UM-Dearborn CECS Mechanical Engineering. Battery Safety Research & Testing Lab

AI-Driven Autonomous Battery Testing Lab

Building an autonomous battery testing center where agentic AI plans the tests, guides the models and directs the robots.

Chat-style animation: the lab assistant walks one round from test plan to model to robot to experiment

One loop of our autonomous lab: AI → model design → experiment + robot → outcome. Robotic build under construction · tests and models ongoing.

RUNNING NOW

ThermalRunawayAI: answers with sources

Ask a thermal-runaway question in plain language. Our assistant searches a curated library of published papers, reads their figures and tables, and cites the source of every answer.

ThermalRunawayAI ranks insulation materials from a published study and cites it.
ThermalRunawayAI ranks insulation materials from a published study and cites it.
ONGOING

AI-guided experimental design

Battery samples are expensive, and a thermal-runaway test destroys them. Our AI turns the literature into a knowledge graph of what drives runaway, then proposes the fewest experiments that answer the question.

Knowledge graph built from published thermal-runaway papers: factors, categories and outcomes.
Knowledge graph built from published thermal-runaway papers: factors, categories and outcomes.
ONGOING

AI-guided modeling design

Before a test, our models screen the options. A 3-D module model shows how runaway spreads from cell to cell, and we check it against our own module tests.

3-D module thermal model (COMSOL).
3-D module thermal model (COMSOL).
Model vs our module test: three cells fail in sequence (simplified redraw).
Model vs our module test: three cells fail in sequence (simplified redraw).
Module simulation: runaway starts in one cell and spreads cell to cell over about an hour (the color scale changes every frame).
Module simulation: runaway starts in one cell and spreads cell to cell over about an hour (the color scale changes every frame).
UNDER CONSTRUCTION

Robotic lab control

Thermal-runaway testing is hazardous. Robots will assemble samples, place sensors and load the calorimeter, so people stay out of the test environment.

This part of the lab is under construction. No robot is in use yet.

RUNNING NOW

ARC testing: component, cell, module

Accelerating-rate calorimeters (ARC) in our lab test materials, cells, modules and electronics.

Our EV+ and Standard ARC calorimeters.
Our EV+ and Standard ARC calorimeters.
Real propagation test in our lab: an ignition flash, sparks, then sustained flames (9 s, no sound).
Real propagation test in our lab: an ignition flash, sparks, then sustained flames (9 s, no sound).
Module propagation test: one cell is triggered, and we measure how runaway spreads.
Module propagation test: one cell is triggered, and we measure how runaway spreads.
MORE CAPABILITIES

Cell level: heat from a growing dendrite

In this simulation there is no outside heat source. As the lithium dendrite grows, the Joule heat it carries rises from under 1 mW to about 108 mW, and the cell warms from 25 °C to about 87 °C within 20 minutes. At the hottest stage, the dendrite begins to thin.

>1000×faster per step than COMSOL
DendriX simulation: dendrite growth (left) coupled with the Joule heat it generates and the cell temperature (right).
DendriX simulation: dendrite growth (left) coupled with the Joule heat it generates and the cell temperature (right).
CONTACT

Partner with us

We offer ARC testing of cells, modules and materials, joint research projects, and model-guided test design.

Email Dr. Chen
Dr. Lei ChenAssociate Professor, Mechanical Engineering, UM-DearbornLab lead · primary contactleichn@umich.edu
Rui WangModeling · data · websitewangrz@umich.edu
Jinrong SuModule modeling · experimentsjinrongs@umich.edu

Meet us at The Battery Show 2026: Booth #6242, Oct 13–15, Huntington Place, Detroit.

University of Michigan-Dearborn · College of Engineering and Computer Science · 4901 Evergreen Road, Dearborn, MI 48128