Conference Tracks and Mini-Symposia

Mini-symposia help bring together the diverse communities that are interested in learning, developing, and applying mechanistic machine learning and digital twins via computational science and engineering tools to a broad range of engineering and scientific problems, and to promote collaborations between engineers, data and computer scientists, and mathematicians from federal agencies, academia, and industry in this field.

Track 1: Multiscale Materials and Engineered Systems


  • George Karniadakis, Brown University

  • Shan Tang, Dalian University of Technology


  • Jian Cao, Northwestern University

  • Wei Chen, Northwestern University

  • Gengdong Cheng, Dalian University of Technology

  • Charbel Farhat, Stanford University

  • Somnath Ghosh, Johns Hopkins University

  • H. Alicia Kim, University of California San Diego

  • Wing Kam Liu, Northwestern University

  • Timon Rabczuk, Bauhaus University Weimar

  • Rekha Rao, Sandia National Laboratories

  • C. T. Wu, ANSYS

  • Wentao Yan, National University of Singapore

  • Julien Yvonnet, Universite Paris-Est

  • Ming Zhou, Altai

MS 1-1: Nonlocal Operators and Machine Learning in Multiscale Modeling


  • Animesh Biswas, University of Nebraska-Lincoln

  • Michael Parks, Sandia Laboratories

  • Petronela Radu, University of Nebraska-Lincoln


Nonlocal models are increasingly becoming a preferred setting for the study of multiscale, multi-physics phenomena. These models offer a distinct mathematical advantage in that nonlocal operators can act on functions with minimal (or no) regularity, which is useful in modeling failure and fracture. A specific nonlocal model, peridynamics, has been shown to unify the mechanics of continuous and discontinuous media within a single, consistent set of equations. Nonlocal models also possess length scales, which motivates their use as multiscale models in capturing microstructure influence on the macroscopic behavior of materials. For example, nonlocal models are useful in the modeling of additively manufactured materials. Recent advances in machine learning present an attractive approach to determine the nonlocal operator directly from data, as one can use a neural network to invert for the nonlocal operator representing the microstructure. This minisymposium will explore all applications of nonlocal modeling to multiscale phenomena.

MS 1-2: Machine Learning in Polymer Science and Chemistry


  • Ying Li, University of Connecticut

  • Lei Tao, University of Connecticut


Artificial intelligence (AI) and, in particular, machine learning (ML) as a subcategory of AI, provides unique opportunities for the discovery and development of innovative polymers and organic molecules. In the past, the development of polymers and organic molecules traditionally has been a trial-and-error process, guided by the experience of experts, human intuition, and conceptual insights. However, such an approach is usually slow, costly, and biased towards certain domains of chemical space, and limited to relatively small-scale studies. In addition, automation of organic molecules and materials design is considerably less developed than that for inorganic materials due to challenges associated with searching the vast design space (on the order of 10^60–10^100) defined by the almost infinite combinations of the molecular constituent, microstructures, and synthesis conditions. Very recently, various ML approaches have emerged, some of which have been successfully employed for the de novo design of polymers and organic molecules. This Special Issue will address recent experimental, computational, and theoretical advances in this burgeoning field. Topics of particular interest include but are not limited to: (a) ML-assisted discovery and design of innovative polymers and organic molecules; (b) data-driven methods for design, synthesis, and characterization of polymers and their composites; (c) ML-enabled physical and mechanistic insights into polymer physics and chemistry; (d) deep insights into chemistry–structure–property–performance relation of polymers revealed by ML techniques; and (e) ML-accelerated multiscale modeling approach for polymers and polymer composites. The goal of this mini-symposium is to bring together researchers from a variety of backgrounds to exchange ideas, identify and address grand challenges, and initiate new areas of research in this burgeoning field.

MS 1-3: Studying Material Response using Machine Learning


  • Qizhi He, Pacific Northwest National Laboratory

  • Ayush Rai, Purdue University

  • J. S. Chen, University of California San Diego


The mini symposium aims to promote the research on material response and material behavior using the data-driven modeling. This includes all deforming mechanisms, creep tests, quasi-static and dynamics studies. Machine Learning with an ability to model very complex functions can replace the traditional analytical studies and can help us better understand the material mechanics.

MS 1-4: Identifying Constitutive Behavior and Dynamics via Physics-Informed Machine Learning


  • Marta D'Elia, Sandia National Laboratories

  • Nathaniel Trask, Sandia National Laboratories

  • Yue Yu, Lehigh University


Constitutive models and closures for materials require derivations and experimental calibration that may be prohibitive as scientific applications demand increasing model complexity. Scientific machine learning provides new data-driven tools for model identification that embed physical laws in the learning algorithm, resulting in physically-consistent learnt models. The purpose of this minisymposium is to bring together experts working on model discovery in the context of multiscale modeling and simulation. Applications of interest include subsurface transport, phase transitions, fracture mechanics and turbulence.

MS 1-5: Data-Driven Approaches in Computational Solid Mechanics


  • Siddhant Kumar, TU Delft

  • Pietro Carrara, ETH Zurich

  • Laura De Lorenzis, ETH Zurich

  • Adrien Leygue, Ecole Centrale de Nantes


Data-driven and machine learning techniques are becoming popular in computational solid mechanics. Recent studies highlight that data-driven approaches may extend and encompass classical approaches and suggest a vast unexplored potential for their applications in computational solid mechanics. As a result, the research in this field has now blossomed into several state-of-the-art directions – e.g., from deep neural networks for highly non-linear and high-dimensional surrogate models, to experimental-data-based model-free approaches that aim to eliminate any modeling bias, and discovery of interpretable constitutive models and governing equations as a departure from black-box techniques. More recently, there is also a push towards reducing data-dependency by integrating physics-based knowledge and modeling with data-driven procedures. The most prominent applications include acceleration of multiscale simulations, material characterization, computational design and optimization of (meta-)materials.

This mini-symposium will discuss advances in data-driven and machine learning approaches in solid mechanics and in their applications, with representative topics that include but are not limited to

  • Model-free data-driven computational mechanics

  • Data-driven discovery of constitutive laws and governing equations

  • Data-driven identification procedures

  • Supervised/Unsupervised data-/physics-driven learning of surrogate models

  • Application to acceleration of multiscale simulations

  • Application to characterization and design of materials and structures

MS 1-6: Machine Learning and Multiscale Modeling for Complex Materials System


  • Nien-Ti Tsou, National Chiao Tung University

  • Chuin-Shan Chen, National Taiwan University

  • C.T. Wu, Ansys


The origin of interesting properties of materials is typically dominated by compositions, microstructures, phase transformation, crack propagation, defect growth etc. A wide variety of computational methods are used to study the underneath physical mechanisms, predict the behavior of materials, and even enable the design/optimization of material properties. For example, finite element method captures the performance of materials, providing design guidelines for the applications at device scale; phase field modeling and molecular dynamics reveal the detail of evolution of the micro/nano structures, exploring the possibility of microstructure engineering; ab initio and related atomistic calculations resolve the quantum mechanical details and energetics of materials.

Recent surge in machine learning and deep learning provides another spectrum for accelerating materials development and solving certain material and device problems with high complexity. For example, the optimized charging protocols for lithium-ion batteries can be found with an extremely high efficiency by an early prediction model and Bayesian optimization algorithm; the performance of dental implants can be estimated by deep learning networks reducing the requirement of animal tests; composite materials with optimized strength and ductility are designed by generative adversarial networks and transformer.

We aim to provide a forum to present and exchange research results featuring contributions on advancing material design and modeling/deep learning techniques, giving in-depth insight of the mechanism of materials.

