Novel deep learning approaches for learning scientific simulations

Scientific Conference (2023, August)

Convolution, aggregation and attention based deep neural networks for accelerating simulations in mechanics

in Frontiers in Materials (2023)

Deep learning surrogate models are being increasingly used in accelerating scientific simulations as a replacement for costly conventional numerical techniques. However, their use remains a significant ... [more ▼]

Deep learning surrogate models are being increasingly used in accelerating scientific simulations as a replacement for costly conventional numerical techniques. However, their use remains a significant challenge when dealing with real-world complex examples. In this work, we demonstrate three types of neural network architectures for efficient learning of highly non-linear deformations of solid bodies. The first two architectures are based on the recently proposed CNN U-NET and MAgNET (graph U-NET) frameworks which have shown promising performance for learning on mesh-based data. The third architecture is Perceiver IO, a very recent architecture that belongs to the family of attention-based neural networks–a class that has revolutionised diverse engineering fields and is still unexplored in computational mechanics. We study and compare the performance of all three networks on two benchmark examples, and show their capabilities to accurately predict the non-linear mechanical responses of soft bodies. [less ▲]

Probabilistic Deep Learning for Real-Time Large Deformation Simulations

in Computer Methods in Applied Mechanics and Engineering (2022), 398(0045-7825), 115307

For many novel applications, such as patient-specific computer-aided surgery, conventional solution techniques of the underlying nonlinear problems are usually computationally too expensive and are ... [more ▼]

For many novel applications, such as patient-specific computer-aided surgery, conventional solution techniques of the underlying nonlinear problems are usually computationally too expensive and are lacking information about how certain can we be about their predictions. In the present work, we propose a highly efficient deep-learning surrogate framework that is able to accurately predict the response of bodies undergoing large deformations in real-time. The surrogate model has a convolutional neural network architecture, called U-Net, which is trained with force–displacement data obtained with the finite element method. We propose deterministic and probabilistic versions of the framework. The probabilistic framework utilizes the Variational Bayes Inference approach and is able to capture all the uncertainties present in the data as well as in the deep-learning model. Based on several benchmark examples, we show the predictive capabilities of the framework and discuss its possible limitations. [less ▲]

Towards real-time patient-specific breast simulations: from full-field information to surrogate model

Scientific Conference (2022, July)

In breast cancer treatment, surgery is one of the most common practices [DeSantis et al., 2019]. The surgery involves a complex pipeline, principally due to the difference between the imaging and the ... [more ▼]

In breast cancer treatment, surgery is one of the most common practices [DeSantis et al., 2019]. The surgery involves a complex pipeline, principally due to the difference between the imaging and the surgical posture [Mazier et al., 2021]. Indeed, because of the stance difference, the surgeon has to rely on radioactive or invasive markers to predict the tumor position in the surgical setup. Biomechanical simulations could predict such complex tumor displacements but often require patient-specific data (material properties, organs geometries, or loading and boundary conditions). Full-field acquisitions coupled with landmark identifications allow obtaining relative deformation between the different configurations. Having this information and assuming a finite element model, an identification procedure of the model parameters can be carried out. Finally, finding a suitable computational model allowing for a compromise between accuracy and speed, one may consider surrogate models for real-time simulations (20 to 50 FPS). In this work, we obtained the patient-specific geometry through micro-computed tomography in 8 different configurations, including 15 bio-markers. Assessing the displacement of the bio-markers enabled us to infer the relative strains between the different configurations. A heterogeneous neo-Hookean model was assumed for simulating soft tissue behavior. Based on the displacements and the position of the biomarkers, model parameters identification was performed to calibrate the experimental data with the finite element method results. To overcome speed performance issues, Convolutional Neural Network (CNN) trained with a synthetic simulation-based dataset generated by applying different gravity directions is used. Preliminary results show that CNN can predict the displacement of anatomical landmarks to millimetric precision and is 100 times faster than the finite element method, satisfying our real-time objective. Plus, the use of Bayesian inferences involves a longer prediction time but allows a 95% confidence interval of the biomarkers' displacements. For a given precision, contrary to CNNs, optimization methods are computationally expensive and depend on an initialization point. Although CNNs require new training for each patient, optimization algorithms can be applied regardless of the patient's geometry. Through this study, we observed that material properties were playing an essential role but not as much as the anatomical structures e.g. infra-mammary or Copper’s ligaments. [less ▲]

Real-time large deformations: A probabilistic deep learning approach

in The 8th European Congress on Computational Methods in Applied Sciences and Engineering (2022, June)

DIGITAL TWINNING FOR REAL-TIME SIMULATION

Poster (2019, November)