Quality inspection applications in industry are required to move towards a zero-defect manufacturing scenario, with non-destructive inspection and traceability of 100% of produced parts. Developing robust fault detection and classification models from the start-up of the lines is challenging due to the difficulty in getting enough representative samples of the faulty patterns and the need to manually label them. This work presents a methodology to develop a robust inspection system, targeting these peculiarities, in the context of solar cell manufacturing. The methodology is divided into two phases: In the first phase, an anomaly detection model based on a Generative Adversarial Network (GAN) is employed. This model enables the detection and localization of anomalous patterns within the solar cells from the beginning, using only non-defective samples for training and without any manual labeling involved. In a second stage, as defective samples arise, the detected anomalies will be used as automatically generated annotations for the supervised training of a Fully Convolutional Network that is capable of detecting multiple types of faults. The experimental results using 1873 Electroluminescence (EL) images of monocrystalline cells show that (a) the anomaly detection scheme can be used to start detecting features with very little available data, (b) the anomaly detection may serve as automatic labeling in order to train a supervised model, and (c) segmentation and classification results of supervised models trained with automatic labels are comparable to the ones obtained from the models trained with manual labels.

Quality control is critical in the production process of solar cells. A small crack in the cell can affect its future performance in energy production. Nowadays, one of the most used techniques to detect these defects is Electroluminescence (EL), which allows obtaining high-resolution images where the defects are highlighted and where a non-invasive inspection can be done. In this way, those defective cells can be removed from production without reaching the final product. In a previous work, the use of a Convolutional Neural Network (CNN) network that was executed multiple times on a EL cell image to obtain a defect segmentation map was proposed. In this paper, we propose to use a Fully Convolutional Network (FCN) to obtain the same defect segmentation map but in a single step. The experiments with the FCN have been performed on the same dataset used in the previous work to compare the results. The comparison comprehends two aspects, the precision to detect the defects and the execution time necessary to process a cell image.

Quality control of solar cells is a very important part of the production process. A little crack or joint failure can cause bad performance of the cell in the future, partly because the defective areas can be electrically disconnected from the active zones. Nowadays, one of the techniques to carry out this control is electroluminescence (EL), which allows obtaining high-resolution images of the cells where a visual and non-invasive inspection of defects can be done. This inspection is mostly performed by trained human operators. However, as the eyes become tired after a working day and the subjectivity of the operators, the accuracy with which the defect detection is done may be compromised. In order to solve this problem, a method to assist the operator in the inspection of polycrystalline silicon solar cells surface from EL images based on Convolutional Neural Networks is proposed. The method would classify the cells as defective and non-defective, and suggest those cells that are defective for re-inspection. Also, it would propose a segmentation map of the defects in the cell. To compensate for the lack of image samples in the dataset, each cell image is divided into regions by a sliding window. Then, each region is classified as defective or non-defective. And finally, all classifications related to the cell are resembled obtaining a segmented image of defective areas in the cell.

<!-- *** Custom HTML *** --> The purpose of this paper is to explore the use of fully convolutional neural networks (FCN) to perform a semantic segmentation of deflectometric recordings for quality control of reflective surfaces. The proposed method relies on a U-net network to identify the location and boundaries of the object and the possible defective areas present on it by performing a pixel-wise classification based on local curvatures and data modulation. Experiments were performed on a real industrial problem using four variations of the architecture. The results demonstrate that the method combining geometric and photometric information enables the identification of a wider variety of shape and texture imperfections, with the resulting segmentations closely correlated with the visual impact of the defects. In addition, several suggestions are presented for near-term industrial utilization.