MS 1-7: Machine Learning for Uncertainty Quantification in Engineering Systems


  • Michael Shields, Johns Hopkins University

  • Krishna Garikipati, University of Michigan

  • Roger Ghanem, University of Southern California

  • James Stewart, Sandia National Laboratories


Machine Learning (ML) and Uncertainty Quantification (UQ) are inextricably linked. By its very nature, the process of uncertainty quantification is one that requires a probabilistic (and sometimes non-probabilistic) learning from data. This learning process is greatly enhanced by advances in computational algorithms supported by ever-improving hardware platforms. Hence, in UQ, a machine is almost always doing the “learning.” Increasingly, the UQ algorithms themselves are being designed to not only perform some human-specified learning operation, but also to make active decisions. These decisions include, but are not limited to, identifying the appropriate data to collect for UQ, specifying the parameters / boundary conditions / initial conditions / etc. of a computational model that will be most informative for UQ purposes, identifying reduced spaces and manifolds on which high-dimensional uncertain data reside, and fitting data-driven surrogate models that emulate more expensive physics-based simulations.

This minisymposium aims to explore the intersection of advanced UQ and ML methods and their applications for engineering systems by exploiting the underlying physical principles. Contributions are welcome exploring the use of novel ML methods for a wide range of UQ challenges from probabilistic inverse problems to propagation of uncertainty, surrogate model construction, sensitivity analysis, and dimension reduction, among others. Of particular interest are approaches that leverage ML for large-scale dimension reduction such as those using manifold learning, approaches that specifically address uncertainties in machine learning pipelines such as artificial neural networks, and those that integrate active/adaptive learning into UQ methodology.

MS 1-8: Geometric Learning for Mechanistic Modeling and Material Designs


  • WaiChing Sun, Columbia University

  • Yusu Wang, University of California San Diego

  • Melvin Leok, University of California San Diego


This mini-symposium invites abstracts related to the applications of machine learning on non-Eucidiean data, such as graphs and manifolds for engineering and material science. Potential topics included but not limited to

  1. Knowledge abstraction and representation

  2. Latent space extraction from graphs with constraints

  3. Machine learning for dynamic graph with data from symplectic and Lie groups

  4. Physical-informed machine learning strategy for solids

  5. Machine learning of constitutive laws from microstructures

  6. Other technologies that accelerate and improve designs of experiments and manufacture process for engineering materials.

Topics that generate breakthroughs and overcome technical bottlenecks through exchanging ideas and interactions among domain experts, data scientists and mathematicians are especially encouraged.

MS 1-9: Machine Learning and Generative Design for Additive Manufacturing


  • Jida Huang, University of Illinois at Chicago

  • Tsz Ho Kwok, Concordia University

  • Yong Chen, University of Southern California


With the rapid advancement of Additive Manufacturing (AM) technologies, it is possible to integrate the AM with mass production mode for the nowadays customization-oriented market. Since the 3D printing can rapidly fabricate a complex physical object, the personalization brings challenges for the design generation in AM mass production. Due to the high complexity and design variations among distinct customization requirements, it is desirable to procure an efficient generative design methodology that can match the customization prerequisites and provide sufficient fabrication integrity for AM processes. Within the context of mass production, the size of design data is explosively increased, the geometry and structure complexity included in the customized products make the problem even more challenging. In the recent decade, machine learning (ML) has been proved a suitable tool for analyzing large and complex datasets. Therefore, it is unsurprisingly to introduce ML methods for the customized design generation problem.

In this track, we invite papers that discuss research in the development and utilization of ML principles and tools to gain a fundamental understanding of generative design for mass customization with AM. Topics in this track include but are not limited to the learning-based method for design specification, customized design formalization, computational design modeling, machine learning-based topology optimization, geometric deep learning methods for design and fabrication in AM, and efficient pre-fabrication computation for 3D printing.

Track 2: Scientific and Engineering Digital Twins


  • Francisco Chinesta, Arts et Métiers Institute of Technology

  • Scott Alan Roberts, Sandia National Lab


  • Charbel Farhat, Stanford University

  • Qizhi He, Pacific Northwest National Laboratory

  • Nathan Kutz, University of Washington

  • Paris Perdikaris, University of Pennsylvania

  • Michael Triantafyllou, Massachusetts Institute of Technology

  • J. S. Chen, University of California San Diego

MS 2-1: Advanced Computational Technologies Enabling Digital Twins


  • Francisco Chinesta, ENSAM Institute of Technology

  • Elías Cueto, Universidad de Zaragoza

  • Charbel Farhat, Stanford University


With this mini symposium we aim at fostering the collaboration in the development of advanced numerical techniques for digital twins. In particular, we are interested in all those techniques that can potentially enable the development of digital twins beyond the state of the art.

Of particular interest is the concept of hybrid and/or cognitive twins, those that make extensive use of artificial intelligence and machine learning to perform their tasks and to learn from the available data.

A non-exhaustive list of topics of interest includes:

  • model order reduction (linear and non-linear)

  • techniques for high-dimensional phenomena

  • machine learning, artificial intelligence

  • physics- and thermodynamics-aware ML

  • statistics, uncertainty quantification

  • data analytics

Especial emphasis will be paid to techniques for information post-processing, such as augmented and virtual reality.

MS 2-2: Statistical Finite Element Methods for Synthesizing Models and Observations


  • Fehmi Cirak, University of Cambridge

  • Mark Girolami, University of Cambridge


The finite element method is one of the great triumphs of modern day applied mathematics, numerical analysis and algorithm development. Engineering and the sciences benefit from the ability to simulate complex systems with FEM. At the same time the ability to obtain data by measurements from these complex systems, often through sensor networks, poses the question of how one systematically incorporates data into the FEM, consistently updating the finite element solution in the face of mathematical model misspecification with physical reality.

This minisymposium will focus on statistical finite element techniques for synthesizing FE models and data from observations with applications, for instance, in Digital Twins, Physics-informed Machine Learning, Bayesian Learning and Data-Centric Engineering. The topics of interest include the mathematical foundations of statistical finite elements, efficient and scalable formulations for linear, nonlinear and nonstationary PDEs, Bayesian uncertainty modelling, sampling and approximation of high-dimensional probability densities and large-scale engineering applications.

MS 2-3: Advances in Data-Driven Methods and Digital Twins for Advanced Manufacturing


  • Mengwu Guo, The University of Texas at Austin

  • Anirban Chaudhuri, Massachusetts Institute of Technology

  • Jaydeep Karandikar, Oak Ridge National Laboratory


The minisymposium focuses on the latest developments in data-driven methods with specific interests in the field of advanced manufacturing. The progress in advanced manufacturing research is inherently multidisciplinary with ties to multiple engineering disciplines, such as computational material science, computational fluid and solid mechanics, and multi-physics, multiscale systems. By integrating machine learning and data analytics into physics-based models, data-driven methods have been developed for the multidisciplinary applications of advanced manufacturing. In addition, the complex structure involved in constructing digital twins for advanced manufacturing systems is an important research question, and there is a close connection between the data-driven methods and the practical implementation of digital twins. Catered towards both methodology development and transitions to applied research, this minisymposium will bring together researchers developing data-driven methods and digital twins for advanced manufacturing, and will address the ties between them. Areas of interest include data-enhanced modeling, data-driven discovery, optimization, uncertainty quantification, Bayesian inference and inverse problems.

MS 2-4: Credible Image-based Simulations for Digital Twins


  • Scott Roberts, Sandia National Laboratories


Traditionally, CAD has been the primary method of describing the geometry of the components and systems that engineers seek to simulate and manufacture. Recently, largely in the quest for digital twins, demand has increased for simulation of as-built materials and components to address the inherent part-to-part manufacturing variability. The abundance and ease of collection of three-dimensional geometric descriptions, including x-ray computed tomography and laser scanning, has inspired scientists and engineers to perform simulations on computational domains derived directly from imaging data. However, the process of converting greyscale three-dimensional image data to a discretized domain suitable for simulation is often arduous and fraught with errors and uncertainties.

