Computing at CEE - Sonny Astani @ USC Viterbi

Data sciences

View Courses

Data processing from 14 different laboratory sessions. Data processing includes using tools such as Matlab or Mathematica, Maple or Excel. Processing includes filtering, sorting, and plotting.
Additional tools are used for creating images, sketches and other illustrations.
Digital image correlation technique is taught in three of the laboratory sessions. Large amounts of digital images are processed to obtain deformation data from test specimens.

Elements of statistical data analysis (sample mean and variance, fitting models to data via maximum likelihood / method of moments, sampling distribution and confidence intervals, hypothesis testing and regression)
Uses (mostly synthetic) datasets from a range of applications in civil and environmental engineering: travel times, annual rainfall in various cities, material properties like concrete strength, pollutant concentrations, conductivity of porous medium…

This course introduces students to fundamental concepts in fluid mechanics including mass and energy conservation, momentum balance, and dimensional analysis. With a focus on a civil or environmental engineering application, a course project asks students to source data gathered on a fluid mechanics system (for instance, pressure and/or velocity measurements from dams and spillways, flow through pipe systems or wells, power output from wind turbines, etc.) and analyze them in context of the physics-based equations learned in class.

This course focuses on Building Information Modeling (BIM), current BIM technologies; coordination of design and construction; information management throughout building lifecycle; project visualization and simulation for integrated practice. Specific to ML, we cover how BIM data (geometry, spatial relations, geographic and semantic information) could be used through ML for various applications including anomaly detection, identification of likely construction delays, better maintenance predictions, and so on.

This course focuses on understanding the capabilities of modeling, simulation, data centric engineering paradigms and technologies for Architecture, Engineering and construction projects or challenges. The course presents a series of expert lectures on the use of these technologies and methodologies across the entire value chain within the built environment. The course provides exposure to the spectrum of technological innovations and use cases.

This course focuses on design and engineering modeling and on process modeling and simulation of built environment projects. The course content includes the teaching of 3D modeling, parametric and associative modeling, and both off site and on site process simulation and dependency modeling. The course teaches and uses technologies that are used in aerospace engineering, mechanical engineering, and manufacturing based industries but are gaining traction in the architecture, engineering, pre-fabrication, industrialized constriction and infrastructure engineering fields. The course teaches how 3D models included geometric, semantic, and time based data types.

This course provides students with the knowledge and analytical skills to cope with exploratory data analysis and data mining in the field of land transportation and logistics. Topics to be covered include descriptive statistics and data visualization, regression analysis, panel data, structural equation modeling, and logistic regression. The data sets used in this course are selected from a wide range of data sources and case studies in various aspects of transportation planning, safety, and economics in order to support a practical understanding of the modeling techniques.

The purpose of this course is to introduce students to the important concepts of computing and programming related to civil and environmental engineering. The course briefly describes the form and operation of modern computational devices, the data structures and algorithms that are used in many of the computations that support CEE software applications and give a fast introduction to Phyton programming.

After taking this course the student should specifically understand the following concepts related to computing in civil and environmental engineering:

How a computer works;
Concepts of data structures and algorithms, and the specific instances of them that are appropriate for CEE; and
Principles of computer programming

This course introduces:

The basics of data acquisition hardware and software.
The basics of databases and database management systems as applied to engineering problems in general and civil and environmental engineering problems in specific. The focus is on the relational data model, with an introduction to object modeling. The course lectures emphasize database concepts and theory.
Introduction to data mining concepts and techniques and knowledge discovery in databases principles applied to engineering problems/data in general and civil engineering problems/data in specific. The course lectures emphasize on issues relating to the feasibility, usefulness, efficiency and scalability of techniques for the discovery of patterns hidden in engineering databases.

Linear/nonlinear regression and connection to models from traffic flow theory, travel demand forecasting, and microeconomics. Basics of network modeling, algorithms, and optimization applied to transportation. Several Matlab/python-based homework problems. Class project involves generation, collection and curation of data, statistical analysis, and hypothesis testing, possibly in conjunction with algorithm design.

