The long read: Finding PV faults faster


From pv magazine 12/2021

Achieving high solar cell and module efficiencies is one thing, but maintaining them in the thousands of PV modules that roll off production lines each year is quite another. Manufacturers are realizing that reaching the levels of consistency their customers demand requires careful monitoring of processes, materials and recipes from start to finish, and are increasingly turning to artificial intelligence to assist with this task.

“Around 2015, manufacturers were all about cutting costs, and in terms of inspection only wanted to do what is really necessary – the flashing and sorting at the end of the manufacturing process,” explained Tom Thieme, business unit manager for solar at inspection equipment supplier Isra Vision. “For some time now, we are seeing a trend for process inspections moving backward along the chain all the way to inspection of incoming wafers or cells.”

Thieme sees two reasons for this. First, manufacturers want to ensure consistent performance on their production lines, and see a higher degree of automation as the best way to achieve this. Second, they are realizing that investment in additional process control and inspection pays off, as customers seek high manufacturing efficiency at the lowest cost of ownership, and cells sorted and binned at the lower end of the efficiency range are increasingly hard to sell.

Along the whole production line, equipment suppliers are increasingly adding software capabilities to their tools that allow for real-time monitoring of processes, and to notify operators when something could be changed or optimized. “We have a continuous and repetitive system for the automatic collection of machine data,” said Michele Caddeo, marketing representative at Ecoprogetti, an Italy-based company that builds out entire module production lines for clients. “With data such as machine status, alarms, counter pieces used and recipes in use we can offer our customers process control and an evaluation of the various sensitive points of the production line, so that they can establish changes to improve quality of the product and the production flow.”

Big data

All of these additional inspections and monitoring within individual machines and processes means PV manufacturers have mountains of data to sift through covering their whole operation, and this presents both a challenge and an opportunity.

German flashing equipment supplier h.a.l.m. sees manual or semi-automated processes to analyze manufacturing data as being laborious for staff and ineffective in producing timely interventions and optimizations on the line. “Manual methods are hard work and time consuming. By the time an error or an opportunity for optimization is spotted, it could already be several thousand cells too late,” said h.a.l.m managing director Michael Meixner. “We got into close discussions with some of our customers to understand how they treat their data: the evaluation methods, the key figures they look at. And we decided to offer a solution which automates this data collection and analysis.”

And as the amount of data and the number of points to track grows, the need for automated solutions to manage it and pull out the points that can save time and money for manufacturers becomes even greater. Given that their tools have long managed and analyzed data produced in the inspection and sorting at the end of production, suppliers of flashing and testing equipment are well placed to develop more comprehensive solutions. Both h.a.l.m. and Isra Vision have this year introduced new software solutions to analyze and manage production line data, and see significant demand for such solutions coming from all regions. “Big data is something that producing companies recognize they will need to understand and make use of more and more in the future,” says Meixner. “And particularly with the growing size of production lines and industry expansions, you cannot monitor or optimize lines like these without a close sight on your data.”

Isra Vision’s Thieme also sees the growing trend for PV manufacturing to make better use of big data, mirroring past developments in the semiconductor industry. “The semiconductor industry has been using these kinds of big data analysis for a long time. And PV is now catching up,” he said. “And in a way the development is similar – first there was a lot of subsidized manufacturing, and then costs fell rapidly and manufacturers were more interested in volume. Now we are again seeing a rise in the value of data, which must come with industry standards and high reliability.”

Beyond MES

PV manufacturers have long made use of centralized “Manufacturing Execution Systems” (MES) to operate and monitor their production lines. However, the latest generation of software solutions on offer goes far beyond the capabilities of these, in terms of the depth of data available and the speed at which analysis can be conducted and optimizations made to processes or recipes.

Such solutions could contribute to the much-discussed comeback of PV manufacturing in Europe, as well as the United States, India, and other regions, by ensuring a smooth ramp-up of production and quick optimization of processes to be able to compete with those of more established rivals already operating large manufacturing bases.

In Germany, this was recognized with the establishment of the Selfab research project, which received around €2 million in funding from the ministry of economics and labor in the state of Baden-Württemberg. The project is a collaboration between five leading research institutes in the region, and focuses on development of artificial intelligence solutions for PV manufacturing.

“Artificial intelligence is a key technology for the future and can also play a decisive role in photovoltaic production,” said Baden-Württemberg Economics Minister Nicole Hoffmeister-Kraut, announcing the project in December 2020. “The self-learning factory significantly increases efficiency and productivity and enables faster implementation of new technologies. Our plant manufacturers in the country can gain decisive competitive advantages.”

The project created a “digital twin” of a PV production line. The generic model takes in all of the processes at work in a factory, allowing plants to trial and improve different process optimizations, and model their potential effect on cell/module efficiency and other parameters. Researchers working on the project have said it could allow for processes to be optimized before having to try them out with actual materials, potentially speeding up the transfer of new advances and even entire new technologies into industrial production.

Using big data in this way, to manage risks in the ramp-up phase and ensure the fastest route to optimized production, means that analytics software backing will play an important role in establishing PV manufacturing in new regions. And equipment suppliers already note a slight regional nuances in how manufacturers are making use of the insights from data.

“Established manufacturers in Asia are tending toward higher degrees of automation to eliminate human influences on process performance. But they know how to ramp up very quickly, they have standard operating procedures for this based on automated inspection data” said Thieme. “Customers in other regions are using big process data more to catch up quickly and safely to gigawatt-scale production at high quality and efficiency levels.”

Growing variation

As PV technology continues to evolve, manufacturers are preparing for increased variation in the products they’ll need to offer – whether that means multiple cell technologies or modules in different shapes and sizes for particular applications. And this requires the regular switching of processes, materials, recipes and other parameters, with a seamless ramp up.

Without a centralized system to manage these changes, switching lines to a new product could mean an operator has to upload new settings manually to each individual tool – both time intensive and leaving room for errors or inconsistency in the settings. And while the switch might still need components and materials going through the line to be physically changed by operators, all of the process parameters can be switched instantly to settings already optimized for a specific product, reducing the need for a more cautious ramp up or optimization cycle to weed out any issues.

For new and emerging cell technologies as well, many of the processes being used to reach higher efficiencies are becoming more complex, with more potential pitfalls. With artificial intelligence (AI) and machine learning to track these, and more data available to dive deep into process details, new processes and research and development cycles can also be shortened.

While AI tools can save a lot of time in manufacturing, it can ultimately only be as smart as the training and tools it is based on. And this is where suppliers of inspection or flashing equipment are able to leverage their experience in analyzing images and spotting quality issues. This information is still at the core of the solution, and provides the high level data to guide operators to the best places to do a deeper dive.

“We have a lot of systems installed, we have a high degree of technology and process experience, which is incorporated in our system software,” said Thieme. “Knowing how to process an image and extract the relevant data for comparison is valued by our customers, and something they can use to achieve their individual objectives.”

This content is protected by copyright and may not be reused. If you want to cooperate with us and would like to reuse some of our content, please contact: