What computer vision algorithms do we support?

Automated inspection algorithms for batteries

This section describes Glimpse's standard automated inspection algorithms. Our standard algorithms are mostly targeted at cylindrical cells, but we support inspection algorithms for all cell form factors and even samples other than battery cells. Glimpse can develop bespoke inspection algorithms upon request. Contact us to discuss further.

How can I trust Glimpse’s high-throughput cell scanning module?

How can I trust Glimpse’s automated inspection results?

Core

Core shape

Core buckling

Core impingement

Can

Anode overhang

Cathode width & positions

Crimp

How can I trust Glimpse’s high-throughput cell scanning module?

Our technology is designed to enable faster CT scanning by improving image quality through advanced image enhancement algorithms. We use trusted, conventional image processing methods that preserve the fidelity of the original data. We intentionally do not use AI or machine learning for image enhancement to avoid any risk of image hallucination or distortion. Please note that our image enhancement algorithms can be disabled upon request.

How can I trust Glimpse’s automated inspection results?

We take pride in ensuring that our automated inspection results are both accurate and precise. For instance, the below plot quantifies the repeatability of some of our standard metrics across 20 scans of the same battery cell, taken across 20 consecutive workdays:

Repeatability of Glimpse's standard inspection metrics across 20 scans of the same battery cell on 20 consecutive workdays

In general, features that are only a few voxels in length will have worse repeatability than features that are tens or hundreds of voxels in length.

We evaluate algorithm repeatability on our own scanner by scanning a reference sample once per week and ensuring the results are consistent over time. We also routinely scan calibration phantoms to evaluate both dimensional accuracy and unsharpness.

That said, no algorithm is perfect, and our algorithms will occasionally make mistakes. That’s why we’ve built the Glimpse Portal in a way that allows you to check the results of any algorithm by visualizing feature overlays over the original CT images.

Note that absolute dimensional measurements are often accurate but not formally guaranteed. Instead, our technology is primarily designed to quantify build quality and variability, detecting deviations between cells or lots with high sensitivity.

Core shape

What we track

Area & effective diameter

Defined by the area captured by the inner contour of the innermost cathode wrap and the diameter of the corresponding circle

Circularity, MCC

Defined by maximum inward deviation from a minimum circumscribed circle (MCC). This metric identifies flat spots in the core

Circularity, d_min/d_max

Defined by eccentricity (min diameter / max diameter, or d_min/d_max). This metric identifies oval-shaped cores

Concentricity

Defined by the distance from the center of the can to the center of the core. This metric measures the centeredness of the jellyroll within the can

Note that we have two different core circularity metrics:

d_min/d_max and MCC. These metrics are complementary and can identify different types of circularity issues (see details below). Neither metric will approach 100% since jellyrolls are actually spirals; for instance, the maximum MCC value we observe in practice is ~93%.

Area & Effective Diameter

Circularity, dₘᵢₙ/dₘₐₓ

Circularity, min circumscribed circle (MCC)

Concentricity

How/where it is measured

Cylindrical cells:

Measured in every radial cross section from top to bottom of the jellyroll

Why it matters