Method for extracting signal characteristics of 6-degree-of-freedom inertial sensor in real time

Abstract

The invention provides a method for extracting signal characteristics of a 6-degree-of-freedom inertial sensor in real time, which comprises the following steps of: acquiring 6 paths of output signals of the 6-degree-of-freedom inertial sensor in real time, continuously refreshing a storage data sequence in a rolling manner, and extracting the characteristics of the 6 paths of data sequences by adopting a time-frequency domain decomposition manner, and comprises the following steps of: respectively determining bandwidths of the 6 paths of signals; the method comprises the following steps: dividing a frequency band into N frequency segmentation points by using a dichotomy method, determining a kernel function cluster corresponding to each frequency segmentation point by taking a singing function as a generating function, constructing a linear equation set based on a kernel function sequence, solving coefficient vectors corresponding to the frequency segmentation points by adopting a least square method, and repeating the steps on each frequency segmentation point to obtain a linear equation set. N coefficient vectors are obtained and form a matrix, and a singular value group of the matrix represents the time-frequency characteristics of the corresponding signals.

Description

Technical Field

[0001] The present invention belongs to the technical field of 6-DOF inertial sensor signal processing and analysis, and specifically relates to a method for extracting time-frequency domain features of an X-axis acceleration signal, a Y-axis acceleration signal, a Z-axis acceleration signal, and a rotation angular velocity signal around the X-axis, a rotation angular velocity signal around the Y-axis, and a rotation angular velocity signal around the Z-axis output by a 6-DOF inertial sensor. Background Art

[0002] Inertial sensor is a common motion sensor used to measure the position and movement of objects in space. It integrates accelerometers and gyroscopes, and can detect the acceleration of people or objects on the X-axis, Y-axis, and Z-axis, as well as the angular velocity of rotation around the X-axis, Y-axis, and Z-axis in real time, and comprehensively record its position and speed changes in space. Inertial sensors are key devices for tracking and navigation. Due to their high accuracy and small size, they are widely integrated into various electronic devices, such as smartphones, vehicle navigation systems, drones, and intelligent mobile robots.

[0003] Fourier transform and Fourier series are the most commonly used tools for processing signals. They are generally applicable to continuous non-time-varying signals. They can convert signals from the time domain to the frequency domain, and comprehensively and profoundly reveal the frequency characteristics of the signal over the entire time range. The 6-channel signals output by the 6-DOF inertial sensor are sampled in real time within a limited time period, and the statistical characteristics of the signals do not change regularly over time. However, Fourier transform does not play its role in this case.

[0004] Wavelet transform is an extension of Fourier transform. It has advantages in analyzing non-stationary signals. It can provide time and frequency information. It uses the Singer function as the wavelet basis function and constructs the Singer function series to perform local analysis of the signal at a specific frequency and time. At the same time, wavelet transform can perform multi-scale analysis of the signal, separate the complex and multi-layered non-stationary signal into frequency bands, and help us identify the frequency signal components with representative characteristics output by the inertial sensor, which can achieve the fitting of the original signal to the greatest extent. In addition, the time-frequency domain decomposition of the inertial sensor output signal can be used to extract features. This feature can be used for data analysis of the signal through machine learning, which is suitable for many mechanical engineering projects such as: signal anomaly analysis, equipment fault diagnosis, signal classification and identification. Summary of the invention

[0005] The present invention proposes a method for real-time extraction of signal features of a 6-DOF inertial sensor. The method samples 6 output signals of the inertial sensor in real time, divides the bandwidths of the 6 signals respectively, constructs a kernel function cluster with a sinc function as a mother function at each frequency division point, constructs a linear equation group based on the kernel function cluster within a sampling time period, and uses the least square method to solve the coefficient sequence that uniquely determines its frequency features. The above operation is performed at each frequency division point to obtain the coefficient sequence corresponding to each frequency, which is then combined into a coefficient vector matrix, and the singular value group of the matrix is ​​the time-frequency feature of the corresponding signal.