In this mini-symposium, we explore techniques for addressing the challenges involved in the domain discretization, meshing, and simulation of as-built parts as well as other alternate (non-CAD) geometry descriptions. We also explore credibility assessments in these techniques, including uncertainty quantification and verification. Topics include, but are not limited to:

  • Computed tomography reconstruction techniques

  • Geometry creation from point cloud data

  • Image segmentation, labelling, and part identification

  • Geometric feature identification and detection

  • Generation of CAD from image/faceted data

  • Discretization/meshing of faceted or voxel data

  • Discretization/meshing of implicit interfaces

  • Numerical algorithms for solving multi-physics problems on as-built geometries

  • Geometric uncertainty quantification and propagation

  • Verification of image-based simulation techniques

  • High performance computing applied to as-built geometries

  • Applications of the above techniques to engineering applications

  • Machine learning-based approaches for all of the above topics

MS 2-5: Data Driven Approaches for Circuit Design and Analysis


  • Pavel Bochev, Sandia National Laboratories

  • Biliana Paskaleva, Sandia National Laboratories

  • Paul Kuberry, Sandia National Laboratories


Circuit simulation (sometimes referred to as SPICE simulation) is foundational to modern circuit design. In circuit simulation, compact models are used to model individual components. A circuit simulator uses compact models to enforce Kirchhoff's laws on a user-defined network. Modern circuits have thousands of components, so it is important that individual compact models be computationally inexpensive and typical compact models consist of a handful of equations. These equations are generally a combination empirical formulas and simplified solutions to semiconductor transport equations. Making such models accurate can be very challenging. Traditional compact model development relies on human expertise and is an expensive, time consuming effort. It requires highly skilled experts combining knowledge of solid state physics, circuit design, model calibration and numerical analysis. Besides the long development times, compact models do not always generalize well and adding new physics may require redeveloping the model from scratch. A data driven approach could help mitigate these issues by providing the means to automate the development of compact semiconductor device models directly from data. Furthermore, machine learning (ML) provides a broad spectrum of regression options that can be leveraged to balance computational cost of the compact models and their generalizability in a way that is best suited for a specific circuit design and analysis stage. This session will focus on recent efforts in the development of data-driven and machine learned models for circuit and device simulations.

MS 2-6: Merging Simulation and Machine Learning for Industrial Applications


  • Veronika Brandstetter, Siemens AG, Technology

  • Frank Naets, KU Leuven

  • Christopher Rackauckas, Massachusetts Institute of Technology


Digital twins incorporate the two fast evolving worlds of data and physical model-based solutions. The virtual replicas unlock unique potential to improve the physical asset during all its lifecycle phases and to support decision making.

Merging physics-based simulation and machine learning (ML) allows to close several gaps in current digital twin technologies such that quality of and coverage by digital twins can be further improved. This combination enables an increased productivity, accuracy, prediction speed, and efficiency of the digital twin. This mini-symposium’s goal is to showcase how the synergetic combination of both is leveraged in industrial applications and contributes to reaching above mentioned aspects.

Examples from different baseline physical modeling approaches, like 1D, FE, CFD and multibody simulation, integrated with ML, e.g. in the area of mechatronic simulation will be presented. The range of contributions covers both methodological and application-oriented presentations right up to combination with commercial software and scaling for industrial problems.

MS 2-7: Physics-based Data-Driven Modeling and Machine Learning for Intelligent and Green Transportation


  • Rajeev Jaiman, The University of British Columbia

  • Jasmin Jelovica, The University of British Columbia

  • Wrik Mallik, Technion Israel

  • Rachit Gupta, The University of British Columbia


This mini-symposium focuses on the integration of computational mechanics with machine learning (ML) towards the development of digital twins for intelligent and green vehicles. New developments and contributions pertaining to algorithms and software development and/or application are sought. Comparison of new efficient, robust and accurate methods, critical assessment and benchmarking with the existing physics-based ML techniques and novel applications are welcome. This mini-symposium aims to provide a platform for investigators to disseminate and discuss physics-based hybrid ML methods and algorithms for multiphysics prediction, control and optimization of aerospace, marine and automotive vehicles. Novel physics-based data-driven and ML technologies for active feedback control, real-time structural monitoring and multidisciplinary design optimization are desired.

MS 2-8: Integration of Models, Data and Artificial Intelligence for Energy and Power Systems


  • Jing Li, Pacific Northwest National Lab

  • Amanda Howard, Pacific Northwest National Lab

  • Qizhi He, Pacific Northwest National Lab

  • Peiyuan Gao, Pacific Northwest National Lab


Machine learning and data science tools have been considered as a promising way to advance our understanding of energy and power systems and make them more accessible, affordable, reliable, and clean. The mini-symposium will provide a forum for presentation and discussion of the state-of-art in fusing physical models, data and artificial intelligence for analysis and design of various types of energy and power systems. The representative topics include:

  • Machine learning enhanced numerical modelling of clean energy and power systems, e.g., power grids, wind energy, solar energy, lithium-ion batteries, flow batteries

  • System identification and parameter estimation

  • Computational design and optimization

  • Machine learning assisted design and discovery of sustainable energy storage materials

  • Battery digital twins

  • Smart grid

MS 2-9: RISE of the Machines: Robust, Interpretable, Scalable, Efficient Decision Support


  • Bart van Bloemen Waanders, Sandia National Laboratories

  • Karen Willcox, University of Texas at Austin


An important challenge in computational science and engineering is enabling decision making in support of complex applications. A promising strategy is to develop digital twins in the service of physical assets. Digital emulations must however be capable of processing data that reflect environmental conditions as well as data that measure the health of subcomponents within a complex system. Moreover, decisions are often required in real time while mitigating uncertainties. This mini-symposium will therefore explore a range of range of computational methods, including inversion, optimal control, neural networks, active and reinforcement learning. The underlying theme will be on the development of algorithms for digital twins that are robust, interpretable, scalable, and provide efficient decision support. Approaches will be presented in the context of different applications, such as epidemiology, unmanned arial vehicles, and wildfire management.

MS 2-10: Deep and Machine Learning Methodology in the Context of Application to Computational Mechanics


  • Yoshitaka Wada, Kindai University

  • Yasushi Nakabayashi, Toyo University

  • Masao Ogino, Daido University

  • Akio Miyoshi, Insight Inc.

  • Shinobu Yoshimura, University of Tokyo


Application of artificial intelligence technology in the field of computational mechanics has been established for a long time. However, many examples of applying deep learning technology currently dominating the world to computational mechanics have not been reported yet. The objective of this mini-symposium is to discuss how to apply artificial intelligence such as deep and machine learning technologies to computational mechanics. We warmly welcome anything related to computational mechanics or artificial intelligence toward uniting both technologies into significant and beneficial applications. Particularly by using deep learning, it is necessary to discuss examples that make it possible to simulate objects that were difficult to simulate in the past, or to improve the accuracy of simulations that have been done in the past.

Track 3: Biomedical Systems, Medical Devices, and Mechanistic Machine Learning Enhanced Diagnostics


  • Ken Loh, University of California San Diego

  • Jessica Zhang, Carnegie Mellon University


  • Suvranu De, Rensselaer Polytechnic Institute

  • Ellen Kuhl, Stanford University

  • Niall Mangan, Northwestern University

MS 3-1: Data-Driven Modeling and Simulation for Computational Biomedicine


  • Jessica Zhang, Carnegie Mellon University

  • Adrian Buganza Tepole, Purdue University

  • Ming-Chen Hsu, Iowa State University

  • Adarsh Krishnamurthy, Iowa State University


The rapid development of computational power has led to significant advances in statistical and machine learning techniques in the past few decades. On the other hand, the plummeting costs of sensors, computational power, and data storage technologies have equipped us with the capability of generating and collecting vast quantities of data. All these developments afford us new opportunities for data-driven discovery in broad engineering applications such as computational biomedicine. In this mini-symposium, we solicit contributions that describe advances in data-driven modeling and simulation for computational biomedicine. Novel modeling methods development across all scales are solicited. This mini-symposium would also highlight interdisciplinary efforts of basic and clinical scientists, biophysicists, engineers, and mathematicians that jointly address the most critical challenges and trends in computational biomedicine, including neuron material transport, soft tissue growth, cardiovascular systems, and implantable devices.

