Artificial Neural Networks for Predicting Springback of aa6061 Sheet Based on Environmental and Forming Data Obtained from Smart Sensors

This research mainly proposes the artificial neural network (ANN) model for predicting the springback of AA6061-T6 sheet using manufacturing process and environmental data, which are collected by temperature, vibration, and force smart sensors. Heat-treated high-strength aluminum alloys, such as AA6061-T6 and AA7075-T6, exhibit low formability and high springback at room temperature, and they are typically formed at elevated temperatures. It is highly necessary to control the elevated temperature of the aluminum sheet before the forming process, as its strength degrades due to the annealing effect at specific temperatures above 250°C. The influencing factors contributing to temperature reduction include ambient temperature, wind speed, and transportation time, among others. Additionally, forming conditions such as forming speed, holding force, and dwell time interact in a complex manner with the environmental conditions to determine the amount of springback in the final product.<br/>To obtain a large dataset for training the ANN model, we newly developed an empirical testing system to simulate the actual warm forming process. This feasibility testing system allows for the systematic application of various combinations of process parameters, such as heating temperatures, transportation time, external wind speed, ambient temperature, and forming conditions. It also enables the attachment of smart sensors to monitor the process status and interconnection. We suggest specific variables from the environmental sensing data as input parameters, along with the forming load-depth curves and vibration data. Various environmental data, including the target temperature, transportation time, wind speed, and duration of exposure to external wind, are used to analyze the characteristics of temperature reduction in the elevated temperature of AA6061-T6 before the forming process. The material properties, maximum forming force, duration time of the maximum force, slope of the load-displacement curve, and FFT analysis of vibration data are used to determine the main influencing factors on the amount of springback in terms of the springback angle of U-channel.<br/>To examine the effect of each input parameter, the performance of each input parameter was evaluated in terms of mean squared error. To confirm the prediction accuracy of the proposed ANN model, the obtained dataset was divided into two subsets: a training dataset and a test dataset, with 30% of the data utilized for the test dataset. Additionally, the proposed ANN model was compared with other conventional machine learning algorithms such as linear regression, decision trees, lasso regression, and ridge regression. Various error statistics, including mean absolute error (MAE), root mean squared error (RMSE), and R-square, were analyzed. The proposed ANN model exhibited remarkable performance in the prediction of the springback angle, which makes it possible to apply in feedback control in the manufacturing process to obtain quick dimensional accuracy when the manufacturing environment undergoes rapid changes.<br/><br/><b><u><u>ACKNOWLEDGEMENT</u></u></b><br/>This research was supported by Korea Evaluation Institute of Industrial Technology (KEIT) grant funded by Ministry of Trade, Industry and Energy (MOTIE, Korea) (No. 20016357, Demonstration platform development of smart sensing unit).