Method for improving data precision of gyroscope array of stabilized platform

Abstract

The invention relates to the field of data processing, in particular to a method for improving the precision of gyroscope array data of a stabilized platform. The method comprises the following steps: S1, carrying out primary screening: carrying out smooth denoising processing on a signal by adopting a low-pass filter, quantifying the performance of a gyroscope unit in a variance form, and showing that the gyroscope jittering degree is more severe when the variance is larger; s2, double screening: drawing a frequency distribution histogram in a normality test, visually fitting a Q-Q diagram, and fitting a normal distribution probability density curve; and S3, temperature drift compensation: performing closed loop on differential speed of the encoder to obtain a gyroscopic drift value, sampling gyroscopic data on the stable two-axis platform when the differential speed of the encoder is 0 in a full temperature range, and then fitting a collected temperature drift curve. The precision of output data of the gyroscope array is improved by utilizing a double screening algorithm and a temperature drift compensation algorithm, and the control effect of the stable platform is greatly improved.

Description

Technical Field

[0001] The invention relates to the field of data processing, and in particular to a method for improving the data accuracy of a gyro array of a stable platform. Background Art

[0002] In recent years, gyro arrays have been widely used in the control of stable platforms, and they can be used as speed feedback devices for closed-loop control of speed loops. Although the traditional MEMS array gyro used in stable platforms can provide relatively accurate speed feedback for stable platforms, its accuracy is still far behind that of gyro with higher accuracy (such as fiber optic gyro). The main reason is that traditional gyro arrays usually rely on simply averaging the output of each gyro unit to obtain the output angular velocity to stabilize the platform. This method has the following obvious disadvantages:

[0003] If there are unstable and jittery cells in the array, even if they only account for a small number of cells in the array, the accuracy of the entire output data will be seriously affected due to the averaging of the final result, thereby affecting the stability of the stable platform. In addition, the temperature drift of the MEMS gyroscope data will eventually lead to serious deviations in the output data. Summary of the invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and propose a method for improving the data accuracy of the gyro array of a stable platform, which uses a double screening algorithm and a temperature drift compensation algorithm to improve the output data accuracy of the gyro array and greatly improve the control effect of the stable platform.

[0005] The technical solution of the present invention is a method for improving the data accuracy of a gyro array of a stable platform, comprising the following steps:

[0006] S1. First screening: Use a low-pass filter to smooth and denoise the signal, and quantify the performance of the gyro unit in the form of variance. The larger the variance, the more severe the gyro jitter;

[0007] S2. Double screening: draw frequency distribution histograms in normality tests, visualize fitted QQ plots, and fit normal distribution probability density curves;

[0008] S3. Temperature drift compensation: The encoder differential speed closed loop obtains the gyro drift value. Within the full temperature range, when the speed obtained by the encoder differential is 0, the gyro data on the stable two-axis platform is sampled with a sampling period of 0.001s. The temperature drift curve after the collection is then fitted. The principle uses the least squares method to determine the goodness of fit R 2 The threshold is used as the segment fitting point and compensation is performed.

[0009] Preferably, in S1, the low-pass filtering algorithm formula is:

[0010] Y(n)=a*X(n)+(1-a)*Y(n-1)

[0011] Among them, a is the filter coefficient, X(n) is the current input value of the gyroscope, Y(n-1) is the output value after the last gyroscope filter, and Y(n) is the output value of the current gyroscope filter.

[0012] Preferably, the variance calculation formula is:

[0013] D(X)=Var(X)=E[XE(X)] 2

[0014] Adopt the modified formula of variance: D(X)=E(X 2 )-[E(X)] 2 The data set is the gyro angular velocity after 10,000 acquisitions. After the real-time acquired gyro data is sent to the low-pass filter model, the filtered gyro value is obtained and the mean difference E(X), E(X 2 ), the variance of each gyroscope is obtained through the variance transformation formula, and the gyroscope with better performance is screened out through the quick sorting method.

