A magnetic induction posture detection system suitable for automation equipment

Abstract

The present invention relates to the field of detection technology, and in particular to a magnetic induction attitude detection system suitable for automation equipment, including a magnetic induction sensor module, an inertial measurement module, and an adaptive calibration module, which establishes a dynamic model to compensate for the offset vector and scale factor errors of the magnetic induction sensor module and the inertial measurement module; a data fusion and processing module, which receives raw data from the magnetic induction sensor module and the inertial measurement module, performs denoising and smoothing processing through a variational Bayesian filter, and fuses the data to obtain attitude information. The present invention can significantly improve detection accuracy, stability and reliability, and provide strong technical support for attitude detection of automation equipment.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and in particular to a magnetic induction posture detection system suitable for automation equipment. Background Art

[0002] Magnetic induction attitude detection is a sensor device that uses the earth's magnetic field or a specific magnetic field to determine the direction, position or attitude of an object. These devices can be used to measure the azimuth relative to the earth's magnetic field, and in some cases, can also provide information on pitch and roll angles. Magnetic induction attitude detection is widely used in automation equipment such as industrial robots, automatic guided vehicles and drones to achieve positioning, orientation and navigation functions.

[0003] The current shortcomings are as follows:

[0004] 1) In order to accurately determine the azimuth angle, it is usually necessary to perform a complex initialization process on the magnetic induction sensor and perform regular calibration to compensate for problems such as hardware offset and scale factor error.

[0005] 2) In practical applications, it is often interfered by various factors, such as environmental factors, sensor internal noise, etc. These noises will reduce the accuracy and reliability of the detection signal, thereby affecting the accuracy of posture detection. Summary of the invention

[0006] The present invention provides a magnetic induction posture detection system suitable for automation equipment, which can significantly improve the detection accuracy, stability and reliability, and provide strong technical support for the posture detection of automation equipment.

[0007] In order to achieve the purpose of the present invention, the technical solution adopted is: a magnetic induction posture detection system suitable for automation equipment, the system includes:

[0008] Magnetic induction sensor module, used to measure magnetic field strength;

[0009] Inertial measurement module, including accelerometer and gyroscope, used to measure gravity acceleration and angular velocity respectively;

[0010] An adaptive calibration module, which establishes a dynamic model to compensate for the offset vector and scale factor errors of the magnetic induction sensor module and the inertial measurement module;

[0011] The data fusion and processing module receives the raw data from the magnetic induction sensor module and the inertial measurement module, performs denoising and smoothing processing through a variational Bayesian filter, and fuses the data to obtain attitude information.

[0012] As an optimization solution of the present invention, the specific steps of the adaptive calibration module establishing a dynamic model to compensate for the offset vector of the magnetic induction sensor module and the inertial measurement module are:

[0013] 1) Obtain the initial offset vector and scale factor error of the magnetic induction sensor, accelerometer and gyroscope;

[0014] 2) Using a variational Bayesian filter to denoise and smooth the initial offset vector and scale factor errors of the magnetic induction sensor, accelerometer, and gyroscope;

[0015] 3) Construct a dynamic model of the offset vector and the scale factor error. The dynamic model is a nonlinear autoregressive model, specifically:

[0016] 31) Define the state variables and observation variables of the magnetic induction sensor, accelerometer and gyroscope;

[0017] 32) Establish the state transfer equations of magnetic induction sensor, accelerometer and gyroscope;

[0018] 33) Establish observation equations for magnetic induction sensors, accelerometers and gyroscopes;

[0019] 34) Estimate the parameters of a nonlinear autoregressive model using historical data along with an initial offset vector and scale factor errors.

[0020] As an optimization solution of the present invention, the magnetic induction posture detection system suitable for automation equipment also includes an environment perception module, which dynamically adjusts the dynamic model of the offset vector and proportional factor error in the adaptive calibration module according to the data provided by the environment perception module.

[0021] As an optimization solution of the present invention, in step 34), specifically:

[0022] 341) Collect historical data of magnetic induction sensors, accelerometers and gyroscopes, including their measurement values ​​and corresponding timestamps;

[0023] 342) setting initial values ​​for parameters of the nonlinear autoregressive model according to the initial offset vector and the scale factor error, the parameters of the nonlinear autoregressive model including coefficients of the state transfer equation, coefficients of the observation equation, and variance of the noise term;

[0024] 343) Define an objective function to measure the difference between the model prediction value and the actual observation value. The objective function is to maximize the log-likelihood function.

