AI-Driven Electrical Fast Transient Suppression for Enhanced Electromagnetic Interference Immunity in Inductive Smart...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Sensor Technology

Background:

• Inductive proximity sensors require electromagnetic interference (EMI) immunity for industrial applications.
• Conventional filters for EMI mitigation limit signal bandwidth (100 Hz–1.6 kHz).

Purpose of the Study:

• To investigate the efficacy of neural networks (NNs) in automatically filtering EMI from sensor signals.
• To compare different NN architectures for EMI denoising.
• To develop an optimized NN model for potential ASIC implementation.

Main Methods:

• Analysis and comparison of 1D convolutional NN, recurrent NN, and hybrid NN models.
• Development of a Gated Recurrent Unit (GRU) based NN model.
• Optimization and compression of the GRU network for memory efficiency.

Main Results:

• The proposed GRU-based NN model successfully denoises EMI-perturbed signals.
• The final optimized network reduces noise by 70% (MSEred).
• The model occupies only 2 KB of memory.

Conclusions:

• NNs, particularly GRU models, offer a powerful solution for EMI filtering in inductive sensors.
• The optimized GRU network meets memory constraints for potential integration into application-specific integrated circuits (ASICs).
• This approach overcomes bandwidth limitations of traditional filtering methods.