The increasing complexity of surface-mount technology (SMT) processes for printed circuit board assemblies (PCBAs) has intensified the demand for advanced inspection and data analysis solutions capable of enhancing quality levels, operational efficiency, and industrial predictability. In this context, this study presents the experimental development and functional validation of an artificial intelligence (AI)-based solution applied to an automated optical inspection (AOI) production line, with a focus on predicting recurrent failures in the SMT assembly process, such as component misalignment and short circuits.

The proposed solution is based on the use of real production data collected from a Yamaha YSi-V AOI system, which undergo preprocessing, intelligent data treatment, and classification stages to ensure consistency and reliability for AI model training. Multi-Layer Perceptron (MLP) neural networks are developed and trained to identify operational patterns associated with the most critical assembly defects, enabling both the automatic classification of detected failures and the estimation of failure occurrence probabilities in future production batches.

The project also includes the implementation of an automated prototype for segregating approved (OK) and rejected (NG) products, consisting of a six-axis robotic manipulator, an adjustable gripper, and a PCBA storage system, thereby integrating the physical layer with the digital intelligence of the system. The results are presented through a graphical visualization interface, providing key performance indicators, trend analysis, and predictive insights to support engineering decision-making.

Experimental validation in an industrial environment demonstrates that the proposed approach enables the evolution of the AOI line to Stage 5 (Predictive Capacity) of the ACATECH Industrie 4.0 maturity model, consolidating a robust technological foundation for data-driven digital manufacturing.