[0006] To achieve the above purpose, the technical solution provided by the present invention is:

[0007] A method for real-time extraction of 6-DOF inertial sensor signal features, characterized in that 6 output signals of the inertial sensor are sampled in real time, and the 6 output signals are stored and refreshed in a real-time rolling manner, wherein the 6 output signals of the inertial sensor are X-axis acceleration signals X a , Y-axis acceleration signal Y a , Z-axis acceleration signal Z a , the angular velocity signal A around the X axis Ω , the angular velocity signal B around the Y axis Ω , the angular velocity signal C around the Z axis Ω ; Form 6 dynamic data sequences X with a length of L a , Y a , Z a , A Ω , B Ω , C Ω , including the X-axis acceleration signal sequence X a , Y-axis acceleration signal sequence Y a , Z-axis acceleration signal sequence Z a , rotation angular velocity signal sequence A around the X axis Ω , rotation angular velocity signal sequence B around the Y axis Ω , rotation angular velocity signal sequence C around the Z axis Ω ; The characteristics of each dynamic data sequence are extracted by time-frequency domain decomposition. The specific process includes: determining the X-axis acceleration signal X a The bandwidth is ω cX , Y-axis acceleration signal Y a The bandwidth is ω cY , Z-axis acceleration signal Z a The bandwidth is ω cZ , the angular velocity signal A around the X axis Ω The bandwidth is ω cA , the angular velocity signal B around the Y axis Ω The bandwidth is ωcB , the angular velocity signal C around the Z axis Ω The bandwidth is ω cC ; The bandwidth of the above 6 output signals is ω cX ,ω cY ,ω cZ ,ω cA ,ω cB ,ω cC Perform frequency point segmentation to form N corresponding frequency segmentation points, and select the Singer function As the mother function, determine the kernel function cluster corresponding to each frequency cut point Each kernel function cluster contains L functions. According to the sampling values ​​of the 6 output signals and their corresponding frequency cut points, a linear equation system is constructed based on the kernel function series. The least squares principle is used to solve the above equation system to obtain the coefficient vector θ related to the frequency cut point. n , repeat the above steps at each frequency segmentation point to obtain N coefficient vectors, and form them into a matrix Θ. The singular value group of the matrix Θ represents the time-frequency characteristics of the corresponding signal.

[0008] The interval sampling period of the inertial sensor is Δ, the sampling length is L, and at the sampling time point {kΔ|k=0,1,2…L-1}, the 6 output signals are sampled in real time to form 6 data sequences: X a =[X a (0),…X a (kΔ),…X a (L-1)Δ],Y a =[Y a (0),…Y a (kΔ),…Y a (L-1)Δ],Z a =[Z a (0),…Z a (kΔ),…Z a (L-1)Δ],A Ω =[A Ω (0), …A Ω (kΔ),…A Ω (L-1)Δ],B Ω =[B Ω (0),…B Ω (kΔ),…B Ω (L-1)Δ],C Ω =[C Ω (0),…C Ω (kΔ),…C Ω L-1)Δ], when k≥L, the sampling time point is {kΔ|k=k-L+1, k-L+2, ...k}, and the real-time rolling acquisition forms a 6-channel data sequence Xa =[X a (k-L+1)Δ,X a (k-L+2)Δ,…X a (kΔ)],Y a =[Y a (k-L+1)Δ,Y a (k-L+2)Δ,…Y a (kΔ)],Z a =[Z a (k-L+1)Δ,Z a (k-L+2)Δ,…Z a (kΔ)],A Ω =[A Ω (k-L+1)Δ,A Ω (k-L+2)Δ,…A Ω (kΔ)],B Ω =[B Ω (k-L+1)Δ,B Ω (k-L+2)Δ,…B Ω (kΔ)],C Ω =[C Ω (k-L+1)Δ,C Ω (k-L+2)Δ,…C Ω (kΔ)].

[0009] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the specific method for dividing the frequency band according to N dividing points is: respectively determining the bandwidths ω of the three acceleration signals of the X-axis, Y-axis, and Z-axis cX ,ω cY ,ω cZ , the acceleration signal is quickly decomposed in the time-frequency domain within each bandwidth according to the formula: Divide each bandwidth to determine the signal X a The bandwidth ω cX The split point is Determine signal Y a The bandwidth ω cY The split point is Determine signal Z a The bandwidth ω cZ The split point is Select the bandwidth ω of the three rotation angular velocity signals around the X axis, Y axis, and Z axis respectively cA ,ω cB ,ω cC , the angular velocity signal is quickly decomposed in the time-frequency domain within each bandwidth according to the formula: Divide the frequency band to determine signal A Ω The bandwidth ωcA The split point is Confirm signal B Ω The bandwidth ω cB The split point is Determine signal C Ω The bandwidth ω cC The split point is

[0010] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the method for constructing the kernel function cluster corresponding to each frequency cut point is: using the Singer function is the generating function, based on which the X-axis acceleration signal X a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the Y-axis acceleration signal Y a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function Generating function, based on this generating function, the Z-axis acceleration signal Z a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal A around the X-axis is Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal B around the Y axis is calculated. Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal C of rotation around the Z axis is calculated. Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions.

