Embedded intelligent bearing fault diagnosis method based on adaptive feature migration fusion

Abstract

The invention relates to the technical field of fault diagnosis, and particularly discloses a bearing fault diagnosis method based on adaptive feature migration fusion. According to the method, an adaptive feature migration fusion algorithm is utilized, and multi-mode signals of acceleration, load, temperature and rotating speed are comprehensively processed, so that accurate fault diagnosis of the embedded intelligent bearing under a complex working condition is realized. Aiming at the problems that a traditional diagnosis method is difficult in signal feature extraction and large in data distribution difference under different working conditions, the invention provides a self-adaptive feature migration fusion algorithm, and the robustness and expression ability of fault features in different working conditions and data sets are improved by constructing a migration learning model and combining a feature selection and fusion mechanism. Compared with the dependence of a traditional deep learning algorithm on large-scale labeled samples, the method can efficiently utilize a small amount of label data, the requirements for hardware computing power and data resources are remarkably reduced, and an efficient and flexible embedded bearing fault diagnosis method is provided for online fault diagnosis of an industrial site.

Description

Technical Field

[0001] The present invention relates to the technical field of large bearing condition monitoring, and particularly to an embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion. Background Art

[0002] Large bearings are important core components of rotating machinery, and their operating status is directly related to the performance, reliability, and safety of the equipment. Although embedded intelligent bearings achieve efficient data acquisition close to the signal source by embedding micro sensors in the bearing body and reduce signal attenuation and improve the signal-to-noise ratio by integrating multiple sensor modules, there are still deficiencies in multi-modal signal processing and feature fusion. The multi-modal signal fusion algorithm lacks the ability to optimize weights under dynamic working conditions. In addition, the existing technologies still have limitations in signal feature extraction and diagnostic robustness, and it is difficult to meet the real-time diagnosis requirements of industrial sites.

[0003] Therefore, the present invention proposes an embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion. By combining the adaptive feature fusion algorithm with transfer learning technology, this method can dynamically optimize the fusion weights of different sensor signals, effectively extract key features, and improve the accuracy and robustness of diagnosis. In addition, the transfer learning technology enhances the adaptability under complex working conditions, and even with only a small number of labeled samples, it can achieve efficient fault diagnosis performance. Compared with traditional methods, the present invention not only improves the accuracy of early bearing fault diagnosis, but also provides an efficient, flexible, and low-cost embedded solution for industrial sites. Summary of the Invention

[0004] The present invention discloses an embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion, aiming to solve the problems of insufficient signal feature extraction ability and poor diagnostic robustness in embedded intelligent bearings. By combining the adaptive feature fusion algorithm and transfer learning technology, this method significantly improves the processing ability of multi-modal sensor signals and the fault diagnosis performance under complex working conditions.

[0005] Aiming at the problems of poor adaptability of traditional methods under complex working conditions and difficulty in extracting early fault features, the present invention designs a dynamically optimized adaptive feature transfer and fusion strategy and develops an efficient embedded fault diagnosis architecture.

[0006] An embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion of the present invention includes the following steps:

[0007] Step S1, multi-modal sensor signal acquisition: Through the acceleration sensor, load sensor, temperature sensor, and speed sensor in the embedded intelligent bearing, multi-modal data acquisition of the bearing operating status is realized, providing a comprehensive data basis for fault diagnosis.

[0008] Step S2, Signal preprocessing: Denoise, normalize, and extract features from the collected multimodal signals. The Kalman filtering technique is used to improve the signal quality, effectively suppress background noise interference, and increase the signal-to-noise ratio of the signals.

[0009] Step S3, Adaptive feature fusion: Propose a feature fusion algorithm based on an adaptive weighting mechanism. Combine the characteristics of multi-sensor signals and dynamically adjust the fusion weights to achieve efficient extraction of key fault features and improve the integrity and accuracy of diagnostic information.

