Data-Driven Strain Sensor Design Based on a Knowledge Graph Framework.

Area of Science:

• Materials Science
• Sensor Technology
• Artificial Intelligence

Background:

• Developing wearable flexible strain sensors demands application-specific performance, but experimental approaches are slow and inefficient.
• Existing methods often lead to suboptimal sensor designs due to knowledge redundancy and limited exploration of design space.

Purpose of the Study:

• To develop a novel framework for intelligent sensor design by integrating knowledge graphs and graph representation learning.
• To overcome the limitations of traditional experimental methods and process-parameter-based machine learning in sensor development.
• To discover new sensor designs with targeted performance characteristics.

Main Methods:

• A framework combining knowledge graphs and graph representational learning was proposed to analyze sensor knowledge.
• Semantic features derived from relationships within the knowledge graph were used, improving prediction precision.
• The framework was validated by designing and testing a novel flexible strain sensor.

Main Results:

• The proposed framework achieved a prediction precision of up to 0.81, outperforming traditional methods.
• A newly designed strain sensor demonstrated a wide working range of 300% strain.
• The tested sensor's performance closely matched predictions and exceeded that of similar materials.

Conclusions:

• The developed framework facilitates intelligent sensor design by managing and utilizing sensor knowledge effectively.
• This approach reduces knowledge redundancy and enables the discovery of high-performance sensors.
• The study paves the way for text-mining-based knowledge management in sensor systems, accelerating innovation.