A MEMS random error analysis method fused with genetic algorithm

Gao, Duanyang; Li, An; Fu, Jun; Zhang, Ruichen

doi:10.1117/12.2504183

Cited by 2 publications

(1 citation statement)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Allan variance method is widely used to observe and quantify random errors of sensors in the time domain by plotting the double logarithmic curve of the inertial sensor output signal and has been widely used to identify random errors in inertial sensors. For MEMS accelerometers, the types of random errors separated by the Allan variance method mainly include Quantization Noise (QN), Velocity Random Walk (VRW), Bias Instability (BI), Acceleration Random Walk (ARW), and Rate Ramp (RR) [ 6 , 7 , 8 , 9 , 10 ]. QN refers to the bias that exists when discretizing an analog signal into a digital signal, where the MSRA output is a quasi-digital signal in the form of frequency without the analog-to-digital conversion process.…”

Section: Introductionmentioning

confidence: 99%

A Random Error Suppression Method Based on IGWPSO-ELM for Micromachined Silicon Resonant Accelerometers

et al. 2023

View full text Add to dashboard Cite

There are various errors in practical applications of micromachined silicon resonant accelerometers (MSRA), among which the composition of random errors is complex and uncertain. In order to improve the output accuracy of MSRA, this paper proposes an MSRA random error suppression method based on an improved grey wolf and particle swarm optimized extreme learning machine (IGWPSO-ELM). A modified wavelet threshold function is firstly used to separate the white noise from the useful signal. The output frequency at the previous sampling point and the sequence value are then added to the current output frequency to form a three-dimensional input. Additional improvements are made on the particle swarm optimized extreme learning machine (PSO-ELM): the grey wolf optimization (GWO) is fused into the algorithm and the three factors (inertia, acceleration and convergence) are non-linearized to improve the convergence efficiency and accuracy of the algorithm. The model trained offline using IGWPSO-ELM is applied to predicting compensation experiments, and the results show that the method is able to reduce velocity random walk from the original 4.3618 μg/√Hz to 2.1807 μg/√Hz, bias instability from the original 2.0248 μg to 1.3815 μg, and acceleration random walk from the original 0.53429 μg·√Hz to 0.43804 μg·√Hz, effectively suppressing the random error in the MSRA output.

show abstract