[0010] Step S4, Application of transfer learning technology: Aiming at the problem of large differences in data distribution in industrial sites, transfer learning technology is used to optimize the diagnostic model. By retaining the feature extraction parameters in the pre-trained model and fine-tuning the target working conditions, the model can quickly adapt to diagnostic tasks under different working conditions and achieve efficient and accurate fault diagnosis with only a small number of labeled samples.

[0011] Step S5, Embedded deployment and real-time diagnosis: Design an embedded processing architecture, deploy the adaptive feature fusion algorithm and the transfer learning model to the embedded hardware to achieve real-time monitoring and fault diagnosis of the bearing operating state, and output results including normal state and rolling element fault diagnosis.

[0012] In step S2, the Kalman filtering algorithm is used to dynamically adjust the filtering parameters according to the real-time working conditions to remove random noise interference in the sensor signals.

[0013] In step S2, the normalization process uniformly converts different modal signals into a standard range for subsequent algorithm processing.

[0014] The extracted feature values in step S2 include, but are not limited to: root mean square value, standard deviation, kurtosis of the vibration signal; mean value, change rate of the temperature signal; instantaneous frequency, fluctuation range of the rotational speed signal; change frequency and load value of the load signal; the extracted feature values constitute the preliminary feature vector of each sensor.

[0015] In step S3, adaptive weighted fusion is performed on the extracted multimodal features, and the weight assignment is based on the following principles: noise level and variance information of the modal signals; signals with smaller variances have higher weights and greater signal contributions.

[0016] In step S4, the construction of the pre-trained model uses a deep learning model based on large-scale bearing operation data as the pre-trained model to learn the association between bearing operating states and fault features through extensive historical data.

[0017] The pre-trained model includes a Convolutional Neural Network (CNN) for extracting time-domain and frequency-domain features, combined with a fully connected layer to achieve a preliminary classification of fault types.

[0018] In step S5, the embedded hardware deployment is to deploy the optimized transfer learning model and the adaptive feature fusion algorithm to an embedded hardware platform, such as an ARM processor or an industrial single-chip microcomputer.

[0019] The embedded hardware receives the data collected by the sensor in real time. First, it passes through the signal and processing module, then extracts the key feature values through the feature extraction module, and uses the adaptive feature fusion algorithm to generate a comprehensive feature vector. The comprehensive feature vector is used as the input of the optimized transfer model, and the diagnostic results, including the normal state and the rolling element fault results, are output through the classification layer.

[0020] In step S1, the roller parameter signals are collected by multiple sensors, and the improved adaptive feature fusion algorithm is used to fuse the roller parameter signals collected by multiple sensors. There are 4 sensors in the embedded intelligent bearing to collect 4 parameters of roller vibration acceleration, load, temperature and speed. The weighting coefficients of each sensor are different, where respectively represent the variances of the 4 sensors. According to the roller parameter signal values Y 1 , Y 2 , Y 3 , Y 4 , the optimal weighting factors V 1 , V 2 , V 3 , V 4 are found in an adaptive form, and the optimal fused value of the roller parameter signals is output. The expression is as follows:

[0021]

[0022] The conditions satisfied by the weighting factors are:

[0023]

[0024] The calculation formula for the total mean square error is as follows:

[0025]

[0026] Since the roller parameter signals Y 1 , Y 2 , Y 3 , Y 4 collected by each sensor are in an independent form, and the estimated value Y has no bias. At this time, E(Y - Y q )(Y - Y o) = 0, (q = 1, 2, 3, 4, o = 1, 2, 3, 4; q ≠ 0), so the simplified formula for the total mean square error is as follows:

[0027]

[0028] When the total mean square error is at its minimum, the calculation formula for the weighting factor is:

[0029]

[0030] The expression for the minimum mean square error is as follows:

[0031]

[0032] When the estimated value is a constant, the mean value of the roller parameter signal collected by the sensor can be used as the optimal estimated value at this time, and its expression is:

[0033]

[0034] Among them: The total number of time points for collecting the roller parameter signal is described by g, and β represents the equalization factor;

