Machine Learning Enhanced Dynamic Response Modelling of Superelastic Shape Memory Alloy Wires.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Superelastic shape memory alloys (SMAs) offer excellent energy dissipation for structural vibration control.
• Accurate material models are crucial for designing SMA-based devices, but parameter identification is complex.
• SMA behavior is highly sensitive to strain rate, complicating thermodynamic modeling.

Purpose of the Study:

• To develop an efficient machine learning-based method for identifying thermodynamic parameters in superelastic SMAs.
• To address the challenges associated with experimental determination of these parameters.
• To facilitate accurate modeling of dynamic SMA behavior for engineering applications.

Main Methods:

• Developed a feedforward artificial neural network (ANN) architecture.
• Adapted a macroscopic constitutive SMA model incorporating strain rate effects to generate training data.
• Trained the ANN using simulated cyclic tensile stress-strain data.
• Validated the ANN's parameter identification capability through experimental tests on SMA wires.

Main Results:

• The proposed ANN approach successfully identified key thermodynamic parameters for superelastic SMA wires.
• The method demonstrated efficiency in parameter estimation compared to traditional experimental techniques.
• The ANN model accurately captured the strain rate-dependent behavior of SMAs.

Conclusions:

• Machine learning, specifically ANNs, provides an efficient and effective solution for identifying complex thermodynamic parameters in SMAs.
• This approach simplifies the design and analysis of SMA-based structural control systems.
• The study validates the use of AI in material modeling for advanced engineering applications.