Development of a Self-Learning Process for Determining Quality in Battery Cell Production Based on Optical Thickness Measurement

In the early production phase of battery cells, the reject rate is particularly high, as conventional end-of-line tests primarily record the overall condition and can overlook minor defects. In the “Quaze” project, we are working with Precitec to develop a non-destructive measurement method that records the spatial expansion of pouch cells during charging and discharging cycles and uses this data to reconstruct a meaningful expansion field. In this way, the “good” expansion behavior that indicates defect-free cells is to be determined. AI evaluates the measurement data, detects deviations from the “good” reference pattern, and provides indications of faulty process steps. This accelerates corrective measures, reduces scrap and costs, and improves the overall environmental balance of cell production.

Initial Situation

The scrap rate in battery cell production, especially in the ramp-up phase, is high. During this ramp-up phase of the production, scrapt rates of 15 to 30 percent can occur. Currently, End-of-Line quality controlprimarily records the overall electrical value of the battery. Small defects might not be detected at this stage. The introduction of innovative, non-destructive quality control measurement technology is therefore essential.

Objective

Our approach is to develop such a measurement method that can be integrated into battery production and offer spatial resolution of cell expansion during charging and discharging cycles. In this way, potential production defects can be detected at an early stage, the responsible production process identified and optimized saving costs and resources.

Approach

For the measurements, a pouch cell is compressed between two metal plates, with the pressure being applied centrally via a pneumatic cylinder and monitored with a force sensor. In order to determine the quality of the cell, we measure its expansion during the charging and discharging cycles. For this purpose, three distance sensors arranged in a triangle distribution at the edge of the cell record the movement of the metal plates. These values are used to reconstruct the spatial expansion of the cell (see adjacent figure). AI-supported processing of the measurement data enables a clear quality assessment: a model is trained with expansion patterns of defect-free cells (“good”). Cells with significant deviations are classified as “defective.”

Results

Pouch cells expand during operation. The expansion behavior can be divided into two different types: reversible expansion occurs during charging/discharging cycles, while irreversible expansion is caused by deposition of degradation products, which causes capacity loss. We use three distance sensors to record the spatial distribution, considering the two forms of expansion separately (see Figure 1). We can derive an expansion reference pattern from the measurements of defect-free cells. Intact cells follow this pattern; significant deviations indicate defects and provide clues about theresponsible production process. For example, for defect-free Li-ion pouch cells, profile of reversible expansion can be observed during a slow discharge cycle (see Figure 2). We are currently working on measuring defective cells in order to identify patterns in their expansion behavior as well.