DOE meets cavity pressure:
February 21, 1999
Editor's note: Design of experiments for injection molding is a powerful tool when used and applied correctly. Medical and automotive molders have been pushed to use DOE by the FDA and QS 9000 while other molders use it simply because it is a powerful tool. Cavity pressure advocate RJG Technologies Inc. and DOE expert Robert Launsby teamed up recently to assess cavity pressure as it applies to DOE. The results proved that without cavity pressure data, even DOE can lead a molder astray. Cavity pressure provides a crucial missing link in the molding process: information on how the plastic behaves in the mold, regardless of machine conditions. The idea is to use DOE to understand cause and effect relationships between the process and parts, to find settings for an optimum process, to establish cavity pressure alarm settings to discriminate acceptable from unacceptable parts, and to validate the process under simulated long-term conditions Following is the case study conducted by Launsby and RJG using DOE and cavity pressure. For the study, a flat test plaque part was used to simplify the example (Figure 1). The part measures 150 mm by 37 mm by .89 mm and was molded using 756A polypropylene from ATC. The critical dimension for this part is the outside grid length, the distance between two grid marks near each end of the part. The target length for this dimension is 138 mm, ±.25 mm. The mold was run on an 85-ton Van Dorn. Cavity pressure was monitored using an RJG Technologies Dartvision cavity pressure monitoring and control system. Cavity pressure sensors were placed behind ejector pins at the post gate and end of fill. |
Figure 1. The part used in this case study is a simple polypropylene plaque. The critical dimension is the one between the grid marks closest to each end of the part. |
The physical model is a hypothesis of what we think will be the main cause and effect relationships. Our physical model is based on an understanding of the four plastics variables used to explain the molding process. During the DOE, our objective is to either confirm or refute any models that are proposed.
For the case study, two possible models were evaluated. The first model suggested more cavity pressure would yield larger parts. If this is the case, any process parameter that increased cavity pressure would increase the part size. This could include hold pressure, fill speed, transfer position, or material viscosity, among others. The other model suggested faster cooling rates would reduce crystallinity, creating larger parts due to less shrinkage. Here, mold and melt temperature would be the main parameters to evaluate.
Designing the Experiment
A screening study was designed based on the physical model and was used to sift out the most important factors. Based on these results, the following experimental parameters were selected for the design: hold pressure, fill rate, and melt temperature. We chose a Box-Behnken design as the orthogonal array for the study. This design, which is one of many design types available, allows us to study each factor at three levels as well as all possible interactions between variables. The Box-Behnken design called for 13 runs. High, medium, and low levels were established for each parameter (Table 1). The above variables were entered into DOE Wisdom software which prescribed the Box-Behnken experimental runs.
One startling observation for those new to orthogonal arrays is that it calls for multiple changes between runs. For instance, between runs two and three, the melt temperature changes from 370F to 410F and the fill rate changes from 5 inches/second to 1 inch/second. How can we make two changes between two runs and be able to distinguish the differences? The answer lies in the use of the orthogonal array. Because of the way an orthogonal array is balanced, we are able to make multiple changes and to sort out their influences using mathematics. Fortunately, software is available which performs the calculations and presents the results graphically.
Parts were molded using a decoupled process-transfer from a velocity-controlled fill to a pressure-controlled pack and hold at a 95 percent full position. As they ran, cavity pressure data was saved to disk for later analysis. Three parts from each condition were measured on a Starrett optical comparator in reflection mode.
Figure 2. The main effects graph shows how the grid length on the part changes as the three variables change. This graph helps the molder separate critical from non-critical variables. |
Optimizing the Process
Next we graphed the results using DOE Wisdom software. Two types of plots were used for this analysis: the main effects plot and the contour plot. The main effects plot shows which variables have the biggest effect on the part dimension. The main effects plot for the case study is shown in Figure 2. It demonstrates that as fill rate and hold pressure increase, the length of the critical dimension also tends to increase. Melt temperature does not appear to have the same magnitude of impact. All three variables appear essentially linear in influence.
The contour plot allows us to determine predicted optimal machine settings so as to hit the target length of 138 mm on the grid dimension. While many different combinations of fill speed and hold pressure result in the same dimension, a high fill speed was chosen; then the matching hold pressure was selected from the contour plot. Using this data, a fill rate of 4 inches/second and a hold pressure of 6500 plastic psi were selected. A melt temperature of 400F was selected because it appeared to have the lowest process variation. With the process center point established, the next step was to establish cavity pressure alarm limits before checking process capability.
Choosing Cavity Pressure Statistics
Using cavity pressure data, cavity pressure alarms detect abnormal shots before the mold opens. In order to maximize their effectiveness, it is useful to determine how well they predict the quality of the part. To do this, we took cavity pressure and part length data from the DOE and plotted it together to see if there was a correlation. A good correlation means the cavity pressure measurement can be used to pinpoint the length of the part; a poor correlation would provide little predictive ability.