[0035] From the above formula, the optimal estimated value of the roller parameter signal is obtained, and the correlation of the roller parameter signal collected by the sensor at time g is used to reduce the random error generated when the sensor measures the roller parameter signal. Therefore, the expression for the estimated value of the comprehensive feature true value is:

[0036]

[0037] The expression for the total mean square error is as follows:

[0038]

[0039] When implementing the true value estimation through Y′ q (g), g is directly proportional to the number of roller parameter signals collected. When the number of roller parameter signals collected increases, g increases, and at this time σ 2 will decrease accordingly; The value of the equalization factor affects the solution efficiency of the optimal estimated value and the signal fusion effect. The above method improves the processing ability of the roller parameter signal.

[0040] The present invention provides an embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion. By comprehensively processing multi-modal signals such as acceleration, temperature, speed, and load, key features are extracted, and fault classification is achieved through an adaptive feature transfer and fusion algorithm, improving the diagnostic stability and accuracy of the model under complex working conditions.

[0041] The present invention introduces an adaptive weighting mechanism to dynamically optimize the weight allocation of modal signals, making full use of the diagnostic potential of multi-modal signals and solving the problems of fixed weights and poor adaptability in traditional algorithms. At the same time, transfer learning technology reduces the dependence on large-scale labeled data and can still quickly adapt to multiple working conditions with a small amount of data, improving the practicality of the model.

[0042] Compared with traditional methods, the present invention has higher classification accuracy under complex working conditions and significantly reduces the computational complexity. When deployed in an embedded hardware system, it has the advantages of low cost, high efficiency, and real-time performance, meeting the real-time and reliability requirements of fault diagnosis in industrial fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is the overall flowchart of the embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion;

[0044] Figure 2 is the schematic diagram of the transfer learning structure;

[0045] Figure 3 is the schematic diagram of the embedded intelligent bearing structure. DETAILED DESCRIPTION OF THE INVENTION

[0046] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention.

[0047] As Figure 1 shown, an embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion realizes real-time monitoring and accurate diagnosis of the bearing operating state by collecting, preprocessing, extracting and fusing multi-modal sensor signals, and optimizing the fault diagnosis task under complex working conditions using a transfer learning model. The specific embodiments of the present invention include the following content:

[0048] An embedded intelligent bearing fault diagnosis method based on adaptive feature transfer and fusion includes the following steps:

[0049] Step S1, multi-modal sensor signal acquisition: Through the acceleration sensor, load sensor, temperature sensor, and speed sensor in the embedded intelligent bearing, multi-modal data acquisition of the bearing operating state is realized, providing a comprehensive data basis for fault diagnosis.

[0050] Step S2, signal preprocessing: Denoising, normalizing, and feature extraction are performed on the collected multi-modal signals. The Kalman filtering technology is used to improve the signal quality, effectively suppress background noise interference, and improve the signal-to-noise ratio.

[0051] Step S3, Adaptive Feature Fusion: An adaptive weighted mechanism-based feature fusion algorithm is proposed, which combines the characteristics of multi-sensor signals, dynamically adjusts the fusion weights, and thus realizes the efficient extraction of key fault features, improving the integrity and accuracy of diagnostic information.

[0052] Step S4, Application of Transfer Learning Technology: Aiming at the problem of large differences in data distribution in industrial sites, transfer learning technology is used to optimize the diagnostic model. By retaining the feature extraction parameters in the pre-trained model and fine-tuning the target working conditions, the model can quickly adapt to the diagnostic tasks under different working conditions and achieve efficient and accurate fault diagnosis with only a small number of labeled samples.

[0053] Step S5, Embedded Deployment and Real-time Diagnosis: An embedded processing architecture is designed to deploy the adaptive feature fusion algorithm and the transfer learning model onto embedded hardware, realizing real-time monitoring and fault diagnosis of the bearing operating state, and outputting results including normal state and rolling element fault diagnosis.

