CN-121475280-B - Error identification method and system for dealing with inertial measurement unit

CN121475280BCN 121475280 BCN121475280 BCN 121475280BCN-121475280-B

Abstract

The invention discloses an error identification method and system for dealing with an inertial measurement unit, and relates to the technical field of inertial measurement, wherein the method comprises the steps of carrying out modal analysis on the inertial measurement unit by adopting finite element simulation, positioning a resonance sensitive area, arranging a distributed sensor array, synchronously acquiring multi-source signals and outputting a standardized signal set; the method comprises the steps of carrying out short-time Fourier transform and multi-channel time-frequency matrix fusion denoising on a standardized signal set, constructing and training a special sparse dictionary based on finite element modal parameters, screening candidate resonance modal characteristics through sparse decomposition, constructing a resonance modal trans-axis error mapping model, outputting a spatial frequency characteristic set, initializing a self-adaptive notch filter set based on the spatial frequency characteristic set, outputting real signals, calculating a pre-filtering and post-filtering signal difference value to obtain trans-axis interference signals, carrying out quantitative analysis and tracing on the trans-axis interference signals by combining the error mapping model, generating an error source positioning list, and obviously improving the accuracy and reliability of an inertial measurement unit.

Inventors

WU ANAN
XU XIAOWEI
ZHU CHUANGLU

Assignees

西安比特联创科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (6)

