CANCELLED: Mixture of Regression Modeling of Image Sequences to Quantify Dendrite Growth in Lithium Batteries

Margaret Johnson, Iowa State University
Event Date: 
Tuesday, February 21, 2017 - 3:00pm
EA 170

NOTE: This talk has been cancelled

Lithium batteries are a very “hot” topic today. Extensive research is currently being conducted to develop a safe Lithium battery, which extends the traditional Lithium-ion battery by replacing the carbon anode in a Li-ion battery with Li metal. These Lithium metal batteries have a much higher energy density than Li-ion batteries, but are not yet commercially produced due to the formation of Li metal dendrites within the battery during charging that do not completely dissipate during discharge. This process causes rapid degradation of the batteries and fire-inducing short circuits. Scientists at Pacific Northwest National Laboratory are currently conducting scanning transmission electron microscopes ((S)TEM) experiments with the aim of understanding and controlling dendrite growth within Li metal batteries. Their experiments produce sequences of images (videos) that must be analyzed to determine where dendrite growth occurs and how it evolves throughout an experiment. To quantify dendrite growth in (S)TEM images, we present methodology utilizing Gaussian mixture of regression models to label image pixels corresponding to dendrite growth. Several forms of the models are presented, as well as methods for merging mixture components and imposing spatial coherence on the pixel labeling. Lastly, we present extensions currently being developed in a Bayesian modeling framework and discuss interesting directions for future research.