Contemporary Amperex Technology Co. Limited (CATL), a lithium-ion battery R&D and manufacturing company established in 2011, has led the world in power battery usage for four consecutive years since 2017 according to SNE Research. In 2019, being keenly aware of potential challenges such as employment difficulties, high labor costs, and profound external changes, the company realized the importance of industrial upgrading. As a result, it embarked on an intelligent path to optimize its R&D and manufacturing processes, and improved the quality inspection line for battery defects in conjunction with Baidu's deep learning open-source platform, achieving a leap in overall inspection accuracy and building an intelligent defense line to assure its power battery quality.

Scenario Analysis

During the production of power batteries for new energy vehicles, it is a must to examine whether the particles resulting from metal welding fall on the surface, whether there are any gaps or voids in the coating, and whether the welding process is consistent, etc. The quality of the power battery is so critical that even a slight defect can lead to major property safety issues if there are no other assistant systems in the end market.

Challenges

• - Challenge one: Large amounts of products. The overall number of test parameters is huge considering its great output and the fact that a single battery possesses more than 3000 parameters to be tested.
• - Challenge two: Requirements of high precision. The required miss rate of battery inspection on the product line has been raised to less than a billion from a million, [zy1] and the average time of defect detection for a single product image is standardized as <2ms.
• - Challenge three: Power batteries have complex defect types with slight differences.
• - Challenge four: Huge input in human and financial resources in the traditional context. An assembly line requires more than a dozen workers, and manual confirmation is needed given the varied standards in different production lines.

Technical Solution

To tackle these industrial difficulties, CATL has improved the algorithm accuracy with the help of PP-YOLO and ResNet series algorithms which are deeply optimized by Baidu's PaddlePaddle; raised the detection accuracy[zy2]  and prediction efficiency by adopting strategies including adjusting and predicting Batch, adopting multi-threaded data pre-processing, and optimizing the data memory copy time, so that it can fully adapt to the actual production needs. In addition, the employment of PaddleX full-process development tools has vastly simplified the company’s industrial implementation with deep learning technologies.

Actual Effects

After the adoption of intelligent battery quality inspection, some procedures have been upgraded based on AI algorithms, with a 66.7% decrease in the over-kill rate of the overall product inspection compared to the traditional ones and a miss rate less than 1DPPB. What’s more, the generalization capability of the algorithm and the efficiency of migration deployment in multiple production lines have also witnessed comprehensive improvements. This not only further guarantees the quality of CATL’s power batteries, but also, to a certain degree, gains low-cost and efficient reuse experiences, which greatly reduces the R&D cost of production lines.

Company Profile

Contemporary Amperex Technology Co. Limited is a leading lithium-ion battery company that focuses on the R&D, production, and sales of power battery systems for new energy vehicles and energy storage systems, and is dedicated to providing first-class solutions for global new energy applications.