Figure 3: This is the pressure graph generated for the plaque and shows cavity pressure and hydraulic pressure. The problematic mold deflection is noted in both curves. As a result, the injection forward integral was selected as the parameter that best corresponds with in-spec parts. |
Summary data were collected for peak cavity pressure, time to peak cavity pressure, cavity pressure cycle integral, and injection forward integral. The integral data is the area under the cavity pressure curve and is interesting because it represents the whole curve, not just one point. While the cycle integral represents the area under the whole cavity pressure curve, the injection forward integral is the area up to the end of hold.
This information was graphed to determine which pressure parameter best correlated with fill rate variations to produce in-spec parts. Of all the correlation plots, both the end of fill cycle integral and end of fill injection forward integral appeared to be best. However, because the cycle integral data was affected by inconsistent mold deflection, the injection forward integral was chosen for further analysis (Figure 3).
Figure 4. This graph was generated during validation and shows how grid length, cavity pressure, and fill rate correlate. It proves the experiment developed a centered process with pressure alarm limits to produce good parts every time. |
Establishing Alarm Limits
Having chosen which pressure parameter to monitor, we next set alarms for these variables. Using the correlation graph for the end of fill injection forward integral, we determined the injection forward integral values at which the upper and lower specification limits (USL and LSL) were met. This was done by plotting the USL and LSL on the correlation graph and finding the intersection with the correlation band (Figure 4). Note the innermost intersections were used, making the alarm settings tighter. While this would cause some good parts to be rejected, it ensures no bad parts are sent to the customer.
Process Validation Study
In most validation studies, the settings are dialed in to the desired process and a number of parts are run. While this shows the effect of short-term process variation, it does not reflect the effects of long-term variation, such as material viscosity changes, check ring wear, and differences in setup from operator to operator. The objective of our process validation study was not only to demonstrate that the process was centered but also to evaluate the potential effects of long term process variation. For comparison, short-term data was also analyzed.
Before conducting the study, we chose variables to be used to simulate long-term variation. These were chosen based on the main effects and the assumption that raw material viscosity variation and check ring inconsistency could cause potential variation as well. With this in mind, the levels of each variable were selected to simulate expected long-term variation (Table 2).
However, during the early stages of the validation study, it was noted that the center level chosen did not achieve the target end of fill injection forward integral of approximately 3500 psi. Analysis of the part confirmed it was too small. Remember, the Box-Behnken results from earlier predicted a dimension of 138 mm using these centered values. In fact, the injection forward integral and part dimension was not matched until the hold pressure was raised to 8000 plastic psi. We searched to determine the cause of this discrepancy. All other variables were at their correct levels. This led us back to a familiar conclusion: Constant machine conditions do not provide constant parts, but constant cavity conditions do. Although the machine settings were unchanged, other variables (such as viscosity), for which these settings did not account, caused the machine to produce out-of-spec parts. The center point was adjusted to the 8000 plastic psi setting in order to raise the cavity pressure to the desired levels.
With the process centered, validation continued. All together, a total of 13 sets were run, with three of these at the center point to evaluate stability. The other sets were used to evaluate long-term variation. Parts were measured on the optical comparator and data was entered into SPSS's QI Analyst, an SQC software package. Statistical measures of process capability, Cpk and Ppk, were 2.0 and 1.1, respectively. While this is acceptable, the Ppk indicates the process will need to be watched carefully over the long run.
For comparison, the short-term capability data were also measured. These results, which represent parts run only at the process center, show a different picture. Here the process capability appears much higher, with a Cpk and Ppk both greater than 2.0. However, it is likely this short-term data is misleading, because even small disturbances in the process result in significant variations in part dimensions.
Finally, in Figure 4, we plotted the end-of-fill injection forward integral against the critical outside grid length to verify its fit with the DOE results. All data falls within the general correlation band. The cavity pressure alarm limits are set conservatively here, as can be seen in the curve. While several datapoints fall close to the upper and lower alarm limits, they are far from the upper and lower specification limits. If experience in production proved it was difficult to hold the process within the alarm limits while part quality was maintained, there may be justification for exploring slightly looser alarm limits.
Conclusions
Our last step was to revisit the proposed physical model. Our first model, which stated outside grid length would be impacted most directly by the amount of pressure in the cavity, is confirmed by the excellent correlation between length and the injection forward integral.
Further, while fill rate plays a role in determining part dimensions, it does so only in that it affects the pressure at the end of fill. The cooling rate/crystallization model, on the other hand, was not substantiated, since neither tool temperature nor melt temperature had a significant effect on part length.
Finally, had we relied solely on the machine settings predicted by the DOE, we would have been less successful in establishing a robust process. However, by including cavity pressure in the DOE, we were able to provide a process easily reproduced not only on the same machine but on other machines as well. Now it is time to transfer the tool, cavity pressure profile, and the information gathered here to production, wherever in the world it may be.
You May Also Like