On November 26-27, 2022, at the Global Artificial Intelligence Technology Conference 2022 (GAITC 2022), Dr. Yongliang Tang, R&D Director of LUSTER, was invited by the China Artificial Intelligence Society to introduce LUSTER's self-developed deep learning platform for industrial quality inspection scenarios - F. Brain and its competitive advantages.

The newly released F.Brain (Fabrication & Factory Brain) deep learning platform is based on LUSTER's years of experience in the industrial field and is a self-developed deep learning platform specifically for industrial quality inspection scenarios. For the industrial quality inspection scene fragmentation, fast delivery, high iteration, low requirements and other characteristics, specially developed a variety of algorithms to achieve the pixel-level inspection of light, weak and small defects, through lightweight, process-oriented model design to achieve rapid deployment, through data enhancement, model pre-training and meta-learning, etc., to solve the "cold start scene" with limited NG defects and fewer training samples ".

3 Functional Platforms

Standardize and platformize the entire application process to significantly reduce the delivery time.

● Data platform: It can interconnect with production equipment and is responsible for data collection, transmission, management, annotation and audit of actual industrial application scenarios.

● Training platform: The biggest advantage is to support the training of deep learning models for industrial scenarios in the cloud and stand-alone version; hundreds of deep learning models are supported internally, covering mainstream general detection models, which have now evolved into an AI algorithm mall with multiple models.

● Inference platform: Deployment of inference platform for the service side and edge side to quickly realize the quantification, deployment and distribution of deep learning models on the "cloud-side" side of industrial scenes.

Through the above three sub-platforms, LUSTER can efficiently realize the standardization and platformization of the whole process of deep learning application and meet the requirements of rapid delivery.

6 Management Modules

Closed-loop process of application data can be more time-saving and efficient

F.Brain has six management modules, such as project management, model management and release dataset management, which can seamlessly connect the three sub-functional platforms, realize the tandem of each step of deep learning application process, cover the whole data process of deep learning application closed loop, and achieve the zero code platform of deep learning industrial scenario application, which is more conducive to the daily operation of the software.

8 Major Features

Lightweight, process-oriented model designed for rapid deployment

Compared with existing deep learning platforms, F.Brain's application scenarios are more vertically focused on the industrial field, realizing the rapid batch replication of deep learning applications in industrial scenarios through process-oriented and lightweight model design.

The new F.Brain platform has already provided the solution for LUSTER's new energy product line of lithium electrodes. Curious of how it works? Stay tuned for the next post!