The use of three different inspection channels in Quality Scanner 3D already provides highly reliable classification results. Viprotron has many years of experience and proven algorithms for correct defect classification.
Nevertheless, there is always potential for improvement, especially in the case of similar defects. In this context, the use of classical methods will not bring any major steps towards a near hundred percent level. The use of AI tools for self-learning is therefore the next step.
First, the data requirements are determined. Thanks to 20 years of experience in image processing, Viprotron has high-quality and reliable data sets. In addition, the company continuously generate new data to close the last of the incorrectly classified defects with compliant data. In doing so, Viprotron uses a reliable filter to add comparable data.
The new Viprotron application software Rel. 9.x also allows customers to report unclear or incorrect classifications directly by sending the error image and the correct classification result. Viprotron qualifies this data and feeds it into the AI tool to develop an even more sophisticated training and evaluation model.
The AI tool uses all existing data, algorithms, new information and the results of the learning process to increase the efficiency and reliability of the control processes. This avoids incorrect scanner results and improves root cause analysis so that no line needs to be stopped.
This leads to higher productivity, better delivery quality and more reliable statistics. The more staff can rely on the scanner’s classification, the shorter the inspection time per glass. This increases productivity. Low-quality glass is removed from the process, resulting in better delivery quality.
Better classification also enables more meaningful statistical reports for the quality manager for root cause analysis. Overall, these improvements help customers simplify processes, save costs and strengthen their image through fewer complaints.