Critical to Quality physical dimensions of glass vials – Part 5. Flange height

Hello everyone – after a brief gap in continuity, I’d like to welcome you back to the final official post in a series on Critical to Quality (CTQ) dimensions in pharmaceutical glass vials.? I use the term “official”, because my introduction to this series listed what might be considered five CTQ dimensions: 1) Vial height, 2) Body outer diameter, 3) Flange outer diameter, 4) Flange inner diameter, and 5) Flange height.? Flange height (also known as Flange thickness, Lip height, Lip thickness, etc.) is therefore the final topic on our list.? With that said, I don’t want to fully ignore the potential importance of other physical dimensions.? For example, a vial that does not meet the ISO 8362-1 specification for bottom concavity (also called the “push up”) is obviously an issue.? I may therefore continue this series to cover other physical dimensions of interest and their associated measurement.

The choice to put this post last in the series was intentional.? I think a majority of experts would agree that flange height measurement can be surprisingly complicated, and so I needed some time to consider my approach to this topic.? In the meantime, I was also very fortunate to begin chatting with Valentin Mayer-Eichberger , COO at Isotronic GmbH , a manufacturer of visual inspection systems for pharmaceutical glass packaging.? We share a common interest in this topic, and so I was thrilled when he agreed to support this article with visuals and data analysis that significantly improved upon what I could have accomplished alone (see Footnote 1).

Figure 1 sets the foundation for our discussion – it’s a simple illustration focused on a cross-sectioned portion of a vial flange.? ISO 8362-1 includes a flange height specification of 3.6±0.2 mm – but what does that really mean?? The illustration in Figure 1 is supplemented by a number of red lines.? These lines demonstrate two examples of methods that could hypothetically be used to define flange height.? For example, one method could consist of: 1) a horizontal line drawn tangent to the top of the vial (i.e., the upper boundary of what also defines vial height), 2) a vertical line drawn tangent to the flange outer diameter, and 3) an angled line drawn tangent to the underside of the flange, which is also specified to be 10°±5° (see Footnotes 2 and 3).? Notice that the vertical line forms two intersection points with the lines drawn tangent to the top of the vial and the underside of the flange.? The vertical distance between these two intersection points is the ISO 8362-1 definition of flange height.? This may all sound straightforward, but the construction of these tangent lines can be difficult in practice. The appearance of vials in the real world (as opposed to the idealized geometry of a drawing) can introduce significant variability in the measurement of flange (more on this later).?

One solution to this variability problem is to use an offset distance, labeled by the quantity X in Figure 1.? The ISO 8362-1 method for measuring flange height would simply be an offset distance of X = 0.00 mm.? In this case, I’m demonstrating the use of an offset distance of X = 0.50 mm, thereby moving the vertical tangent line into the flange.? For simplicity, I’m retaining the upper intersection point defined by the vertical line (the tangent to the flange OD) and the horizontal line (the tangent to the top of the vial).? Let’s instead look at changing the lower intersection point.? The lower boundary of our flange height is now defined by the intersection of the offset vertical line with the actual underside surface of the vial flange (see Footnote 4).? Because we’re still working with an idealized drawing of the vial, the intersection point at an offset distance of 0.50 mm still falls perfectly on the line drawn tangent to the underside of the flange.? The table included in Figure 1 shows you various scenarios for this case.? For example, let’s assume that the ISO 8362-1 definition of flange height (i.e., an offset distance of X = 0.00 mm) for an idealized vial having an underside flange of θ = 10° exactly equals 3.60 mm.? If we apply an offset distance of X = 0.50 mm to this same vial, we would instead determine the flange height to be 3.68 mm (still within the ISO spec having a tolerance of ±0.2 mm).? So what have actually gained here? – nothing, because we’re working with an idealized vial.? In a real manufactured vial, the underside of the flange may have a more curvilinear appearance that complicates the construction of a tangent line.? As a result, it’s entirely possible that the best-fit tangent line to the underside of the flange does not perfectly coincide with the intersection point defined by applying an offset distance.? Stated simply, the use of an offset distance (in conjunction with good cameras and appropriate analysis software) can significantly reduce the ambiguity and variability of flange height measurements.? But don’t just take my word for it, let’s look at some real data.

