Resumen de Decision Support System Based on Deep Learning for Improving the Quality Control Task of Rifles: A Case Study in Industry 4.0
Luca Romeo, Riccardo Rosati, Emanuele Frontoni
The quality control (QC) procedure was usually constrained by the high demands in time and resources, as well as a limitation in performance mainly due to the intra- and inter-operator variability and the risk of reproducing unwanted bias. The quality control (QC) procedure was usually constrained by the high demands in time and resources, as well as a limitation in performance mainly due to the intra- and inter-operator variability and the risk of reproducing unwanted bias. Accordingly, the increasing amount of collected data open the realm of possibilities for designing and implementing a Decision Support System (DSS) empowered by Deep Learning algorithms for solving the QC task by overcoming these challenges. The work proposes a Deep Learning model for predicting the aesthetic quality classification (QC task) of rifles based on the analysis of wood grains. The task and the collected dataset originated from a collaboration with an industrial company. The higher performance (up to 0.86 of F1 score) by the proposed VGG-16 based model and the validation with respect to human annotator suggest how the proposed approach represents a solution to automate the whole QC procedure in a challenging industrial case scenario. The proposed solution allows to (i) speed up the QC procedure, (ii) minimize intra/inter-operator variability, and (iii) detect and mitigate unwanted bias by forcing the network to learn the features that correctly describe shot quality, rather than other confounding geometric features.
