Activity News

Publish Date : 2022/08/26

Pure GPU lineage Leadtek AI x AOI platform enhances 3D machine vision optical inspection performance


Leadtek continues to invest in the application and innovation of automated AI management platforms, integrating artificial intelligence, NVIDIA GPU, NVIDIA certified workstation/server and years of professional image processing technology to create an AI management system (AIDMS) required for automated optical inspection (AOI). AIDMS subverts the traditional "dynamically split hardware resource function" and "multi-person collaborative AI model function", which can flexibly fine-cut hardware resources such as GPU memory at the same investment cost, significantly reducing costs and shortening project development time with multi-person collaborative AI model. On the other hand, the innovative "AI model automatic deployment and update" function realizes process automation and standardization, establishes a robust AI model for AOI equipment suppliers to uniformly manage the image data in different optical inspection equipment, and effectively reduces the machine error of the equipment, to ensure the consistency of machine quality inspection and enhance the competitiveness of products in the market.

Maximize the use of hardware resources under AIDMS

Leadtek Business Development Manager Wayne Huang pointed out that the traditional AOI image processing algorithm can meet the basic needs of image positioning, measurement, object detection, etc. through artificial programming rule-based algorithm for the image features with a single and simple background. Taking the metal surface defect detection as an example, due to the diversity and uncertainty of defect features, coupled with the sandblasting coating on the surface of the product, in a high-resolution optical system, defect features with complex backgrounds and variations are formed. As a result, it is difficult to manually program traditional rule-based algorithms. Engineers have to continuously adjust and optimize parameters but will fail to meet the customer's requirement that the yield rate is generally above 99.8%, which makes the maintenance of the AOI defect detection system difficult and high cost of factory maintenance and operation inevitable.

Under the requirements of high efficiency and high quality, the introduction of Leadtek AIDMS is regarded as one of the solutions. Richard Chou, general manager of Leadtek PC BU, said that to assist enterprises in digital transformation, a codeless interface is an essential requirement. The introduction of Leadtek's AI total solution - AIDMS management system can reduce the AI environment construction time by more than 50%. In addition, AIDMS allows users to upload AI models and labeled data that were developed before AIDMS were introduced uploaded to the AIDMS platform. With the help of GPU, the verification and retraining can be accelerated and AI maintenance can be well executed.

Different from the general scattered stand-alone equipment, Leadtek uses the AIDMS AI management platform to centralize resources for configuration management based on the NVIDIA certification system. Image data management, AI model building, training, verification, deployment and model analysis reports are all completed on the same platform. Even the inspection hosts are managed at the same time, so that the data between the devices can be systematically managed, which is convenient to deal with the tricky leakage and overkill problems in real time, quickly retrain the model, and then deploy the update on all the devices with one click, making the process automation and standardization a reality.

It is worth mentioning that AIDMS strengthens hardware resource management. In the AI model development stage, not only can multiple people collaborate to jointly develop training models, but also split CPU, GPU memory and other configurations according to the complexity of the model architecture. For example, in the past a developer had to have a GPU to work independently. With the same hardware, AIDMS allows multiple developers to develop and train models at the same time, maximizing production value under AIDMS regardless of hardware resources and manpower.

Dual GPU Edge Inference Device

In edge computing, Leadtek launched the EG1090 inference device equipped with the latest NVIDIA GPU, which is used as a real-time image recognition host. A single device can be equipped with dual GPUs, coupled with excellent heat dissipation and low power consumption optimized design, it can provide 7~10 times faster model inference performance.

Richard Chou, General Manager of the PC BU of Leadtek, emphasized that Leadtek has more than AI and GPU expertise. One of the important reasons why customers trust us is that Leadtek is familiar with the operation mode of the AOI industry and the problems it faces, and is able to overcome the problems during the AI application process one by one, greatly reducing the transformation time. In the future, Leadtek will continue to invest in the research and development of AIDMS, provide industry-leading innovative services, and cope with the ever-changing future with customers.