AI-supported Ultrasonic Testing for Safe and Long-lasting Batteries

In the SAMBA research project, we are working with industry partners to develop a non-destructive testing method for quality control of battery pouch cells based on scanning acoustic microscopy (SAM). The aim is to build and scientifically validate a demonstrator that enables rapid random inspections of cells and automates the evaluation of measurement data using an AI-supported routine.

Initial Situation

Inhomogeneities in battery cells, such as gas inclusions, foreign particles, or defects in the separator, impair their service life and, in the worst case, can jeopardize their safe functioning. For manufacturers, such defects pose the risk of extensive recalls and considerable economic damage.
Currently, industrial battery cell production lacks fast and cost-efficient methods for detecting such defects reliably and reproducibly—both at the end of production and before assembly into modules. There is therefore an urgent need for improved non-destructive quality control methods that can increase the safety and acceptance of battery technology.

Objective

The SAMBA project is developing and validating a novel, ultrasound-based, non-destructive testing method for the spatially resolved detection and description of inhomogeneities and safety-critical defects in pouch cells.

The core objective is to build a demonstrator – initially as an offline solution with a target technology readiness level (TRL) of 6 – that enables rapid random sampling analysis of cells. The evaluation of the generated SAM image data is largely automated: an AI-based algorithm identifies, localizes, and classifies defects. In addition, the project supports the development of a measurement setup that can be integrated into cell production by industry partners and evaluates the suitability of the method for industrial quality control.

Approach

We use scanning acoustic microscopy (SAM) to generate high-resolution ultrasound image data of battery pouch cells. To this end, the measurement setup is being further developed and optimized in the project specifically for the requirements of cell testing. Commercial battery cells with deliberately introduced defects are systematically examined to test the detection capability of the method. The SAM data is evaluated using AI. Specifically, we are developing algorithms and using machine learning methods to automatically detect anomalies, locate defects, and categorize them. At the same time, project partners PVA TePla AS GmbH and Microvast GmbH are supporting the development of a test set-up that can be integrated into production processes in the future; both technical requirements and feasibility for industrial use are being evaluated here.

Results

The SAMBA project delivers a functional demonstrator of a SAM-based testing system for random, rapid inspections of battery pouch cells (target TRL ≈ 6). A validated, partially to fully automated AI algorithm is being developed that processes SAM data, detects anomalies, determines their position, and classifies defect types. Together with industry partners, concepts for a set-up that can be integrated into cell production are being developed, taking cycle times, handling, and automation into account. Overall, SAMBA improves end-of-line quality assurance, thereby increasing the operational reliability of batteries for competitive European battery cell production.

The "SAMBA" project is funded by the Federal Ministry of Economic Affairs and Energy (BMWE).