After less than five years of operation, Senvol of New York City has become the go-to company for nitty-gritty data behind additive manufacturing.
Database Helps You Find Materials and AM Systems
Need to find a powder-bed AM system that can run ceramics and create parts at least 6 inches tall? The searchable Senvol Database will list four possibilities, even noting that one of them has been discontinued. You just saved yourself a day of web research.
Want to identify an AM metal with ultimate tensile strength greater than 900 Mpa? The Senvol Database shows 132 possibilities – unless you also require greater than 30% elongation at break – and then the number goes down to 21. Make that a week of research done in a few minutes.
Senvol also offers detailed material characterization datasets, a guide/manual for gathering and documenting consistent AM test data, and access to a data API that can go into third-party packages. But perhaps the most eye-opening use of data shows in a current company project: data-driven machine learning (ML) that analyzes the relationships between AM process parameters and final material performance, to help users quickly zoom in on best operational settings.
Now in its second phase of funding from the U. S. Navy’s Office of Naval Research, the Senvol ML project is aimed at reducing the trial-and-error approach to defining AM material-process procedures. Zach Simkin, co-president of Senvol, says the company has developed a proprietary algorithm that can be “trained” to predict an unlimited variety of performance goals.
“Any material property or mechanical performance target can be addressed,” explains Simkin, “such as density, surface roughness or fatigue life. Because our software is data-driven, we can analyze just about any variable as input and/or output.”
Data Plus Analysis Yields Desired Properties
One of the many beauties of this software tool is the fact that the math behind the algorithm is material/machine/process agnostic. For input, users quantify the desired outcome, such as a final material density of 8.4 g/cm3 then name the input parameters that can be varied. For a chosen AM system, the algorithm already includes default, optimized process parameters. The power of Senvol ML is that the algorithm has already “learned” and continues to learn from previously input datasets.
For example, the analysis result for a powder-bed system could show the range of values for laser power, laser dwell time and point distance that, in any combination, would create parts with that density. The reason for a range (instead of a point), presented as a 3D graphical envelope of possible settings, is because there are trade-offs to be made among the parameters (e.g. high laser power must be compensated with high scan speed so that a user is not putting too much energy into the powder bed) – and users need to work in the real world.
“The Senvol ML algorithm also has the capability of showing the AM user which parameters are sensitive and which are robust based on where they’re located in the dimensional space,” says Annie Wang, co-president of Senvol. “Users might choose to run the system with a faster scan speed, to shorten print time, if they know they can increase the laser power and still end up with a part of the desired density.”
Moving Ahead with Machine Learning
Senvol’s proprietary algorithm has been developed specifically for AM applications; it uses either empirical or simulated data and can be applied to any AM process, any AM machine and any AM material. The work has been validated across multiple datasets during the Phase I Base and Option periods of the company’s Small Business Technology Transfer (STTR) project with the Navy.
Eventually the Senvol ML software will include the following four capabilities:
- Forward Prediction – predicting mechanical performance (such as fatigue life) from a given set of process parameters
- Inversion – given a target value (e.g., desired tensile strength), the algorithm will determine what process parameters to use
- Machine Learning – continuing to learn from previous datasets and applying that knowledge to new datasets, improving output accuracy and reducing the amount of needed data for new builds
- Recommended data collection – suggesting to users just which data points are needed to improve prediction accuracy, again saving time and effort
Adding value to the Senvol ML product is a computer-vision algorithm that analyzes, in real time, in-situ monitoring data. Simkin describes how two technologies come into play here: “With high resolution photos of the build surface, the algorithm is able to detect irregularities on the build surface (e.g., hills, valleys, streaks) and quantify those irregularities. We’ve also developed an algorithm to process thermal data so we can visualize and quantify areas of the build surface that are hotter or cooler.”
In both cases, he explains, the tools analyze data from each layer of the build. Users can then correlate relationships between irregularities in the build and the resulting mechanical performance.
With the announcement of Phase II STTR funding, Senvol says its software will be made commercially available to any company looking to qualify AM parts. Contact Senvol (firstname.lastname@example.org) for more details, especially if you are interested in the beta-stage program.