Machine learning and specifically, deep learning, is a powerful tool to establish the presence (or absence) of such correlations with its ability to flesh out relationships and trends that are difficult to establish otherwise. Watch this webinar to learn how deep learning tools—in this case, convolutional neural nets—can more accurately:
If you have any questions about these or any of our other solutions, please contact us. You can also follow @BrukerNano on Twitter for event information and updates.
As microscopists, we often hope – and intuit – that the microstructure and local mechanical properties we measure correlate somehow to bulk properties, but this correlation is often difficult to quantify.
In this webinar, we use deep learning tools such as convolutional neural nets (CNNs) to analyze AFM images of impact copolymers (polymer blends of polypropylene with micro-sized domains of rubber) collected using TappingMode and PeakForce QNM. First, CNNs are used to successfully classify TappingMode phase images of a variety of ICPs where the rubber morphology and distribution vary. The model shows that the PF-QNM deformation channel provides the best accuracy in classification.
Next, we use a regression-based CNN to correlate the AFM images with bulk mechanical properties; both phase images and the PF-QNM deformation channel were used separately in the model. The results show that the deformation data exhibited a superior correlation with the bulk mechanical properties, with high accuracy for predicting flexural and Young’s modulus, ultimate elongation, and impact toughness.
Dalia Yablon, Ph.D.
Research Chemist, SurfaceChar LLC
Ishita Chakraborty, Ph.D.
Stress Engineering
John Thornton,
Senior Application Scientist, Bruker
Bede Pittenger, Ph.D., Sr. Staff Development Scientist, AFM Applications, Bruker Nano Surfaces