Figure 2(a) shows a camera image of the shoulder and flange regions of a vial.? The red box within Figure 2(a) defines the boundaries of the magnified view shown in Figure 2(b).? You’ll notice a number of colored lines overlaid on this screen captured image provided by Valentin.? There are numerous algorithms that can be defined for measuring flange height, each of which depend on differing combinations of these lines (more on these in a moment).? For example, you could take my prior example of applying an offset distance of X = 0.50 mm one step further by eliminating the use of the horizontal line drawn tangent to top of the vial.? Instead, the intersection of the offset vertical line with the actual underside and topside surfaces of the vial flange could be used to determine flange height – this is approach is demonstrated by the vertical yellow line visible in Figure 2(b)).

So what is the impact of various algorithms on the measured flange height?? Do we observe any systematic differences?? To answer these questions, Valentin’s team at Isotronic was able to analyze inspection data for a relatively large sample population (>360,000 vials) and applied four different algorithms that I will call Methods 1 through 4 for the sake of convenience.? The key features of each method are illustrated in Figure 3 – to briefly summarize:

·?????? Method 1 uses a single offset distance from a tangent line to the flange edge.? The distance between intersections of the offset line with: 1) the underside of the flange and 2) and tangent to the top of the flange defines the flange height.

·?????? Method 2 uses a single offset distance from a tangent line to the flange edge.? The distance between intersections of the offset line with: 1) the underside of the flange and 2) the topside of the flange defines the flange height.

·?????? Method 3 uses two offset distances (Offset 1 and Offset 2).? The intersections of these offsets with the underside of the flange (shown as black dots) defines a tangent line to the underside of the flange.? The distance between intersection of: 1) the tangent lines to the flange edge and the top of the flange and 2) the tangent lines to the flange underside and the flange edge defines the flange height.

·??????Method 4 uses two offset distances (Offset 1 and Offset 2).? The intersections of these offsets with the underside and the topside of the flange (shown as black dots) define two tangent lines.? The distance between the intersections of: 1) the tangent lines to the flange underside and the flange edge and 2) the tangent lines to the flange topside and the flange edge defines the flange height.

The results of applying each of the four algorithms to the same data set of vial images are shown in Figure 4.? A key aspect of these results is that the individual histograms describing the flange height distribution from a given algorithm have been “re-centered” to the same value to aid in visualization and comparison (see Footnote 5).? You might conclude that there are some apparent differences amongst the distributions produced by each method from a simple visual comparison – but are they truly different?? The results of a Kolmogorov-Smirnov test indicate that we can readily reject with 95% or better confidence the null hypothesis that any two distributions are equal.

This outcome also has some potentially important implications for determining whether a manufactured vial is within specifications or not. Comparing these re-centered distributions to allowable tolerances for flange height, we find that the hypothetical ejection rate (meaning the rate at which a vial is determined to be out of specification) can be dependent on the chosen algorithm, as shown below:

·?????? Ejection rate for Method 1: 0.5065%

·?????? Ejection rate for Method 2: 0.6170%

·?????? Ejection rate for Method 3: 0.5157%

·?????? Ejection rate for Method 4: 2.4052%

We have an important outcome here – Method 4 produces an apparent ejection rate that is around 4 to 5 times higher than the other three methods.? And remember that we are applying these four methods to the same set of vial images.? These results aren’t necessarily being provided to convince you that one algorithm is definitively superior to another.? It’s more about demonstrating that the methodology can have a statistically significant effect on our determination of flange height.? If nothing else, it’s critical that vial suppliers and their customers are aligned on the methodology used to determine flange height.

A perhaps more interesting question is the extent to which the flange height algorithm impacts the risk assessment of container-closure integrity (CCI) based on dimensional stack-up analysis.? If you’re new to the topic, a “stack-up analysis” is a process of considering the accumulated variation in the tolerances of multiple interacting components.? For example, the CCI of a container-closure system consisting of a glass vial, an elastomeric stopper, and an aluminum seal will depend in part on whether these components are dimensionally compatible with a given set of tolerances.? The simplest analysis considers the extreme ranges of allowable tolerances for each component. ?For example, the nominal dimensions plus/minus the allowable tolerances along the vertical direction with a container-closure system (CCS) are:

·?????? Vial flange height = 3.6 ±0.2 mm per ISO 8362-1

·?????? Stopper flange height = 2.00 ±0.25 mm (for 13 mm OD vial flange) or 3.30 ±0.25 mm (for 20 mm OD vial flange) per ISO 8362-2

·?????? Crimp seal skirt length = 6.3 ±0.2 mm (for 13 mm OD vial flange) or 7.3 to 7.8 ±0.2 mm (for 20 mm OD vial flange) per ISO 8362-3

·?????? Crimp seal thickness = 0.168 mm to 0.242 mm

?