1. An error recognition method for an inertial measurement unit, comprising: performing modal analysis on the inertial measurement unit by adopting finite element simulation, positioning a resonance sensitive area, arranging a distributed sensor array, synchronously acquiring multi-source signals, preprocessing the signals, and outputting a standardized signal set; Carrying out short-time Fourier transform and multi-channel time-frequency matrix fusion denoising on a standardized signal set, constructing and training a special sparse dictionary based on finite element modal parameters, and screening candidate resonance modal characteristics through sparse decomposition, wherein the method comprises the steps of constructing an initial dictionary based on resonance modal parameters of finite element simulation, wherein dictionary atoms are frequency and space position two-dimensional characteristic vectors; the method comprises the steps of adopting an improved K-SVD algorithm to train a dictionary, taking a fused time-frequency matrix as a training sample, introducing resonance mode constraint items based on cosine similarity, iterating for a plurality of times, stopping training when a joint objective function is smaller than a target threshold, adopting an improved orthogonal matching pursuit algorithm to carry out sparse decomposition, taking the space consistency constraint in the decomposition process, obtaining a sparse coefficient matrix after the decomposition is completed, screening dictionary atoms corresponding to the sparse coefficient with the sparse coefficient amplitude larger than the sparse coefficient threshold to obtain a candidate resonance mode feature set, wherein the feature set comprises frequency features and space features of each candidate mode, combining a semi-physical simulation platform to train a BP neural network, constructing a resonance mode cross-axis error mapping model, outputting the space frequency feature set, comprising constructing an IMU semi-physical simulation platform, simulating resonance modes with different frequencies and different vibration modes to obtain cross-axis coupling error data of different resonance modes, taking the space and the frequency features of the candidate resonance modes as input parameters, training the BP neural network, adopting a random gradient descent algorithm to optimize the network weight, stopping when the value of the training error is smaller than 0.001, constructing a cross-axis resonance mode cross-axis error mapping model, obtaining a candidate resonance mode error mapping model, outputting a candidate resonance mode error mapping model, the method comprises the steps of obtaining a spatial frequency characteristic set comprising a resonant frequency, a spatial position and a reference error value by using a reference accelerometer trans-axis error value and a reference gyroscope trans-axis error value; The method comprises the steps of initializing a self-adaptive notch filter bank based on a space frequency characteristic set, performing FFT on a gyroscope preprocessing signal, calculating an actual trans-axis error initial estimation value by combining an acceleration preprocessing signal through a frequency domain matching method, calculating an error deviation of the initial estimation value and a reference error value, if the error deviation is larger than an error deviation threshold value, adjusting filter center frequency, bandwidth and inter-axis cooperative weight by a PID algorithm, inputting the preprocessing signal into the adjusted filter bank, performing cooperative filtering according to inter-axis cooperative suppression weight, checking physical logic matching of a filtered angular velocity signal and a filtered acceleration signal based on a rigid body kinematics principle, outputting the filtered acceleration signal as a real acceleration signal when the mean square error is smaller than the mean square error threshold value, outputting the real signal, calculating a trans-axis interference signal by calculating a difference value of the pre-filtered acceleration signal, combining an error mapping model to quantitatively analyze and source, wherein the difference value of the preprocessed acceleration signal and the real acceleration signal is the actual acceleration trans-axis interference signal, the pre-axis interference signal is the actual acceleration interference signal, the pre-processed signal is the actual trans-axis interference signal, the error is the actual interference signal of the main axis interference signal, the error is the actual interference signal, the error is generated by the fact that the error is the actual interference signal, and the error is the real-axis interference signal is the actual interference signal, and the real channel is the real channel, and the error is the real channel is the error, and the real channel is the error is the real channel, and the error is formed, and the real channel is the harmonic channel and has the largest.
2. The method for recognizing errors of an inertial measurement unit according to claim 1, wherein the method is characterized in that mode shape and amplitude values of the mode shapes at different frequencies are calculated through simulation, a plurality of resonance sensitive areas with the amplitude values of the mode shapes being larger than a preset amplitude value threshold value are positioned, and N piezoelectric film sensors are distributed in each sensitive area according to a 2 x 2 spatial gradient.
3. The method for recognizing errors of a coping inertial measurement unit according to claim 2, wherein the method is characterized by synchronously collecting multi-source signals including vibration signals, acceleration signals and gyroscope signals, preprocessing the collected multi-source original signals, performing direct current component self-adaptive cancellation on the vibration signals, the acceleration signals and the gyroscope signals by adopting a moving average method, performing filtering processing on the multi-source signals by adopting a 5-order Butterworth low-pass filter, and outputting a standardized vibration signal set, an acceleration preprocessing signal and a gyroscope preprocessing signal after preprocessing is completed.
4. The method for identifying errors of an inertial measurement unit according to claim 1 is characterized by comprising the steps of carrying out short-time Fourier transform and multi-channel time-frequency matrix fusion denoising, namely, carrying out STFT on each channel of a standardized vibration signal set, adopting a Hanning window and a variable window a long term plan, setting the length of a resonance estimated frequency band window to 128 points, setting the length of a non-resonance estimated frequency band window to 256 points, unifying the overlapping rate of all frequency band signals to 60%, obtaining N high-resolution time-frequency matrixes, calculating covariance matrixes between any two channel time-frequency matrixes, extracting covariance coefficients among the channels from the covariance matrixes, screening effective signal channels with the covariance coefficients larger than a coefficient threshold, and fusing the effective channel time-frequency matrixes by adopting a weighted average fusion algorithm.
5. The method for identifying errors of an inertial measurement unit according to claim 1, wherein the method is characterized by initializing an adaptive notch filter bank, specifically initializing m notch filters based on a spatial frequency feature set, aligning center frequencies of the filters to resonance frequencies of corresponding resonance modes, adapting filter bandwidths by adopting error weight, and setting inter-axis cooperative suppression weights of the filter bank according to inter-axis association relations corresponding to reference errors of the modes in the spatial frequency feature set.
6. An error recognition system for handling inertial measurement units for implementing the method of any one of claims 1 to 5, comprising: the signal acquisition module adopts finite element simulation to perform modal analysis on the inertial measurement unit, positions a resonance sensitive area, distributes a distributed sensor array, synchronously acquires multi-source signals, and outputs a standardized signal set after preprocessing the signals; The resonance mode feature extraction module performs short-time Fourier transform and multi-channel time-frequency matrix fusion denoising on the standardized signal set, builds and trains a special sparse dictionary based on finite element mode parameters, screens candidate resonance mode features through sparse decomposition, trains a BP neural network by combining with a semi-physical simulation platform, builds a resonance mode trans-axis error mapping model, and outputs a spatial frequency feature set; And the error filtering and tracing module is used for initializing a self-adaptive notch filter group based on the space frequency characteristic set, carrying out collaborative filtering and motion consistency check on signals, outputting real signals, calculating the difference value of the signals before and after filtering to obtain a trans-axis interference signal, carrying out quantization analysis and tracing by combining an error mapping model, and generating an error source positioning list.