MS 3-2: Numerical Twins for Biological Systems


  • Elisa Budyn, ENS Paris-Saclay/ University Paris-Saclay

  • Shaofan Li, University of Berkeley

  • Roger Sauer, RWTH Aachen University

  • Marino Arroyo, Universitat Politecnica de Catalynia

  • Mandadapy Krhanti, University of Berkeley


When the human body experiences large loss of tisssue, regenerative therapies using stem cells, primary cells, and various cellularized substrates can now be considered while they could not be technically envisioned in the past. 3D in vitro of human tissues in organ-on-chips can be used to test pharmaceuticals, cell reprogtamming or engineer cellularized systems for transplantation. These technologies open new avenues to investigate human cells in situ physiology and push patient-specific designs of grafting systems. These technologies offer the opportunity for studying almost any biological systems and improving transplation techniques with patient-specific scaffolds and/or patient-specific engineering cell for cellularized systems.

Recent technological advances in high resolution imaging makes it possible to create imaging-based computational models and numerical twins to investigate cell mechanotransduction pathways and matrix formation down to the nano scale in parallel to experimetal measurements of the cell response. Despite the tremendous challenges these methods are facing, they can advance tissue engineering and regenerative medicine.

This symposium will consider multi-scale multi-physics numerical twins of human cell and tissue mechanics and mechanobiology in biological systems promoting tissue growth. This symposium will particularly focus on numerical twins for regenerative medicine and organ-on-chip using native, synthetic or 3D printed scaffolds. Preprocessing algorithms for image processing of high resolution microscopy and algorithms for 3D volume reconstruction and segmentation are welcome. Growth models to describe the mechanics of cells at the interface between the newly formed matrix or with biocompatible materials are very welcome.

Targeted themes:

  • Numerical twins of cells, tissues and organs, Multiscale multiphysics models, cell mechanobiology, cell interactions, cell-substrate and cell-cell interaction, cell growth, cell imaging, stem cell differentiation.

  • Tissue mechanics and characterization, tissue/tumor growth, tissue pathological evolution, tissue imaging, tissue engineering.

  • Regenerative medicine, organ-on-chip.

Elisa Budyn, Shaofan Li, Roger Sauer, Marino Arroyo, Mandadapy Krhanti

Track 4: Reduced-order Modeling for Fluids, Solids, and Structures


  • Charbel Farhat, Stanford, University

  • James Stewart, Sandia National Laboratories


  • J. S. Chen, University of California San Diego

  • Elias Cueto, Universidad de Zaragoza

  • Zhengtao Gan, Northwestern University

  • Qizhi He, Pacific Northwest National Laboratory

  • George Karniadakis, Brown University

  • Wing Kam Liu, Northwestern University

  • Ye Lu, Northwestern University

  • Masayuki Yano, University of Tokyo

MS 4-1: Machine-Learning-based Models for Forward and Inverse Problems in Computational Science and Engineering


  • Tan Bui-Thanh, Oden Institute of Computational Engineering and Sciences

  • Timothy Wildey, Sandia National Laboratories

  • Anh Tran, Sandia National Laboratories

  • Hwan Goh, Oden Institute of Computational Engineering and Sciences


The fast growth in practical applications of machine learning (ML) in a wide range of contexts has fueled a renewed interest in machine learning methods over recent years. Subsequently, scientific machine learning (SciML) is emerging as a multidisciplinary field that combines computational sciences and machine learning. This mini-symposium is focused on SciML approaches that facilitate physics-based and/or data-driven methods to derive more efficient and accurate reduced-order models (ROM), with an emphasis on the utilization of these ML-ROM to accelerate forward, inverse and UQ problems.

While computational sciences and engineering focus on large-scale models that are derived from physics laws, machine learning focuses on developing data-driven models that require minimal knowledge and prior assumptions. SciML endeavors to combine the two disciplines in order to develop explainable models that are data-driven but require less data than traditional ML methods through the utilization of physics. The resulting models, therefore, possess physical knowledge that can prevent overfitting, often reduces the number of hyper-parameters, and promotes a more robust ability to extrapolate, while retaining outstanding speed and accuracy of ML techniques.

We cordially invite researchers to submit work related to one of the following areas:

  • Bayesian statistics and machine learning

  • Deep learning for forward and inverse problems

  • Deep learning for UQ: Methodologies and applications

  • Data-driven reduced-order models: Methodologies and applications

  • Data-driven inverse problems: Methodologies and applications

  • Explainable artificial intelligence applications

MS 4-2: Hybrid Data-Driven and Physics-based Model Reduction in Mechanical Systems


  • Bogdan I. Epureanu, University of Michigan

  • Amin Ghadami, University of Michigan

  • Sean Kelly, University of Michigan


Recent years have witnessed a significant shift towards far more complex and large-scale engineering systems than ever before. Despite remarkable increases in computational power, many real-world systems are still far too complex to simulate with digital twins based on high fidelity physics-based numerical methods. Model order reduction presents a promising way to tackle the computational bottleneck related to the computational intensity and model complexity. Nevertheless, such techniques face challenges when used on systems exhibiting a wide variety of parameter-dependent nonlinear behaviors or localized features. Recent advances in data-driven analysis of systems and machine learning approaches have revolutionized how we model engineering systems. Consequently, one can augment or optimize model reduction techniques through hybrid methods that combine data-driven learning processes with physics-based models to tackle previously unattainable challenges in modeling and analysis of complex engineering systems and structures. This symposium provides a platform to share the most recent developments on the integration of data-driven and physics based models for model reduction across the fields. Recent advances in mechanistic model reduction techniques in computational fluids and structural mechanics, model reduction-based prediction and control, and novel reduced-order modeling methods are welcome.


  • Data-assisted reduced-order modeling

  • Reduced-order prediction

  • Reduced-order models guided by machine learning

  • Data-driven reduced-order control of fluids and structures

MS 4-3: Machine Learning and Data Diversity


  • Ramin Bostanabad, University of California, Irvine

  • Hongyi Xu, University of Connecticut

  • Anton Van Beek, Northwestern University


Data-driven learning has introduced a paradigm shift in many fields including solid mechanics, fluid dynamics, and materials science. Across different applications, the acquired datasets can greatly vary based on, for instance, size (big, small, …), dimensionality, input type (mixed, variable-length, integer, conditional, …), output type (mixed, multi-label, quantitative, …), and input-output relation (static, dynamic, deterministic, stochastic…). This high diversity in data characteristics calls for developing different learning mechanisms that enable researchers to maximally distill information from the dataset.

The objective of this mini symposium is to introduce different learning mechanisms that leverage the recent advances in deep learning, statistics, high-performance computing, and mathematics to bring contributions to emerging applications in solid mechanics, materials science, manufacturing, fluid dynamics, multi-scale and/or multi-physics modeling, uncertainty quantification, inverse problems, and other relevant topics.

MS 4-4: Extreme Events in Complex Systems: Statistical Prediction, Quantification, and Mitigation


  • Antoine Blanchard, Massachusetts Institute of Technology

  • Themis Sapsis, Massachusetts Institute of Technology


Extreme events are short-lived episodes occurring due to exogenous causes or internal instabilities during which observables significantly depart from their mean values. A great deal of effort has been devoted to predicting and statistically quantifying extreme events because they can have catastrophic consequences (e.g., structural failure, rogue waves, earthquakes). This mini-symposium provides a venue to review the latest advances in the field and discuss implications for today's engineering challenges.

