Machine learning-based process quality control of screen-printed titanium dioxide electrodes

Abstract

AI-based quality control has gained attention in the manufacturing industry due to its ability to improve speed and accuracy. AI can analyze a printed electrode and classify it as either good or bad quality within milliseconds, much faster than humans and conventional methods (random sampling and control charts). Herein, machine learning methods including Random Forest (RF), Support Vector Machine (SVM), and Feedforward Neural Network (FNN) are used to address a quality control problem involving the classification of screen-printed TiO₂ electrodes based on image data. Multivariate data analysis techniques such as factor analysis were employed to evaluate the effectiveness of the features extracted from these images. Characterization techniques like FTIR, 4-point probe, and microscopy were used to study the printed electrodes and provide accurate labeling. A dataset comprising ∼300 electrodes was created to train the AI models. The SVM model demonstrated the best performance, achieving 100 % accuracy and recall, followed by the FNN model with 99 % accuracy. Models were optimized and accelerated through feature engineering and extraction techniques, allowing them to be trained in under 1 min. This rapid training capability makes these models highly suitable for real-world quality control applications where hundreds of electrodes are produced per minute.