Industry: Manufacturing / Solar Manufacturing

Defect Detection Vision System

Manufacturing defects plaguing your production line? Having issues with quality control? Always wanted to automate detection for out of specification.The customer, a solar manufacturing company based in Singapore, worked with us to achieve those goals.

The customer’s fully automated wafer to panel production line suffered from significant reject rates at the last panel assembly stage. Before the solar panel was laminated, they had no way to detect micro cracks in the solar cells. Unable to apply quality control process, these micro cracks resulted in larger cracks forming when heat and pressure was applied in the final lamination stage. These solar panels were subsequently classified as lower grade and commanded a lower selling price.

The customer turned to Red Vector to provide a customized solution to solve their problem. Through consultations and technical deep dives, a rules based vision system was designed. Built on Microsoft Visual Basic, the software used captured images from any type of camera (even off the shelf Logitech camera) to identify micro cracks before lamination began. Common image file types such as JPEG and PNG can be processed and analyzed. Production line solar cells were benchmarked against good quality solar cells. This ensured that solar cells with micro-crack could be removed before the final step of lamination. Rework could thus be carried out, resulting in many more solar panels being classified as higher grade.

Red Vector recommended rules based analytics over machine learning and AI. Significant amount of time and good/bad solar cell images would have been needed to train the vision system. Moreover, all images had to be classified to properly train the machine learning model. The probabilistic nature of machine learning did not lend itself well to ensuring quality control.

Rules based analytics leveraged on deep technical expertise which the customer possessed. The vision system relied on specific threshold levels which were determined scientifically. Accuracy was determined by a fixed set of rules which was easily explainable to management and the production team. These qualities made rules based analytics the clear choice for quality control in the manufacturing line.

Productivity and cost savings were achieved on top of better quality control. Now every solar cell can undergo micro crack inspection and rework can be done. Automation of the quality control process also resulted in higher efficiency and productivity of the workers. Manual inspection, which was tedious and prone to error, was eliminated. The workers could also be freed up for other processes. As with any automation, the vision system could run 24/7 with no loss in accuracy or productivity.