AI meets Siemens first digital factory in China

Siemens AI understands the industry.

Challenges for industrial AI

AI applications in the industrial field are faced with greater and more complex challenges than those in the consumer field – due to limited data volume, unbalanced data annotations and uneven data quality, coupled with the facts that there are not clearly-defined rules and boundaries for many issues in industry and that the knowledge on vertical industries is highly specialized, industrial AI is faced with challenges. Siemens AI better understands the industry, and Siemens expert team has successfully helped customers from many sectors reduce costs and improve efficiency by providing customized AI solutions, including SEWC – a well-renowned digital factory.

SEWC: unlock the power of industrial AI

Siemens Electronics Works Chengdu (SEWC) is Siemens’ first digital factory outside of Germany and also a prestigious name in China’s industrial community. In SEWC, industrial AI technologies have been implemented and shown results. The Siemens expert team helped SEWC reduce costs and boost efficiency by deploying AI technologies including computer vision, small sample learning and multi-modal fusion in the factory. Meanwhile, SEWC is building the first cloud and edge computing-based AI system to unlock the application potential of AI in industry and lead where industrial AI will go.

Game-changing quality inspection of electronic components

Printed Circuit Board (PCB) is one of the key components in the electronics industry. The conventional quality inspection comprised two parts: firstly, after leaving the Surface Mounting Technology (SMT), PCB would go to the THT step for through-hole welding, and then the Automated Optical Inspection (AOI) equipment which made a preliminary judgement of product quality based on the preset inspection parameters; and next, the products that might have quality problems screened by the AOI unit would be sent to quality inspectors for manual judgement. In the conventional approach, the AOI equipment only made judgements based on some simple rules, and the results were not that satisfactory. In fact, up to 80% of the products preliminarily judged defective, were proved of good quality after manual judgement, i.e. the so called “false positive”. This meant that there were still a large number of products needing re-judgement by quality inspectors after the preliminary judgement by AOI, which generated very high manpower costs.


The broad definition of product quality was a key cause of the low accuracy of preliminary judgement. Nowadays, Siemens AI expert team defines two categories of product quality data – data with clear limits, like size and quantity of solder bead, and data without clear limits like welding quality, and for different data categories, image feature-specific extraction and clustering AI algorithms, and supervised deep learning method are used to train machines. As a result, the accuracy of AOI’s preliminary judgement is significantly improved, and the number of products needing manual judgement is reduced by 75%, considerably lowering manpower costs while ensuring zero escape rate of defective products. In addition, AI module as a non-invasive application will not change the built-in programs of AOI, so it is suitable for all AOI software.

A new breakthrough in sorting of industrial wastes

AI is also applied in the automated sorting of industrial wastes in SEWC. When the wastes reach the treatment workstation, AI can distinguish types of waste, such as hazardous, general, recoverable and carton, by acquiring and analyzing the image data of wastes, thus guiding mechanical arms to send wastes to corresponding treatment processes.


Previously, wastes sorting was based on rules – every time when the waste category, form or definition changed, manpower was required to judge the characteristics of a new category, set new rules and guide mechanical arms to adapt to the sorting changes. Now, thanks to the core technology of AutoML and the deployment of multi-modal fusion technology, the sorting accuracy is boosted from about 70% to 97%. In addition, the sorting process is more intelligent and flexible – when the waste category, form or definition change, machines can automatically learn, adapt to and optimize the sorting rules, thus avoiding retraining model and manual adjustments of parameters.

Uncover data mystery to reduce production costs

Previously at the HMI function inspection workstation, the product test time was longer than other workstations, which became the bottleneck of the entire production line. Now the Siemens expert team leverages such techniques and technologies as data analysis, statistics and AI to make test data transparent, thus facilitating parameter optimization, test process optimization and predictive maintenance. These optimization suggestions help shorten the inspection time of HMI products, delivering annual cost saving of about CNY 400,000.


The AI system can provide new insights of and suggestions on improvement of inspection speed by identifying data linkages that could not be spotted previously, for example, the relationship between the components of the testing machine and the inspection accuracy, thus helping boosting efficiency. The AI application is also observed in the PCBA part – the data at all levels of enterprise is leveraged to make closed-loop data analysis that creates higher value.

Brave challenges: small sample learning in the age of big data

The Siemens expert team has deployed the small sample learning technology in SEWC’s quality inspection activities to enhance the ability of machines to identify new defects. As a result, when a new scenario occurs, machines only need a very small sample base to adapt the algorithm model to the new scenario. This is significant to the electronics industry that features fast production changes and high flexibility. In the future, when products are observed with unknown and new quality defects, AI can help deliver accuracy of more than 90% even if there is an extremely small data volume, for example, only a picture available. When the product types of the production line are adjusted, only a small sample set is required to achieve high-accuracy quality inspection of the new products.

The first industrial IoT-based AI system

To further explore the industrial potential of innovative technologies, SEWC is building the first industrial IoT-based AI system. Built on Siemens MindSphere, the cloud-based, open IoT operating system, it can upload the production data to the cloud where machine learning model trainings are carried out, and the trained algorithm model is allocated to the edge side and used to implement AI-based real-time data analysis and reasoning. In the coming days, this AI system of high demonstration significance will deploy AI applications such as product quality inspection, automated sorting of industrial wastes, and others in the factory in a centralized way.


These solutions enabled by advanced AI technologies, are deployed in such applications as product quality inspection, sorting of materials (wastes), and optimization of production process parameters, successfully improving the production quality and efficiency of SEWC and providing a good reference for AI-based production technology innovations in Siemens and the global electronics manufacturing industry. In addition, the successful applications also drive implementation of industrial data analysis, edge computing and cloud technology, marking that Siemens goes even further in integrating AI and industrial manufacturing.


“SEWC as ‘one of the world’s most advanced factories’ and also a practitioner of Industry 4.0, has been working to make continuous breakthroughs and striving to set benchmarks in the innovation and digitalization fields and contribute more to China’s new infrastructure development.” Remarked Mr. Li Yongli, Managing Director of SEWC.

May, 2020