Machine learning and deep learning based modeling of aging and mechanical behavior in AA 7075 alloys

This study investigates the artificial aging behavior of AA 7075 aluminum alloy and presents a new method based on machine learning and deep learning for non-destructive prediction of applied heat treatment conditions. In the study, aluminum AA 7075 alloy samples were solution treated at 470°C for 3 hours, then artificially aged at temperatures ranging from 100°C to 225°C with 15°C increments for 1, 2, 3, 6 and 9 hours each. The vibration responses of the treated samples were recorded using a piezoelectric accelerometer device, and frequency domain data were obtained to reveal the vibration response features associated with different heat treatment conditions. To classify artificial aging conditions, firstly, a Convolutional Neural Network (ESA) model was developed using vibration frequency images, and feature extraction was performed with the ESA image processing method, and machine learning models were created. With the ESA prediction model, a classification accuracy of 89% with a macro average of 0.80 was achieved, and the model showed strong performance in distinguishing between sensitive heat treatments. In parallel, performance analysis of hybrid models combining ESA-based feature extraction with classical machine learning classifiers was performed with a comparative method. Among the models created, the Random Forest prediction model showed the highest performance with an accuracy of 89%, leaving behind Decision Trees and K-Nearest Neighbor prediction model (83%) and Support Vector Machine (72%). The conducted mechanical analysis provides information about the tensile strength, hardness and wear properties of the material. In addition, prediction models of mechanical properties were created with different machine learning models, and comparative performance analyzes of the models were performed. Tensile tests revealed that the highest ultimate tensile stress was observed at 100°C for 4 hours with 590.9 MPa, while 130°C for 3 hours and 175°C for 3 hours gave tensile stress values of 548.1 MPa and 549.1 MPa, respectively. It was observed that prolonged aging times generally resulted in decreased strength due to excessive aging. Hardness measurements followed similar trends, with a peak value of 145 HB at 1 hour at 175°C and 190°C, decreasing to 109.67 HB at 225°C for 9 hours, reflecting microstructural coarsening. The lowest wear rate of the material was obtained at 100°C for longer times, while higher temperatures (e.g., 225°C) were found to lead to deteriorating wear resistance. Among the regression models trained and tested to predict hardness and wear rate values based on aging process parameters, the Random Forest prediction model consistently outperformed the others in both training and testing performance and achieved the lowest mean square error. This comprehensive study confirms that the vibration data of heat-treated aluminum 7075 alloy contains sufficient spectral information to provide accurate classification of heat treatment parameters. The proposed machine and deep learning models offer a potential new avenue for heat treatment prediction in industrial quality control and optimization processes.