Using SEM image analysis and AI to accurately predict final product performance from intermediate products
Hitachi, in collaboration with Hitachi High-Tech, has developed a new process informatics technology as further advancement of their earlier Manufacturing Process Improvement Solution*1 that helps find optimal manufacturing processes, to support manufacturing line launch and improve yield. The newly developed technology utilizes measurement data of intermediate products obtained during manufacturing to predict final product performance with high accuracy. The performance predictions use the microstructural features*2 extracted from SEM*3 image data of the intermediate products as input parameters. The predicted results can then be fed back into the update of manufacturing conditions (Fig.1). While conventional performance prediction has been based on manufacturing conditions set at the start of the manufacturing process, the utilization of microstructural features from intermediate products enables highly accurate prediction with closer correlation to actual performance.
Verification trials on a lithium-ion battery prototype manufacturing line have demonstrated that the new technology predicts performance with higher accuracy than conventional methods, even with limited datasets. Determining product performance at an intermediate stage of the manufacturing process is expected to reduce trial-and-error steps in the line and ease the burden on workers, while shortening manufacturing line launch time and helping improve yield. In the future, through co-creation activities with our potential customers, we will proceed with demonstrations to improve the efficiency of manufacturing lines using this technology. In addition, we will promote research and development with a view to applying it to various manufacturing industries and contribute to the realization of a carbon-neutral society by reducing manufacturing loss and environmental burden.
*1 Developing a Manufacturing Process Improvement Solution that Helps Finding Optimal Manufacturing Processes, and Enhancing the Informatics Business: January 23, 2025.
*2 Numerical values that quantitatively express the characteristics of the internal structure of an analyte.
*3 Scanning Electron Microscope.
Figure 1. Illustration of a lithium-ion battery manufacturing process in which microstructural features are extracted from intermediate products and undergo process informatics
Background and issues
The manufacturing industry faces growing demands for faster launching of manufacturing lines and more efficient manufacturing processes, to meet rapidly changing market needs. To address these demands, Hitachi and Hitachi High-Tech have announced and are conducting verification trials of the Manufacturing Process Improvement Solution, which proposes highly efficient manufacturing processes by using our proprietary database specialized for manufacturing processes and problem-solving generative AI. One issue with using AI for highly accurate predictions of product quality and performance is the need for large amounts of training data, presenting time and cost challenges. Hitachi and Hitachi High-Tech have therefore developed this process informatics technology, focusing on the intermediate products obtained during the manufacturing process and merging microstructural features extraction technology with informatics technology, enabling highly accurate prediction of final product performance even with limited training data.
Technologies developed to solve issues
1. Performance prediction model using microstructural features of intermediate products
It was confirmed, on a prototype manufacturing line, that battery performance can be predicted with high accuracy even with limited datasets by extracting microstructural features of electrode sheets*4, which are intermediate products during lithium-ion battery manufacturing, and then by applying these to a machine learning model. Previously, manufacturing conditions set at the time of electrode manufacturing were used as explanatory variables*5 in the model. Now, by utilizing the microstructural features as explanatory variables, the prediction error has been reduced (Fig. 2). Moreover, by evaluating the contribution of explanatory variables to the objective variables*6, the correlation between microstructural features and battery performance has been clarified. Being able to use evidence-backed prediction results in the manufacturing process should help to reduce trial-and-error steps during manufacturing line startup and ease the burden on workers.
*4 One of the constituent elements of batteries, in the form of a sheet coated with the active materials of anodes and cathodes and conductive additive, etc. on a current collector.
*5,6 Explanatory variables are used as input in a machine learning model, which have impacts on the variables to be predicted (objective variables).
Figure 2. Examples of performance prediction: Results of prediction accuracy verification for (a) conventional machine learning model and (b) the newly developed model. The objective variable of prediction is ion conductive resistance in the electrode. The red dotted line indicates matching with the prediction. MAE (mean absolute error) indicates mean values of prediction error.
2. Technology for extracting microstructural features by analyzing top-view SEM images
Using a SEM developed by Hitachi High-Tech that does not require high vacuum conditions, top-view images of the electrode sheet are taken in a few minutes. Following this, through image analysis, we have developed a technique to extract important microstructural features reflecting the agglomerate structure, porosity, and the distribution of constituent elements. This technology makes it possible to determine the quality of the electrode sheets that are intermediate products, without having to wait for performance measurement of the final battery product. Early detection of quality defect risks in the manufacturing process enables quick remedial treatment, which should lead to more efficient manufacturing processes.
Future prospects
Hitachi continues R&D to further refine this technology, and through co-creation activities with potential customers, intends to help make manufacturing processes more efficient in a wide range of fields. Moreover, while seeking to ease the burden of on-site workers, we will contribute to the realization of a carbon neutral society by reducing manufacturing loss and environmental impacts.
The details of the developed technology are scheduled to be presented in part at AABC(Advanced Automotive Battery Conference) Europe 2025, Global Battery Manufacturing Production, to be held June 25-26 in Mainz, Germany.
For more information, use the inquiry form below to contact the Research & Development Group, Hitachi, Ltd. Please make sure to include the title of the article.
https://www8.hitachi.co.jp/inquiry/hitachi-ltd/hqrd/news/en/form.jsp