Methodologies to infer unknown input-output models of dynamical systems from data. Nonparametric methods include frequency response and spectral analysis. Parametric methods include regression, least squares, maximum likelihood estimation, and instrument-variable methods. Subspace identification methods and neural networks are also covered. Primary focus is on linear systems, with extensions to nonlinear systems through Volterra series representation and generalized frequency response systems. Finally, fundamental requirements on experiment design for identifiability are also discussed. The course involves working with the Matlab Systems Identification Toolbox. Class project typically involves learning model of a complex system from large scale data.

Felipe de Barros

My group’s research leverages physics and data driven methods to predict the dynamics of fluid flow in porous media and solute transport processes.

The physical model predictions can be conditional on field data like geostatistics. In addition to the physics-driven focus, some data-driven methods have been employed to analyze problems in fluid flows and scalar transport. Data-driven approaches have been employed to the following topics:

analyze the value of data acquisition in the context of subsurface structure identification using deep learning and geostatistics
physics-informed identification of governing equations (ODEs/PDEs) in multiphysics (in collaboration with Sami Masri)
image recognition algorithms for evaluating the performance of data assimilation methods
estimation of the dispersion tensor through deep learning (in collaboration with Muhammad Sahimi).
enhance the performance of physical models with parameters inferred from ML.

Research Website: https://www.felipedebarros.com/

Felipe de Barros

My group’s research leverages bioinformatics tools to analyze DNA and RNA sequencing data from environmental samples.

Anaerobic biotechnologies such as anaerobic membrane bioreactors are capable of matching treatment performance of existing aerobic systems while generating methane-rich biogas, a renewable energy resource. Researching these technologies encompass broader environmental protection by developing systems that minimize energy demand and other inputs while recovering resources from water streams.

Some additional tools used in analysis include high-throughput amplicon sequencing, metagenomics, and metatranscriptomics.

Research Website: https://www.smithresearchusc.com/

David Gerber

My work focuses on the development of intelligent design modeling and simulation methods and the development of digital twins to improve design and engineering phases of projects and eventually actual performance of built projects. My research is focused on supporting the undergraduate and graduate curriculum and the development of skills sets used by industry including modeling, rapid prototyping, robotic construction, and the automating of sustainability and other performance criteria into the design of the built environment. We use 3D and 4D modeling, simulation, data visualization, and develop algorithms to support applications in the Architecture, engineering, and construction fields.

Erik Johnson

My group’s research leverages machine learning to monitor structural health, develop models, and study dynamic systems.

Machine learning is used in the following ways:

Using neural networks to learn part of the solution to an optimal control problem so that the optimization required in real time while using the control is computationally tractable
Using normalizing flows, a type of generative model, to learn a near-optimal importance sampling distribution to estimate the probability of rare events such as structural failure
Using deep neural network Q-learning to find near-optimal solutions to the NP-hard problem of placing heterogeneous sensor arrays on a structure to maximize the information gain.
Generative adversarial networks with a normalizing flow model to solve inverse problems, with application to heat conduction and x-ray images (with Assad Oberai)

Audrey Olivier

Use of Bayesian inference techniques to infer parameters in both physics-based and data-driven models

Joint state-parameter estimation of physics-based nonlinear dynamical systems using online data assimilation techniques for damage detection applications
Enhancements of Bayesian neural networks for various applications including surrogate materials modeling, probabilistic ambulance travel time prediction, structural health monitoring.

Burcin Becerik-Gerber

We use computing in various ways in our research

Use of ML in predicting/modeling occupant related information (e.g., occupancy schedules to optimize building system’s operation, thermal comfort preferences of occupants, choosing energy-efficient contort-driven set points for building operations, contribution of IEQ factors to sick building syndrome).
Extensive sensing of both buildings and occupants (e.g., environmental sensing for understanding indoor environmental quality (IEQ), localization of occupants, physiological sensing to understand human response (stress, comfort, etc.).
Use of visualization techniques (e.g., virtual reality, augmented reality) to explore various research questions related to human-building interactions (e.g., emergency response) and construction safety while working in a robot-assisted sites.

