A MEMS IMU De-Noising Method Using Long Short Term Memory Recurrent Neural Networks (LSTM-RNN)

Area of Science:

• Robotics and Control Systems
• Signal Processing
• Artificial Intelligence

Background:

• Microelectromechanical Systems (MEMS) Inertial Measurement Units (IMUs) are crucial for navigation but suffer from inherent errors.
• Standalone Inertial Navigation Systems (INS) based on MEMS IMUs experience significant error divergence over time.
• Integration with Global Positioning Systems (GPS) enhances navigation but fails under signal-challenged conditions.

Purpose of the Study:

• To propose and evaluate an Artificial Intelligence (AI) method for de-noising MEMS IMU output signals.
• To improve the accuracy and reliability of MEMS IMU-based Inertial Navigation Systems (INS).
• To specifically filter MEMS gyroscope outputs using a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN).

Main Methods:

• Utilized a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), a variant of Recurrent Neural Network (RNN), to process gyroscope data as time series.
• Collected 2 minutes of raw gyroscope data from a MEMS IMU (MSI3200) at a 400 Hz sampling rate.
• Compared the performance of the LSTM-RNN de-noising method against raw signals and an Auto Regressive and Moving Average (ARMA) model.

Main Results:

• The LSTM-RNN method reduced gyroscope data standard deviation by up to 60.3% and attitude errors by up to 51.3%.
• Compared to the ARMA model, LSTM-RNN achieved greater reductions in gyroscope STD (up to 42.4%) and attitude errors (up to 52.0%).
• The proposed de-noising scheme effectively enhanced MEMS INS accuracy, with LSTM-RNN demonstrating superior performance.

Conclusions:

• The proposed LSTM-RNN de-noising method is effective in improving the accuracy of MEMS IMU-based INS.
• AI-driven signal filtering offers a promising approach to mitigate errors in navigation systems.
• The LSTM-RNN technique is preferable for enhancing MEMS INS performance, especially in challenging environments.