Random forest assisted vector displacement sensor based on a multicore fiber.

Area of Science:

• Optoelectronics
• Fiber Optics Sensing
• Machine Learning Applications

Background:

• Accurate measurement of two-dimensional vector displacement is crucial in various engineering applications.
• Traditional sensors often face limitations in simultaneously determining displacement direction and amplitude accurately.
• Optical fiber sensors offer advantages in remote and harsh environment sensing.

Purpose of the Study:

• To develop and evaluate a two-dimensional vector displacement sensor capable of simultaneous direction and amplitude measurement.
• To investigate the performance enhancement of the sensor using a random forest machine learning algorithm.
• To compare the sensor's performance with a theoretical model versus a machine learning approach.

Main Methods:

• Design and fabrication of a sensor based on a seven-core multi-core fiber with inscribed Bragg gratings.
• Acquisition of displacement data under random circumstances.
• Development and application of a theoretical model for data analysis.
• Implementation of a random forest algorithm for enhanced data processing and performance evaluation.

Main Results:

• The theoretical model demonstrated good performance within a limited linear range (0-9 mm).
• The random forest-assisted sensor achieved a significantly wider measurement range (0-45 mm).
• Mean absolute errors for direction and amplitude reconstruction were reduced by 60% and 98%, respectively, with the random forest model.

Conclusions:

• The proposed multi-core fiber sensor effectively measures two-dimensional vector displacement.
• Random forest machine learning significantly improves the sensor's measurement range and accuracy.
• This work highlights the potential of machine learning in point-based optical multi-parameter sensing systems.