[0015] Preferably, in S2, the process of drawing the QQ graph is as follows:

[0016] A1. Sort the data set and calculate the empirical quantile;

[0017] A2. Compare all the calculated empirical quantiles with the theoretical quantiles of the standard normal distribution;

[0018] A3. Plot the point pairs on the graph, where the x-axis represents the empirical quantile and the y-axis represents the theoretical quantile;

[0019] A4. Assess the normality of the data by observing the distribution pattern of these points.

[0020] Preferably, when fitting the normal distribution probability density curve, the gyro data set is first preprocessed to remove outliers;

[0021] The maximum likelihood estimation is used to estimate the mean μ and standard deviation σ of the normal distribution. By constructing the likelihood function of the data set and maximizing the likelihood function using the gradient descent method, the optimal mean and standard deviation are solved.

[0022] The probability density function of the normal distribution is:

[0023]

[0024] Dataset X = {x1, x2, …, x n}, the likelihood function is the joint probability density function, which is the product of the probability density functions of each data point:

[0025]

[0026] Log-likelihood function:

[0027]

[0028] Set the objective function:

[0029] Maximize the log-likelihood function -l(μ,σ 2 ) In optimization problems, it is usually transformed into minimizing its negative value, that is, maximizing l is equivalent to minimizing -l:

[0030]

[0031] Use gradient descent to calculate J with respect to μ and σ 2 The partial derivative of :

[0032]

[0033] Update μ and σ 2 :

[0034]

[0035]

[0036] Among them, α is the learning rate, which is used to control the step size of the update;

[0037] The fitting process is an iterative optimization process, starting from the initial parameter value, and continuously optimizing the parameter value through multiple iterations until the preset convergence conditions are met.

[0038] Preferably, in S3, the gyroscope drift value is obtained for the encoder differential speed closed loop, and when the speed is 0, the gyroscope data is collected, which is the current gyroscope drift angular velocity.

[0039] Preferably, the optimal solution of k and b is determined by calculating the optimal extreme point to compensate, and the algorithm formula is:

[0040] The linear function is:

[0041] Among them, x i is the original data collected by the gyroscope, k is the slope of the fitted linear function, and b is the intercept of the function;

[0042] Linear loss function mean square error:

[0043] in, It means finding the value of the independent variable that makes the function reach the minimum value;

[0044] According to the principle of mean square error of the above linear loss function, Convert to The minimum value of Find the extreme point of the L derivative and calculate the optimal solution of k and b. The calculation formulas for their respective partial derivatives are as follows:

[0045]

[0046] Compared with the prior art, the present invention has the following beneficial technical effects:

[0047] The present invention uses a double screening algorithm and a temperature drift compensation algorithm to improve the accuracy of the gyro array output data. Experimental verification shows that the screening algorithm reduces the jitter amplitude of the gyro angular velocity output by about 10 times, and the real angular velocity after temperature drift compensation is also stable near zero in a static state. The results show that after the gyro data is processed by the screening algorithm and temperature drift compensation, the control effect of the stable platform can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a structural flow chart of an embodiment of the present invention;

[0049] Figure 2 This is the original output data diagram after the fusion of 8 gyro data;

[0050] Figure 3 The original data of gyroscope 1 and the data after filtering and denoising;

[0051] Figure 4 The original data of Gyro 2 and the data after filtering and denoising;

[0052] Figure 5 The original data of Gyro 3 and the data after filtering and denoising;

[0053] Figure 6 The original data of Gyro 4 and the data after filtering and denoising;

[0054] Figure 7 The original data of Gyro 5 and the data after filtering and denoising;

[0055] Figure 8 The original data of Gyro 6 and the data after filtering and denoising;

[0056] Fig. 9 The original data of Gyro 7 and the data after filtering and denoising;

[0057] Fig.10 The original data of Gyro 8 and the data after filtering and denoising;

[0058] Fig.11 The frequency distribution histogram, QQ graph and normal distribution graph of gyroscope 1;

[0059] Fig.12 The frequency distribution histogram, QQ graph and normal distribution graph of Gyro 2;

[0060] Fig.13 The frequency distribution histogram, QQ graph and normal distribution graph of Gyro 3;

[0061] Fig.14 The frequency distribution histogram, QQ graph and normal distribution graph of gyro 4;

