image: (A) Inverse control architecture, which can reproduce the strain field by inputting target strain field images. (B) Forward control architecture, which can predict the strain field by inputting the voltage input array for DEA array.
Credit: Jue Wang, School of Mechanical Engineering, Purdue University.
A research paper by scientists at Purdue University presented a deep learning method that enables the customization of complex strain fields according to specific requirements.
The new research paper, published on Aug. 14 in the journal Cyborg and Bionic Systems, used a deep learning method based on image regression and achieved to predict and customize complex strain fields.
Traditional bioreactors, powered by pneumatic actuators or motors, struggle to generate complex strain fields due to limited control over individual actuators. However, fields like cardiovascular biomechanics and tissue engineering require more advanced customization. “In the field of biomechanics, customizing complex strain fields according to specific requirements poses an important challenge for bioreactor technology, primarily due to the intricate coupling and nonlinear actuation of actuator arrays, which complicates the precise control of strain fields.” Explained study author Jue Wang, a PhD candidate at Purdue University. “To address the challenge, we employ image regression-based machine learning for both replicating target strain fields through inverse control and rapidly predicting feasible strain fields generated by the bioreactor in response to control inputs via forward control.”
The entire process is bifurcated into four sequential stages: Initially, authors collected data through the establishment of a finite element analysis (FEA) simulation model. In the FEA, the device was pre-stretched, followed by the random assignment of voltages to each pixel, yielding 10,000 distinct output strain field images for the training set. In the training phase, the authors employed a multilayer perceptron (MLP) to achieve inverse control, enabling the device to replicate a specified target strain field based on input images. Furthermore, the authors combined MLP with a super-resolution generative adversarial network (SRGAN) to facilitate rapid prediction of strain field images from input voltage arrays. In the demonstration section, we input 2 biomechanically significant strain fields, and the proposed method successfully enabled the virtual device to replicate these fields. Subsequently, by introducing various tumor-stroma interfaces as inputs, the virtual device adeptly replicated these strain fields. “Therefore, our method demonstrates its capacity to customize strain fields,” said Jue Wang.
This study successfully demonstrates the use of a 9 × 9 array of independently controllable DEAs to achieve precise control over strain fields, overcoming the limitations of traditional bioreactor technologies. By replicating biomechanically significant strain fields and customizing strain fields based on tumor-stroma interface, this bioreactor demonstrates its potential as an advanced testbed for research in mechanobiology, tissue engineering, and regenerative medicine. “this paper represents a significant step forward in the customization of strain fields for biomechanical research, showcasing the potential of combining advanced materials, machine learning, and simulation techniques to address complex challenges in the field of biomechanics and beyond.” said Jue Wang.
Authors of the paper include Jue Wang, Dhirodaatto Sarkar, Atulya Mohan, Mina Lee, Zeyu Ma, and Alex Chortos.
This work was supported by the Purdue startup funding to A.C. and by NSF award 2301509.
The paper, “Deep Learning for Strain Field Customization in Bioreactor with Dielectric Elastomer Actuator Array” was published in the journal Cyborg and Bionic Systems on Aug 14, 2024, at DOI: 10.34133/cbsystems.0155.
Journal
Cyborg and Bionic Systems