Could AI-powered defect detection boost battery manufacturing?

Bay Area startup Liminal Insights is betting the energy transition needs a better way to run quality control on batteries.

• The nuts and bolts: Liminal’s technology, called EchoStat, sends ultrasound pulses into battery cells, collecting information about their physical properties. Machine learning models use that information to predict performance quality, and detect both defects and potential failures. The company’s first gigascale factory deployment is currently underway.
• The market grounding: The startup is focused on an issue that has cost automakers millions of dollars and years of time: faulty batteries. Time to market for high-quality batteries is slow, even for major producers, and Liminal is hoping to drastically reduce that time and smooth over potential market bottlenecks. 

Andrew Hsieh co-founded Liminal Insights in 2015 with the goal of catalyzing the global transition to electric vehicles, he told Latitude Media: “In order to achieve rapid EV adoption around the world, we obviously need more batteries,” Hsieh said. “More than that, we need better quality batteries.”

Today, in the battery manufacturing industry, even major players like Panasonic and LG require up to five years to ramp up from an initial production run to a high-yield, profitable production, he said.

Liminal’s EchoStat technology sends non-destructive ultrasound pulses into battery cells, collecting information about their physical properties. It essentially creates an image of a cell, including any potential anomalies created by production errors. The hardware itself is modular and, according to Liminal, is easily integrated into factory lines.

Then comes the artificial intelligence layer. ML models analyze the information collected via ultrasound to predict cell life performance and quality, and detect specific defects and product failures. The result is a greatly shortened quality testing timeline that, according to Hsieh, takes feedback time from weeks down to mere minutes.

As drivers increasingly transition to EVs, faulty batteries and related issues have already cost automakers both time and money. And, Hsieh said, they have resulted in increased pressure on battery manufacturers to ramp up quality assurances.

He pointed to a series of Chevy Bolt recalls in 2020 and 2021 — in which General Motors cited manufacturing defects as the cause of battery fires — as the “biggest evidence” that the status quo of cell manufacturing isn’t sufficient. And in recent years, battery defects have multiplied. In January 2023 BMW recalled more than 14,000 electric vehicles in January 203 and a month later Ford paused production of its F-150 Lightning in light of a battery fire.

While Liminal is currently laser-focused on the EV industry, Hsieh said EchoStat can also be applied to grid storage, including other chemistries: “Manufacturing will be similar enough that we can develop solutions for sodium ion too.”

Contracting market, expanding moves

While demand for EVs has reportedly slowed and certain automakers have pulled back from their ambitious manufacturing plans, Hsieh said those shifts don’t hurt Liminal’s strategy: “There’s a little bit of a correction happening, and it’s not necessarily a bad thing,” he said.

During market contractions, Liminal’s prospective customers likely have more availability on their production lines or more downtime, which Hsieh sees as an opportunity for the company to install pilots and do much-needed education around ultrasound and machine learning.

“What we’re doing is pretty new in the industry and it’s not intuitive to a lot of customers, at least on the surface, why ultrasound and machine learning can create value,” he said. “It’s getting easier, but we still have to do that side-by-side comparison.”

Getting those “technology bakeoffs” up and running is a little simpler during slower periods for the industry, he added.

Hsieh said he’s seen an uptick in investor interest in support systems for battery manufacturing, as opposed to interest in the manufacturing itself. (He added that this phenomenon could be attributed in part to high-profile problems like the implosion of Britishvolt.)

In early 2023, Liminal raised $17.5 million, including from Swedish battery maker Northvolt, to help get EchoStat into factories. Deployment of Liminal’s first gigafactory installation, for a Tier 1 battery supplier in Europe, also started last year.

So far, Europe has been a more willing market for Liminal’s tech than the United States; Hsieh said a lot of the company’s current customers — early adopters, so to speak — are based on the continent. 

“Europe’s industrial policy is a lot more favorable, generally speaking, and the funding environment is a lot more favorable to these heavy manufacturing companies,” he said, though he added that the status quo is starting to change, with increased interest among U.S. investors and manufacturers that is driven in part by federal incentives and funding.

Today, Liminal’s standard business model involves offline pilots in the form of a paid proof of concept. 

While Hsieh said retrofitting existing factories is “still valuable,” the biggest opportunity for Liminal is greenfield deployment.Inking out partnerships to outfit new factories with EchoStat requires establishing relationships pretty early in a company’s factory planning process. That’s the best way to get Liminal’s technology — which measures approximately one by 1.5 meters, around the size of a standard production cell — on a blueprint, Hsieh said.