Electrospray mode discrimination with current signal using deep convolutional neural network and class activation...

Area of Science:

• Materials Science
• Chemical Engineering
• Electrical Engineering

Background:

• Electrospray is vital for generating micro/nanoparticles in energy, display, sensor, and biomedical fields.
• Controlling electrospray modes is crucial for particle quality, but real-time discrimination remains challenging.
• Current methods for observing electrospray modes lack automation and detailed analysis.

Purpose of the Study:

• To develop a simple, automated method for discriminating electrospray modes.
• To utilize current signals and deep learning for real-time electrospray mode analysis.
• To enhance understanding and control of the electrospray process.

Main Methods:

• A deep convolutional neural network (CNN) was employed for mode classification.
• Current signals from the collector were analyzed to detect droplet deposition.
• Frequency data derived from current signals were used to train a 1D CNN model.
• Class Activation Map (CAM) was utilized to visualize and interpret the CNN's decision-making process.

Main Results:

• The automated method successfully classified electrospray into dripping, pulsating, and cone-jet modes.
• The 1D CNN model achieved a high accuracy of 96.30% in mode discrimination.
• CAM provided insights into the features the CNN used for classification, aiding interpretability.

Conclusions:

• The developed CNN-based method offers an effective and automated approach for electrospray mode discrimination.
• This technique can improve the quality control of micro/nanoparticles generated via electrospray.
• The interpretability offered by CAM enhances trust and understanding of the automated system.