Heterogeneous Structure Omnidirectional Strain Sensor Arrays With Cognitively Learned Neural Networks

Area of Science:

• Materials Science
• Robotics
• Artificial Intelligence

Background:

• Stretchable strain sensors are crucial for bioinspired skin and AI tactile sensing.
• Current sensors struggle with accurate detection of both strain intensity and direction.

Purpose of the Study:

• To develop an omnidirectional strain perception platform for high-accuracy strain detection.
• To overcome limitations in simple device structures for comprehensive strain analysis.

Main Methods:

• Utilized a stretchable strain sensor array with a triangular-sensor-assembly (three sensors at 45°).
• Employed a machine learning (ML)-based neural network classification algorithm for data interpretation.
• Designed sensors with strain-insensitive electrodes and strain-sensitive channels for reliable sensing.

Main Results:

• The strain sensor demonstrated a gauge factor (GF) of ≈8, a sensing range of 0-35%, and reliability over 3000 cycles.
• The triangular-sensor-assembly coupled with the ML algorithm achieved ≈98% accuracy in recognizing arbitrary strain direction and intensity.
• The platform showed high accuracy across various surface stimuli environments within a 0-30% strain range.

Conclusions:

• The proposed omnidirectional strain perception platform effectively integrates stretchable sensors and ML for advanced tactile sensing.
• This system offers a significant advancement in accurately detecting complex strain patterns, paving the way for more sophisticated artificial skin and tactile sensors.