[0054] In step S2, the Kalman filtering algorithm is adopted to dynamically adjust the filtering parameters according to the real-time working conditions, removing the random noise interference in the sensor signals.

[0055] In step S2, the normalization process uniformly converts different modal signals into the standard range for subsequent algorithm processing.

[0056] The extracted feature values in step S2 include, but are not limited to: the root mean square value, standard deviation, and kurtosis of the vibration signal; the mean value and change rate of the temperature signal; the instantaneous frequency and fluctuation range of the rotational speed signal; the change frequency and load value of the load signal; the extracted feature values form the preliminary feature vector of each sensor.

[0057] In step S3, the extracted multi-modal features are adaptively weighted and fused, and the weight assignment is based on the following principles: the noise level and variance information of the modal signals; signals with smaller variances have higher weights and greater signal contributions.

[0058] In step S4, the construction of the pre-trained model uses a deep learning model based on large-scale bearing operation data as the pre-trained model, and learns the association between the bearing operating state and fault features through extensive historical data.

[0059] The pre-trained model includes a convolutional neural network (CNN) for extracting time-domain and frequency-domain features, combined with a fully connected layer to achieve a preliminary classification of fault types.

[0060] In step S5, the embedded hardware deployment is to deploy the optimized transfer learning model and the adaptive feature fusion algorithm onto an embedded hardware platform, such as an ARM processor or an industrial single-chip microcomputer.

[0061] The embedded hardware receives the data collected by the sensors in real time. First, the data passes through the signal processing module, then the key feature values are extracted by the feature extraction module, and the adaptive feature fusion algorithm is used to generate the comprehensive feature vector. The comprehensive feature vector is used as the input of the optimized transfer model, and the diagnostic results are output through the classification layer, including the normal state and the rolling element fault results.

[0062] After preprocessing the roller parameter signals collected by each sensor, the feature values that can characterize the bearing operation state are extracted.

[0063] The preprocessing includes denoising and normalizing the collected multi-modal signals, and extracting the key feature values to improve the signal quality and feature expression ability. Specifically, it includes: Denoising: The Kalman filtering algorithm is used to dynamically adjust the filtering parameters to adapt to the signal fluctuations under different working conditions, effectively suppressing the background noise interference and improving the signal-to-noise ratio.

[0064] Normalization: Map the signal values of different modalities to a unified range to reduce the influence of data distribution differences on the feature fusion process. The normalization formula is:

[0065] Then feature extraction is carried out. The feature values include but are not limited to: the root mean square value, standard deviation, and kurtosis of the vibration signal; the mean value and change rate of the temperature signal; the instantaneous frequency and fluctuation range of the rotational speed signal; the change frequency and load value of the load signal; the extracted feature values constitute the preliminary feature vector of each sensor.

[0066] The acquisition frequency of the sensor is set to 20Hz to ensure the integrity and fineness of the signal. The acquired data is transmitted to the data processing module through the embedded processing platform.

[0067] The roller parameter signals are collected by multiple sensors, and the roller parameter signals collected by multiple sensors are fused by improving the adaptive feature fusion algorithm. There are 4 sensors in the embedded intelligent bearing that collect 4 parameters of the roller vibration acceleration, load, temperature, and rotational speed. The weighting coefficients of each sensor are different, where respectively represent the variances of the 4 sensors. According to the roller parameter signal values Y 1 , Y 2 , Y 3 , Y 4 , the optimal weighting factors V 1 , V 2 , V 3 , V 4 are found in an adaptive manner, and the optimal fusion value of the roller parameter signal is output The expression is as follows:

[0068]

[0069] The weighting factor satisfies the condition that:

[0070]

[0071] The formula for calculating the total mean square error is as follows:

[0072]

[0073] Since the roller parameter signals Y 1 , Y 2 , Y 3 , Y 4 collected by each sensor are in an independent form, and at the same time, there is no bias in the estimated value Y. At this time, E(Y - Y q )(Y - Y o ) = 0, (q = 1, 2, 3, 4, o = 1, 2, 3, 4; q ≠ 0). Therefore, the simplified formula for the total mean square error is as follows:

[0074]

[0075] When the total mean square error is at its minimum, the formula for calculating the weighting factor is:

[0076]

[0077] The expression for the minimum mean square error is as follows:

[0078]

[0079] When the estimated value is a constant, the mean value of the roller parameter signals collected by the sensor can be used as the optimal estimated value at this time, and its expression is:

[0080]

[0081] Where: the total number of time points for collecting the roller parameter signals is described by g, and β represents the equalization factor.

