A Temperature Compensation Approach for Micro-Electro-Mechanical Systems Accelerometer Based on Gated Recurrent...

Area of Science:

• Sensor Technology
• Signal Processing
• Machine Learning

Background:

• Micro-Electro-Mechanical Systems (MEMS) accelerometers suffer from accuracy degradation due to temperature fluctuations and inherent noise.
• Existing methods often struggle to effectively address both noise and temperature drift simultaneously, limiting accelerometer precision in dynamic environments.

Purpose of the Study:

• To propose and validate a parallel denoising and temperature compensation fusion algorithm for MEMS accelerometers.
• To significantly improve the accuracy and reliability of MEMS accelerometer data by mitigating noise and temperature-induced errors.

Main Methods:

• Signal decomposition using Robust Local Mean Decomposition (RLMD) into product function (PF) signals and a residual.
• Signal classification via Sample Entropy (SE) to identify noise, mixed, and temperature drift segments.
• Denoising of noise and mixed segments using Time-Frequency Peak Filtering (TFPF) with adaptive window lengths.
• Temperature compensation using a novel GRU-MLP-attention model (GMAN), incorporating accelerometer output time series as input.

Main Results:

• The RLMD-SE-TFPF algorithm effectively denoises the accelerometer signal by classifying and processing different signal segments.
• The GMAN model successfully compensates for temperature drift, leveraging the accelerometer's output time series for enhanced accuracy.
• The integrated fusion algorithm reduced acceleration random walk by 96.11%, from 0.23032 g/h/Hz to 0.00895695 g/h/Hz.

Conclusions:

• The proposed parallel denoising and temperature compensation fusion algorithm offers a robust solution for enhancing MEMS accelerometer accuracy.
• The combination of RLMD-SE-TFPF for denoising and GMAN for temperature compensation demonstrates superior performance compared to original signals.