Get the Facts on… Curve Generation
Generating material data is an advantageous endeavor, regardless of whether you are looking to use additive or conventional manufacturing methods. The right type of data can tell you a lot about your material and how it will behave in different application environments, but the way we obtain material data for additive can sometimes be different from conventional manufacturing (although they do share some similarities as well).
We caught up with Kelsey Rainey, materials engineer, to discuss how data is generated for additive parts and some of the key pointers when comes to obtaining the right type of data.
What is the relationship between alloy selection and curve generation?
Overall, alloy selection helps how you approach the curve-generation process. For example, if you’re interested in nickel alloys, then you’re going to need to consider testing at temperatures above 1000°F. However, if you’re working with aluminum, the tests are going to be more in the low to mid hundreds of degrees F. So, the choice of material is going to change the approach you’re choosing for curve generation and the way that you approach creating your test matrix.
Once you select an alloy, that choice is going to drive the component development. Alloy selection and curve generation is a big intertwined, iterative process. On one hand, you may know what your component is and you’re looking for a material that works, but on the other hand, you may already know a material that works and you’re trying to pair that material with a suitable component. By combining the need for alloy selection and curve generation, it’s going to help drive what you need to test, the field conditions your component will see and what kind of data you need to capture to address that.
What are some of the assumptions additive users make about curve generation?
There are two misconceptions that we regularly encounter. The first is that material property curves are just math. Some users believe that you come up with the test matrix, send the bars off to get tests and the data that you receive is plugged into the computer and your curve is ready to go and design toward. However, at GE Additive, we approach curve generation as a combination of statistical analyses and engineering judgment. Pure statistics can tell us one thing but having engineering experience with certain alloys and test conditions can pick out points in the statistical analysis that might not tell the whole story. You need to marry both approaches before creating your final curves.
The second misconception is that you can generate the material property curves faster with additive than conventional manufacturing methods. In additive manufacturing, there is often a push to get things done faster and reduce time and cost. But when it comes to testing, it’s not always that simple. For many tests, there is a limit to how much you can reduce the time spent. Take creep testing for example. A common creep test is a thousand hours long, and that test will run for the full thousand hours, regardless of whether you’re doing additive or a conventional manufacturing method. Testing takes time and cost investment, and there are only so many places where you can realistically cut down your timeframe.
How do you make a curve and how do additive curves differ from the curves generated for conventional materials?
There are two main approaches to generating material property curves. One way is to take an existing equation and fit your data to that equation. One example is the Coffin-Manson equation, which can be used to fit fatigue data.
We approach curve fitting from a regression analysis perspective. So, we’ll develop a test matrix, test the bars, get raw data points back and fit the data using regression analysis. We look at the data using a few different regression fits, and then we’ll combine those regression fits with our engineering judgment to determine what the final curve will look like. We also have several different criteria for determining whether or not the fit of the data is good.
The additive manufacturing curve generation process is similar to those implemented for conventional materials; however, the big difference comes when you start to look at what independent variables you need to consider. Once a component has been selected, or we at least know what environment the material will be in, we create a test matrix that consists of different property tests across a range of temperatures.
This is where the process starts to differ. In conventional materials, the independent variables you would typically consider are temperature, grain orientation and material heat/lot. In additive manufacturing, we want to consider these variations as well, but they present in different ways. For example, with grain orientation, rather than capturing bars oriented differently within an ingot like we do with conventionally manufactured material, we print bars horizontally and vertically on our print bed to capture any material dependence on orientation. We also need to consider how we’re capturing chemistry variations in our powder and variations across the build plate.
There are different independent variables that we now need to consider in additive that we wouldn’t have to consider in conventional materials, but the general process of taking the test data and generating the curves from that remains essentially the same.
Speaking of variables, what are some of the end-use variables that you need to consider?
One of the questions we get asked frequently is about the type of data collected. There is a baseline set of tests that we would perform if we wanted to develop a material, but there’s no specific component in mind. Once you have your alloy family, we can then tailor the tests. For example, we always run a tensile program and fatigue testing for nickel and aluminum, but the test conditions (temperature for example) will vary depending on the material selected.
Tensile and fatigue tests are the two main tests, so even if there’s no specific component in mind, we will nearly always do some variation on those tests. Once the component has been decided, that will then drive further tests based on the intended environment. The more specific and detailed test programs start to get fleshed out as we figure out what a customer’s component is going to see in the field. The test matrix becomes much more specific and tailored to specific component requirements when the customer can understand what the component is going to see, which will ultimately drive what kind of testing you will do.
How much data, and what type of data, do you need to make a curve?
It ultimately depends on how you want to use your curves and whether you control your process or not. At GE Additive, we make sure that we control our process and put limits in place to make sure that our process is stable (stable parameter sets, materials specification and machine specification). If those steps are complete, and the process is controlled, then there’s not a massive amount of variation in our test data, so you might not need as much data as you might think.
If you don’t have a stable process, don’t have a locked parameter set or if you’ll be testing across different machine types, then you will need a lot of data to ensure that you have a statistically significant quantity of data to create curves. This can get prohibitively expensive, and quickly.
However, if you have those controls in place, obtaining the required data is much more cost efficient and economically achievable. So, it’s a better use of time and resources to first control your process and understand the limits on that, and then generate your data.
The quality of the data is also important as well, isn’t it?
It is. The most important thing to consider when collecting data is controlling the process, followed by understanding and capturing all the pedigree information. Pedigree information in this context would be anything that could impact your test results. This might be from the powder lot to the location on the build plate, the parameter set you used to build the bars or where the bars were treated, tested and/or machined. You want to have a clear traceability to all this information, because data without pedigree is worthless. So, the quality of your data is important. By controlling your process and maintaining traceability of the pedigree of all the bars, you will get quality data points.
What can data and curves tell us about as different material properties, such as surface roughness, as-printed surfaces, or fatigue results?
There are many ways to characterize your surface—from optical to tactile measurements—but the curve generation and test data are great ways to quantify what effect that surface has on your material. You can spend a lot of time in the lab coming up with the actual roughness of a surface, and while those values are useful to know, they don’t tell you what impact that has on the properties of your material. So, by undergoing a range of surface-specific testing programs, we can quantify these values if a customer has an as-printed surface on their component.
These testing programs can tell us what is going to happen with parts, and you can then life your parts that have as-printed surfaces. This is important from an application perspective as you can take the characterization of the surface and put a quantifiable debit on what you can expect from a machined surface compared to an as-printed surface.
Generating data curves for your materials and components can be a much simpler and economically viable option if you have your process controls and characterizations in place first. If you have these controls in place, you have all your pedigree data, and you have considered the additive-specific independent variables, then you don’t need to obtain as much data as you might think.
There are different approaches to generating material data curves, depending on both the material/component itself, where the part will be used and where a user is in their additive journey.
If you’re at any stage of additive development and design or want to know more about it, please contact us.