MS 4-5: Data-Driven Reduced-Order Methods for System Control


  • Boris Kramer, University of California San Diego

  • Benjamin Peherstorfer, New York University

  • Wayne Uy, New York University


The control of complex high-dimensional dynamical systems can be computationally expensive due to the high cost of solving the optimality systems. Model reduction has become a key enabler for control of high-dimensional dynamical systems, and the vast availability of data and high-fidelity simulation results has propelled the field of data-driven model reduction in recent years. Data-driven model reduction describes a class of low-dimensional surrogate modeling techniques that are motivated by their intrusive counterparts, yet learn the models solely from data. These learned models can at times be equipped with guarantees on learning error and convergence towards their projection-based equivalents. The learned reduced models can then be used to design (feedback) controllers, thus providing an attractive framework to design high-performance controllers from system-response data only. The speakers in this minisymposium will cover a range of recent results in this direction and point to future directions in the field.

MS 4-6: Machine Learning Closures for Multiphase Flows


  • Tim Colonius, California Institute of Technology

  • Spencer Bryngelson, California Institute of Technology

  • Themis Sapsis, Massachusetts Institute of Technology


Machine learning is revolutionizing simulations of fluid flows [Brunton et al. Ann. Rev. Fluid Mech. 2020]. However, attention is required to translate these developments to multiphase flows. For example, dispersions of particles, bubbles, or droplets and their collective behaviors introduce additional scale separations. Physically faithful simulations must reconcile these behaviors via, e.g., sub-grid closures for the coupled flow equations and model reduction for their dynamics. This minisymposium focuses on recent developments that leverage machine learning to enable multiphase flow simulations.

MS 4-7: Physics-Informed Data-based Model Reduction


  • Juan Manuel Lorenzi, Siemens AG

  • Benjamin Peherstorfer, New York University

  • Felix Dietrich, Technical University of Munich


Real-time models are a key technology for realizing the next generation of Digital Twins for the digital industries and smart infrastructure of the future. They have the potential to enable new value-streams across a product’s life cycle, such as design optimization, model-based control, predictive maintenance, or virtual commissioning.

Traditional projection-based model order reduction techniques represent a tested way of generating real-time digital twins. These use physics-based models built for the design phase, which are typically large and slow to compute, and generate compact, fast models based on the governing physical equations. These methods can be highly successful when applied to well-studied linear and nonlinear problems such as heat conduction or small-deformation elastic problems. However, they become increasingly difficult to derive when dealing with more complex phenomena such as fluid flow, chemistry, or phase transitions, relevant in industrial fields such as chemical, energy, healthcare, or aerospace.

An alternative method of generating real-times models is the use of machine learning and artificial intelligence techniques, such as deep neural networks. While these can capture extremely complex behavior, they also can require very large amounts of training data, typically not available for the industrial cases discussed above.

Recently, a variety of model reduction methods have been developed that seek to combine these two approaches. They incorporate knowledge of the governing physical equations into the reduced models and also benefit from machine learning techniques by learning from data generated by the full model. This symposium is open to contributions that highlight such physics-informed data-based model reduction methods focusing on methodological aspects and/or presenting applications in industrial use cases.

MS 4-8: Model Reduction and Machine Learning for Fluids and Fluid-Structure Interactions


  • Gianluigi Rozza, SISSA

  • Annalisa Quaini, University of Houston

  • Giovanni Stabile, SISSA


Numerical simulation is nowadays used to solve a huge variety of problems in Fluid and Structural mechanics, as well as Fluid-Structure Interaction. Often, the discretisation of this type of problem results in high-dimensional systems of equations and requires a high computational effort. For this reason, standard numerical techniques such as the finite element method, the finite volume method, or the finite difference method are not viable when time is tight or a high number of system configurations need to be tested. Typical examples of this case can be found in shape optimisation, uncertainty quantification, or real-time control. Reduced order models (ROMs) offer a possible way to reduce the computational burden. Many different types of ROMs have been developed over the years and a possible distinction is between those of them that are intrusive and require the knowledge of the underlying full order model and those that are merely data-driven and therefore non-intrusive. In the first category fall the reduced-basis method, the POD-Galerkin approach, and the Proper Generalised Decomposition. In the second category, one can find truncation-based methods, the dynamic mode decomposition, neural networks, and in general all the models based on just input-output data. This mini-symposium will be devoted to both types of ROMs with a special focus on those specifically tailored to fluid dynamics and fluid-structure interaction problems.

MS 4-9: Advances in Reduced Order Modeling of Solids, Fluids and Porous Media


  • Nikolaos Bouklas, Cornell University

  • Youngsoo Choi, Lawrence Livermore National Laboratory

  • Francesco Ballarin, SISSA


Reduced order modeling has proven to be an efficient tool for analyzingpartial differential equations that correspond to large discretized systems thatneed to be investigated over a wide range of loading and/or system parameters.Recent advancements in Machine Learning are enabling i) the investigation ofpreviously unfeasible problems due to restrictions of the traditional methodologies, or ii) an unprecedented speed-up to the inquiry of data-rich problems. Some examples of the above are non-intrusive model order reduction techniquesusing Neural Networks as well as the nonlinear manifold reduced-order models (NM-ROM), which proves useful when a low-dimensional linear subspacepoorly approximates the solution. This minisymposium aims to showcase newtechniques in reduced order modeling, focusing on solids, fluids, and porous media problems.

MS 4-10: Model Reduction of Dynamical Systems with Deep Learning


  • Kevin Carlberg, Facebook

  • Eric Parish, Sandia National Laboratories


Model reduction for dynamical systems is a critical technology that can enable high-fidelity simulations to be employed for a range of time-critical applications; it is thus highly salient to developing high-fidelity digital twins that can be continually updated and employed for high-fidelity prediction. Recently, researchers have investigated ways of applying innovations in deep learning (e.g., deep autoencoders, convolutional architectures, generative models) to learn nonlinear embeddings that can overcome performance limitations incurred by classical linear-subspace techniques (e.g., POD, DMD). This mini symposium will cover a range of contributions in this topic, including both projection-based and non-intrusive formulations for deep-learning-based model reduction, application to inverse problems, and integration with uncertainty quantification.

Track 5: Geosystem, Geostatistics, and Petroleum Engineering


  • WaiChing Sun, Columbia University

  • John T. Foster, University of Texas at Austin


  • Shabnam Semnani, University of California San Diego

MS 5-1: Advances in Machine Learning and Model Reduction for Inversion in Geophysical and Geological Applications


  • David Pardo, University of the Basque Country (UPV/EHU), BCAM, and Ikerbasque

  • Victor Calo, Curtin University

  • Sergey Alyaev, Norce


We welcome contributions on Machine Learning and Model Reduction algorithms intended to efficiently invert measurements relevant for geophysical and geological applications. Relevant topics include, but are not limited to, the following:

  • Data cleansing and clustering

  • Synthetic data generation for training

  • Pattern recognition for target and facies identification

  • Deep learning for inference from data and data assimilation

  • Support-vector machines for site classification

  • Production and net-present-value estimates from seismic data

  • Real-time inversion of borehole resistivity measurements using Machine Learning

  • Design of efficient measurement acquisition systems using Machine Learning for geophysics.

MS 5-2: Advances in Machine Learning Algorithms in Geosciences and Reservoir Engineering Applications


  • Eduardo Gildin, Texas A&M University

  • Hector Klie,


Embedding physics into data-driven models and /or extracting physical insights from data are becoming a critical component for improving the modeling of complex Oil and Gas reservoir systems. There is a pressing need for reducing interpretation, simulation, and optimization cycles through approaches that are interpretable, generalizable, and that can ultimately lead to the discovery of unexposed reservoir dynamics. The multiscale and multiphysics nature of reservoir systems in addition to large uncertainties poses particular challenges to recently proposed physics-informed machine learning models. This mini-symposium explores some successes and short-comings involving the combined application of physics-based and data-driven models designed to embed physical constraints, conservation laws, and constitutive relations within data. The discussion will be showcasing several key challenges aimed at finding the best compromise between accuracy, efficiency, and interpretability of a new generation of models in Geosciences and Reservoir Engineering applications.

