The international consulting firm McKinsey forecasts that the entire process chain for producing lithium-ion batteries, from extracting the materials to recycling, will reach more than $400 billion by 2030. Roughly 30% of the value created is attributable to cell production. The process of battery cell production is divided into three steps: electrode production, cell assembly, and formation and aging. Maximum precision is required for all three steps in the process, which means there is a high potential for error. Machine vision can help mitigate such errors. Product quality is an essential part of the production of batteries. In the manufacture of electrodes, copper or aluminum foil is coated on both sides with an active material. The coating must be within a narrow tolerance range and precisely oriented on both sides. Any impurities, inconsistencies, or defects will impair the quality, safety, and performance of the entire battery. Machine vision is essential for inspecting the coating surface on the top and bottom and ensuring that the coating is applied uniformly and is free of defects (Figure 1). The thickness of the coating is a critical parameter that can affect the battery’s performance. Since the thickness of the film varies between 5 and 25 μm, depending on the cell type, it is crucial to identify any deviations early in the process. A particular challenge for machine vision — aside from the high processing speeds during coating (up to 80 m/min at a coil width of up to 1.5 m) — is the overlay area between coated and uncoated areas because the actual gray value lies below 5. Figure 1. Machine vision is used to check the surface, dimensional accuracy, and alignment of the coated electrode tracks to ensure their quality. Courtesy of MVTec Software GmbH. This processing step is followed by a drying process, then a step called calendering, in which the coated foils are compacted. In the separation steps that follow, including slitting and cutting, the coated electrode foil is separated into narrower strips or individual electrode sheets that have the desired dimensions for subsequent cell assembly. Depending on the battery cell format, laser or shear cutting methods are used to separate the sheets. Cut geometry, cutting edge quality, and particle contamination are crucial criteria that can be optimized through machine vision. In electrode manufacturing, there are several applications in which machine vision is essential for precise orientation and positioning, along with surface inspection and measuring. Aside from pure quality control, results delivered by the machine vision software permit consistent process optimization, sometimes in combination with other data points. This enables the early determination of whether, for example, the coating was not applied uniformly, the width of the electrodes is inconsistent, or coating residue adheres to the edges (Figure 2). Figure 2. A wide variety of defects can occur during electrode production. Machine vision is used to check the quality of the coating of the copper or aluminum foil, and to detect defects. Courtesy of MVTec Software GmbH. Machine vision and the corresponding hardware, such as line-scan cameras, can robustly detect and classify the smallest defects even under these challenging production circumstances, including high-speed and problematic contrast conditions. Methods used to recognize anomalies can also be used to identify hard-to-define defects, such as burrs, impurities, cracks, and unclean cuts. Avoiding errors in cell assembly Following electrode manufacturing, the next production step is cell assembly, in which the electrode foil and a separator are stacked either in prismatic or pouch cell packs or wound in cylindrical cells. Next, they are inserted into the battery housing. Most of the housing is subsequently welded and filled with electrolyte via a remaining opening. Once the filling process is completed, the last opening is closed. Once again, machine vision software is used to check the individual work steps to assure the necessary quality. To produce pouch and prismatic cells, the precise and damage-free positioning of stacked electrode sheets is crucial. A stack comprises the sequencing of an anode, separator, cathode, separator, and so on. The stacking process follows very tight parameters to ensure the optimal performance of the battery. In Z-folding, individual anode and cathode sheets are placed laterally in the Z-folded separator web. In single-sheet stacking, the separator is available as a sheet used to form the stack. Both Z-folding and single-sheet stacking take place at high speeds of one second per sheet. Depending on the cell specification, cell stacks can comprise up to 120 individual layers with a stacking accuracy of ±200 to 300 μm. The exact positioning of the individual sheets in a stack is considered the central quality criterion. There are several steps during the stacking process in which machine vision helps to reduce faulty stacks. It is used to inspect electrode surfaces and tabs as well as to measure cut geometry. Tab inspection after slitting and cutting has especially proved to be a major issue, because tabs have flexible characteristics. Deformations such as bending, twisting, and tearing are frequent and diverse, which makes the selection of criteria to determine whether a tab is OK or not OK a challenging task. In this step, accuracy is almost more important than speed. While a deep-learning approach could be feasible here as well, not enough bad images exist in comparison to good images. Thus, the task is usually carried out through specifically precise classic rule-based methods. Machine vision is also used to read laser-etched Data Matrix codes on the cell sheets, which are usually found on the tabs. In the stacking process, the length of the anode and cathode sheets are inspected with the use of, for example, x-ray and CT scans. X-ray images, however, are difficult to inspect due to their defocused nature. While CT scans can generate exceptionally high-quality images, the procedure itself is time-consuming. Given the limited window available for battery inspection, only low-resolution images can be generated in this context, posing challenges in interpretation. Machine vision helps to recognize defects in these images that a human eye would not be able to spot. It also supports the detection