[0011] Furthermore, the method and device for real-time extraction of 6-DOF inertial sensor signal features are characterized in that the specific method of constructing a linear equation system based on a kernel function cluster at each frequency cut point is: selecting an X-axis acceleration signal X a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0012]

[0013] Select Y-axis acceleration signal Y a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0014]

[0015] Select the Z-axis acceleration signal Z a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0016]

[0017] Select the rotation angular velocity signal A around the X axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0018]

[0019] Select the rotation angular velocity signal B around the Y axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0020]

[0021] Select the angular velocity signal C around the Z axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

[0022]

[0023] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the least squares principle is used to solve the above equations, comprising the following steps:

[0024] Step 1: For the X-axis acceleration signal X output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0025]

[0026] For the Y-axis acceleration signal Y output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0027]

[0028] For the Z-axis acceleration signal Z output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0029]

[0030] For the rotation angular velocity signal A around the X axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0031]

[0032] For the rotation angular velocity signal B around the Y axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0033]

[0034] For the rotation angular velocity signal C around the Z axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix

[0035]

[0036] Step 2: Collect the real-time X-axis acceleration signal X a The data sequence X a =[X a (0),…X a (kΔ),…X a (L1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time acquired Y-axis acceleration signal Y a The data sequence Y a =[Y a (0),…Y a (kΔ),…Y a (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time collected Z-axis acceleration signal Z a The data sequence Z a =[Z a (0),…Z a (kΔ),…Z a (L-1)Δ], regarded as a column vector, coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient column vector The real-time collected A-axis angular velocity signal A Ω The data sequence A Ω =[A Ω(0),…A Ω (kΔ),…A Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time acquired B-axis angular velocity signal B Ω Data sequence B Ω =[B Ω (0),…B Ω (kΔ),…B Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time collected C-axis angular velocity signal C Ω The data sequence C Ω =[C Ω (0),…C Ω (kΔ),…C Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector

[0037] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the coefficient matrix Θ is composed as follows: The coefficient vector obtained at the point uniquely represents the time-frequency domain component at the frequency point. For the X-axis acceleration signal X output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the Y-axis acceleration signal Y output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the Z-axis acceleration signal Z output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the A-axis angular velocity signal A output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the B-axis angular velocity signal B output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the C-axis angular velocity signal C output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix

[0038] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the specific solution method of the eigenvalue group of the coefficient matrix Θ is: for the X-axis acceleration signal X output by the inertial sensor a The coefficient matrix of

[0039]

[0040] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group The singular value group represents the time-frequency characteristics of the currently collected X-axis acceleration signal. For the Y-axis acceleration signal Y output by the inertial sensor a The coefficient matrix of

[0041]

[0042] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected Y-axis acceleration signal Y a The time-frequency domain characteristics of the inertial sensor output Z axis acceleration signal Z aThe coefficient matrix of

[0043]

[0044] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected Z-axis acceleration signal Z a The time-frequency domain characteristics of the inertial sensor output about the X-axis rotation angular velocity signal A Ω The coefficient matrix of

[0045]

[0046] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal A around the X axis Ω The time-frequency domain characteristics of the inertial sensor output about the Y-axis rotation angular velocity signal B Ω The coefficient matrix of

[0047]

[0048] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal B around the Y axis Ω The time-frequency domain characteristics of the inertial sensor output about the Z axis rotation angular velocity signal C Ω The coefficient matrix of

[0049]

[0050] right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal C around the Z axis Ω time-frequency domain characteristics.

[0051] Furthermore, the method for real-time extraction of 6-DOF inertial sensor signal features is characterized in that the extracted real-time features of the 6-DOF inertial sensor signal can be used for signal classification and recognition through machine learning methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The present invention is a flow chart of a method for real-time extraction of 6-DOF inertial sensor signal features. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] like Figure 1 The present invention provides a method for real-time extraction of 6-DOF inertial sensor signal features.