[0025] 344) Use gradient descent method to update parameters, divide the data into training set and validation set, use validation set to evaluate model performance, and adjust parameters;

[0026] 345) When the iteration reaches the preset convergence condition, the iteration is stopped. The parameter estimation value obtained at this time is the final parameter of the nonlinear autoregressive model.

[0027] As an optimization solution of the present invention, the data fusion and processing module receives the raw data from the magnetic induction sensor module and the inertial measurement module, and performs denoising and smoothing processing through a variational Bayesian filter. Specifically, the denoising and smoothing processing is as follows:

[0028] A. Preprocess the raw data of the magnetic induction sensor module and the inertial measurement module;

[0029] B. Establish a variational Bayesian filtering model for the preprocessed data and set the initial parameters of the model, including the prior distribution, the covariance matrix of the observation noise and the covariance matrix of the process noise;

[0030] C. Use the variational Bayesian filtering algorithm to perform iterative calculations, update the probability distribution of the target state, and output the denoised and smoothed data.

[0031] As an optimization solution of the present invention, preprocessing of the raw data of the magnetic induction sensor module and the inertial measurement module includes timestamp synchronization, missing value processing and abnormal value detection.

[0032] The present invention has positive effects: 1) By introducing an adaptive calibration module and an environmental perception module, the system can compensate for the offset vector and scale factor errors of the magnetic induction sensor and the inertial measurement unit (IMU) in real time. In particular, when environmental conditions change, such as when there is external magnetic field interference or temperature fluctuation, the data provided by the environmental perception module enables the dynamic model to be adjusted in real time, thereby ensuring that the high accuracy of posture detection is not affected by external factors.

[0033] 2) The present invention uses a variational Bayesian filter to denoise and smooth the sensor data, and combines it with a nonlinear autoregressive model for parameter estimation and optimization. This not only reduces noise interference, but also improves the robustness of the system in complex environments and ensures stable performance during long-term operation. In addition, the use of gradient descent to update parameters and the evaluation of model performance through a validation set further enhance the reliability and generalization ability of the system.

[0034] 3) The present invention greatly simplifies this process through an automated adaptive calibration mechanism. The system can self-adjust and optimize during operation, reducing the need for manual intervention and also reducing maintenance costs caused by hardware aging or environmental changes.

[0035] 4) The present invention receives raw data from multiple sensors through a data fusion and processing module, performs preprocessing (such as timestamp synchronization, missing value filling and outlier detection), and then uses a variational Bayesian filtering algorithm to achieve efficient data fusion. This method not only improves the quality of single sensor data, but also integrates multi-source information, providing more complete and accurate posture information support for automation equipment.

[0036] 5) The magnetic induction attitude detection system of the present invention is particularly suitable for use in automation equipment such as industrial robots, automatic guided vehicles (AGVs), and drones. It can not only meet the requirements of these devices for high-precision positioning, orientation, and navigation, but also adapt to various working environments. Whether it is indoor or outdoor operation, static or dynamic operating conditions, the system can perform well, providing a solid guarantee for the attitude control of automation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0038] Figure 1 It is a principle block diagram of the present invention.

[0039] Figure 2 It is a flow chart of the adaptive calibration module establishing a dynamic model to compensate for the offset vectors of the magnetic induction sensor module and the inertial measurement module. DETAILED DESCRIPTION

[0040] like Figure 1 As shown, the present invention discloses a magnetic induction posture detection system suitable for automation equipment, comprising:

[0041] Magnetic induction sensor module, used to measure magnetic field strength;

[0042] Inertial measurement module, including accelerometer and gyroscope, used to measure gravity acceleration and angular velocity respectively;

[0043] Accelerometer: used to measure gravity acceleration and determine attitude angles (pitch and roll angles).

[0044] Gyroscope: used to measure angular velocity and provide fast dynamic response in a short period of time.

[0045] The adaptive calibration module establishes a dynamic model to compensate for the offset vector and scale factor errors of the magnetic induction sensor module and the inertial measurement module; Figure 2 As shown, the specific steps of the adaptive calibration module to establish a dynamic model to compensate for the offset vector of the magnetic induction sensor module and the inertial measurement module are:

[0046] 1) Obtain the initial offset vector and scale factor error of the magnetic induction sensor, accelerometer and gyroscope;

[0047] The initial offset vector of the magnetic induction sensor is B offset , which is a three-dimensional vector, representing the output value of the magnetic induction sensor in the absence of a magnetic field. offset =[b mx , b my , b mz ] T , b mx represents the initial offset of the magnetic induction sensor on the x-axis, b my represents the initial offset of the magnetic induction sensor on the y-axis, b mz Indicates the initial offset of the magnetic induction sensor on the z-axis.

