Automated Metal Cleanliness Analyzer (AMCA): Improving Digital Image Analysis of PoDFA Micrographs by Combining Deterministic Image Segmentation and Unsupervised Machine Learning

Quality control of aluminum is critical for a wide range of applications across different industries. The main method for assessing aluminum cleanliness is PoDFA. The manual nature of the method imposes limitations in speed and statistical robustness that made aluminum producers and suppliers call for alternative methods with higher degrees of standardization and automation in recent years. We previously demonstrated the Automated Metal Cleanliness Analyzer (AMCA) method as a feasible way of assessing metal cleanliness from PoDFA micrographs using digital deterministic image segmentation techniques. Here, we continue this work by combining the deterministic approach with unsupervised machine learning for decreasing false-positive detections and achieving a higher degree of automation. Our results show that this approach generates metal cleanliness data closer to PoDFA reference data than previous implementations on the one hand and decreases algorithm setup time for new types of micrographs (e.g., alloys) by automating parts of the algorithm.