[0062] Fig.15 The frequency distribution histogram, QQ graph and normal distribution graph of gyro 5;

[0063] Fig.16 The frequency distribution histogram, QQ graph and normal distribution graph of Gyro 6;

[0064] Fig.17 The frequency distribution histogram, QQ graph and normal distribution graph of gyro 7;

[0065] Fig.18 The frequency distribution histogram, QQ graph and normal distribution graph of gyro 8;

[0066] Fig.19 This is the output data graph of the gyro fusion after screening;

[0067] Fig. 20 This is a comparison chart between the original output data and the filtered output data;

[0068] Fig.21 It is a comparison chart before and after temperature compensation, as well as a data chart of temperature changes over time. DETAILED DESCRIPTION

[0069] like Figure 1 As shown, a method for improving the accuracy of gyro array data of a stable platform proposed in this embodiment includes the following steps:

[0070] Step 1: First screening: Use a low-pass filter to smooth and denoise the signal, and quantify the performance of the gyro unit in the form of variance. This can not only roughly screen out gyro units with poor noise resistance and performance, but also greatly reduce the amount of calculation for subsequent normal distribution curve fitting. The low-pass filter algorithm formula is:

[0071] Y(n)=a*X(n)+(1-a)*Y(n-1)

[0072] Among them, a is the filter coefficient, X(n) is the current input value of the gyroscope, Y(n-1) is the output value after the last gyroscope filter, and Y(n) is the output value of the current gyroscope filter.

[0073] Variance calculation formula:

[0074] D(X)=Var(X)=E[XE(X)] 2

[0075] Adopt the modified formula of variance: D(X)=E(X 2 )-[E(X)] 2 The data set is the gyro angular velocity after 10,000 acquisitions. After the real-time acquired gyro data is sent to the low-pass filter model, the filtered gyro value is obtained and the mean difference E(X), E(X 2 ), the variance of each gyroscope is obtained through the variance deformation formula, and the gyroscope with better performance is screened out through the quick sorting method. The larger the variance, the more severe the gyroscope jitter.

[0076] Step 2: Double screening: In statistics, normality test is crucial for the analysis of data sets because it ensures the validity of many statistical methods. This article uses two methods to visualize the data for normality test: drawing a frequency distribution histogram and visually fitting a QQ graph in the normality test. The QQ graph (Q stands for quantile) is a probability graph.

[0077] The process of drawing the QQ graph is as follows:

[0078] A1. Sort the data set and calculate the empirical quantile;

[0079] A2. Compare all the calculated empirical quantiles with the theoretical quantiles of the standard normal distribution;

[0080] A3. Plot the point pairs on the graph, where the x-axis represents the empirical quantile and the y-axis represents the theoretical quantile;

[0081] A4. Assess the normality of the data by observing the distribution pattern of these points.

[0082] Fit a normal distribution probability density curve:

[0083] Data preprocessing: First, the gyro data set is preprocessed to remove outliers. The outliers are deleted when the absolute value of the difference between the measured value and its average value exceeds 3 times the standard deviation (|x-μ|>3σ);

[0084] Optimization algorithm selection: Maximum likelihood estimation (MLE) is used to estimate the mean μ and standard deviation σ of the normal distribution. By constructing the likelihood function of the data set and maximizing the likelihood function using the gradient descent method, the optimal mean and standard deviation are solved;

[0085] The probability density function of the normal distribution is:

[0086]

[0087] Dataset X = {x1, x2, …, x n}, the likelihood function is the joint probability density function, which is the product of the probability density functions of each data point:

[0088]

[0089] Log-likelihood function:

[0090]

[0091] Set the objective function:

[0092] Maximize the log-likelihood function -l(μ,σ 2 ) In optimization problems, it is usually transformed into minimizing its negative value, that is, maximizing l is equivalent to minimizing -l:

[0093]

[0094] Use gradient descent to calculate J with respect to μ and σ 2 The partial derivative of :

[0095]

[0096] Update μ and σ 2 :

[0097]

[0098] Among them, α is the learning rate, which takes a small positive number (such as 0.01) to control the step size of the update;

[0099] The fitting process is an iterative optimization process, starting from the initial parameter value, and continuously optimizing the parameter value through multiple iterations until the preset convergence condition is met (such as the parameter change is less than a certain threshold or the likelihood function value increase is less than a certain threshold). After each iteration, the change of the parameter value is checked. If the convergence condition is met, the iteration is stopped and the final mean and standard deviation are output as the fitting result.