MS 5-3: Machine Learning Enhanced Multi-scale Modeling and Imaging Technologies for Geosystems Applications


  • Shabnam Semnani, University of California San Diego

  • J. S. Chen, University of California San Diego


Geomaterials such as soils and rocks are heterogeneous and multi-phase porous materials. The macroscopic behavior of these materials in response to physical phenomena in various geoengineering applications is determined by the heterogeneities existing at different scales, such as grain size distribution, particle shapes, mineralogy, pore space and fractures. Advanced multi-scale techniques are needed to fully capture the complex behavior of geomaterials in geosystems and geoengineering applications. These multi-scale techniques often require information regarding the constituents and microstructure of geomaterials, which can be obtained from various imaging techniques, e.g., X-ray tomography, scanning electron microscopy, and optical microscopy. This mini-symposium aims to provide a forum to discuss recent advances in applications of artificial intelligence (AI), machine learning (ML), and data-driven and data-centric methods to enhance multi-scale modeling of geomaterials and application of imaging technologies. The topics of interest include, but are not limited to: 1) multi-scale fluid flow or mechanistic simulations; 2) multi-scale image processing; 3) image-based simulations; 4) dimensionality reduction methods; 5) ML-enhanced feature identification; 6) multi-scale data fusion.

MS 5-4: Data-driven Methods in Geophysics


  • Shabnam Semnani, University of California San Diego

  • Yuri Fialko, Scripps Institute of Oceanography

  • Peter Gerstoft, Scripps Institue of Oceanography

  • Matthias Morzfeld, Scripps Institue of Oceanography


Many of today’s societal needs such as mitigation of natural hazards, energy and environmental sustainability and accessing natural resources require studying the physical properties and processes of the Earth and geophysical systems across all scales from both scientific and technological perspectives. This involves application of advanced computational methods and mathematical models along with extensive integration of data obtained from a range of sources, e.g., satellites, laboratory experiments, seismic data, in-situ instrumentation, and geodetic measurements. This mini-symposium aims to provide a forum to discuss recent advances in applications of artificial intelligence, machine learning, and data-driven and data-centric methods for improved modeling and investigation of geophysical systems across scales. The topics of interest include, but are not limited to:

  1. Data analytics in geophysical systems: data fusion, data mining, denoising and feature extraction

  2. Data-driven modeling of geophysical systems across scales

  3. Data-driven uncertainty quantification in geophysics

  4. Detection of damage and disturbances

  5. Data-driven geohazards prediction and assessment

  6. Data augmentation for geophysics applications

  7. Data-driven inverse modeling

MS 5-5: Machine Learning for Petroleum Engineering Applications


  • John Foster, The University of Texas at Austin

  • Masa Prodanovic, The University of Texas at Austin

  • Michael Pyrcz, The University of Texas at Austin


Petroleum Engineering has long been a field where data-driven models and decisions and are widely used. Subsurface data contains vast uncertainties which make predictive simulation difficult even when the underlying physics are well understood. Compared to the scale of hydrocarbon reservoirs, sampled data from the subsurface is always sparse and biased and there are tremendous opportunities to develop new machine learning methodologies for petroleum engineering applications that account for these and other deficiencies in data and/or knowledge of physical processes.

The purpose of this MS is intended to provide a forum for researchers to discuss recent advances in artificial intelligence, machine learning, and data-driven modeling as applied to petroleum engineering applications. Research topics in data science and machine learning related to geospatial modeling, reservoir engineering, formation evaluation, automated drilling and geosteering, reservoir, fracture, and drilling fluid property evaluation, as well as fundamental processes in petroleum engineering are welcome.

MS 5-6: Multiscale Machine Learning for Geotechnical, Geophysics and Geomechanics Applications: from Grain- to Field-scales


  • WaiChing Sun, Columbia University

  • Kane Bennett, Los Alamos National Laboratory

  • Hari Viswanathan, Los Alamos National Laboratory

  • Paul Johnson, Los Alamos National Laboratory


This mini-symposium focuses on data-driven machine learning approaches that enable ideas, methodologies and algorithms, and techniques that enable predictions for geomechanics, geohydrology, rock physics and geophysical problems. Special emphasis is placed on the applications that may link predictions or models across multiple length and spatial scales. Topics of interest may include but are not limited to applications of convolutional neural networks to predict seismic activities, big data applications for landslide predictions, constitutive modeling of geomaterials, physical-informed neural networks for poromechanics and multi-phase flow simulations, machine learning/data-driven accelerated multiscale homogenization, data-driven/model-free methods, as well as uncertainty quantification and digital twin approaches for reservoir management and waste disposal.

Track 6: Infrastructure and Cyberinfrastructure Systems


  • Joel Conte, University of California San Diego

  • David Mascarenas, Los Alamos National Laboratory


  • Fangxin Fang, Imperial College London

  • Marco Giometto, Columbia University

  • Qizhi He, Pacific Northwest National Laboratory

  • Gilberto Mosqueda, University of California San Diego

  • Haeyoung Noh, Stanford University

  • Michael Todd, University of California San Diego

MS 6-1: Machine Learning and Data Centric Engineering in Civil Engineering Applications


  • Qizhi He, Pacific Northwest National Laboratory

  • P. Benson Shing, University of California San Diego

  • Roger Ghanem, University of Southern California

  • J. S. Chen, University of California San Diego


Recent advancements in machine learning and data-intensive computing provide new possibilities for the modeling and simulation of complex civil engineering problems. The mini-symposium aims to solicit research developments and applications that involve integrating physics-based models with data-driven and machine learning approaches for model updating, identification, prediction, and uncertainty quantification that pertain to civil materials, large-scale structures, and infrastructure systems. Topics relevant to this mini-symposium include but are not limited to:

  • Data collection, manipulation, and analysis

  • Data-assisted multi-scale modeling of heterogeneous materials, including composites, concrete, wood, and others

  • Physics-informed machine/deep learning computational methods for civil engineering applications

  • Data assimilation, data fusion, Bayesian inference, uncertainty quantification, and digital twin approaches

  • Real-time simulation for large-scale dynamic systems

  • Damage diagnosis and prognosis, structural health monitoring, and hazard mitigation of real-world structures using machine learning

  • Cyber-physical infrastructure and Internet of Things (IoT)

  • Hazard-resistant control, optimization, and design

MS 6-2: Human-Society-Machine-Infrastructure Interfaces


  • David Mascarenas, Los Alamos National Laboratory

  • Yongchao Yang, Michigan Technological University

  • Fernando Moreu, University of New Mexico

  • Alessandro Cattaneo, Los Alamos National Laboratory


Sensing, imagers, networking, embedded computing, machine learning, and cloud computing resources have advanced to the point where it is now possible to envision the deployment of large-scale, distributed sensor systems capable of capturing high-resolution measurements of critical infrastructure that can be used to generate high-fidelity digital twin models. It is now important to consider how these distributed sensor systems will interface with public and private stakeholders as well as the general public. In the last three years, smart city projects in San Diego, Hong Kong, and Toronto have all experienced severe setbacks on account of not addressing privacy concerns upfront. We must also consider the security implications associated with digital twins that encode intimate engineering details associated with critical infrastructure. Questions also arise concerning how sensor data can be shared between stakeholders in a manner that respects their interests and sensitivities. This mini symposium will solicit papers that address the interconnections between humans, society, machines and infrastructure from the individual scale to the city/society scale. Topic of interest include data visualization, human-machine interfaces, security, block chain, health and usage monitoring, smart cities, wireless sensor networks, robotics, cloud computing, cross-organization data sharing, federated/collaborative machine learning, differential privacy, zero-knowledge proofs/ZK-SNARKS, and secure multi-party computation.