Let’s assume we’re assembling a CCS using a 2R vial, meaning a vial with a 13 mm OD flange.? Now consider the thickest possible vial flange (3.8 mm) and stopper flange (2.25 mm) on top of each other.? A crimp seal is then placed on top and a capping machine compresses the entire assemblage such that the stopper is 20% shorter – i.e., 1.8 mm.? The total height of the (incompressible) vial flange and (compressible) stopper flange is now 5.6 mm.? Finally, let’s assume the crimp seal has the longest possible skirt length (6.5 mm) and greatest thickness (0.242 mm).? This particular combination means that there will be 0.658 mm (i.e., 6.5 mm - 0.242 mm – 5.6 mm = 0.658 mm) of available aluminum skirt to crimp underneath the vial flange during the capping operation.? Is this good or bad? – that’s best left to a separate discussion for a future post.? Instead, let’s recognize that the probability of assembling a CCS from this exact combination of components (or any other combination at the edges of allowable tolerance ranges taken from product specifications) is presumably quite small.?

Assuming the dimensional data are available, a better approach might consider how the size distributions of real manufactured components stack together under an assumed level of stopper compression (see Footnote 6).? This approach includes knowing the distribution in flange height for a given population of glass vials.? However, we also just discussed how there are multiple algorithms available for vial flange height measurement, none of which are necessarily “most correct” (see Footnote 7).? I have yet to see any studies considering how a given algorithm might influence the reported flange height distribution for a vial and, by extension, what this could mean for a CCS stack-up analysis – let’s call it a potentially interesting research topic for a curious packaging engineer.

Questions or comments? – please leave them below or feel free to contact me directly.? I’d also like to close by thanking Valentin for the fruitful discussions and the willingness to share the images, data, etc. that greatly contributed to this post.

?

Footnotes

1.?????? I’ve been pleasantly surprised by all of the new connections, the questions, people sharing their specific experiences, etc. since starting this writing project on LinkedIn about a year and a half ago.? In particular, there’s been an increased number of conversations over the past couple months about the possibility of working on collaborative posts – this article is the first such example.? I’m completely open to the idea provided that the original intent of this project is maintained – i.e., the content is intended to have a non-commercial, technical focus with the goal of educating the users of pharmaceutical glass packaging.

2.?????? How are these lines being drawn?? It’s not as if we can take an actual vial and begin marking off lines with a straightedge tool and a marker.? Instead, we use an image acquired under magnification with a camera.

3.?????? Prior posts in this series have demonstrated the use of relatively simple hand tools such as calipers for making measurements.? Calipers can in theory be used to measure flange height, but I personally don’t recommend it.? I find the underside flange angle makes it difficult to obtain a consistent measurement.? Similarly, you could also consider using some sort of a go/no-go gauge for testing flange height.? Just remember that such gauges only produce a Pass/Fail result – we don’t get actual flange height data.

4.?????? The intersection of the offset vertical with the underside surface of the flange could in principle be determined manually by someone viewing the image.? This approach might work when evaluating a relatively small number of samples, but it would not be possible when performing inspections in real time during high volume manufacturing of vials.? Instead, additional edge detection algorithms are used to find the underside flange surface.? It’s also worth noting here that the detecting the top edge of the flange can be more difficult than you might expect.? This is because you are trying to pull an edge out of the flange sealing surface when viewed through a lens.? In an ideal world, this view would product a crisp, perfectly straight line.? However, a real flange sealing surface has geometric variability that introduces uncertainty.

5.?????? Let’s explore this re-centering approach a bit further.? What we’re essentially doing is re-calibrating the output of each algorithm by assuming the mean value of a given distribution is equal to the nominal flange height and then offsetting the entire distribution accordingly.

6.?????? Refer to the following article for an in-depth discussion of using the dimensional distributions of real packaging components to aid in risk assessment: Bucci A, Ho L, Orme L, Zeng Q (2020).? A vial container closure system performance optimization case study using comprehensive dimensional stack-up analyses.? PDA Journal of Pharmaceutical Science and Technology, 74: 368-376.

7.?????? I’ve been entirely focused on the flange height of vials in this post.? I’m less familiar with the details of characterizing the stopper dimensions.? Readers please chime in – are there similar issues with measuring the flange height of stoppers??