The threshold for manufacturing to embrace AI has become so low

In actual production, factories cannot guarantee that products are 100% perfect. The screws may not be tightened, or there may be small scratches on the surface, or the product label may be forgotten to be affixed... A seemingly insignificant defect or flaw may cause loss of the company's reputation and property, or even lead to casualties.

As the old saying goes, "A thousand-mile dam can be breached by a single ant hole." The truth contained in this proverb is equally applicable to modern manufacturing.

In actual production, factories cannot guarantee that products are 100% perfect, or the connecting screws are not tightened, or there are small scratches on the surface, or the product label is forgotten to be affixed... Especially in industries such as display screens, electronic products, automobiles, aerospace, etc., a seemingly inconspicuous defect or flaw can cause damage to the company's reputation and property, or even cause casualties.

In order to free manufacturing companies from the worries of defects and flaws, ThunderSoft is working with Amazon Web Services (AWS) to help customers further improve product yields, release production capacity, and enhance product competitiveness based on the advantages of both parties on the device and cloud sides.

Smart Industrial ADC Systems: Detecting Defects

Traditional manufacturing mainly relies on manual inspection of product surface defects. However, as industrial products become more and more sophisticated and their quantities increase, this traditional defect detection method is becoming increasingly unsustainable.

One of Thundersoft's clients, one of the largest smart panel companies in China, once faced various challenges in defect detection. First, some product defects would lead to unstable panel yields, which put the company at a disadvantage when competing with its competitors. Second, traditional detection methods were inefficient and costly, which was not conducive to the company's long-term development. The smart panel company was eager to upgrade to digitalization and intelligence, but it had no idea where to start because of the lack of relevant IT capabilities and infrastructure.

As a world-renowned provider of intelligent operating system products and technologies, Thundersoft has a deep understanding of the development needs and trends of the traditional manufacturing industry represented by this company. Based on its deep technical accumulation in the fields of intelligent operating systems, graphics and image processing, and artificial intelligence, it launched a one-stop solution for industrial vision inspection in 2018 - the Smart Industrial ADC (Automatic Defect Classification) system.

The system includes three subsystems: automatic defect classification, data cleaning for new product iteration, and certification of business operators. From operator skill certification, data set update to new product introduction, it runs through the entire life cycle of industrial inspection, effectively helping manufacturing companies reduce workload by 75% and increase production capacity by 35 times. Compared with manual inspection, the missed detection rate is reduced by 3% and the accuracy rate is increased by 99%.

At present, the smart industrial ADC system has been successfully implemented in many cases in the LCD panel industry and has been widely praised by customers. In the LCD panel industry, Thundersoft has the experience of implementing the ADC defect automatic classification system, which has been successfully launched on the first actual production line in China and has been running stably for more than one year. In addition to LCD panels, Thundersoft has further expanded the system to industries such as automobile manufacturing, electronic products, cosmetics manufacturing, and rubber manufacturing, helping many customers improve the level of industrial automation and intelligence.

It is worth noting that any customer application of smart industrial ADC systems requires the implementation of machine learning, which requires the help of AWS. On June 4, 2020, when the Amazon SageMaker machine learning service was launched in the AWS China region, Thundersoft took the lead in announcing that it had integrated Amazon SageMaker into its own ADC system, allowing manufacturing customers to easily obtain  AI  quality inspection capabilities in industrial production.

Amazon SageMaker: Lowering the threshold for enterprises to embrace AI

With the amazing performance of AlphaGo and Boston Dynamics Dog, people have reached a consensus on the transformative potential behind AI technology. In many cases, AI can replace humans to complete the corresponding work, and it is highly efficient, rarely makes mistakes, and never gets tired.

Many companies have been eager to try AI, but they can only sigh in despair when it comes to actual applications. This is because the implementation of machine learning is a very complex and expensive task, involving a lot of trial and error, and requires professional skills - in other words, the "threshold" is very high.