MS 6-3: Civil Infrastructure Digital Twins for Life Cycle Asset Management


  • Michael Todd, University of California San Diego

  • Joel Conte, University of California San Diego

  • Ken Loh, University of California San Diego

  • Alessandro Cattaneo, Los Alamos National Laboratory


The concept of a digital twin is primarily defined by its use. Digital representations or manifestations of some structural asset may be used for a number of activities including (but not limited to) visualization/curation, state awareness quantification, assessing as-built geometry or other properties, optimization of in-situ monitoring systems, prediction of performance (e.g., when critical limit states or even failure are expected to occur), or retrofit design. Consequently, digital twins take a variety of forms ranging from point clouds to machine learners to physics-based simulators (such as finite element models). This mini-symposium is soliciting contributions related to the development, implementation, and applications of digital twins for civil infrastructure systems. Topics that investigate how civil infrastructure digital twins incorporate autonomous structural health monitoring systems, multi-modal and distributed sensor data fusion, effective multi-scale modeling strategies, predictive performance modeling including treatment of uncertainties, damage diagnosis and prognosis algorithms, and data-to-decision frameworks, among others, are especially welcome.

MS 6-4: Physics-Informed Machine Learning for Smart City


  • Susu Xu, State University of New York at Stony Brook

  • Shija Pan, University of California Merced

  • Hae Young Noh, Stanford University


The rapid development and deployment of smart cities is resulting in the generation of vast amounts of data at unprecedented rates. For example, smart devices deployed on buildings, transportation systems, civil infrastructures are collecting real-time data for monitoring, prediction, and control of urban systems. However, given complex urban dynamics and sensing constraints, the generated data are often noisy, incomplete, unlabeled, and subject to heterogeneous distributions. This makes traditional machine learning models difficult to be directly applied. There is a compelling need to couple physical domain knowledge to enable more flexible, adaptable, and explainable machine learning approaches for smart city ecosystems.

This mini symposium is soliciting contributions related to theory, algorithms, systems, applications of physics-informed machine learning in or for smart cities. Topics of interest include but are not limited to: (1) incorporating prior physical knowledge into machine learning algorithms, (2) integrating simulation-based models, mechanics, and data-driven models, (3) identifying and controlling complex physical dynamics and conditions using big data, (4) physics-informed edge computing or federated learning, (5) understanding interactions between human beings and physical systems, (6) physics-informed multimodal learning, (7) innovative applications, laboratory studies, or field validation in smart cities.

MS 6-5: Recent Advanced in Hybrid Numerical and Experimental Simulation


  • Gilberto Mosqueda, University of California San Diego


Hybrid simulation provides a cost-effective approach to experimentally examine the performance of large-scale structural systems under dynamic loads. It combines the advantages of computational techniques and physical testing methods to provide a versatile platform for experimentation that captures the system level response of a structural system. As part of the experiment, new data is generated during the test and can be used to update and improve the computational models. While hybrid simulation has certain advantages over other simulation approaches, it also faces challenges as it is prone to both numerical and experimental sources of errors that can propagate throughout the simulation if not effectively mitigated. This session will present recent advancements in hybrid simulation applied to determine the dynamic response of structural systems to seismic, wind and other dynamic loads. Of interest are new algorithms to take advantage of data generated during the test and challenges encountered in preparing and conducting hybrid simulations.

MS 6-6: Technologies for the Rapid Assessment and Monitoring of the Integrity of Structures Exposed to Earthquakes and other Hazards


  • David Mascarenas, Los Alamos National Laboratory

  • Yongchao Yang, Michigan Technological University

  • Fernando Moreu, University of New Mexico


This mini-symposia will solicit papers in relevant to emerging technologies for the rapid assessment and monitoring of structures exposed to earthquake excitation and other hazards. Advances in sensing, robotics, commercial satellite imagery, and wearable computing have opened up a variety of new opportunities for improving our ability to assess the integrity of structures exposed to earthquakes and other hazards. These systems must be robust enough to survive earthquake events in order to enable post-event surveillance. This session will solicit papers on technologies for structural assessment in the face of earthquakes including tools that make use of risk assessment, drones, augmented reality, video, robust distributed sensor networks, photogrammetry, remote sensing, synthetic aperture radar, and time-of-flight imagery. This symposium is also interested in papers on experiences and lessons-learned during post-earthquake assessments, as well as experience and lessons-learned associated with currently in-place earthquake/hazard monitoring systems.

Track 7: Technology Transfer for Innovative Scientific and Engineering Applications


  • C. T. Wu, ANSYS

  • Jay Pathak, ANSYS

  • Prith Banerjee, ANSYS


  • Frederic Barbaresco, Thales

  • Dirk Hartmann, Siemens

  • Dean Ho, National University of Singapore

  • Wing Kam Liu, Northwestern University

  • Lars Ruthotto, Emory University

MS 7-1: Progress of Machine Learning and Digital Twins in Computational Mechanics for Industrial Applications


  • C. T. Wu, Ansys

  • Jay Pathak, Ansys

  • Prith Banerjee, Ansys

  • David (C. S.) Chen, National Taiwan University

  • Wing Kam Liu, Northwestern University


Computer aided engineering (CAE) is a well-known and effective analysis tool in industry, but machine learning and digital twin technologies have recently emerged as scientific disciplines. They are useful for addressing the challenges of the traditional CAE simulation and reshape its foundations in real-world applications. In particular, as modern products become more complex, high-fidelity, model-based CAE simulations are often too slow to obtain in the design cycle. This affects the verification and troubleshooting in the overall product lifecycle management (PLM).

The combination of machine learning and mechanistic models has made CAE simulation more intelligent. Mechanistic machine learning encodes physical information into the architecture and loss functions of the deep neural network. It engenders reasonably large data set, but removes the possible bias for preventing the “loss of physics” and for better predictions. This data-driven approach not only opens up possibilities to discover previously unseen patterns in the data, but also creates a unique learning-system that can manage complex data for product design and analysis. Leveraging the machine learning technique to improve digital twins has become an important step to advance model-based CAE simulation tools and stay technically top-notch. It is believed that mechanistic machine learning and digital twins can greatly enhance the engineering process that allows industrial companies to develop products with desired speed, efficiency, quality and flexibility in an unprecedented way.

This mini-symposium aims to offer industrial and academic experts a special forum for direct communication and firsthand information exchange in these new R&D fields. In this mini-symposium, we not only wish to share the cutting-edge research works of machine learning and digital twins, but also to identify the emergent needs of industry to make more rapid progress in practical applications.

MS 7-2: Functional Digital Twins for DoD Applications


  • John Michopoulos, Materials Science and Technology Division, U.S. Naval Research Laboratory

  • Geoffrey Cranch, Optical Sciences Division, U.S. Naval Research Laboratory

  • Charbel Farhat, Stanford University

  • Madan Kittur, Naval Air Systems Command

  • Eric Tuegel, Air Force Research Laboratory

  • Mark Robeson, U.S. Army DEVCOM Aviation & Missile Center


The technology of functional digital twins (FDT) has been proposed as a means to predict the short- and long-term behaviors of DoD-relevant materials, and that of the platforms and systems made from them. Whether used in the context of long-range platform sustainment and qualification or that of real-time material and structural heath assessment and management, FDTs require the integration and deployment of many component technologies including:

  1. Accurate and computationally efficient models of the multiscale multiphysics material performance;

  2. Physics-aware or physics-agnostic machine learning methodologies for constructing these models;

  3. Model order reduction methodologies for both continuum and particle-based systems;

  4. Multi-material composition interaction behavior models for interacting platform subsystems;

  5. Sensor sub-systems (including electrical and fiber optic) for monitoring and acquiring data associated with quantities of interest characterizing the state space of such systems;

  6. Data aggregation management and assimilation methods and technologies;

  7. Data-driven model update and adaptation schemes to account for service or mission induced changes of the behavior of the system;

  8. Technologies for estimating and taking advantage of epistemic and aleatory uncertainty for model-based predictions;

  9. Control logic implementation and associated analytical, numerical, software and hardware technologies for real time compensatory system actuation;

  10. Actuation sub-systems capable of reacting to real-time operational and mission-defined requirements;

  11. Decision support systems associated with both FDT data and models.

This mini-symposium welcomes the participation of researchers on any of these topics, as well their combination and integration, as they relate to DoD platforms and applications.