[0055] Step 1: The inertial sensor collects the X-axis acceleration signal X in real time a , Y-axis acceleration signal Y a , Z-axis acceleration signal Z a , the angular velocity signal A around the X axis Ω , the angular velocity signal B around the Y axis Ω , the angular velocity signal C around the Z axis Ω ; Form 6 dynamic data sequences X with a length of L a , Y a , Z a , A Ω , BΩ , C Ω , including the X-axis acceleration signal sequence X a , Y-axis acceleration signal sequence Y a , Z-axis acceleration signal sequence Z a , rotation angular velocity signal sequence A around the X axis Ω , rotation angular velocity signal sequence B around the Y axis Ω , rotation angular velocity signal sequence C around the Z axis Ω .

[0056] Step 2: Determine the X-axis acceleration signal X a The bandwidth is ω cX , Y-axis acceleration signal Y a The bandwidth is ω cY , Z-axis acceleration signal Z a The bandwidth is ω cZ , the angular velocity signal A around the X axis Ω The bandwidth is ω cA , the angular velocity signal B around the Y axis Ω The bandwidth is ω cB , the angular velocity signal C around the Z axis Ω The bandwidth is ω cC ; The bandwidth of the above 6 output signals is ω cX ,ω cY ,ω cZ ,ω cA ,ω cB ,ω cC Frequency point segmentation is performed to form corresponding N frequency segmentation points.

[0057] Step 3: Select the Single function As the mother function, determine the kernel function cluster corresponding to each frequency cut point Each kernel function cluster contains L functions.

[0058] Step 4: According to the sampling values ​​of the 6 output signals and their corresponding frequency cut-off points, a linear equation system is constructed based on the kernel function series, and the least squares principle is used to solve the above equation system to obtain the coefficient vector θ related to the frequency cut-off point n .

[0059] Step 5: Repeat the above steps at each frequency segmentation point to obtain N coefficient vectors, and form them into a matrix Θ.

[0060] Step 6: The singular value group of the matrix Θ represents the time-frequency characteristics of the corresponding signal.

[0061] An embodiment of the present invention provides a method for real-time extraction of 6-DOF inertial sensor signal features.

[0062] Reference Figure 1 The method for real-time signal feature extraction described in the flowchart is applicable to each signal of the 6-DOF inertial sensor. The method of collecting 6 signals of the inertial sensor and performing time-frequency domain decomposition of the signals to extract the characteristics of the signals can be applied to a variety of engineering and scientific projects, especially when analyzing non-stationary signals.

[0063] Embodiment 1, the present invention can be applied to the diagnosis of mechanical faults. By analyzing the six signals generated by the mechanical vibration collected by the six-degree-of-freedom inertial sensor, the characteristics of the machine vibration signal are extracted using the time-frequency domain decomposition method. The characteristics can uniquely represent the original vibration signal. By comparing the characteristic sequence of the normal signal, it can be identified whether the machine is in a healthy working state; the collected six groups of feature series can be used through machine learning, using the neural network method to classify the fault signal and identify the type of mechanical fault.

Claims

1. A method for real-time extraction of 6-DOF inertial sensor signal features, characterized in that The 6-channel output signal of the inertial sensor is sampled in real time, and the 6-channel signal is stored and refreshed in a real-time rolling manner. The 6-channel output signal of the inertial sensor is the X-axis acceleration signal X a , Y-axis acceleration signal Y a , Z-axis acceleration signal Z a , the angular velocity signal A around the X axis Ω , the angular velocity signal B around the Y axis Ω , the rotational angular velocity signal C around the Z axis Ω ; Form 6 dynamic data sequences X with a length of L a , Y a , Z a , A Ω , B Ω , C Ω , including the X-axis acceleration signal sequence X a , Y-axis acceleration signal sequence Y a , Z-axis acceleration signal sequence Z a , rotation angular velocity signal sequence A around the X axis Ω , rotation angular velocity signal sequence B around the Y axis Ω , rotation angular velocity signal sequence C around the Z axis Ω ; The characteristics of each dynamic data sequence are extracted by time-frequency domain decomposition. The specific process includes: determining the X-axis acceleration signal X a The bandwidth is ω cX , Y-axis acceleration signal Y a The bandwidth is ω cY , Z-axis acceleration signal Z a The bandwidth is ω cZ , the angular velocity signal A around the X axis Ω The bandwidth is ω cA , the angular velocity signal B around the Y axis Ω The bandwidth is ω cB , the angular velocity signal C around the Z axis Ω The bandwidth is ω cC ; The bandwidth of the above 6 output signals is ω cX ,ω cY ,ω cZ ,ω cA ,ω cB ,ω cC Perform frequency point segmentation to form N corresponding frequency segmentation points, and select the Singer function As the mother function, determine the kernel function cluster corresponding to each frequency cut point Each kernel function cluster contains L functions. According to the sampling values ​​of the 6 output signals and their corresponding frequency cut points, a linear equation system is constructed based on the kernel function series. The least squares principle is used to solve the above equation system to obtain the coefficient vector θ related to the frequency cut point. n , repeat the above steps at each frequency segmentation point to obtain N coefficient vectors, and form them into a matrix Θ. The singular value group of the matrix Θ represents the time-frequency characteristics of the corresponding signal.

