Ongoing Research
Unsupervised Deep Learning of Turbulent Heat Transfer
via Generative Adversarial Networks
Nicholas J. Ward*1, Stephen Ekwaro-Osire*1
Introduction
- Extensive importance of fluids simulations
- Atmospheric flows, heat exchangers
- Current methods are costly
- Correlation between temperature and velocity
- Machine learning for fluid dynamics [3]
- Deep learning
- Generative adversarial networks (GANs)
- Application of machine learning for studyingturbulence
Motivation
- Previous works provided insight on modeling turbulence
- Fluids modeling: direct numerical simulation (DNS),near-wall physics
- Artificial intelligence: data-driven models,supervised/unsupervised learning
- Simulate at reduced computational expense
- Improve data-driven models in turbulence research
- Apply deep learning for predicting turbulent heattransfer
Research question
Can turbulent heat transfer be eficiently predicted with wall variables?
- Predict heat transfer from wall variables
- Extrapolate temperature gradients to higher Reynolds numbers
1: Texas Tech University, Department of Mechanical Engineering
Prognostics and Health Management of Wind Energy Infrastructure Systems
Yasar Yanik^1, Stephen Ekwaro-Osire^1, João Paulo Dias^2, Edgard Haenisch Porto^3, Diogo Stuani Alves^3, Tiago Henrique Machado^3, Gregory BregionDaniel^3, Helio Fiori de Castro^3, Katia Lucchesi Cavalca^3
Introduction
- A digital twin (DT) is a combination of integrated multi-physics, multi-scale, probabilistic simulation of a complex product
- Verification and validation (V&V) is the method of determining if the criteria for apart or a system are accurate and complete
- Applications; aerospace, manufacturing, construction, healthcare, automobile, city, agriculture, electricity
Motivation
- Implementing V&V of a single componentinstead of a full system in DT
- Enhancing the model's credibility which is carried out by V&V as an essential initialservice in DT
- Providing quick access to asset data or information and a reduction in manual data transfer effort
Research question
Does implementing V&V using DT improve access to data and reduce the effort of data exchange?
- To conduct a code and calculation verification
- To validate simulations against experiments
- To demonstrate easy access to assetinformation or data
1 : Texas Tech University, Department of Mechanical Engineering, 2 : Shippensburg University of Pennsylvania, Departmentof Mechanical and Civil Engineering, 3 : University of Campinas, Department of Mechanical Engineering
Prediction of the remaining useful life of a battery with limited data and no temporal identifier
Camilo Lopez-Salazar^1, Stephen Ekwaro-Osire^1, Shweta Dabetwa^r2, and Fisseha Alemayehu^3
Introduction
- Li-ion batteries are ubiquitous
- Unavailability of battery data makes it difficult to perform prognostics (deep learning)
- Is it possible to predict the remaining use fullife (RUL) of batteries having only three data points without a temporal identifier?
Motivation
- High unpredictability of battery state of healthparameters and experimental data unavailability
- Learning methods can handle complex regression tasks but they require large amounts of high-quality data
- There is a literature gap on predicting the RUL of batteries without temporal information
- By estimating RUL of batteries on an ongoing basis, adaptative actions can be made during the mission to extend the maintenance-free operation window
Research question
Can the remaining useful life(RUL) of a battery be predictedwith no temporal identifier?
1 Dept. of Mechanical Engineering, Texas Tech University, 2 Dept. of Mechanical Engineering, University of MassachusettsLowell, 3 College of Engineering, West Texas A&M University
Can entropy be used to predict the incipient failure of systems?
Nazir L. Gandur^1, Camilo A. Lopez-Salazar^1, Stephen Ekwaro-Osire^1
Motivation
- Entropy is used to quantify dynamic complexity from an arbitrary time series
- Incipient failure (IF) is an imperfection in the condition of an item from which a degraded or critical failure can be expected to result if corrective action is not taken
- Remaining useful life (RUL) is an estimate of the amount of time that an item is estimated to be able to function before replacement
- Data can be used to better predict IF and RUL
- Avoid catastrophic failure
Research question
Can Multi-scale Divesirty Entropy be used to develop a framework to predict the IF of a system?
1 Dept. of Mechanical Engineering, Texas Tech University,
Product Design & Development Lab
-
Address
Texas Tech University, 2500 Broadway, Lubbock, TX 79409 -
Phone
806.742.2011 -
Email
webmaster@ttu.edu