A central issue in the performance of a blow-fill-seal machine is the achievement of class 100 conditions in the critical fill zone (for the purposes of this discussion, class refers to the number of particles -0.5 pm/ft). These conditions must be met to comply with both FDA and international standards for aseptic filling.
Blow-fill-seal (BFS) is an automated process by which plastic containers are formed, filled, and sealed in one continuous operation. Aside from the economic advantages, BFS is considered a favorable method for aseptic packaging of sterile liquid products because of the limited need for human intervention and, hence, minimal opportunity for microbial contamination. One limitation, however, is the generation of nonviable particles during the plastic extrusion and container formation process. Such nonviable particles provide a potential means of transport for viable microorganisms into the open container before the sealing operation. In efforts to protect the product from that potential contamination, BFS machine designers and fabricators have installed shrouds around the critical fill zone (CFZ).
An article describing the iterative process used to develop a particulate control system (PCS) appeared in the February 1998 issue of Pharmaceutical Technology (1). This article documents the process used to optimize the design of the PCS. (The testing conducted in this study focused on particulate levels in the CFZ. There were no microbial challenges or media fill runs in these tests.)
Background
The objective of the first case study was to design a PCS that would satisfy the regulatory requirements of maintaining class 100 conditions in the CFZ while operating in the dynamic state (1). The case study discussed how the obstacles encountered were addressed and eventually resolved. In all, three major design revisions were made to meet the objective. That was accomplished primarily through a combination of smokestick studies, particle-count testing, and trial-and-error methods. Results of the studies illustrated that particulate levels in the CFZ are a function of, but not limited to, environmental particulate levels, unit design, high-efficiency particulate air (HEPA) velocity, damper opening, and chimney air velocity. The case study concluded that PCS development is a complex undertaking that requires a substantial amount of custom design, testing, and refinement.
Although the PCS had been shown at that point to be effective, it remained unclear whether the system could be improved or optimized. What were the cause-and-effect relationships between the system variables? Could the system settings be altered to improve performance? Were there interactive effects between any two or three variables? Those questions are the focus of this article.
Because of the large number of variables involved and the possibility of interactions (the unexpected effects that result when two or more input factors are changed simultaneously) between those variables, conventional testing methods were scrapped in favor of a more sophisticated test method known as statistical design of experiment (DOE). That method of experimentation is widely used and is well suited to the type of test in which the results can be quantifiably measured. The primary benefit of the DOE method is the identification of cause-andeffect relationships between the system variables. That facilitates the determination of the optimal design settings.
Design of Experiment
The statistical DOE method can be broken down into the following six-step process.
1. Define the experimental objective.
2. Identify the input factors and their levels (settings) to be evaluated.
3. Determine the type of experiment to run based on the number of factors and levels selected. Determine whether to run a full factorial or fractional factorial experiment.
4. Conduct the experiment.
5. Analyze data and interpret results.
6. Implement improvements.
The experimental objective was to identify all of the critical input factors and their settings on the PCS while operating in the dynamic state.
The input factors chosen were based primarily on the experience gained from the previous case study. In all, six factors (main effects) were identified and believed to be critical to the design. Table 1 lists the input factors and their test values. Figure 1 shows front and side views of the PCS and the location of the input factors. Figure 2 shows a top view of the two isolation plates. The holes in the isolation plate are slightly larger than the filling needles.
Experiment type. To better understand the cause-and-effect relationship of this system, a 1/4 fractional factorial experiment (FFE) was performed to examine the main effects and to identify any two-factor interactions. Higher-order (three-, four-, five-, and sixlevel) interactions do occur but rarely are significant when compared with main effects and two-factor interactions.
The advantage of running a % FFE instead of a full factorial test is that the number of test runs is reduced by 75%. In that case, a full factorial test with six variables tested at two levels would require 26, or 64, runs.
Using the 1/4 FFE, only 16 runs were required. In addition, each of those tests was repeated three times, which means that only 48 runs were required instead of 192.
The advantage of running a full factorial (64-run) experiment is the amount of information gathered. Analysis of a full factorial test would yield an estimation of 63 effects: all 6 main effects, 15 two-factor interactions, 20 three-factor interactions, IS four-factor interactions, 6 five-factor interactions, and 1 six-factor interaction. Although that information would have been helpful, my primary interests were in the main effects and the existence of two-factor interactions.
The 1/4 FFE chosen for testing is a Resolution IV design. That means that the main effects can be detected free and clear of two-factor interactions; however, twofactor interactions are confounded with each other. That type of test, referred to as a "screening" design, screens for large main effects and possible two-factor interactions with a minimum number of runs.