[0048] The scale factor error of the magnetic induction sensor is k B , which represents the proportional relationship between the magnetic induction sensor output and the actual magnetic field strength.

[0049] The initial offset vector of the accelerometer is A offset , which is also a three-dimensional vector, represents the output value of the accelerometer in a static state. offset =[a ox , a oy , a oz ] T , a ox represents the initial offset of the accelerometer on the x-axis, a oy represents the initial offset of the accelerometer on the y-axis, a oz Indicates the initial offset of the accelerometer on the z-axis.

[0050] The scale factor error of the accelerometer is k A , which represents the proportional relationship between the accelerometer output and the actual acceleration.

[0051] The initial offset vector of the gyroscope is G offset , represents the output value of the gyroscope in a non-rotating state. G offset =[g ox , g oy , g oz ] T , g ox represents the initial offset of the gyroscope on the x-axis, g oy represents the initial offset of the gyroscope on the y-axis, g oz Indicates the initial offset of the gyroscope on the z-axis.

[0052] The scale factor error of the gyroscope is k G , which represents the proportional relationship between the gyroscope output and the actual angular velocity.

[0053] 2) Using a variational Bayesian filter to denoise and smooth the initial offset vector and scale factor errors of the magnetic induction sensor, accelerometer, and gyroscope;

[0054] 3) Construct a dynamic model of the offset vector and the scale factor error. The dynamic model is a nonlinear autoregressive model, specifically:

[0055] 31) Define the state variables and observation variables of the magnetic induction sensor, accelerometer and gyroscope; the state variable x k Includes offset vector and scale factor errors.

[0056]

[0057] Observed variable z k , raw data obtained directly from the sensor. z m，k is the observation value of the magnetic induction sensor at time step k; z a，k is the observed value of the accelerometer at time step k; z g，k is the observation value of the gyroscope at time step k.

[0058] 32) Establish the state transfer equations of magnetic induction sensor, accelerometer and gyroscope;

[0059] The state transfer equation is: k =Fx k-1 +w k , where: F is the state transfer matrix, W k ~N(0,Q k ) is the process noise, and Q k is the process noise covariance matrix.

[0060] 33) Establish observation equations for magnetic induction sensors, accelerometers and gyroscopes;

[0061] The observation equation is: k =H(x k )+v k ; where H(·) is the observation function, v k ~N(0,R k ) is the process noise, and R k is the observation noise covariance matrix.

[0062] H(·) is the observation function, specifically:

[0063] Magnetic induction sensor: z m，k =k B，k (m true，k +B offset，k )+v m，k, where z m，k is the observed value of the magnetic induction sensor at time step k, m true，k is the real magnetic field intensity vector at time step k, v true，k is the observation noise vector at time step k, representing the inevitable random error in the measurement process. m，k ~N(0,R m ), where R m is the observation noise covariance matrix.

[0064] For the accelerometer:

[0065] z a，k =k A，k (a true，k +A offset，k )+v a，k

[0066] For gyroscope: z g，k =k G，k (ω true，k +G offset，k )+v g，k

[0067] 34) Estimate the parameters of a nonlinear autoregressive model using historical data along with an initial offset vector and scale factor errors.

[0068] In step 34), specifically:

[0069] 341) Collect historical data of the magnetic induction sensor, accelerometer and gyroscope, including their measurement values ​​and corresponding timestamps, and set the sampling frequencies of the magnetic induction sensor, accelerometer and gyroscope.

[0070] 342) setting initial values ​​for parameters of the nonlinear autoregressive model according to the initial offset vector and the scale factor error, the parameters of the nonlinear autoregressive model including coefficients of the state transfer equation, coefficients of the observation equation, and variance of the noise term;

[0071] The estimation of the parameters of the nonlinear autoregressive model directly determines whether the module can successfully compensate the offset vector of the magnetic induction sensor module and the inertial measurement module (IMU) and ensure that high-precision measurement results are continuously provided over a long period of time.

[0072] 343) Define an objective function to measure the difference between the model prediction value and the actual observation value. The objective function is to maximize the log-likelihood function.