[0100] Step 3, temperature drift compensation: On the two-axis platform, temperature drift compensation uses the angle differential of the angle encoder to calculate the angular velocity to replace the MEMS gyro for closed-loop speed control. The reason is that the angular velocity calculated by the encoder is the velocity value in the base coordinate system. When the angular velocity is zero, it can be considered that the theoretical output of the gyro should be 0, but the gyro data will cause the speed in the base coordinate system to be not zero due to temperature changes. Therefore, the encoder is used to temporarily replace the gyro for closed-loop speed control. When the speed is 0, the gyro data is collected, which is the current gyro drift angular velocity.

[0101] Drift algorithm: In the full temperature range, when the speed obtained by encoder differentiation is 0, the gyro data on the stable two-axis platform is sampled with a sampling period of 0.001s, and then the temperature drift curve after collection is fitted. The principle is to use the least squares method to determine the goodness of fit R 2 The threshold is used as a segmented fitting point, and the optimal solution of k and b is determined by calculating the optimal extreme point to compensate. The algorithm formula is:

[0102] The linear function is:

[0103] Among them, x i is the original data collected by the gyroscope, k is the slope of the fitted linear function, and b is the intercept of the function;

[0104] Linear loss function mean square error:

[0105] in, It means finding the value of the independent variable that minimizes the function value, because the minimum sum of squared deviations can ensure that each deviation is not large. So the problem comes down to determining The constants k and b in the equation are minimized. The method of determining the coefficients k and b in this way is called the least squares method.

[0106] According to the principle of mean square error of the above linear loss function, Convert to The minimum value of Find the extreme point of the L derivative and calculate the optimal solution of k and b. The calculation formulas for their respective partial derivatives are as follows:

[0107]

[0108] Similarly, we get:

[0109]

[0110] Experimental verification:

[0111] In order to verify the proposed method of improving the accuracy of gyro array data, 8 MEMS gyroscopes of the same model are used to form an array. They have the same manufacturing process and materials, and the same acquisition environment. Because the gyroscope used in this experiment is a two-axis gyroscope, and because its two axes are orthogonal, only the data of one axis is collected to verify the conclusion. The original data of the 8 gyroscopes measured in the experiment are directly averaged, as shown in Figure 2 As shown, its variance is 1.043492. (All the graphs in this experiment are data from 10,000 gyroscope samplings, with a sampling period of 0.001s, gyroscope angular velocity units of rad / s, and graph abscissa units of 0.001s).

[0112] Table 1. Variance of the original data of each gyroscope and the variance after filtering and denoising

[0113] Gyro number variance Variance after low-pass filtering Gyro 1 0.219589 0.080640 Gyro 2 0.0756 0.1101 Gyro 3 0.595654 0.175323 Gyro 4 0.206116 0.003668 Gyro 5 0.642437 0.094620 Gyro 6 0.054598 0.008973 Gyro 7 0.060056 0.010126 Gyro 8 0.880444 0.012839

[0114] The above data show that for the variance of the unfiltered gyroscope, gyroscope 8 has the worst performance and gyroscope 6 has the best performance. However, it is far from enough to sort and screen gyroscopes by this method alone. For the low-frequency jitter of gyroscope 2, sorting and screening only by variance will result in the performance of gyroscope 2 being second only to gyroscope 6 and gyroscope 7. From the data fluctuation chart, it can be seen that the performance of gyroscope 4 is far better than that of gyroscope 2. Therefore, only by variance can we roughly screen out gyroscopes with poor noise resistance and severe high-frequency jitter, but it may not reflect the performance of gyroscopes under low-frequency jitter. In the variance data after low-pass filtering, low-pass filtering does play a good role. Gyroscope 2 is ranked at the back, but the data shows that the performance of gyroscope 7 and gyroscope 8 is similar. From the data fluctuation chart, it can be seen that the jitter of gyroscope 7 is much smaller than that of gyroscope 8, but because of the low-frequency jitter of gyroscope 8, the calculated variance is very small. In summary, it is inaccurate to judge and screen gyroscopes only by variance, so the second screening algorithm is necessary. By setting the elimination rate of the top under its variance to 12.5%, it can be obtained that the second top is eliminated first.

