Identification of Performance-Relevant Optically Detected Defects by Correlative Data Analysis in Solar Cell Production

The quality control of a solar cell production relies on a large variety of sensors, cameras, and measurements. Typically, each individual measurement is analyzed by its own criteria. For example, an optical inspection yields the number of optically detectable artefacts, such as finger interruptions. On the other hand, performance parameters are available for each cell. Thus, linking these different data sources in a quantitative manner is a major objective in quality control. However, the impact of a single defect found in optical inspection on a single cell’s performance can be much smaller than other process-induced effects. Therefore, we evaluate three different statistical approaches to identify whether defects observed by optical inspection quantitatively relate to the performance of larger cell sets. This analysis results in a separation of relevant optically detected defects from defects without impact on the cell’s performance. From a data analysis point of view, this is challenging as the optical data set is rather sparse, e.g. only few cells show many optically detected defects, while the performance data set is described by a smooth distribution of values around some average value. The presented methods are rather universal and can thus be transferred to establish quantitative relations between any two characterization methods in a production line.