2. The method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1, characterized in that: The interval sampling period of the inertial sensor is Δ, the sampling length is L, and at the sampling time points {kΔ|k=0, 1, 2…L-1}, the 6 output signals are sampled in real time to form 6 data sequences: X a =[X a (0),…X a (kΔ),…X a (L-1)Δ],Y a =[Y a (0),…Y a (kΔ),…Y a (L-1)Δ],Z a =[Z a (0),…Z a (kΔ),…Z a (L-1)Δ],A Ω =[A Ω (0), …A Ω (kΔ),…A Ω (L-1)Δ],B Ω =[B Ω (0),…B Ω (kΔ),…B Ω (L-1)Δ],C Ω =[C Ω (0),…C Ω (kΔ),…C Ω (L-1)Δ], when k≥L, the sampling time point is {kΔ|k=k-L+1, k-L+2, ...k}, and the real-time rolling acquisition forms a 6-channel data sequence X a =[X a (k-L+1)Δ,X a (k-L+2)Δ,…X a (kΔ)],Y a =[Y a (k-L+1)Δ,Y a (k-L+2)Δ,…Y a (kΔ)],Z a =[Z a (k-L+1)Δ,Z a (k-L+2)Δ,…Z a (kΔ)], A Ω =[A Ω (k-L+1)Δ,A Ω (k-L+2)Δ,…A Ω (kΔ)], B Ω =[B Ω (k-L+1)Δ,B Ω (k-L+2)Δ,…B Ω (kΔ)], C Ω =[C Ω (k-L+1)Δ,C Ω (k-L+2)Δ,…C Ω (kΔ)]。 3. The method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1, characterized in that: The specific method of dividing the frequency band according to N dividing points is: respectively determine the bandwidth ω of the three acceleration signals of the X-axis, Y-axis and Z-axis cX ,ω cY ,ω cZ , the acceleration signal is quickly decomposed in the time-frequency domain within each bandwidth according to the formula: Divide each bandwidth to determine the signal X a The bandwidth ω cX The split point is Determine signal Y a The bandwidth ω cY The split point is Determine signal Z a The bandwidth ω cZ The split point is Select the bandwidth ω of the three rotation angular velocity signals around the X axis, Y axis, and Z axis respectively cA ,ω cB ,ω cC , the angular velocity signal is quickly decomposed in the time-frequency domain within each bandwidth according to the formula: Divide the frequency band to determine signal A Ω The bandwidth ω cA The split point is Confirm signal B Ω The bandwidth ω cB The split point is Determine signal C Ω The bandwidth ω cC The split point is 4. The method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1, characterized in that: The method for constructing the kernel function cluster corresponding to each frequency cut point is: using the Singer function is the generating function, based on which the X-axis acceleration signal X a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the Y-axis acceleration signal Y a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function Generating function, based on this generating function, the Z-axis acceleration signal Z a Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal A around the X-axis is Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal B around the Y axis is calculated. Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions, with the Singer function is the generating function, based on which the angular velocity signal C of rotation around the Z axis is calculated. Ω Perform fast time-frequency domain decomposition and construct kernel function clusters corresponding to each frequency segmentation point Each kernel function cluster contains L sub-functions.