Description

Error identification method and system for dealing with inertial measurement unit Technical Field The invention relates to the technical field of inertial measurement, in particular to an error identification method and system for an inertial measurement unit. Background The accuracy of an Inertial Measurement Unit (IMU) serving as a core component of a navigation and motion control system directly determines the performance of the system, in the prior art, the error identification of the IMU mainly depends on two modes of signal filtering and hardware compensation, a signal filtering method generally adopts a notch filter or a Kalman filtering algorithm with fixed parameters, the error is reduced by inhibiting interference signals of a specific frequency band, and the influence of resonance on the output of a sensor is reduced by optimizing the mechanical structural design of the IMU or adding damping materials. However, the prior art has obvious limitations, firstly, the traditional filtering method depends on fixed parameters, cannot dynamically adapt to resonance frequency changes of the IMU under different working conditions, so that an error suppression effect is unstable, secondly, hardware compensation can partially reduce resonance influence, but lacks accurate positioning of an error source, structural design is difficult to be optimized pertinently, the cost is high, furthermore, the prior art cannot effectively solve the problems of identifying and tracing a coupling error of a trans-axis, the resonance mode of the IMU often causes signal interference among multiple axes, the prior art is limited to single-axis analysis, and the modeling capability of spatial frequency characteristics and an error mapping relation is lacking, so that the error source positioning is inaccurate, reliable basis is difficult to be provided for a subsequent compensation algorithm, and further improvement of the performance of the IMU is limited. Disclosure of Invention (One) solving the technical problems Aiming at the defects of the prior art, the invention provides the error identification method and the system for dealing with the inertial measurement unit, which solve the problems that the traditional method cannot dynamically adapt to the change of the resonant frequency, the identification of the coupling error of the trans-axis is inaccurate and the positioning of the error source is difficult by adopting the finite element simulation positioning resonance sensitive area, the multi-source signal collaborative acquisition and processing, the self-adaptive filtering and the error tracing technology, and obviously improve the precision and the reliability of the inertial measurement unit. (II) technical scheme In order to achieve the above object, the present invention is realized by the following technical scheme that an error recognition method for coping with an inertial measurement unit includes: performing modal analysis on the inertial measurement unit by adopting finite element simulation, positioning a resonance sensitive area, arranging a distributed sensor array, synchronously acquiring multi-source signals, preprocessing the signals, and outputting a standardized signal set; Carrying out short-time Fourier transform and multi-channel time-frequency matrix fusion denoising on a standardized signal set, constructing and training a special sparse dictionary based on finite element modal parameters, screening candidate resonance modal characteristics through sparse decomposition, combining a semi-physical simulation platform to train a BP neural network, constructing a resonance modal trans-axis error mapping model, and outputting a spatial frequency characteristic set; initializing a self-adaptive notch filter bank based on a space frequency feature set, carrying out collaborative filtering and motion consistency verification on signals, outputting real signals, calculating difference values of signals before and after filtering to obtain a trans-axis interference signal, carrying out quantization analysis and tracing by combining an error mapping model, and generating an error source positioning list. Furthermore, the mode shape and the amplitude value of the vibration mode at different frequencies are calculated through simulation, a plurality of resonance sensitive areas with the amplitude value of the vibration mode being larger than a preset amplitude value threshold are positioned, and N piezoelectric film sensors are distributed in each sensitive area according to a 2 multiplied by 2 spatial gradient. Further, synchronously collecting multi-source signals including vibration signals, acceleration signals and gyroscope signals, preprocessing the collected multi-source original signals, namely adaptively eliminating direct current components of the vibration signals, the acceleration signals and the gyroscope signals by adopting a moving average method, filtering the multi-source signals by adopting