Application of Constitutive Models and Machine Learning Models to Predict the Elevated Temperature Flow Behavior of...

Area of Science:

• Materials Science
• Metallurgy
• Mechanical Engineering

Background:

• Understanding the hot deformation behavior of titanium-aluminum (TiAl) alloys is crucial for optimizing manufacturing processes.
• Accurate prediction of flow stress is essential for simulating and controlling plastic deformation during high-temperature forming.
• Existing constitutive and data-driven models have limitations in predicting flow stress under diverse conditions.

Purpose of the Study:

• To investigate the hot deformation characteristics of a Ti46Al2Cr2Nb alloy.
• To evaluate and compare the predictive accuracy and generalization capabilities of various constitutive and data-driven models for flow stress.
• To develop an improved predictive model with enhanced generalization for TiAl alloys.

Main Methods:

• Hot compression tests were conducted on Ti46Al2Cr2Nb alloy across a temperature range of 910-1060 °C and strain rates from 0.001 to 0.1 s^-1.
• Three constitutive models (including SCS) and three data-driven models (including ANN) were employed to predict experimental flow stress data.
• Model performance was assessed based on accuracy (coefficient of determination, R²) and generalization capability under varied and new deformation conditions.

Main Results:

• The activation energy for hot deformation of the TiAl alloy was determined to be 319 kJ/mol.
• The Artificial Neural Network (ANN) model demonstrated high accuracy (R² > 0.98) for interpolated and extrapolated strains under known conditions.
• The Sellars (SCS) model showed lower accuracy (R² < 0.5) at extrapolated strains, and both ANN and SCS models performed poorly under new deformation conditions.
• A hybrid model integrating SCS and ANN predictions exhibited superior generalization capabilities across different deformation regimes.

Conclusions:

• The ANN model excels at predicting flow stress within known deformation parameters, while the SCS model is less effective for extrapolation.
• Neither ANN nor SCS models alone possess sufficient generalization for predicting flow stress under entirely new processing conditions.
• A hybrid SCS-ANN model offers a promising approach for robust and accurate flow stress prediction in TiAl alloys across a wider range of conditions.