Enhancing Micro Electro Mechanical System Gyroscope Data Precision Through Autoencoder-Based Denoising

Micro Electro Mechanical System gyroscopes, used predominantly in the aerospace industry, utilize a vibrating structure to measure the rate of rotation. The performance of these gyroscopes is affected by random drift noises. Various methods have been employed to denoise the gyroscope output. This paper proposes a robust Deep Learning model, leveraging auto encoders implemented in both Linear and Convolutional Neural Network architectures, to denoise the gyroscope data. In a post-processing analysis using the sensor's original data, the DL based denoising model shows a significant improvement in short term stability, demonstrating its effectiveness in enhancing the overall robustness and accuracy of the gyroscope measurements. Metrics such as standard deviation and Peak Signal to Noise Ratio are employed to measure the effectiveness of the denoising models in enhancing the quality of gyroscope data. The robustness of the proposed architecture is also assessed in terms of its computational complexity and efficiency. The performance of the model is compared with conventional methods like the Kalman Filter, Principal Component Analysis, and Empirical Mode Decomposition. The results show that the autoencoder model exhibits superior denoising and better computational efficiency compared to the conventional approaches.