Track 8: Mechanistic Data Science for High School and Undergraduates STEM Education and Applications


  • Wing Kam Liu (Northwestern University)

  • Zhengtao Gan (Northwestern University)

  • Mark Fleming (Northwestern University/Fusion Engineering)

  • Jack Chessa (University of Texas at El Paso)

  • Olivia Graeve (University of California San Diego)

  • Alicia Kim (University of California San Diego)

  • Ken Loh (University of California San Diego)

  • Saura Naderi (Tentative, University of California San Diego)

  • Shabnam Semnani (University of California San Diego)


This track promoting mechanistic data science (MDS) in STEM education is offered during the three-day conference. As shown in the table below, the course consists of three two-hour lectures with exercises. It is supplemented with six two-hour panel discussion sessions focusing on how similar MDS examples can be used in high school and undergraduate education.

Download the tentative schedule.

Lecture 1 (Monday, September 27, 2021): 45 min with 15-min Q&A

Learning Objectives:

  1. Motivation and overview of MDS (Mechanistic Data Science) with applications

  2. Introduction to some of the mathematical “tools” to be employed including Feature Engineering, data collection, PCA (Principal component analysis), frequency domain.

  3. Introduction of the six steps of MDS

  4. Give overview of possible career paths and future of these technologies.

Overview and Motivation of Mechanistic Data Science and relation to STEM

  • In this lecture, we will discuss what MDS is and how it compares with Machine Learning (ML) and Artificial Intelligence (AI). No prior knowledge of these areas is assumed. Every day we experience AI/ML through things like weather forecasting, traffic prediction, UBER & Lyft ride-sharing, spam filters, and obstacle avoidance for driverless cars. We will discuss how the introduction of some basic physics into these types of applications and improve the performance and efficiency.

  • From this point-of-departure, we will discuss specific cases where physics-infusion through MDS has shown benefit. The primary example for this will be COVID-19 detection from chest X-rays.

Feature Engineering will be introduced for handling mechanistic features from the large datasets

  • Understanding the features of the data is critical for including physics into ML/AI. Examples will include reading information from signs and billboards, medical data, motion detection and obstacle avoidance.

  • MDS will be formalized into six steps.

  • Basic knowledge of ML/AI systems and MDS will be critical for all competitive careers in the future. Many exciting and rewarding careers paths will depend on individuals who are well-versed in applying these technologies: autonomous vehicles, robotics and intelligent systems, medical diagnosis and patient interface, space-exploration, consumer marketing and pricing, etc. An overview of these areas will be discussed, as well as how best to prepare for possible careers in these areas.

Exercise 1 (Monday, September 27, 2021): 60 min

Exercise Objectives:

  1. Give experience in basic analysis and manipulation of data

  2. Give students some basic comfort using the Graphical User Interface (GUI)

Students will work directly with customed-made “hands-on/mini-apps” Matlab/Python applications, with “graphical user interface (GUI)” illustrating some of the topics covered in Lecture 1. These applications will take advantage of the GUI (user can interface with the mini-apps”) that will require minimum technical knowledge of ML algorithms, data-science or Matlab/Python programing.

  • These will walk students through key steps in analyzing large data.

  • Students will be given a dataset of diamond pricing based on various parameters (4 Cs - cut, color, clarity, and carat weight, and other important features) and will perform some basic regression and significant parameter extraction (Feature Engineering)

Lecture 2 (Tuesday, September 28, 2021): 45 min with 15-min Q&A

Learning Objectives:

  1. Learn about the basics of the spring-mass-damper vibrating system

  2. Data generation and collection

  3. Introduction to PCA (Principal Component Analysis) data reduction

  4. Acoustic: basic physics of string vibration and sound generation

  5. Applications of Fourier analysis, short time Fourier transform, and Fast Fourier transform

To look more deeply into MDS we will cover some of the basic science, mathematics, and knowledge-driven dimension reduction methods employed together with the generated data to develop highly accurate engineering system models. We will consider the synthesis of sound and musical instruments using MDS in the system and design module of this class.

  • Introduction to spring-mass-damper (S-M-D) systems and vibration. This system is one of the simplest physical systems, but it can be used to describe the physics of many other dynamics systems, including vibrations, sound generation. It can also describe other systems such as price fluctuations.

  • Introduce data reduction and data re-construction/collection with this S-M-D system. Three cell phones are used to record the motion of this S-M-D dynamical system. These dynamical images will be translated into digital signals for data analysis of motion. The basics of Principal Component Analysis (PCA) will be shown. PCA will then be used to extract parameters for large data sets and reduce the size of the model that is being developed.

  • Introduction to string vibration and how sound is generated and perceived. How we describe sound from the standpoint of frequency, amplitude, and time. This will motivate mathematical concepts such as frequency transform and Fourier/Spectrum analysis. These concepts are very similar to how an equalizer on a stereo or audio device works.

Exercise 2 (Tuesday, September 28, 2021): 60 min

Exercise Objectives:

  1. Give the students experience with dealing with preparing data for MDS analysis.

  2. Look at data in time and frequency domain to help reinforce this concept.

Students will work directly with some “hands-on/mini-apps” applications illustrating the concepts in Lecture 2.

  • The first application will involve single and two degree of freedom spring mass damper systems. This will allow students to interact with the system parameters and see how they affect the motion, both in time and frequency. We will show how multiple spring-mass systems can be used to approximate the vibration of a string as in a piano.

  • Experimental data of the motion for and actual dynamical system will also be provided and analyzed. A software, Tracker, will be used to track the motion of an object, i.e., a spring-mass system.

  • The second application will be frequency analysis on the experimental data to determine natural frequencies as well as the magnitude of the vibrations.

Lecture 3 (Wednesday, September 29, 2021): 45 min with 15-min Q&A

Learning Objectives:

  1. See how a simple spring-mass-damper model can be used in the development of a model of a piano acoustic system.

  2. Introduction to feature extractions using Fourier analysis, Short time Fourier transform, and feed forward neural network.

  3. Understand how we can use the extracted features (amplitudes, frequencies, damping coefficients, and phase angles) of a piano key and the same guitar key can be trained for sound conversion from piano to guitar based on the mechanistic mathematical model.

  4. Discussion on ideas on the creation of sounds of any musical instrument from a cheap keyboard.

  5. Transition to forensic engineering.

In this final lecture we will discuss how some of the concepts from the previous lectures can be used to develop a very accurate mathematical model of a piano.

Exercise 3 (Wednesday, September 29, 2021): 60 min

Exercise Objectives:

  1. Give the students a summative experience in MDS.

  2. Challenge the students to think of other ways of employing MML and MDS in similar applications.

In this final “hands-on” exercise the students we will show many of the concepts previously presented in a practical example of a physics-based model of the sound synthesis of a piano. The students will work directly with the piano synthesis application as well as have some time to consider how to possibly extend it to other uses.

  • The students will be given sound signals (.wav files) from a piano. They will use custom-made mini-apps to extract important parameters for the simplified spring-mass based model.

  • The students will then use these parameters in the Matlab piano application to generate realistic sounds and then use those notes to play a melody. They will then be able to interactively modify the model as see how this may improve (or not improve) the piano model.

  • Students will be challenged to propose ideas for extending this piano model to model other instruments, voices or possibly to create new instruments.

  • SUGGESTED PANEL IN DEPTH DISCUSSIONS DURING THE CONFERENCE (1-2 sessions): Continuation: Illustration of the use sine and cosine series, definitions, and interpretations; trigonometry relation of contraction the sum of sine and cosines into a sine series with phase angle. Application for Fourier transform, Short time Fourier transform to the S-M-D systems, estimation of damping and its meaning. Introduction to feedforward neural networks.

UC San Diego, Department of Structural Engineering, 9500 Gilman Dr MC 0085, La Jolla, CA 92093-0085
Copyright (C) Regents of the University of CaliforniaAll Rights Reserved