[0115] like Figure 11-Figure 18 As shown in the figure, the gyro data after low-pass filtering is visually fitted. It can be seen from the frequency distribution histogram and the QQ diagram that the normality test of the gyro data after noise reduction is reasonable. Except for gyro 2, the other gyros are very close to normal distribution, so the gyros with better performance can be screened out by setting the threshold of the error interval of their normal distribution. Assuming the standard deviation interval is ±0.1, the probability of each gyro data in this error interval can be obtained, as shown in Table 2 below:

[0116] Table 2. Probability of each gyro data within the ±0.1 error range

[0117] Gyro number Variance after 10 Hz filtering Probability (±0.1 error) Gyro 1 0.080640 27.27％ Gyro 2 0.1101 23.69％ Gyro 3 0.175323 16.64％ Gyro 4 0.003668 86.02％ Gyro 5 0.094620 20.44％ Gyro 6 0.008973 71.98％ Gyro 7 0.010126 59.98％ Gyro 8 0.012839 9.74％

[0118] From Table 2, we can see that gyro 4 has the best performance, and gyro 8 has the worst. From the variance after low-pass filtering, we can see that the variances of gyro 7 and gyro 8 are approximately equal, but the probability of the error intervals of the two gyro is very different, and the performance is very different. The probability of the interval of the normal distribution curve can well reflect the degree of fluctuation of the data near the zero position. The results show that by fitting the normal distribution curve, we can well screen gyro with good performance. After the gyro is screened and the data is fused, the original output data is compared with the data after screening and fusion. Fig. 20 As shown, its variance is 0.000961.

[0119] By setting the gyro elimination rate under its normal distribution curve to 57%, it can be obtained that except for the second gyro, gyro 1, 3, 5, and 8 are eliminated first, leaving gyro 4, 6, and 7 with better performance.

[0120] Before and after temperature drift compensation, comparison chart of the two, temperature change over time Fig.21 As shown in the figure, it can be seen that the data value of the compensated gyro data in the static state is stable near the zero line compared with the uncompensated data.

[0121] The following conclusions are drawn from the above experimental verification process:

[0122] In order to improve the accuracy of the gyro array data of the stable platform and reduce the gyro jitter and drift under temperature changes, the gyro array composed of 8 gyros was screened and the temperature drift curve was fitted using a double screening algorithm of low-pass filtering variance and normal curve fitting, which effectively reduced the gyro jitter and drift problems. From the experimental data, it can be seen that the standard deviation is reduced by 32.97 times in the static state, and the data after temperature drift compensation is also stable at zero. The experiment shows that after screening and compensation, the output data of the gyro array is more accurate. The experimental results show that this screening method and temperature drift compensation method are effective.

[0123] This embodiment uses a double screening algorithm and a temperature drift compensation algorithm to improve the accuracy of the gyro array output data and greatly improve the control effect of the stable platform. First, through the double screening algorithm, the performance indicators of the gyro unit are quantitatively evaluated, and the gyro units with low static jitter are screened out. Secondly, combined with the law of MEMS gyro data changing with temperature, a temperature drift compensation algorithm based on least squares curve fitting is proposed. By modeling and predicting the trend of temperature change, the gyro drift is effectively suppressed. Finally, it is experimentally verified that the screening algorithm proposed in this embodiment reduces the jitter amplitude of the gyro angular velocity output by about 10 times. In the static state, the real angular velocity after temperature drift compensation is also stable near zero. The results show that after the gyro data is processed by the screening algorithm and temperature drift compensation, the control effect of the stable platform can be greatly improved.