[0082] From the above formula, the optimal estimated value of the roller parameter signal is obtained, and the correlation of the roller parameter signals collected by the sensor at time g is used to reduce the random error generated when the sensor measures the roller parameter signals. Therefore, the expression for the estimated value of the comprehensive feature true value is:

[0083]

[0084] The expression for the total mean square error is as follows:

[0085]

[0086] When through Y′ q(g) In the case of implementing true value estimation, g is directly proportional to the number of collected roller parameter signals. When the number of collected roller parameter signals increases, g increases, and at this time, σ 2 will decrease accordingly. The value of the equilibrium factor affects the solution efficiency of the optimal estimated value and the signal fusion effect. The above method improves the processing ability of roller parameter signals.

[0087] Transfer learning involves two important concepts: domain and task. The domain can be divided into the source domain D s and the target domain D t , and the task can be divided into the source domain task T s and the target domain task T t . Its purpose is to reduce the generalization error of the target domain prediction model ft(x) under the condition that D s and D t , T s and T t are all different.

[0088] Aiming at the problem of large differences in the data distribution in industrial sites, the present invention also adopts transfer learning technology to map the comprehensive eigenvalue from the source domain to the target domain. Find better representative features in the source domain, and through feature transformation, transform the features of the source domain and the target domain into the same space, in which the data in the target domain has the same distribution.

[0089] As Figure 2 shown, there are large distribution differences between the data in the source domain and the target domain. The mechanical fault knowledge learned from the source domain cannot directly and accurately identify the fault categories in the target domain. Therefore, it is also necessary to use the maximum mean discrepancy (MMD) distance metric function to reduce the differences between the two domains before a better transfer effect can be obtained.

[0090] As a non-parametric distance metric, MMD is usually used to measure the distribution differences of transferable elements. The specific operation process is as follows: MMD is used to measure the distance between two different but related distributions. The MMD distance between the marginal distributions of the source domain and the target domain is:

[0091]

[0092] represents the 2-norm operation in the reproducing kernel Hilbert space, is the mapping function that maps the source domain kernel and target domain features into RKHS data. Rewrite formula (1) into matrix form, then solving is converted into the problem of optimizing the transfer matrix W.

[0093] D(P(X s ), P(X T )) = tr(KWW T KL) = tr(W T KLKW)

[0094] In the formula, tr represents the tracking operation, is K ss , K st the kernel matrix of the element in the i-th row and j-th column, K tt are respectively (K ss ) if = f(x i , x j )(x i , x j ∈ X s ), (K tt ) if = f(x′ i , x′ j )(x′ i , x′ j ∈ X T ),

[0095] where f(x i , x j ) = exp(-||x i - x j || 2 / γ) is the Gaussian kernel function, the bandwidth is the median of the distances between the source domain and the target domain data, q (< p) is the feature dimension extracted by the network, and the element in the i-th row and j-th column of the coefficient matrix L is:

[0096]

[0097] To optimize the transfer matrix W, the minimization problem of MMD is formulated as:

[0098]

[0099] In the formula, λ is the adjustment parameter, I is the identity matrix, e is a column vector, tr(W T W) is the regularization term. To solve the MMD minimization problem, the Lagrangian method is used to construct:

[0100]

[0101] In the formula, KLK + λI -1 The q generalized eigenvectors corresponding to the q non-zero minimum generalized eigenvalues of KHK can be used to obtain the optimal transfer matrix W. Finally, the sample data in the source domain and the target domain are transferred into the data in the same subspace by using the optimal transfer matrix W corresponding to the minimum MMD value.