Because the chosen test had 16 runs, a total of 15 effects or groups of confounded effects could be estimated. The main effects accounted for six effects, the two-level interactions accounted for seven, and the three-factor interactions accounted for two. That test design yields the following structure for the main effects and confounded two-factor interactions. (The confounding structure for three-factor interactions is not shown.)
chimney air velocity (ft/min)
HEPA flow rate (%)
damper (% open)
height of HEPA chamber (in.)
knife cut setting (single or double)
type of isolation plate (slotted or holes)
chimney air velocity X HEPA flow rate = knife cut X isolation plate
chimney air velocity X damper = height of HEPA chamber X isolation plate
chimney air velocity x height of HEPA chamber = damper X isolation plate
chimney air velocity x knife cut = HEPA flow rate X isolation plate
chimney air velocity X isolation plate = EPA flow rate X knife cut = damper x height of HEPA chamber
HEPA flow rate X damper = height of HEPA chamber x knife cut
HEPA flow rate X height of HEPA chamber = damper x knife cut
chimney air velocity x HEPA flow rate x damper
chimney air velocity x HEPA flow rate x height of HEPA chamber.
For simplicity, only the interactions listed in bold type will be discussed. However, it is possible that the confounded interaction (regular type) is being observed. That is what is sacrificed by fractionating (that is, reducing the number of tests) - the ability to evaluate each interaction independently.
Table 1 shows that there were three factors to be tested at three levels (low, midpoint, and high) and three to be tested at two levels (low and high). The purpose of testing at three levels is to determine whether there is a nonlinear response over the specified range. A two-level test will estimate a linear response over the test range. The three midpoint tests add eight runs to the original 16-run test, for a total of 24 tests.
Table 2 shows the 24-run orthogonal test matrix developed to test the main input factors (see the "Orthogonal Experimental Design" box). Note that for each row, the six input factors are altered in a systematic manner (from high to low) an equal number of times. That is one requirement for an orthogonal matrix. Although that test matrix does not show the interactions, they are accounted for in the analysis of the test data. Materials and Methods
Materials. The BFS machine used for testing (model 912, Vital Pharma, Inc., Riviera Beach, FL) was equipped with a five-cavity, 3-mL mold. Other equipment included a variable-speed HEPA blower rated from Q to 250 cfm (model 250, International Portland Corporation, Hillsboro, OR); a certified 99.99% efficiency rated, 0.3-gm HEPA filter (Flanders Filter, Riverhead, NY); and a digitally controlled, variable-speed blower (Reliant Electric, Sumner, IA), which was used to maintain negative pressure in the chimney. During a recent calibration, the blower supplied 150 cfm of air at a 50% setting. The PCS was designed to easily change hardware settings.
Instrumentation included a calibrated particle counter with a volumetric sampling rate of 1 cfm and particle-size resolutions of >0.3, >0.5, >0.7, >1, >5, and >10 gm (model CI-7300, Climet Instruments, Redlands, CA). A resolution of >0.5 gm was chosen for this experiment. A digital anemometer was also used (model HHF300, Omega Engineering, Inc., Stamford, CT).
Methods. Testing was conducted in a machine acceptance area - an uncontrolled environment. Particulate levels measured in several locations around the BFS machine were an average of class 300,000. Each of the 24 tests (patterns) shown in Table 2 were repeated three times for a total of 72 test runs. The results are shown in columns Yl-Y3 of the test matrix (Table 2).
Results and Discussion
Results were analyzed using analysis of variance (ANOVA), a statistical method used to mathematically quantify sources of variation, and the results are shown in Table 3. Of particular interest is the column F ratio. Statistically, that represents the ratio of the variability caused by changing a factor (for example, chimney air velocity) to the variability caused by random error alone. Any value larger than 4.0 is significant, with a 95% confidence level. Of the 15 factors analyzed, only five factors were statistically significant (that is, F ratio >4.0). Of those five factors, four were main factors (damper, height of HEPA chamber, knife cut, and isolation plate) and one was a two-level interactive factor (HEPA flow rate x height of HEPA chamber).
Figures 3-8 illustrate the average independent effect on the response of changing the input factor. Note that the figures are graphed using the natural log of the particle count.
Of all the interactive effects analyzed, the most significant was HEPA flow rate X height of HEPA chamber. Figure 9 illustrates that the amount of HEPA supply (20% compared with 80%) had a significant effect on lowering particulate levels (from 1,823 to 606 particles/ft3/min) when the HEPA chamber was in the low position (0 in.). However, when the HEPA chamber was in the high position (0.375 in.), the HEPA supply had virtually no effect on particulate levels.
Figure 10 summarizes the combined effects of the three most significant input factors (that is, those with the largest F ratios). Eight combinations are shown (three factors at two levels: 2^sup 3^ = 8). Note that there are 64 possible combinations (26 = 64). When the three input factors were set at the worst condition, the average particle count was 7,332 particles/ft3/min.
