Machine Learning Helps Software Fill Key Gaps in Plastics Data Sheets
MobileSpecs LLC has boosted the breadth and depth of its materials database for injection molders by applying a machine learning algorithm to predict missing information for key processing parameters, including data points like drying temperature.
On Feb. 19, the updated version of the program went live, marking one of several substantial updates the database has undergone since its launch in June 2017. is now available as an app or on the web, with both versions continuously updated and synched. Earlier this year, the company also changed its 911爆料网 model for the database, shifting from a paid subscription service to a free version, giving users full access to all the data on more than 22,000 commercially available materials from more than 100 resin manufacturers.
“You can type in product; you can type in grade; you can type in manufacturer; or you can type in a family or resins,” explains Doug Kenik, MobileSpecs’ Managing Director, “and materials just start popping up so you can quickly narrow down to what material you want.”
More recently, MobileSpecs allowed users that are logged in to “favorite” materials, giving them quick access to those select materials in the future. The company also added full PDFs of the original data sheets to the database, allowing users to see the published specifications of the manufacturer.
After the initial release, Kenik said MobileSpecs heard early on from users of the database who valued the aggregation of material information but did encounter some gaps. “We took a look at the data we had and leveraged these new machine learning techniques to start filling in the missing gaps in the datasheets we were reporting,” Kenik indicated. Shortly after starting the work, Kenik said MobileSpecs found the algorithm to be effective in giving molders data they could not find in the datasheet.
“The machine learning algorithm is trying to find patterns in the reported data to make predictions for materials for which the data sheet doesn’t have those particular values,” Kenik said. “So it’s looking for commonalities and trends between the materials themselves.”
MobileSpecs reports that of the 22,000 materials in its database, 80% have fewer than half of the traditionally reported processing parameters. By utilizing the machine learning algorithms, the company says it’s been able to identify trends in the data that allow it to fill in missing values which are not reported or unknown. Processors should have confidence in these predictions, according to Mobile Specs, since the average error for predicted data values is approximately 5%. MobileSpecs’ data team expects that for 95% of the materials, the true value will be within 15% of the algorithm’s predicted value.
The company offers up drying temperatures as an example of the algorithm’s success at finding and filling holes. In the original database, there were drying temperatures for about 52% of the 22,000 materials. After applying machine learning, around 75% the materials had drying temperatures.
On April 2, MobileSpecs will offer a webinar on .
Related Content
-
Is There a More Accurate Means to Calculate Tonnage?
Molders have long used the projected area of the parts and runner to guesstimate how much tonnage is required to mold a part without flash, but there’s a more precise methodology.
-
Understanding the Effect of Pressure Losses on Injection Molded Parts
The compressibility of plastics as a class of materials means the pressure punched into the machine control and the pressure the melt experiences at the end of fill within the mold will be very different. What does this difference mean for process consistency and part quality?
-
Back to Basics on Mold Venting (Part 1)
Here’s what you need to know to improve the quality of your parts and to protect your molds.