[0073] The function that maximizes the log-likelihood is:

[0074]

[0075] Among them, θ is the model parameter, xi is the state vector at time step i, z i is the observation vector at time step i, R i is the observation noise covariance matrix. N is the number of data points, n is the observation vector z i Dimension.

[0076] Initialization parameters: Set initial values ​​based on the initial offset vector and scale factor error.

[0077] State transfer matrix F;

[0078] Observation function: H;

[0079] Process noise covariance matrix: Q;

[0080] Observation noise covariance matrix: R;

[0081] Compute Log-Likelihood: Computes the value of the log-likelihood function using the current parameters.

[0082] Gradient calculation: Calculate the gradient of the log-likelihood function with respect to each parameter.

[0083] Parameter update: Update parameters according to the gradient, using the gradient descent method:

[0084] α is the learning rate.

[0085] Until the change in the log-likelihood function is less than a certain threshold or the maximum number of iterations is reached.

[0086] L(θ|D): Log-likelihood function, which measures the degree of match between the model prediction value and the actual observation value.

[0087] 344) Use gradient descent method to update parameters, divide the data into training set and validation set, use validation set to evaluate model performance, and adjust parameters;

[0088] 345) When the iteration reaches the preset convergence condition, the iteration is stopped. The parameter estimation value obtained at this time is the final parameter of the nonlinear autoregressive model.

[0089] The specific steps of the adaptive calibration module establishing a dynamic model to compensate for the scale factor error of the magnetic induction sensor module and the inertial measurement module are the same as the steps of compensating the offset vector, which will not be described in detail here.

[0090] Compensation for offset vector and scale factor errors: By accurately obtaining the initial offset vector and scale factor errors and using a variational Bayesian filter for denoising and smoothing, the accuracy of the sensor output can be effectively improved. A nonlinear autoregressive model is constructed to compensate for the offset vector and scale factor errors of the magnetic induction sensor module and the inertial measurement module in real time, ensuring that high-precision measurement results are continuously provided for a long time. The nonlinear autoregressive model can adapt to changes in different working environments and automatically adjust parameters to maintain optimal performance, thereby enhancing the reliability of the system. The model parameters are optimized by maximizing the log-likelihood function to ensure the best match between the model prediction value and the actual observation value, thereby improving the model's fit and prediction ability. The gradient descent method is used to gradually optimize the parameters so that the model can be continuously improved during the iteration process and finally converge to the optimal solution. The data is divided into a training set and a validation set to ensure that the model not only performs well on the training data, but also has strong generalization ability on new data to avoid overfitting. During the iteration process, the parameters are dynamically adjusted according to the feedback of the validation set to ensure that the model is always in the best state.

[0091] The data fusion and processing module receives the raw data from the magnetic induction sensor module and the inertial measurement module, and performs denoising and smoothing processing through a variational Bayesian filter. Specifically, the denoising and smoothing processing is as follows:

[0092] A. Preprocess the raw data of the magnetic induction sensor module and the inertial measurement module; Preprocess the raw data of the magnetic induction sensor module and the inertial measurement module, including timestamp synchronization, missing value processing and outlier detection. Use linear or spline interpolation methods to synchronize timestamps, fill missing values ​​with the previous valid value or the next valid value, and use the mean and standard deviation to detect outliers. Timestamp synchronization ensures that the data of all sensors are synchronized in time. Missing value processing fills or interpolates missing data points. Outlier detection identifies and handles possible outliers to avoid negative impacts on subsequent processing.

[0093] B. Establish a variational Bayesian filtering model for the preprocessed data and set the initial parameters of the model, including the prior distribution, the covariance matrix of the observation noise and the covariance matrix of the process noise;

[0094] Prior distribution: Set the initial probability distribution p(x0) of the state variable. Usually it is assumed to be a Gaussian distribution:

[0095] p(x0)=N(μ0,Σ0)

[0096] Where: μ0 is the initial mean vector and Σ0 is the initial covariance matrix.

[0097] C. Use the variational Bayesian filtering algorithm to perform iterative calculations, update the probability distribution of the target state, and output the denoised and smoothed data.

[0098] Process noise covariance matrix Q k : Describes the random error introduced during the state transition process. It can be estimated based on empirical values ​​or by analyzing historical data.

[0099] Q k =diag([qB x , qB y , qB z ,qk B , qA x , qA y , qA z ,qk A ,qG x ,qG y ,qG z ,qk G ])Observation noise covariance matrix R k : Describes the random error introduced during the observation process. It can also be estimated based on empirical values ​​or by analyzing historical data.