Developers and data scientists must first visualize, transform, and preprocess the data before it can be formatted for algorithms to train models. Even for simple models, companies need to spend a lot of computing power and a lot of training time, and may need to hire a dedicated team to manage the training environment that includes multiple GPU servers. From selecting and optimizing algorithms to adjusting the millions of parameters that affect the accuracy of the model, all stages of training the model require a lot of manpower and guesswork. Then, when deploying the trained model in the application, customers need another set of professional skills in application design and distributed systems. In addition, as the data set and the number of variables increase, the model will become outdated, and customers must retrain the model again and again to let the model learn and evolve from new information. All of this work requires a lot of expertise and consumes huge computing power, data storage, and time costs.

Amazon SageMaker can lower the threshold for enterprises to embrace AI, help customers remove the confusion and complexity involved in machine learning, and enable customers to quickly build, train, and deploy models. The entire process is simple and efficient. Especially at the edge, which is commonly involved in the smart industrial field, with Amazon SageMaker Neo, developers only need to train the machine learning model once and run it anywhere in the cloud and on the edge.

Amazon SageMaker Neo optimizes models to run twice as fast, using only 1/10 the memory, without compromising accuracy. Amazon SageMaker Neo optimizes models deployed on Amazon EC2 instances, Amazon SageMaker endpoints, and devices managed by AWS Greengrass, enabling industrial vision inspection applications to seamlessly connect with other applications.

According to Thundersoft CTO Zou Pengcheng, in the implementation of the ADC system in the electrical industry, by integrating Amazon SageMaker, the end user's one-time investment cost was reduced by 42%, the software development workload was reduced by 39%, the system's online time was shortened by 50%, and the system's operating efficiency is 35 times that of traditional testing, which solves the obstacles to the implementation of the ADC system in industrial scenarios.

End-to-end cloud integration: fusion system concept

"In a sense, AWS is a cloud operating system, and Thundersoft is a terminal operating system." Zou Pengcheng described the two in this way. However, Thundersoft and AWS took the lead in joining hands not only because of the complementarity of technology, but also because of the fit of concepts. "The main reason why we can impress customers and stand out in this field is our customer-centric concept, which is consistent with AWS. The solution we help customers meet challenges is the concept of integrated systems, that is, the integration of terminals and clouds, scenarios and technologies, products and services, hardware and software, and vision and AI, and ultimately provide customers with a complete end-to-end cloud solution."

To understand what the concept of integration is, we must first understand what is a non-integrated state. When answering questions from the IoT Think Tank, Zou Pengcheng gave an example, such as the end side uses its own system, the cloud side uses its own architecture, and the engineers on the end side and the cloud side each write their own code, which does not intersect with each other, and communicates through some traditional protocols in the middle. But this method is obviously inefficient. A better state should be that the cloud and the terminal have a consistent system and architecture, and developers do not have to worry about whether it is the terminal, the cloud or the edge side, and the same code can be seamlessly distributed.

This feature is also well reflected in Amazon SageMaker, especially the Amazon SageMaker Studio integrated development environment (IDE), which provides a unified interface for the entire machine learning workflow, making it easier and faster to build, train, explain, inspect, monitor, debug, and run machine learning models.

AWS 中国区生态系统及合作伙伴部总经理汪湧表示：“中科创达是非常优秀的 APN（AWS 合作伙伴网络）合作伙伴，在 IoT、人工智能方面的实力尤其突出。Amazon SageMaker 一个重要的特点在于能够与各类行业应用进行集成，来进一步赋能各行业的应用场景。我们非常高兴中科创达能够成为首批在 AWS 中国区域利用 Amazon SageMaker 的 APN 合作伙伴。基于 Amazon SageMaker，中科创达能够打造更加优秀的智慧工业视觉检测 AI 系统，满足更多客户的需求，助力他们实现智能化转型。”

Digital transformation is a systematic project. In the future, Thundersoft will continue to work closely, firmly and deeply with AWS to help more industries achieve intelligent transformation.