Calibration of low-grade inertial measurement unit sensors using real-coded genetic algorithms
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This paper presents a calibration process that implements a wide variety of solutions, including error modeling, platform design, algorithms, and measurement methods for estimating errors of a low-grade inertial measurement unit sensors (LIMUS), while ensuring cost savings for both instrumentation and computation. Error elements that include bias, non-orthogonality, misalignment, and scale factors are modeled as sensor error models (SEM). A do-it-yourself (DIY) integrated platform is capable of multi-position data collection that can reach up to 48 orientations for accelerometers (ACCE) and rotational velocity for gyroscopes (GYR), crafted using low-cost and available means. The local Earth’s gravity is utilized to reference signals of ACCEs, while GYR’ readings are compared to encoder samples. Then, real-coded genetic algorithms (R-CGA) are designed to estimate the SEMs. Root mean square residual (RMSR) between readings and reference signals is calculated and considered as the fitness function. Three different populations of 300, 500 and 1000 individuals are studied for evaluating the quality of error estimation. Additionally, analyzing the effect of individual selection probability in the selection operators provide an effective solution that reduces the number of initial individuals in the population and saves computing cost.
Abstract
This paper presents a calibration process that implements a wide variety of solutions, including error modeling, platform design, algorithms, and measurement methods for estimating errors of a low-grade inertial measurement unit sensors (LIMUS), while ensuring cost savings for both instrumentation and computation. Error elements that include bias, non-orthogonality, misalignment, and scale factors are modeled as sensor error models (SEM). A do-it-yourself (DIY) integrated platform is capable of multi-position data collection that can reach up to 48 orientations for accelerometers (ACCE) and rotational velocity for gyroscopes (GYR), crafted using low-cost and available means. The local Earth’s gravity is utilized to reference signals of ACCEs, while GYR’ readings are compared to encoder samples. Then, real-coded genetic algorithms (R-CGA) are designed to estimate the SEMs. Root mean square residual (RMSR) between readings and reference signals is calculated and considered as the fitness function. Three different populations of 300, 500 and 1000 individuals are studied for evaluating the quality of error estimation. Additionally, analyzing the effect of individual selection probability in the selection operators provide an effective solution that reduces the number of initial individuals in the population and saves computing cost.