Kelly Sanders

Running regional climate models on HPC to gain insight into how local climate and air pollution is affected by scenario design
Running large hydrology models to analyze water availability and irrigation requirements across SoCal region
Analyzing huge datasets of hourly smart meter data on HPC to gain insights about residential electricity consumption
Using energy modeling software (EnergyPlus) to analyze energy consumption
Manipulating large energy, emissions, air pollution, weather, traffic datasets, etc. to gain insights about the environmental & climate impacts of energy systems
Developing/running machine learning models to predict future residential electricity consumption
Predicting emissions and energy demand using statistical techniques and neural networks

Thomas Petersen

Phase separation of non-equilibrium thermodynamic systems are often solved using density-functional theory and phase-field equations that are phenomenological in nature, often suffering in predictive power when adjusting parameters to fit specific material systems. Part of our research focuses on capturing experimental images of transient density patterns of systems undergoing sol-to-gel transitions on encountering a change in the solvent. To enhance the predictive power of our models, the physics-based ODEs and PDEs will be fit to time dynamics and spatial information gleaned from experimentally collected information. Research will focus primarily on materials relevant to civil engineering applications: Clays/ geopolymers, calcium-silicate-hydrate, cellulose, silica, calcium carbonate, etc.

Tools used:

HPC and numerical solvers to solve ODEs and PDEs of the density patterns in geomaterials.
Statistical analysis of modeled and experimentally observed material domains: Determination of radial distribution functions, structure factors, pore-chord-length distributions, relaxation dynamics from image time-correlation.
Training on loss functions that match experimentally collected and modeled gel dynamics.

Berok Khoshnevis

Automation of construction using 3D printing involves motion control applicable to giant robots which have significant inertia as well as structural deflection. Many sensors including servosystem encoders, laser range finders and vision systems generate massive amounts of data which must be processed in real-time in order to stay as close as possible to planned motion trajectories. On the input side, architectural designs must be received and processed to create motion commands which involve numerous geometrical coordinates on the basis of which inverse kinematics is performed to determine coordinated motion commands for all actuators involved.

Vision systems and real-time image recognition software are used for quality assurance. We have also made attempts to determine the state of cementitious material mixture using visual data and deep learning.

Roxana Javid

Developing statistical models to study the impacts of different transportation strategies including alternative fuel vehicles and high occupancy vehicle lanes on energy, economy, and emission.
Using spatiotemporal data to evaluate the effectiveness of different regional sustainable transportation policies, at different levels of aggregation.

Chukwuebuka C. Nweke

My research employs the use of regression analysis to develop and validate models focused on estimating site response in sedimentary basins and surrounding non-basin locations. These regression analysis include linear and non-linear mixed-effect regression as well other Bayesian approaches. I plan to employ enhanced statistical approaches which involve the incorporation of machine learning algorithms for feature selection and determination of causation between said features and physical mechanisms (subsurface interactions/mechanisms between seismic energy and variations in geologic materials). My current projects focused employing data science/ML include:

Quantifying path effects in sedimentary basin site response using ML clustering algorithms and residual analysis based on mixed effect regression.
Geometric parameterization of sedimentary basins in southern California for site response analysis and modeling (using causal ML algorithms to derive causative and effect relationships)
Validating simulated earthquake platforms for basin site response predictions (using ML algorithms to assess large amounts of synthetic data to discern consistencies and inconsistencies with real world observations)
Validating automated processing of earthquake waveform records

Geotechnical projects include:

Quantifying uncertainty of the mechanical characteristics of biocemented sand using numerical simulations and deep learning ML algorithms to derive yield evolution at different stress ranges.
Characterization of soil erosion hazard using data derived methods and ML algorithms.