[0100] Rk=diag([rm x , rm y , rm z ,ra x ,ra y ,ra z , rg x , rg y , rg z ])

[0101] 3. State transfer equation

[0102] x k =Fx k-1 +w k

[0103] Where: F is the state transfer matrix, wk~N(0,Qk) is the process noise.

[0104] 4 Observation equation

[0105] z k =H(x k )+v k

[0106] Where: H(·) is the observation function, vk~N(0, R k ) is the observation noise.

[0107] Use the variational Bayesian filtering algorithm to perform iterative calculations, update the probability distribution of the target state, and output denoised and smoothed data

[0108] 1. Initialization

[0109] Set initial parameters:

[0110] State transfer matrix: F; observation function H; process noise covariance matrix Q; observation noise covariance matrix R, initial state distribution p(x0)=N(μ0,Σ0).

[0111] 2. E step (expectation step)

[0112] Compute the expectation and covariance of the hidden states:

[0113]

[0114] Where: μ k and Σ k The states x are k The expectation and covariance of .

[0115] 3.M step (maximization step)

[0116] Update the model parameters θ to maximize the ELBO:

[0117]

[0118] The ELBO is defined as:

[0119]

[0120] 4. Iterative Update

[0121] Repeat the E and M steps until the convergence condition is met (the log-likelihood change is less than a certain threshold or the maximum number of iterations is reached).

[0122] 5. Output denoised and smoothed data

[0123] Posterior distribution: The final state distribution q(x1:N) provides the denoised and smoothed data.

[0124] Extract results: Extract the state estimate for each time step from the posterior distribution as the final output.

[0125] Letters meaning:

[0126] x k : The state vector at time step k.

[0127] z k : The observation vector at time step k.

[0128] μ0: Initial state mean vector.

[0129] Σ0: Initial state covariance matrix.

[0130] Q k : Process noise covariance matrix.

[0131] R k : Observation noise covariance matrix.

[0132] F: state transfer matrix.

[0133] H(·): observation function.

[0134] w k :Process noise, subject to N(0,Q k ).

[0135] v k : Observation noise, subject to N(0, R k ).

[0136] L(q, θ): Evidence lower bound (ELBO), a measure of model fit.

[0137] θ: set of model parameters.

[0138] N: number of data points.

[0139] The state estimate at time step k.

[0140] The magnetic induction attitude detection system for automation equipment also includes an environment perception module, which dynamically adjusts the dynamic model of the offset vector and the scale factor error in the adaptive calibration module according to the data provided by the environment perception module. k Represents the environmental perception data vector at time step k, including but not limited to: external magnetic field strength e m，k , temperature e T，k and vibration / shock v，k The environmental perception module detects significantly increased external magnetic field interference and temperature changes. The system automatically adjusts the dynamic model in the adaptive calibration module based on these changes to compensate for the offset vector and scale factor errors in the new environment, thereby ensuring that the attitude detection accuracy is not affected.

[0141] Example:

[0142] Industrial robot arms need to know their posture precisely in order to operate accurately. The robot is equipped with the above-mentioned magnetic induction posture detection system:

[0143] At the beginning of installation, the system is initialized according to known environmental conditions, including setting the initial offset vector and scale factor error.

[0144] When the robot starts working, the magnetic induction sensor, accelerometer and gyroscope continuously collect data. The data first goes through a preprocessing step to ensure that the timestamps are synchronized, missing values ​​are correctly filled, and outliers have been removed.

[0145] The adaptive calibration module constantly monitors changes in sensor output and automatically adjusts internal parameters to compensate for any new offsets or errors, ensuring long-term stable performance.

[0146] The data fusion and processing module receives the pre-processed data from each sensor, applies the variational Bayesian filtering algorithm for further denoising and smoothing, and finally outputs high-precision posture information.

[0147] Based on these high-precision posture data, the robot control system can make more accurate motion instructions, improving work efficiency and operating accuracy.

[0148] The environmental perception module dynamically adjusts the model parameters to maintain high-precision posture detection and keep the error within a small range even in the presence of significant external magnetic field interference (such as an increase of 100μT) or temperature fluctuations (±20℃).

[0149] The variational Bayesian filtering algorithm is used for denoising and smoothing, which can effectively reduce noise interference, improve the robustness of the system in complex environments, and make the output signal more stable and reliable.

[0150] By preprocessing the raw data of the magnetic induction sensor module and the inertial measurement module (such as timestamp synchronization, missing value processing and outlier detection), and then using the variational Bayesian filtering algorithm to achieve efficient data fusion, the quality of a single sensor data is improved, and at the same time, multi-source information is integrated to provide more complete and accurate attitude information support for automation equipment.

