Optel Chairman Wang Zhen: AI lights up the road to innovation in photovoltaic testing

Optai Equipment Manufacturing Workshop Photo provided by the company

Optech, located in Tiandi Software Park, Putuo District, Shanghai, is a Beijing Stock Exchange listed company focusing on the photovoltaic industry and with AI visual inspection system enabling solutions as its core. Wang Zhen, chairman of the company, recently said in an exclusive interview with a reporter from China Securities Journal that as the first intelligent photovoltaic testing company in the industry to integrate software and hardware, the company has accumulated a considerable amount of image data to train and develop better AI models. After the development of intelligent operation and maintenance of photovoltaic power stations in the second growth curve business is completed, the company will launch a new model of charging for services and share with customers the digital intelligence achievements of AI-enabled photovoltaic fields.

Introducing AI to solve industry pain points

In the context of the accelerated global energy transformation, the demand for mass production of photovoltaic products has increased, and photovoltaic testing equipment also needs to be capable of large-scale batch testing and processing. However, the manual inspection methods used in the traditional photovoltaic inspection industry have become a bottleneck restricting the production efficiency and product quality of photovoltaic products.

It is understood that since the efficiency, speed and accuracy of manual inspection depend on the subjective judgment of the inspector, the use of manual inspection in the production of photovoltaic products will lead to low overall production efficiency and yield rate of component manufacturers. At the same time, due to the need for a large number of inspection personnel, further pushing up the labor cost of component production.

As a veteran who has been deeply involved in the photovoltaic testing industry for more than ten years, Optai is well aware of the pain points of this industry. Wang Zhen told reporters that in 2017, the company decided to introduce AI machine vision technology into photovoltaic inspection. Through the visual defect detection system, it can achieve batch, stable and accurate inspection of photovoltaic products, reduce labor costs for component manufacturers, and improve production efficiency.

But the seemingly simple replacement logic is very difficult to operate. Wang Zhen said frankly: "A module with an area of ​​1m x 2m may have more than 100 defects. If AI is used to completely replace manual inspection, in case of missed inspection, the risk and liability of such loss will not be easy to define, so at the beginning, companies were not interested in AI. The attitude is more cautious.”

In order to gain the trust of customers, Optai established a dedicated team in 2019 to conduct on-site deployment of equipment at the customer's LONGi Green Energy Chuzhou base factory. "We verified the AI ​​model for each piece of equipment. It took more than a year to go from one piece of equipment to a production line, from a workshop to a base. Until October 2020, Chuzhou Longi's 10GW factory fully deployed OPPO Tai component appearance EL (electroluminescence)-AI detection product." Recalling this past event, Wang Zhen couldn't help but sigh with emotion.

At present, Optei’s overall AI solution has been widely used in large-scale and mature applications in global leading companies in photovoltaic module shipments such as Longi Green Energy, JA Solar Technology, JinkoSolar, and Canadian Solar. Wang Zhen said that the quality of the AI ​​model is only the first step and a basic condition. In addition, supporting system software development is also required and can withstand the long-term test of the production line. Only such AI products can be put into production and achieve the purpose of improving quality and saving manpower.

Massive data to create excellent AI models

Data, computing power and algorithms are the three core elements of AI. Among them, data is equivalent to the "feed" of the AI ​​algorithm. Supervised learning and semi-supervised learning in machine learning must be trained with labeled data. Only after a large amount of training and covering as many scenarios as possible can a good model be obtained. The same principle applies to photovoltaic testing.

As the first photovoltaic testing company in the industry to integrate software and hardware, Optai has expanded its market share by fully deploying hardware testing equipment in the customer market. All original component clients have reserved AI system interfaces. "Customers are willing to provide us with defect data to train models, and better models can also help customers continue to improve detection efficiency and recognition rates." Wang Zhen said.