Lucio Soibelman

A huge amount of information is generated for an infrastructure element/system during its lifecycle, from designs, specifications, plans, construction documents, process control plans, inventory management reports, cost estimates and control reports, schedules, and maintenance plans and records. As infrastructure designers, builders and owners adopt new computer technologies, computerized data are becoming more and more available. There exist numerous opportunities to exploit and extract knowledge from this vast amount of data. Unlike much of the previous research in machine learning that has been successfully applied in several domains, infrastructure management data are of multiple types, are generated from many different sources, and sometimes is of very low quality. The general direction of Dr. Soibelman research has been the development of frameworks, and algorithms, that support the acquisition, modeling, management, and analysis of such diverse infrastructure-oriented data.

Dr. Soibelman research concentrates on: (1) studying the increasing amount of available construction/infrastructure management data; (2) developing methods, processes, and tools for big data in Civil Engineering to generate novel knowledge from large construction/infrastructure management databases; and (3) better management of the large amount of available construction/infrastructure management data with the development of improved data acquisition systems (sensors, drones, cameras, lidar, etc.) and information retrieval methods by developing frameworks and technologies that combine techniques from machine learning, artificial intelligence, pattern recognition, statistics, databases, and visualization. By developing new technologies and by integrating and adapting existing ones he was able to derive new principles and methodologies that allow infrastructure and construction managers to better manage, and to extract concepts, causal relationships, and patterns of interest from complex infrastructure-oriented data.

Together with PhD students Dr. Soibelman developed the following data acquisition tools, systems, frameworks and methods (partial list):

A framework for construction knowledge discovery.
Methods for integrating unstructured text documents and images in A/E/C model based systems.
A knowledge assisted search engine for A/E/C products.
A decision support framework for electricity production vulnerability assessment.
Automated defect detection for sewer pipeline Inspection and Condition Assessment.
A geospatial clustering tool to support the detection of patterns in water distribution systems.
Data driven techniques for ultrasonic monitoring of gas, oil, and water pipes.
A learning system for electrical consumption of buildings.
An Estimation Algorithm that leverages building information and sensing infrastructure for localization during emergencies.
immersive virtual learning for workers-robot teamwork on construction sites.
Civil Infrastructure Systems Open Knowledge Network
Real time 3D reconstruction of war zones from drone images
Urban Energy Efficiency modeling from infrared data acquired by drones.
Data driven economic modeling tools for evaluating cost and impact of extreme events.

Ketan Savla

Algorithms for efficient and robust operation/planning of societal scale infrastructure (e.g., transport, energy) and autonomous systems, through a combination of tools from control and dynamical systems, optimization, stochastics, networks, game theory, learning, and experiments. Recent projects include:

Analytical characterization of sparsity in feedback control structure of large-scale infrastructure networks and use it for sample-efficient learning of optimal control
Plug and play distributed operation protocols for connected and autonomous transportation networks
Joint estimation and control algorithms for data-driven operation of infrastructure systems with performance guarantees
Theoretical and experimental studies on nudging decisions of dynamically interacting agents in a large population towards social optimum, through information and incentives

Some prior work on data driven traffic management licensed to a startup co-founded by Savla.

Bora Gencturk

Data and computational methods are heavily used to study material and structural behavior. Data from various experiments including images from microscopes, scans and cloud point data from CT scans and digital image correlation methods are processed to obtain information about the material behavior and failure mechanisms. Additional standard vector data from conventional sensors such as strain gages and linear potentiometers are analyzed and graphed.

Finite element modeling, lattice discrete particle modeling and smooth particle hydrodynamics, among other techniques are used to understand the extreme loading behavior of materials and structural systems. Additional coding is performed in various languages for specific analysis of physical, chemical or electrical processes in materials (such as diffusion, corrosion or aging induced deteriorations).

Computing @ CEE

Curriculum

Transport Service Systems

Extreme Habitats

Environmental Stewardship

Disaster Resilience

Urban Livability

RESEARCH

Felipe de Barros

Adam Smith

David Gerber

Erik Johnson

Audrey Olivier

Burcin Becerik-Gerber

Kelly Sanders

Thomas Petersen

Berok Khoshnevis

Roxana Javid

Chukwuebuka C. Nweke

Lucio Soibelman

Ketan Savla

Bora Gencturk

Computing @ CEE