tensile testing.

1. Make and pour several standard test bar molds. Most believe iron poured after being held different lengths of time will have different tensile properties. Pouring several test bars in a short enough period so that the properties will not be significantly different, however, can be done. In most operations, at least 25-30 molds can be poured fast enough so the properties will be the same. To minimize variability, make the molds as uniform as possible. 2. Verify the bars are normally distributed. Each mold will produce two test bars. All bars should be sequentially numbered. Several bars should be randomly selected and tested to determine if the results are normally distributed. The full range of variables used in production testing should be used in this phase. If several machines and machinists are normally used to prepare test bars, they should also prepare these tests. Likewise, if different technicians pull the production tests, they should be included.

The results of these tests should be carefully studied to see if any variables cause statistical abnormality. Do the bars poured last fit the same distribution as those poured first? If more than one machine is used, do they all fit the same distribution? If any of these variables causes abnormal distributions, the cause should be found and corrected.

This information will be used as the basis of comparison of the information obtained in Step 3.

3. Test remaining bars periodically. Remaining bars can now be used as standards. Periodic testing using the "standard bars" and statistical comparison of the results can tell if the testing process remains stable. When results of such tests suggest a change in the process, investigate the cause. As with other in-house standards, these steps aren't a good way to check if you're getting the "right" answer. It does tell you whether you're getting the same answer you would have gotten before. Round-robin testing can be helpful to check if your answer is "right."

Obviously, having to prepare in-house standards is not ideal. But since we sell castings on the basis of tensile tests, it's logical that we should monitor the process control of the testing.

Practical Ways to Apply Steps

Procedure 1

Step 1. Start with 25 molds the first time the procedure is used. If Step 2 above reveals normal distribution, make and pour an additional mold the next time. This process is repeated until physical limitations are reached to reduce costs per "standard" bar.

Step 2. Test at least 25-30 bars. The minimum number that shows normal distribution is 30. Smaller numbers will reduce the statistical certainty of the distribution being normal. Optimally, this would be done when the tensile machine has just gone through certification.

Step 3. Quarterly, test five of the remaining bars. Results of average and variance should be compared to the results obtained in Step 2 using the Student 't' and/or the "F" distribution tests. The process has changed if a statistically significant difference is indicated in the average or the variance. If a change is noted, investigate.

Procedure 2

Step 1. Same as procedure 1.

Step 2. Same as procedure 1, except the information is collected so it can be used to construct an individual X moving R chart.

Step 3. Every other week, test one bar and chart results on the individual X moving R chart. If charting shows an out-of-control situation, investigate.