[0124] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A method for improving the accuracy of gyro array data of a stable platform, characterized in that: The following steps are involved: S1. First screening: Use a low-pass filter to smooth and denoise the signal, and quantify the performance of the gyro unit in the form of variance. The larger the variance, the more severe the gyro jitter; S2. Double screening: draw frequency distribution histograms in normality tests, visualize fitted QQ plots, and fit normal distribution probability density curves; S3. Temperature drift compensation: The encoder differential speed closed loop obtains the gyro drift value. Within the full temperature range, when the speed obtained by the encoder differential is 0, the gyro data on the stable two-axis platform is sampled with a sampling period of 0.001s. The temperature drift curve after the collection is then fitted. The principle uses the least squares method to determine the goodness of fit R 2 The threshold is used as the piecewise fitting point and compensation is performed.

2. A method for improving the accuracy of gyro array data of a stable platform according to claim 1, characterized in that: In S1, the low-pass filtering algorithm formula is: Y(n)=a*X(n)+(1-a)*Y(n-1) Among them, a is the filter coefficient, X(n) is the current input value of the gyroscope, Y(n-1) is the output value after the last gyroscope filter, and Y(n) is the output value of the current gyroscope filter.

3. A method for improving the accuracy of gyro array data of a stable platform according to claim 2, characterized in that: Variance calculation formula: D(X)=Var(X)=E[XE(X)] 2 Adopt the modified formula of variance: D(X)=E(X 2 )-[E(X)] 2 The data set is the gyro angular velocity after 10,000 acquisitions. After the real-time acquired gyro data is sent to the low-pass filter model, the filtered gyro value is obtained and the mean difference E(X), E(X 2 ), the variance of each gyroscope is obtained through the variance transformation formula, and the gyroscope with better performance is screened out through the quick sorting method.

4. A method for improving the accuracy of gyro array data of a stable platform according to claim 3, characterized in that: In S2, the drawing process of the QQ graph is as follows: A1. Sort the data set and calculate the empirical quantile; A2. Compare all the calculated empirical quantiles with the theoretical quantiles of the standard normal distribution; A3. Plot the point pairs on the graph, where the x-axis represents the empirical quantile and the y-axis represents the theoretical quantile; A4. Assess the normality of the data by observing the distribution pattern of these points.

5. A method for improving the accuracy of gyro array data of a stable platform according to claim 4, characterized in that: When fitting the normal distribution probability density curve, the gyro dataset is first preprocessed to remove outliers; The maximum likelihood estimation is used to estimate the mean μ and standard deviation σ of the normal distribution. By constructing the likelihood function of the data set and maximizing the likelihood function using the gradient descent method, the optimal mean and standard deviation are solved. The probability density function of the normal distribution is: Dataset X = {x1, x2, …, x n }, the likelihood function is the joint probability density function, which is the product of the probability density functions of each data point: Log-likelihood function: Set the objective function: Maximize the log-likelihood function -l(μ,σ 2 ) In optimization problems, it is usually transformed into minimizing its negative value, that is, maximizing l is equivalent to minimizing -l: Use gradient descent to calculate J with respect to μ and σ 2 The partial derivative of : Update μ and σ 2 : Among them, α is the learning rate, which is used to control the step size of the update; The fitting process is an iterative optimization process, starting from the initial parameter value, and continuously optimizing the parameter value through multiple iterations until the preset convergence conditions are met.

6. A method for improving the accuracy of gyro array data of a stable platform according to claim 5, characterized in that: In S3, the gyro drift value is obtained for the encoder differential speed closed loop. When the speed is 0, the gyro data is collected, which is the current gyro drift angular velocity.

7. A method for improving the accuracy of gyro array data of a stable platform according to claim 6, characterized in that: The optimal solution of k and b is determined by calculating the optimal extreme point to compensate. The algorithm formula is: The linear function is: Among them, x i is the original data collected by the gyroscope, k is the slope of the fitted linear function, and b is the intercept of the function; Linear loss function mean square error: in, It means finding the value of the independent variable that makes the function reach the minimum value; According to the principle of mean square error of the above linear loss function, Convert to The minimum value of Find the extreme point of the L derivative and calculate the optimal solution of k and b. The calculation formulas for their respective partial derivatives are as follows:

Images