Towards Reliable 3D Printing: Hybrid CNN-KNN Model for Automated Defect Identification

This paper aims to design a model for classifying defects during the process of extrusion in manufacturing. Here, the work is limited to only four major classes: Deformed Edges, Under-Extrusion, Weak Fiber Bonding, and Base Bulging. Accuracy, recall, and F1-score are the metrics used for testing the model's performance for the four classes. The model obtained impressive results for all four classes. The dataset used for this study had 10,080 images, and the overall accuracy of the model was 98.21%. Precision was between 97.94% for Base Bulging and 98.42% for Weak Fiber Bonding, and the recall ranged between 98.02% for Deformed Edges and 98.46% for Under-Extrusion. F1 scores were equally impressive, with scores between 98.14% for Base Bulging and 98.27% for Under-Extrusion. It was also found that the model achieved a 99% accuracy across all categories of defects in support and proportion values. The support values for each class varied between 2420 for Base Bulging to 2600 for Under-Extrusion, and proportional support values varied between 0.24 and 0.26, thus indicating a balanced dataset. Further, True Positive, False Positive, False Negative and True Negative counts were used for every class to determine the results of the model classification, and values were indicative of reliable predictions. Overall, the results demonstrate the model’s effectiveness in accurately detecting extrusion defects, making it a valuable tool for quality control in manufacturing.