Mainboard repair is an area that requires the inheritance of skills and the accumulation of experience. Recently, LUSTER has successfully precipitated the valuable experience of the master by using AI technologies such as knowledge graphs and large models, so that employees who have been in the company for three months can have maintenance capabilities equivalent to five years of experience.

Experience Gap in Dealing with Massive Data

As the "heart" of electronic products, the reliability and stability of the mainboard are directly related to the performance and life of the device. During the manufacturing process, there are often failed or defective mainboards, which require experienced FA (Failure Analysis) engineers to perform failure analysis and repair. However, individual differences and subjectivity lead to inaccurate maintenance results and unstable maintenance results.

Troubleshooting a mainboard with unknown issues is a highly complex task, and newcomers can only rely on Standard Operating Procedures (SOPs) to gradually analyze the problem, which may require thousands of steps. On the other hand, experienced technicians can quickly and accurately find solutions based on their expertise and skills.

"Just like solving a Rubik's Cube, beginners need to follow step-by-step methods to solve it, while Cube masters can use techniques to find quicker solutions. For example, when certain detection parameters fall within a specific range, they can skip certain subsequent steps. They can even determine which parts to test first through visual inspection," said Bao, Head of LUSTER Research and Development.

In addition, with each new product iteration, manufacturing companies require a significant amount of manpower and resources to develop ESOPs for new mainboard models. This task is typically performed manually, resulting in low efficiency and the potential for omissions or inaccuracies. Furthermore, ESOPs need constant optimization and adjustments in practical applications, a complex and time-consuming process.

"For example, after the production of mainboards is completed, operators place them on testing machines for inspection, generating a large number of test items, typically in the range of hundreds of thousands. The testing machines store test logs during the mainboard inspection. If these test logs are manually compared and analyzed, it would incur significant manpower and time costs," added Bao.

Both the development of ESOPs for new products and the testing and operation processes during repairs heavily rely on the experience of FA engineers. However, frequent personnel turnover makes it difficult to accumulate and pass on expert knowledge and know-how. New employees often have to start from scratch and accumulate their own experience through trial and error.

Now, they have finally retained an experienced "master technician."

Million-graph Knowledge Intelligence

Mainboard repair is a knowledge-intensive business that involves knowledge from multiple technical domains. LUSTER AI solution addresses this by building a vast knowledge graph that integrates explicit and implicit knowledge in mainboard repair. With the GLM (Generalized Linear Model) large-scale model that reinforces domain-specific knowledge, it offers a new and innovative mainboard repair solution, revolutionizing the traditional operational mode.

Explicit knowledge in mainboard repair includes information such as mainboard models, technical documents, and procedural guidelines, which are historical data. Implicit knowledge, on the other hand, encompasses insights and patterns derived from repair engineers' historical repair records, fragmented notes, experiences, typical cases, and knowledge transfer between mentors and apprentices. It may even include intuitive perceptions that are difficult to describe in words. For example, visually inspecting the shape of the mainboard and components can often help in identifying the most probable areas to start the troubleshooting process.

"To uncover this implicit knowledge, in-depth research and analysis are required, delving into detailed repair process logs and data from each past mainboard. It took us nearly half a year to construct a knowledge graph based on various mainboard models over the years. This graph includes hundreds of concepts such as components, test results, and test steps, along with hundreds of thousands of entities, relationships such as Pass, Fail, Contains, Connects, and Judgments, and millions of attribute values. We also collected millions of data points and historical repair records. By building the mainboard repair knowledge graph, we convert both explicit and implicit knowledge into structured knowledge entities and relationships. Through the application of large-scale models and intelligent assistive tools, we achieve the long-term accumulation and inheritance of knowledge," said Bao, Head of LUSTER Research and Development.

AI Decision-Making: One Step Solution

Optimizing repair steps, accelerating mainboard repair:

By leveraging algorithms such as knowledge graph reasoning and ESOP process optimization, combined with the synergistic enhancement of knowledge graph and large-scale models, AI-powered repair systems possess the capability for professional knowledge inference. By inputting a description of the mainboard issue, the system can understand the possible faults and provide targeted guidance for each step of the diagnostic process through real-time interaction and process optimization. This enables the system to offer fast and accurate troubleshooting solutions. In practical applications, repair technicians can skip irrelevant steps, rapidly and accurately identify the cause of the issue, and take effective measures for repair. The unit output per hour (UPPH) has increased by 37.4%, significantly improving repair efficiency.

Automatic generation of repair ESOP documents, with a 50% speed increase:

The intelligent ESOP generation tool can automatically generate new ESOPs based on the input of new mainboard information, utilizing technical documents and repair history from past mainboard models. As repair tasks progress, the ESOP is dynamically optimized and adaptively adjusted based on real-time repair data, ensuring accuracy and practicality. The speed of ESOP generation has increased by 50%, saving a significant amount of manpower and time costs.

Continuous experience accumulation, building a long-term knowledge system:

As repair tasks continue, the knowledge base continues to grow and evolve. Key knowledge, such as failure analysis and repair experiences, can be continuously accumulated, leading to iterative improvements in the knowledge graph and models. This process helps build a long-term knowledge system and application framework. With advancements in large-scale model technology and artificial intelligence, this "master technician" will continue to learn and progress, assisting researchers and practitioners in mainboard repair to work more efficiently, conveniently, and accurately.

Currently, in the industrial field, the use of large-scale models for vertical applications is still in the exploratory stage. LUSTER AI will actively collaborate with industrial customers in exploring other scenarios, aiming to solve industry challenges and provide more intelligent and efficient solutions for the industrial sector. This will contribute to the innovation and development of the industry as a whole.