Specifically, Optai defect image data is provided on-site by the customer. It uses the photovoltaic defect data cleaning and enhancement technology program independently written by the company's core technical personnel, and uses AI model training and reasoning to improve the efficiency of algorithm updates. Batch cleaning and modification of erroneous data contained in large-scale data sets. Use defect sample enhancement technology to extract more features for model training, thereby making the data set more accurate and defect samples richer. Therefore, under the same data set, the recognition accuracy is improved. At the same time, this technology can significantly improve defect labeling efficiency.

Optech’s listing prospectus shows that mainstream component manufacturers’ requirements for detection failure rates have increased rapidly from less than 3% in 2017, to less than 1% in 2019, to less than 0.1% in 2021. The misjudgment rate indicator has also gradually increased from less than 5% in 2017 to less than 2% in 2021. The downstream yield rate has gradually increased from 90% in 2017 to 99.90% in 2021.

With the support of massive data and image training, Optai can dynamically update the sample capacity and image data, continuously train and optimize the AI ​​model, and the "flywheel effect" becomes more and more obvious. Wang Zhen told reporters: "Although the current trend of software and hardware integration in the industry is gradually emerging, many equipment companies that used to do hardware are forming teams to do AI, and many companies that used to do AI software are seeking cooperation with hardware equipment companies, but we were the first to start Layout, the data accumulated in this process is Optel’s first-mover advantage.”

From the perspective of financial performance, in the past three years, the operating income of Optai visual defect inspection system has grown rapidly, from 5.1177 million yuan in 2020 to 32.9913 million yuan in 2022, and the revenue proportion increased from 5.43% to 24.81%. The gross profit margin has exceeded 80% in the past two years.

Draw the second growth curve

With the rapid penetration of AI, big data, cloud computing and other technologies in the photovoltaic field, the intelligent development blueprint for photovoltaic power station operation and maintenance is slowly unfolding. Optel has chosen to use this as the company’s second growth curve and launch a SaaS platform-based Cloud artificial intelligence data analysis services include data collection and analysis related to drone infrared, EL and other power station operation and maintenance.

It is understood that defects such as module appearance shading and EL cracks will cause hot spots to form on the modules, which will lead to serious fire accidents. In addition, it will also cause the power of the modules to drop, affecting the income level of the power station. Therefore, the "physical examination" of photovoltaic power stations is very important, especially for water surface power stations, mountain power stations, and BIPV/rooftop power stations. Due to terrain restrictions, manual inspection is difficult and only drones can be used.

However, currently drones can only perform infrared and appearance inspections. The EL part with the highest component failure rate can only be photographed piece by piece with a portable EL, and the defective photos can be screened one by one manually, which is inefficient. Extremely low. Wang Zhen said that the Optai drone’s automatic cruise, precise autofocus, and high-definition EL imaging can accurately map the two-dimensional images captured to the actual three-dimensional terrain and landforms, enabling a comprehensive physical examination of EL, infrared, and appearance, greatly improving the efficiency of inspections.

The Optai power station cloud platform is based on SaaS architecture. Through cloud computing and cloud deployment, the power station operation and maintenance party can upload pictures anytime and anywhere for AI cloud deduction, without being restricted by the site, and the efficiency of defect screening is greatly improved; after the picture analysis is completed, Reports can be issued with one click to view the overview of the entire area, details of local areas, and statistical analysis of defects.

"At present, AI-powered photovoltaic inspection has only completed the preliminary work, which is to find defects and then classify them. As the scale of data continues to accumulate, it is particularly important to analyze these data and even output inspection analysis reports." Wang Zhen told reporters , "For example, AI can find out the cause of the problem by analyzing defects or trace it back to a certain production link/equipment, allowing companies to carry out a series of production and operation optimization and adjustments based on this."

Wang Zhen said that in the future, we plan to develop an AI decision support system based on existing data and share the digital intelligence achievements of AI-enabled photovoltaics with downstream customers.

The above is the detailed content of Optel Chairman Wang Zhen: AI lights up the road to innovation in photovoltaic testing. For more information, please follow other related articles on the PHP Chinese website!