of electrode sheet positions and the alignment of stacks. The improper alignment of layered electrodes can lead to problems, such as the inability to insert them into the cell housing or electrolyte leaks. To prevent such issues, the alignment system must ensure precise control over electrode positioning (Figure 3). Figure 3. Machine vision ensures the proper alignment of stacked electrode sheets during battery cell assembly. Courtesy of MVTec Software GmbH. Additional failure types that can occur in the stacking process that can be discovered early with machine vision include unreadable Data Matrix codes, bent stacked cells, scratches or folds on the electrode surface, low-positioning accuracy of the anode and cathode sheets or foils, damaged electrode surfaces and edges, or damage to the tabs at the material transition zone. To finish the cell assembly, the cell stack is inserted into the cell housing and is partly sealed. During this process, it is crucial to cut the tabs to a defined length without damaging the material, to insert the cell correctly into the cell housing, and to flawlessly weld seams free of damage to ensure a tight seal of the battery. Welding is a critical step in the production of batteries. For example, several welds are needed to connect the current collectors inside the cells. The quality of these connections is crucial to the quality of the battery, where even a 1% fail rate is a serious issue. Various methods can be used to inspect the weld seams and recognize anomalies. The high variation of potential errors resulting from this process makes AI approaches particularly suitable for detecting these errors (Figure 4). In contrast to an appropriately trained deep-learning network, the human eye would not be able to detect and quantify patterns or make decisions based on these patterns. Furthermore, due to machine vision, Data Matrix codes can be read robustly, even during motion and changing lighting conditions. Figure 4. Checking the quality of the weld seam with the naked eye is often impossible. An appropriately trained deep-learning network provides a remedy and enables nondestructive in-line inspection. Courtesy of MVTec Software GmbH. End-of-line tests Formation and aging represent the final stages in the production of battery cells, encompassing charging, discharging, and quality testing. Preceding formation and aging, during the cell assembly phase, the cells undergo final contacting and welding, in which a protective coating or insulation foil is applied onto the battery cells. This protective layer could exhibit imperfections, such as trapped bubbles with foreign particles beneath them, scratches that cut through the foil, and an unevenly applied coating. When these cells are packed tightly into a battery module, several factors can lead to an electrical shortage or overheating. Battery cell coatings might display a range of minor imperfections that do not affect their functionality as well as apparently insignificant scratches that can negatively affect safety and functionality. Within formation and aging, machine vision helps spot these defects while reducing the rejection of cells that, despite having flaws, still serve their purpose. Additional tasks performed by machine vision include the reading of Data Matrix codes on the battery cells, the measurement of the outer diameter, and cell dimensions of cylindrical cells. Reading the Data Matrix code can be challenging due to its small size and highly reflective surfaces. Before the battery cells leave the factory, they undergo a variety of end-of-line tests, including visual inspections. After the tests, the cells are sorted according to their power capacity, packaged, and prepared for shipping. For visual surface inspections and measurements, machine vision can quickly determine whether cells need to be rejected. If a cell is deformed, does not meet the specified cell diameter, or has any other type of surface damage, the machine vision software classifies it accordingly. Unreadable codes, impurities, or foreign bodies can also be robustly detected, even on highly reflective surfaces. This ensures that cells classified as damaged can still be screened out before they leave the factory. Achieving in-line quality control As illustrated, battery cell production is divided into many complex individual steps, where machine vision enables 100% in-line quality control and significantly contributes to process optimization. Both classic rule-based methods and deep-learning-based methods are employed. The rule-based techniques are especially suitable for very fast production steps, such as coating and calendering, or when maximum precision and dimensional accuracy is required — for example, during the measuring and stacking of individual electrode sheets for the production of prismatic battery cells. On the other hand, deep-learning technologies increasingly come into play in the later phases of cell production, particularly for defect detection and surface inspection. They permit the robust detection of even the smallest defects and contaminants. In addition, the deep-learning technology known as anomaly detection allows previously unknown defects to be identified. Another advantage of this method is that only good images are needed to train the model. Sustainable battery production The use of machine vision in battery production offers numerous benefits for manufacturers. The consistent integration of machine vision as part of a comprehensive digitalization and integrated process optimization strategy holds significant potential. It enables a reduction in rejection rates and makes a measurable contribution to sustainability due to the more efficient use of valuable raw materials. At the same time, it increases output because, as the “eye of production,” machine vision facilitates fully automated and seamlessly traceable production processes. Meet the author Klaus Schrenker is the business development manager for MVTec. He previously worked at Dropbox for two years and before that worked for French tech companies where he was responsible for internationalization, partnerships, and customer acquisition. He holds a bachelor’s degree in industrial engineering and management at FAU Erlangen-Nürnberg and a master’s in innovation and industrial management at the University of Gothenburg; email: klaus.schrenker@mvtec.com.