Machine learning-based self-sensing of the stiffness of shape memory coil actuator

Area of Science:

• Materials Science
• Robotics Engineering
• Machine Learning

Background:

• Self-sensing actuation (SSA) enables monitoring of shape memory coils (SMCs) for smart equipment and robotics.
• SMC stiffness is crucial for intelligent robotics in defense, but lacks accurate analytical models.
• Electrical resistance changes in SMCs during martensitic phase transformation can indicate mechanical properties.

Purpose of the Study:

• To propose a machine learning-based soft model for accurate autosensing of SMC stiffness during actuation.
• To develop an automated method for predicting SMC stiffness, overcoming the difficulty of experimental data collection.
• To compare the performance of different machine learning models for SMC stiffness prediction.

Main Methods:

• Developed an experimental facility to collect data on SMCs under varying Joule heating currents.
• Proposed soft computing-based methods, specifically Classical Polynomial and Bayesian optimization-based Feedforward Neural Network (FFNN) models.
• Utilized electrical resistance changes to infer mechanical properties like stiffness.

Main Results:

• A hybrid FFNN model achieved 95.2650% accuracy in predicting SMC stiffness.
• The developed FFNN model demonstrated strong correlation with experimentally recorded stiffness values.
• The proposed automated method simplifies the prediction of stiffness compared to traditional experimental determination.

Conclusions:

• Machine learning, particularly hybrid FFNN models, offers a viable solution for accurate SMC stiffness autosensing.
• The developed model can significantly improve the monitoring and control capabilities of intelligent robotics.
• This approach facilitates the development of advanced defense systems leveraging smart materials.