Training Module #14: Quantify Producibility with HyperXpert (Click-Along)

Create a Metal Tee Design Property that mirrors the definition below and assign it to the "Skins Upper" Structure. This defines the design space of the upper wingskin. The metal tee cross section has a total of 6 dimensions defining the cross-section, and 2 Material inputs to define the skin and stiffener materials, respectfully.

Notice that we are manually setting 4 cross sectional variables (indicated in Green) to be the same across all panels - i.e. forcing them to be constant across the part. This leaves the remaining 4 variable (i.e. not set to a constant value). To better understand why this is important, see Variation as a Measure of Producibility.

A Load and an Analysis Property must also be applied to the upper skin Zones prior to Sizing. Reference the definitions below for both, respectfully:

After assigning the Design Property above to the upper skin (along with a Peak Load Load Property and Stiffened Panel Analysis Property) the Structure is Sized using a traditional HyperX optimization algorithm. By definition, HyperX will find the lightest weight design that attains all positive margin for each individual panel in the upper skin. While lightest in weight, in many cases this solution is not the most producible.

This idea is illustrated in the image below. The Design Results corresponding to two different panels along the wing are shown. As you can tell, while both results fall within the given design space, the designs are very different - even in Zones located right next to each other.

The burden now falls on the engineer to resolve how to take the optimum solution and massage it into something more producible - i.e. something with less design variation over the part surface. Traditionally this would involve some combination of trial and error or a one-factor-at-a-time approach, with no true guarantee of finding the lightest, and producible combination of design variables.

Enter: HyperXpert - Design Insights.

Run HyperXpert on the current upper skin setup using the HyperXpert run form, indicated in the image below.

Once the design of experiments is finished running, the results can be viewed in the HyperXpert Viewer, launched from the Results tab of the Ribbon. Once the viewer is open, open the Run Set Library to select the most recent DOE solution.

Each point represents a workable (i.e. all positive margin) part-level design, characterized by the number of variables that are changing throughout all Zones in the part. By quantifying the number of variables in each solution, we are assigning a number value to the relative Design Variation of each solution. Therefore, each solution can be characterized by its weight and relative producibility. See Variance as a Measure of Producibility.

In general, the upper lefthand side of the plot contains the higher weight, but easier to produce design solutions (because there are less variables changing along the part). Whereas the lower righthand side of the plot contains the lower weight, but harder to produce design solutions (because there are more variables changing along the part). The pareto optimum points are indicated by the pareto frontier, which is automatically plotted. See Understanding Trends in the Point Cloud for more information.

Notice that the maximum number of Variables (x-axis) is equal to 4, as we’d expect from our Design Property definition. The rightmost point that appears at this value is known as the “Anchor Point.” This solution, which includes the most variability in design, is also the lightest. By definition, this is the same result that our typical HyperX Sizing provided - prior to running the DOE. Hovering your cursor over the point you can confirm the Structure weight matches what we saw in Step #2. Click on this point to add it to the Point Grid on the bottom of the interface.

The leftmost points – at 0 Variables – represent the most producible solutions. These are also inherently the heaviest. The leftmost solution – added to the bottom row of the table – is ~70lb heavier than the HyperX optimum solution, but has a constant design (characterized by the boxed dimensions and materials) over the entire skin.

This example lends itself well to several mechanisms for identifying the "best" design. 2 recommended approaches are summarized below.

Using the Tree to Identify Points by Important Linked Variables

Expand the Panel section of the Tree. Notice that for each cross-sectional variable, each possible dimension in the design space is a node on the Tree. Clicking the circle corresponding to a particular value will identify all Points have this value constant. So, if you click Spacing = 2in, the red series that is indicated (and appears in the Legend) is all of the solutions in which the part has a consistent 2in spacing between stiffeners across all Zones.

The filled teal circle in the Tree indicates dimensions that are identified. The Legend in the upper righthand corner of the viewport identifies which series is which. Clicking on one of the spacing dimensions a second time will hide all solutions that contain that value (whether it's constant or not). These same Tree options are available in the right-click menu as well.

Tip

The "Hide/Show" operations operate on Points that contain the selected dimension/material, whereas the "Identify" operation only operates on Points that are constant at that dimension/material. So, more Points will likely be hidden/shown than those that are colored by the identify option. This is by design.

By eye-balling the pareto frontiers for each series, it is easy to see that a stiffener spacing of 2.8in seems to generate lighter optimum solutions.

Now, thinking like an engineer, stiffener spacing is the most advantageous variable to link in terms of manufacturability of a part. So, let’s leave these points (2.8in) identified and double click the 2in circle to unidentify (but still show) the other points.

Maybe – based on the manufacturing processes – it is also advantageous for stiffeners to all be the same height. So, identify solutions based on stiffener height values as well. Notice that selecting heights adds three more series to the Plot, some identified by multiple Constants.

By comparing the pareto frontiers, you can see that the light green (bottom, leftmost point) is the optimum solution. This design has a linked stiffener spacing and stiffener height, increasing the overall producibility of the part. But, because the thickness of the skin and the stiffeners are allowed to vary as needed depending on the load path, this solution is a fairly reasonable weight (only 4lbs over HyperX solution at Anchor Point). So, if you’re willing to give up 4lbs, you can have drastically increased producibility.

Using the Point Grid to Compare Pareto Optimum Points

Another way to get to this same solution would be using the Point Grid. By proceeding from right to left along the pareto curve and selecting each Point, you can build the table shown below – 1 row in the table to show the data corresponding to each Point. This compares the lightest weight solution(s) at each x-axis value.

Being able to directly compare these weight values side-by-side allows you to make an educated decision on which design to select. Using the same logic as above, pick the indicated point. It links the two priority variables, and is the lightest solution that does so. You can also see that after this Point, linking additional variables becomes a lot more costly in terms of part weight.

Once you’ve selected your preferred design, click the “Create Design Property” option – either in the Point Grid or the Point Info window.

This will automatically create a Design Property corresponding to the design you selected - with constant inputs for the variables that are linked. See stiffener spacing and height variables are forced to be constant at the dimensions identified.

Select the Skin Upper Structure to populate the "Zones Selected" dialog box, then click "Assign" to assign this new Design Property to the corresponding Zones.

Size the upper skin Structure and review the results. Notice that: