Machine-Learning-Based Automatic Metallographic Grading System for High-Gloss Anodized Aluminum Profiles
Open Access
Abstract
The excellent “mirror” effect of medium and high-strength aluminum alloy profiles from the 6-series, achieved through anodizing, is highly valued by customers. Metallographic analysis is a key method for predicting the anodizing effect. However, traditional metallographic analysis methods suffer from unstable accuracy and low efficiency. To address these issues, this paper successfully develops a metallographic grading system by constructing a dataset and integrating computer vision with machine-learning techniques. Based on grain classification, the system automatically determines the metallographic grade by analyzing the proportion of good grain areas. After applying SMOTE sampling and 10-fold cross-validation to the machine-learning algorithm, we conducted a comparative analysis of the model’s performance from the perspectives of accuracy, good grain recall rate, bad grain recall rate, and AUC. The XGBoost model, selected as the final predictive model from 18 machine-learning models due to its superior performance, achieved a grain classification accuracy of 96.21% and a good grain recall rate of 98.07%. Both the accuracy and good grain recall standard deviations were less than 0.02. These results indicate that the model can effectively distinguish between good and bad grains with high robustness. Additionally, the average time for metallographic grading is less than 9 s. In comparison to the instability of traditional manual grading, this method significantly enhances both the accuracy and efficiency of metallographic analysis while also reducing grading costs.Keywords
This publication has 28 references indexed in Scilit:
- Classification of graphite particles in metallographic images of cast irons – Quantitative image analysis versus deep learningMaterials Characterization, 2024
- ECG signal classification via ensemble learning: addressing intra and inter-patient variationsInternational Journal of Information Technology, 2024
- Metallographic image segmentation using feature pyramid based recurrent residual U-NetComputational Materials Science, 2024
- Wire melted mark metallographic image recognition and classification based on semantic segmentationExpert Systems with Applications, 2023
- Landslide Susceptibility Prediction based on Decision Tree and Feature Selection MethodsJournal of the Indian Society of Remote Sensing, 2023
- A theoretical distribution analysis of synthetic minority oversampling technique (SMOTE) for imbalanced learningMachine Learning, 2023
- Machine Learning for Varietal Binary Classification of Soybean (Glycine max (L.) Merrill) Seeds Based on Shape and Size AttributesFood Analytical Methods, 2022
- A tutorial on the segmentation of metallographic images: Taxonomy, new MetalDAM dataset, deep learning-based ensemble model, experimental analysis and challengesInformation Fusion, 2022
- Effect of AlSiFe on the anodizing process of 6063 aluminumSurface and Interface Analysis, 2016
- Median Filtering: Statistical Properties2006