[0151] Through this magnetic induction posture detection system for automation equipment, the performance improvements are reflected in:

[0152] Azimuth error: reduced from ±2° to within ±0.5°.

[0153] Pitch / roll angle error: reduced from ±1° to within ±0.3°.

[0154] Magnetic induction intensity error:

[0155] Offset error: reduced from ±50μT to within ±10μT.

[0156] Proportional factor error: from ±5% to ±10% controlled within ±1%.

[0157] Total error range: reduced from ±60μT to ±80μT to within ±20μT.

[0158] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A magnetic induction posture detection system suitable for automation equipment, characterized in that: The system comprises: Magnetic induction sensor module, used to measure magnetic field strength; Inertial measurement module, including accelerometer and gyroscope, used to measure gravity acceleration and angular velocity respectively; An adaptive calibration module, which establishes a dynamic model to compensate for the offset vector and scale factor errors of the magnetic induction sensor module and the inertial measurement module; The data fusion and processing module receives the raw data from the magnetic induction sensor module and the inertial measurement module, performs denoising and smoothing processing through a variational Bayesian filter, and fuses the data to obtain attitude information.

2. A magnetic induction posture detection system suitable for automation equipment according to claim 1, characterized in that: The specific steps of the adaptive calibration module to establish a dynamic model to compensate for the offset vector of the magnetic induction sensor module and the inertial measurement module are as follows: 1) Obtain the initial offset vector and scale factor error of the magnetic induction sensor, accelerometer and gyroscope; 2) Using a variational Bayesian filter to denoise and smooth the initial offset vector and scale factor errors of the magnetic induction sensor, accelerometer, and gyroscope; 3) Construct a dynamic model of the offset vector and the scale factor error. The dynamic model is a nonlinear autoregressive model, specifically: 31) Define the state variables and observation variables of the magnetic induction sensor, accelerometer and gyroscope; 32) Establish the state transfer equations of magnetic induction sensor, accelerometer and gyroscope; 33) Establish observation equations for magnetic induction sensors, accelerometers and gyroscopes; 34) Estimate the parameters of a nonlinear autoregressive model using historical data along with an initial offset vector and scale factor errors.

3. A magnetic induction posture detection system suitable for automation equipment according to claim 2, characterized in that: The magnetic induction attitude detection system suitable for automation equipment also includes an environment perception module, which dynamically adjusts the dynamic model of the offset vector and scale factor error in the adaptive calibration module according to the data provided by the environment perception module.

4. A magnetic induction posture detection system suitable for automation equipment according to claim 3, characterized in that: In step 34), specifically: 341) Collect historical data of magnetic induction sensors, accelerometers and gyroscopes, including their measurement values ​​and corresponding timestamps; 342) setting initial values ​​for parameters of the nonlinear autoregressive model according to the initial offset vector and the scale factor error, the parameters of the nonlinear autoregressive model including coefficients of the state transfer equation, coefficients of the observation equation, and variance of the noise term; 343) Define an objective function to measure the difference between the model prediction value and the actual observation value. The objective function is to maximize the log-likelihood function. 344) Use gradient descent method to update parameters, divide the data into training set and validation set, use validation set to evaluate model performance, and adjust parameters; 345) When the iteration reaches the preset convergence condition, the iteration is stopped. The parameter estimation value obtained at this time is the final parameter of the nonlinear autoregressive model.

5. A magnetic induction posture detection system suitable for automation equipment according to claim 4, characterized in that: The data fusion and processing module receives the raw data from the magnetic induction sensor module and the inertial measurement module, and performs denoising and smoothing processing through a variational Bayesian filter. Specifically, the denoising and smoothing processing is as follows: A. Preprocess the raw data of the magnetic induction sensor module and the inertial measurement module; B. Establish a variational Bayesian filtering model for the preprocessed data and set the initial parameters of the model, including the prior distribution, the covariance matrix of the observation noise and the covariance matrix of the process noise; C. Use the variational Bayesian filtering algorithm to perform iterative calculations, update the probability distribution of the target state, and output the denoised and smoothed data.

6. A magnetic induction posture detection system suitable for automation equipment according to claim 5, characterized in that: Preprocessing of the raw data from the magnetic induction sensor module and the inertial measurement module includes timestamp synchronization, missing value processing, and outlier detection.

Images