Computationally intelligent calibration framework for durable soft strain sensors.

Area of Science:

• Materials Science
• Computer Science
• Robotics

Background:

• Soft strain sensors are crucial for ubiquitous electronics, but their non-degradable nature and signal inconsistencies (nonlinearity, hysteresis, attenuation, batch variation) necessitate durability innovations to minimize waste.
• Current approaches often rely on empirical material or structural engineering, which can be time-consuming and inefficient for addressing sensor performance limitations.

Purpose of the Study:

• To develop an efficient computational calibration framework for soft strain sensors that addresses signal predicaments and enhances device durability.
• To utilize hierarchical physics-guided machine learning (ML) models for real-time sensor signal calibration without empirical trial-and-error.

Main Methods:

• Developed a computational framework employing hierarchical physics-guided machine learning (ML) models.
• Applied the framework to an eco-friendly carbon waste-based strain sensor as a case study.
• Trained ML models for real-time sensing signal calibration to correct nonlinearity, hysteresis, cycling attenuation, and batch inconsistency.

Main Results:

• The physics-guided ML models demonstrated high computational efficiency and learning accuracy in calibrating the strain sensor's signals.
• The calibration framework automatically corrected the sensor's performance, achieving linearity, non-hysteresis, long-term stability, and batch consistency.
• ML-driven calibration extended the sensor's reliable working lifetime by over 3000 times compared to its uncalibrated counterpart.

Conclusions:

• The proposed physics-guided ML framework offers an efficient and effective solution for overcoming signal predicaments in soft strain sensors.
• This computational approach significantly enhances sensor durability and performance, reducing waste and facilitating long-term practical applications in robotics and electronics.
• The study highlights the potential of physics-guided ML in advancing sustainable and reliable sensor technologies.