Digitalisation of PV manufacturing
Rationale for support
Today’s modern PV factories produce several gigawatts of wafers, solar cells and modules every year, and the trend is rising sharply. This equates to about one billion solar cells produced annually per factory. During production, extremely large and high-dimensional data is generated: for example, from the produc- tion equipment and the inline measurement devices which monitor the process and classify the products. Digitization helps to collect and evaluate these large amounts of data. Production can be optimised in terms of efficiency, durabil- ity and manufacturing costs of cells and modules. Equipment manufacturers and suppliers can generate important customer benefits in the generation of machine data, the definition of interfaces, machine control and the digital op- timisation of analysis and (predictive) maintenance concepts. Heatmap and ex- perienced trend analysis can lead to manual and automated closed circle opti- misations.
Only partial quantities of the diverse data are centrally collected in databases and processed in a standardized manner using MES software systems (Manu- facturing Execution Systems) in most factories. In addition, the data is evaluated manually, for example by means of data correlations. First factories already use wafer tracking methods, which are based on virtual tracking or active tracking. Active wafer tracking may be realised either by encoding each ingot with a slice code leading to an edge code at each wafer or by encoding each wafer by a sur- face code. This coding is read out with a camera before each production step, which allows the production data of the respective solar cell to be assigned. In the future, this process will make it possible to accurately record and evaluate the entirety of production data for each wafer, which can be used to establish a correlation between the efficiency and longevity data. Solar cell companies are already researching software based on artificial intelli- gence (AI) with which production data is evaluated to sup- port the automatic analysis of large amounts of data and optimise production.
The fully automatic identification and quantification of the measurement data for data analysis, production control and process optimisation using modern AI is possible but not yet widely adopted. Experienced machine vision com- panies and R&D institutes have the expertise to develop AI methods that enable meaningful data compression and theory-based data analysis. This adds value to the existing procedures and especially enables companies producing in Europe to benefit from local support and protection of in- tellectual property (IP). Regarding the operation and main- tenance of production machines, IT-based remote mainte- nance systems already exist today, but these are still used by people and are carried out according to schedules or in the event of plant malfunctions. Predictive or predicted machine maintenance is not yet state-of-the-art.
Targets, Type of Activity and TRL
Research and development for digitalisation in photovolta- ics combine two megatrends and thus offer a great oppor- tunity for our climate and the PV industry. By 2030, Euro- pean Companies and research institutions will have seized the opportunity to improve cost efficiency in the manufac- turing of PV cell, module, inverter, and mounting system.
Digitisation applied to manufacturing will allow mainte- nance intervals or failure rates of modules to be better predicted. Systematically comparing PV power plant and component quality will enable high learning effects. The long-term vision is to evaluate and link the data from component production to the construction and oper- ation of PV power plants. By 2030, the first automated self-learning and self-optimising factories with very little downtime will exist.
Generated data will be stored centrally requiring standard data representations and interfaces. Workpiece tracking will link single-wafers or carriers to their particular produc- tion parameters. These will vary greatly as greater diversity in PV products is expected. Legal hurdles for gathering us- ing and sharing and exploiting data must also be overcome.
AI-supported software algorithms will scour data volumes for new connections and correlations to optimise produc- tion. The AI software will be self-learning to handle the large volumes.
AI kits and largely standardized AI application packages illustrating sample solutions for typical AI-based machine applications will be available.
In particular for new module and cell concepts - such as multi-junction technologies, combining thin-film com- pound semiconductors, perovskites with silicon, recycling strategies need to be developed and findings of these de- velopments need to be shared upstream to improve de- sign concepts.
Regarding digitization of PV manufacturing, the following research actions are required:
- Develop intelligent, self-sufficient multi-sensors for the acquisition of relevant data and suitable application of the generated data for AI-support- ed control and optimisation
Use digital twins (a digital twin is a “virtual rep- resentation of a real object or process”)
- Develop multi-scale and meta-models of manu- facturing processes, production and products as well as their components and their evaluation for optimisation of PV production through AI methods
- Develop digital twins of the entire production as the basis of a self-learning factory (vision) to ac- celerate optimisation cycles through automated data analysis
AI-based data analysis
- Develop self-learning AI-based software that au- tomatically analyses the large amounts of data during production, resulting in increased cell effi- ciency and reliability
- Improve human-computer interaction to support the adaptation of process parameters, e.g. auto- matic setup of measuring systems
- Consumable procurement triggered by the pro- duction plant
- Develop virtual and active identification processes and intelligent logistics components for material and device tracking across the value chai
- Develop fast and cost-efficient in-line measurement technology for real-time process control to widen the data base for machine learning in production.
Exchange and storage of data:
- Develop a range of common and standardised da- tabases, including cloud services, to ensure data ex- change across all segments.
- Develop plant interfaces for simplified and flexible connection of production machines to the existing data infrastructure of the factory and extension for bi-directional communication for real plant control by Advanced Process Control (APC) algorithms.
- Further development of object- and graph-based databases for production control and development of processes for automated context acquisition and assignment to expand the database (including un- structured data) and improve data quality (collec- tion of metadata) for AI-supported production op- timisation individualized production environments
- Further development of machine modelling, specif- ically of parts subject to wear, to implement predic- tive maintenance
- Realisation of a central simulation platform, which is multi-user-ready with proprietary shares to pro- tect core competencies for lowering barriers to data exchange
- Development of standards for simulation interfaces for significant acceleration of the adaptation of sim- ulation modules into overarching models
- Methods for improvement of the image of digital twins to allow accurate modelling based on the pre- cision of measurement data to improve the value of the digital twin due to increased imaging sharpness
AI-based data analysis
- Identification of relevant machine parameters that influence customer targets and development of suitable self-optimisation algorithms
- Develop AI-supported concepts for predictive main- tenance
- Use of existing system sensors for advanced pro- cess/maintenance monitoring to benefit from short-term potential for improved monitoring of solar cell production
- Development of best practice examples of digital twins to visualise its benefits
- Demonstration of transfer of digitization methods to industry. For this AI kits/application packages that present the application possibilities and ben- efits of various AI software solutions and illustrate them with sample solutions are required. New busi- ness models for the provision of equipment such as “pay per use” or “production as a service” and reduction of investment costs of new factories/pro- vision of production know-how by mechanical engi- neering companies need to be implemented.