Second, find the breakthrough point: AI “into the factory” to achieve double liters of quantity and quality

With the vigorous development of technology such as machine vision and the Internet of Things, the manufacturing enterprises at the forefront of The Times are building smart factories based on the existing IT architecture and combining cutting-edge technology through expansion solutions.

At the Fifth Global Intelligent Industry Conference 2021, Professor Jia Jia, tenured professor of the Chinese University of Hong Kong, member of the International Association of Electrical and Electronics Engineers, and chairman/founder of Simo Technology, said that the core of intelligent manufacturing lies in “intelligence” : Intelligence is not simple automation, intelligence is to make the manufacturing has a “brain” and a variety of nervous systems that let the brain make decisions, EVM32-BASE only with a smart “brain” can maximize the role of automation “arms” – and AI is the core factor that makes the manufacturing of industrial machines and equipment that can think.

Intelligent sorting, intelligent fault prediction, AI surface defect detection and even intelligent quality control, timing prediction… These applications highlight the key role played by AI in different stages and fields of manufacturing, so it has become the most typical case of intelligent manufacturing in recent years.

In the case of the aforementioned enterprise, Ningde Times, its demand for capacity and quality improvement based on the continuous growth of global market demand has prompted the demand for AI power battery defect detection solutions. Such a solution should not only meet the requirements of the EVM32-BASE headquarters layer by layer control, but also need to have a more efficient real-time defect detection capability, that is, to achieve more than 400FPS in a single process and achieve the goal of zero leakage detection in image processing speed.

EVM32-BASE

Defect detection is a highly detailed and time-consuming project. The traditional artificial defect detection method is not only slow and poor accuracy, but the traditional digital image processing technology has poor generalization ability, which requires parameter adaptation according to each machine and is disconnected from the division of labor and headquarters, lack of overall deployment control ability, and processing capacity cannot match the continuous growth of market demand.

“When Ningde era and Intel communicate such confusion, they clearly put forward two needs, one is to capture the pictures of different production lines, with real-time image data analysis to achieve product defect detection during battery production, and the other is to help EVM32-BASE Intel to establish a unified AI platform, which can make defect detection more real-time and accurate.” Said Wu Zhijing, account manager of Intel’s industry solutions business group in China.

Before this, Ningde era has deployed multiple cameras on each battery production line in the battery product manufacturing factory, which can produce hundreds of pictures per second, and a factory has at least a dozen production lines, so a factory has thousands or even tens of thousands of pictures every second. Therefore, Ningde Times urgently needs to introduce a set of technical solutions to analyze and process the above-mentioned massive pictures in real time to determine whether there are product defects in the production process, so as to solve the industry problems in quality control.

After several investigations, the company chose to introduce a new generation of Intel ® Xeon ® scalable platform product portfolio with integrated AI acceleration capabilities, and built a set of AI battery defect detection solutions across the “cloud-edge-end”, integrating computer vision (CV), deep learning EVM32-BASE (DL) and machine learning (ML) technologies.

The industrial vision platform is the core system of this new AI defect detection solution, which is built in the form of cluster, the “cloud platform” in the Ningde Times headquarters, the “edge” system is set up in the division of labor factory, and the “terminal” is set up in the production line, which is not only easy to unified control, but also can reduce the processing pressure through distributed deployment.

But distributed reasoning often suffers from a lack of cohesion. To solve this problem, the choice of Ningde era is to respond with a unified big data analysis and AI platform, while choosing the PyTorch deep learning framework optimized for Intel ® architecture for AI processing, and Intel’s open source OpenVINO™ tool suite to further accelerate AI reasoning performance.