Visual inspection is used to assess a product’s quantitative characteristics (physical inspection) and/or to assess a product’s qualitative characteristics (sensory inspection). Due to the complexity of the product, inspection tasks are often performed by humans and are therefore prone to errors. It is particularly the case when controllers have to detect aesthetic anomalies, to evaluate them and decide if a product must be rejected or not. The paper details how to improve visual inspection.
This paper details how the performance of visual inspection can be measured. It then lists the actions which can be carried out to improve the detection and the evaluation of aesthetic anomalies. Finally, it describes how can be made the knowledge about visual inspection more explicit in order to be shared by controllers. The methods we propose are illustrated with a concrete example detailed throughout the paper.
The gage R2E2 we developed can be used to decide which corrective actions to carry out. The four generic descriptors and the list of their attributes we list are usable by a controller to both describe and characterize any aesthetic anomaly on the surface of any product. The paper details then how evaluate an anomaly with a grid or with a neural network when the link between attributes values and the overall intensity of the anomaly is not linear. Finally, a method to formalize the expertise of controllers is described.
The proposed approach has been applied in companies which are part of an european research program (INTERREG IV). The practices we suggested have significantly reduced the variability of the visual inspection results observed up to now.
The paper shows how to improve inspection vision of products.
CitationDownload as .RIS
Emerald Group Publishing Limited
Copyright © 2016, Emerald Group Publishing Limited