Optimizing Industrial Etching Processes for PCB Manufacturing: Real-Time Temperature Control Using VGG-Based Transfer Learning

Accurate temperature control in Printed Circuit Board (PCB) manufacturing is essential for maintaining high-quality etching results. Automated monitoring us-ing machine vision and deep learning offers an effective approach for this task. This study investigated a feature-based transfer learning technique for classifying temperature readiness in infrared images of the etching process. The captured da-taset containing 470 ‘Production-Ready’ and 480 ‘Not-Ready’ infrared images of the etchant tank was utilized. Pre-trained Visual Geometry Group (VGG) Convo-lutional Neural Network (CNN) models, specifically VGG16 and VGG19, were employed to extract discriminative features from these images. Logistic Regres-sion (LR) classifiers were then trained on these features to classify the infrared images. The performance of the VGG16-LR and VGG19-LR pipelines was evaluated on training, validation, and test sets using a 60:20:20 split. While both pipelines achieved 100% accuracy on the training sets, the VGG19 pipeline showed exceptional performance, achieving a validation accuracy of 95%, and a test accuracy of 99%. The VGG16 pipeline also demonstrated robust perfor-mance, achieving 96% accuracy on both the validation and test sets. Considering the dimensions and the overall efficiency of the pipeline, it was determined that the VGG19-LR model was appropriate for the captured dataset. The high accura-cy indicates that transfer learning is suitable for categorizing temperature fluctua-tion in infrared thermography, as opposed to training a deep neural network from scratch. Computer vision and deep learning provide automated and precise tem-perature management during the etching process, leading to enhanced efficiency in PCB manufacturing.