However, when the three input factors were set at the optimized condition, the average particle count was 3 particles/ft3/min. Each of the points in between are combinations of the worst and optimized settings.
The optimized settings for order 8 in Figure 10 were
height of HEPA chamber = 0.375 in.
isolation plate = slot
knife cut = single.
The final optimized settings for the other three input factors, HEPA flow rate, chimney air velocity, and damper, were
HEPA flow rate = 20% (Because of the interaction between the HEPA flow rate and the height of HEPA chamber, the HEPA flow rate could be set anywhere from 20% to 80% if the HEPA chamber height is 0.375 in. The 20% setting was chosen in view of operating costs. That is also the case with the chimney air velocity.)
chimney air velocity = 300 ft/min
damper = 80% open
isolation plate = slot
knife cut = single
height of HEPA chamber = 0.375 in.
A Taylor series mathematical model was developed to predict the particle levels as a function of the input variables. The following response equation for the natural log of the particle count (Inpc) was developed from the results of the test data.
1n^sub PC^ = 4.575 - (2.513 X height of HEPA chamber) + (0.856 x isolation plate) - (0.425 x knife cut) - (0.277 X HEPA flow rate) + (0.271 X HEPA flow rate X height of HEPA chamber)
The value for each of the input factors is either + 1 or -1. Using the height of HEPA chamber = 0.375 in. (+ 1), isolation plate = hole (+ 1), knife cut = double (-1), and HEPA flow rate = 20% (-1), the response equation becomes
1n^sub PC^ = 4.575 - (2.513 x 1) + (0.856 x 1) - (0.425 X -1) - (0.277 X -1) + [0.271 x (-1) x 1] = 3.35 PC = e^sup 3.35^ = 28.
Discission. The design optimization of the PCS is an involved process that requires the majority of the work (experimental design) be completed before testing. However, this is a cost-effective method because the experiment is predefined on paper, which minimizes the actual costs associated with testing.
The DOE process revealed that four of the six main input factors were significant (as measured by the F ratio using ANOVA.) That is not to say that the remaining two factors (HEPA flow rate and chimney air velocity) are not important; it simply indicates that their influence on particulate levels is not as great over the specified range when compared with the other four factors.
Of the nine interactions investigated, only one two-level interaction (HEPA flow rate x height of HEPA chamber) was significant as indicated by the F ratio in the ANOVA. Although that interaction was confounded with the damper X knife cut interaction, it was evident that it was the HEPA flow rate X height of HEPA chamber interaction that was observed. The knowledge of that interaction was quite beneficial because it revealed that the particulate levels were independent of the HEPA flow rate if the HEPA chamber was in the upper position.
This study revealed three unexpected results. Chimney air velocity has a very small effect on particulate levels over the given range. The slotted plate is significantly more effective than the hole plate. The height of the HEPA chamber is by far the predominant factor. For reference, other tests affecting the particulate counts were conducted outside of the test matrix. In one test, the HEPA flow rate was set to 0% and the chimney air velocity was 0 ft/min. Using the hole isolation plate and the HEPA chamber in the lower position, the particle levels were measured at an average of class 1,100,000 particles/ft3/min. That worst-case result illustrates the effectiveness of the PCS.
An additional analysis confirmed the linearity of the first three input factors (listed in Table 1) over the specified range (not shown).
Conclusion
The study met the design objective of minimizing the particulate levels while the PCS operated in the dynamic state. In addition, a more thorough understanding of the cause-and-effect relationship between the critical input factors and the particulate levels was obtained using the DOE method.
Acknowledgment
I wish to thank Kim Vukovinsky, the statistician who provided assistance in experiment design and data analysis.
[Sidebar]
ORTHOGONAL EXPERIMENTAL DESIGN
Orthogonal experimental design, which is a part of DOE, is a testing strategy that uses the unique properties of an orthogonal array to efficiently pinpoint the source(s) of variation. The unique quality of orthogonality means that factors can be evaluated independently of one another; the effect of one factor does not influence the estimation of the effect of another factor (2). In other words, although several input factors may be altered simultaneously during testing, the analysis will reveal the effect of each individual factor.
[Reference]
References
[Reference]
(1) J. Price, "Blow-Fill-Seal Technology: Part I, A Design for Particulate Control," Pharm Technol. 22 (2), 62-72 (1998).
(2) P.J. Ross, Taguchi Techniques for Quality Engineering (McGraw-Hill, London, 1998), p. 66. BP
[Author Affiliation]
Jeff Pri P is the engineering manager at Vital Pharma, Inc., 1006 W. 15th Street, Riviera Beach, FL 33404, (561) 844-3221, fax (561) 844-3661, email jeff@vitalpharua.com. This article appeared in the February 1999 issue of BioPharm 's sister publication, Pharmaceutical Technology.
