Detecting vibrations in digital holographic multiwavelength measurements using deep learning.

Area of Science:

• Optical Metrology
• Machine Learning
• Industrial Sensing

Background:

• Digital holographic sensor systems are crucial for quality control in production lines.
• Complex vibrations in multi-axis systems (robots, machine tools) degrade measurement accuracy.
• Early detection of vibrations is essential for maintaining sensor performance.

Purpose of the Study:

• To develop a deep learning approach for detecting vibrations in digital holographic sensor systems.
• To improve the reliability and accuracy of holographic measurements in industrial environments.
• To enable real-time vibration monitoring during hologram reconstruction.

Main Methods:

• A deep neural network was trained to predict the standard deviation of the hologram phase.
• The model was evaluated using training data and data simulating production environments.
• Performance was compared against classical machine learning algorithms.

Main Results:

• The deep learning model achieved 96.0% accuracy on training-like data.
• An accuracy of 97.3% was achieved on data simulating a production environment.
• The model demonstrated performance comparable to or exceeding classical machine learning methods.
• Individual predictions were processed in 35 microseconds on a GPU.

Conclusions:

• Deep learning offers a robust solution for vibration detection in digital holographic sensors.
• The proposed method enhances measurement quality and reliability in industrial production.
• Fast GPU-based predictions enable real-time application in dynamic manufacturing settings.