5. A method and device for real-time extraction of 6-DOF inertial sensor signal features according to claim 1 or claim 4, characterized in that: The specific method of constructing the linear equations based on the kernel function cluster at each frequency cut point is: select the X-axis acceleration signal X a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows: Select Y-axis acceleration signal Y a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows: Select the Z-axis acceleration signal Z a The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows: Select the rotation angular velocity signal A around the X axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows: Select the rotation angular velocity signal B around the Y axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows: Select the angular velocity signal C around the Z axis Ω The first frequency At this frequency point, define the coefficient sequence Definition and Kernel Function Cluster The corresponding incidence matrix Setting up the equation Constructed at the first frequency The linear equations of the sampling values ​​at each moment are as follows:

6. A method for real-time extraction of 6-DOF inertial sensor signal features according to any one of claims 1 to 5, characterized in that: The method of solving the above equations by the least squares principle comprises the following steps: Step S1: For the X-axis acceleration signal X output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix For the Y-axis acceleration signal Y output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix For the Z-axis acceleration signal Z output by the inertial sensor a , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix For the rotation angular velocity signal A around the X axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix For the rotation angular velocity signal B around the Y axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix For the rotation angular velocity signal C around the Z axis output by the inertial sensor Ω , according to the first frequency point The kernel function cluster corresponding to the location Constructing the correlation matrix Step S2: The real-time collected X-axis acceleration signal X a The data sequence X a =[Y a (0),…X a (kΔ),…X a (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time acquired Y-axis acceleration signal Y a The data sequence Y a =[Y a (0),…Y a (kΔ),…Y a (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time collected Z-axis acceleration signal Z a The data sequence Z a =[Z a (0),…Z a (kΔ),…Z a (L-1)Δ], regarded as a column vector, coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient column vector The real-time collected A-axis angular velocity signal A Ω The data sequence A Ω =[A Ω (0),…A Ω (kΔ),…A Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time acquired B-axis angular velocity signal B Ω Data sequence B Ω =[B Ω (0),…B Ω (kΔ),…B Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector The real-time collected C-axis angular velocity signal C Ω The data sequence C Ω =[C Ω (0),…C Ω (kΔ),…C Ω (L-1)Δ] is regarded as a column vector, and the coefficient sequence Considered as a coefficient column vector, based on the least squares principle, according to the formula: Solving for the coefficient vector 7. A method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1 to claim 6, characterized in that: The coefficient matrix Θ is composed as follows: The coefficient vector obtained at the point uniquely represents the time-frequency domain component at the frequency point. For the X-axis acceleration signal X output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the Y-axis acceleration signal Y output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the Z-axis acceleration signal Z output by the inertial sensor a , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the A-axis angular velocity signal A output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the B-axis angular velocity signal B output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix For the C-axis angular velocity signal C output by the inertial sensor Ω , at the other frequency cut points Repeat the above steps to obtain N coefficient vectors It is composed into a coefficient matrix 8. The method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1 or claim 7, characterized in that: The specific solution method for the eigenvalue group of the coefficient matrix Θ is: for the X-axis acceleration signal X output by the inertial sensor a The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group The singular value group represents the time-frequency characteristics of the currently collected X-axis acceleration signal. For the Y-axis acceleration signal Y output by the inertial sensor a The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected Y-axis acceleration signal Y a The time-frequency domain characteristics of the inertial sensor output Z axis acceleration signal Z a The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected Z-axis acceleration signal Z a The time-frequency domain characteristics of the inertial sensor output about the X-axis rotation angular velocity signal A Ω The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal A around the X axis Ω The time-frequency domain characteristics of the inertial sensor output about the Y-axis rotation angular velocity signal B Ω The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal B around the Y axis Ω The time-frequency domain characteristics of the inertial sensor output about the Z axis rotation angular velocity signal C Ω The coefficient matrix of right Perform singular value decomposition: In the formula is an L×L orthogonal matrix, is an N×N orthogonal matrix, is an L×N diagonal matrix, whose main diagonal elements are coefficient matrices Singular values ​​of, arrange the non-zero singular values ​​in descending order to get the singular value group This singular value group represents the currently collected rotation angular velocity signal C around the Z axis Ω time-frequency domain characteristics.

9. The method for real-time extraction of 6-DOF inertial sensor signal features according to claim 1, characterized in that: The extracted real-time features of the 6-DOF inertial sensor signal can be used to classify and identify the signal through a machine learning method.

Images