SENSORS PRODUCTION: PROCESS STATISTICAL ANALYSIS AND ARTIFICIAL INTELLIGENCE FOR QUALITY CONTROL

Screen printing is an economical, practical, and sustainable alternative for sensor production. Screen-printed carbon electrodes (SPCEs) are a fundamental component of biosensors, which act as diagnostic devices, and therefore need to meet high quality and performance standards. This work aimed to study a company production process of SPCEs through statistical analyses of the process’ failures and the final product's thickness. In addition, it aimed to apply Artificial Intelligence for visual inspection based on machine learning, to automate and accelerate the decision to approve or discard the product. The company provided production data totaling five production batches from March to July (2024). Generalized linear mixed models with a binomial family were used to analyze yield and manufacturing errors, performed in RStudio software (a=0.05). Success proportions were evaluated through their logit values, variance, and confidence interval obtained from the model and logistic distribution. Control and Pareto charts were used to analyze product thickness and manufacturing errors, respectively. Additionally, a computer vision model based on a pre-training neural network was developed in Python for automating the product inspection process using sensor’s images to identify manufacturing errors. It was observed that the screen-printing process had a yield of 41.5%, with the main errors being rejections due to unprinted tracks, tracks wider than standard and paste accumulation during printing. The developed computational model presents itself as a promising tool with 83% efficiency in precision, recall, and F1-Score metrics when tested in the production process.