[0102] Deploy the optimized transfer learning model and the adaptive feature fusion algorithm to an embedded hardware platform (such as an ARM processor or an industrial single-chip microcomputer) to achieve real-time fault diagnosis.

[0103] The above shows and describes the basic principles, features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion, characterized in that: The following steps are involved: Step S1, collecting bearing operation status signals including acceleration, temperature, speed and load through a multimodal sensor in an embedded intelligent bearing, and establishing a multimodal signal data set; Step S2, preprocessing the collected multimodal signals, including denoising, normalization and feature extraction, wherein the denoising adopts a Kalman filter algorithm; Step S3, extracting characteristic values ​​of each modal signal, including the root mean square value, standard deviation, and kurtosis of the vibration signal, the mean value and change rate of the temperature signal, the instantaneous frequency and fluctuation range of the speed signal, and the change frequency and load value of the load signal; Step S4, dynamically adjusting the weight of each modal signal through an adaptive feature fusion algorithm to generate a comprehensive feature vector, wherein the weight allocation is based on the noise level and variance information of the modal signal; Step S5, optimizing the diagnostic model based on transfer learning, using the maximum mean difference (MMD) method to reduce the distribution difference between the source domain and the target domain, and fine-tuning the model in combination with the labeled samples of the target domain; Step S6, deploying the optimized transfer learning model and the adaptive feature fusion algorithm to the embedded hardware platform, and outputting the operating status diagnosis results of the bearing based on the multimodal sensor signals received in real time, including normal status and rolling element fault.

2. The embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion according to claim 1 is characterized in that: In step S4, the adaptive feature fusion algorithm realizes the comprehensive processing of each modal signal by weighted summation; dynamically adjusts the weight distribution of each signal to optimize the effect of signal fusion, generates a more representative comprehensive feature vector, and provides high-quality input data for subsequent fault diagnosis.

3. The embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion according to claim 1 is characterized in that: In step S5, the transfer learning model is constructed based on a deep learning pre-training model, and the pre-training model includes a convolutional neural network CNN for feature extraction and a fully connected layer for fault classification.

4. The embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion according to claim 1 is characterized in that: In step S5, the maximum mean difference (MMD) method is used to reduce the distribution difference between the source domain and the target domain.

5. The embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion according to claim 1 is characterized in that: The embedded hardware platform is an ARM processor or an industrial-grade single-chip microcomputer. The hardware deployment includes embedding a signal preprocessing module, a feature extraction module, an adaptive feature fusion algorithm and a transfer learning model into the embedded hardware platform to realize data collection, processing and real-time fault classification.

6. The embedded intelligent bearing fault diagnosis method based on adaptive feature migration and fusion according to claim 1 is characterized in that: In step S1, roller parameter signals are collected by multiple sensors, and the roller parameter signals collected by the multiple sensors are fused by improving the adaptive feature fusion algorithm; there are 4 sensors in the embedded intelligent bearing to collect 4 parameters of roller vibration acceleration, load, temperature and speed, and the weighting coefficient of each sensor is different, among which Represents the variance of the four sensors respectively, and collects the roller parameter signal value according to each sensor , find the optimal weighting factor through adaptive form , output the optimal roller parameter signal fusion value , The expression is as follows: ； The weighting factors satisfy the following conditions: ； The total mean square error is calculated as follows: ； The roller parameter signals collected by each sensor Independent form, simultaneous valuation There is no deviation at this time , so the simplified formula of the total mean square error is as follows: ； When the total mean square error is the minimum, the weighting factor calculation formula is: ； The minimum mean square error expression is as follows: ； When the estimated value is a constant, the mean value of the roller parameter signal collected by the sensor can be used as the optimal estimated value at this time, and its expression is: ； Among them: The total number of time points for collecting roller parameter signals is describe, represents the equalization factor; The optimal estimated value of the roller parameter signal is obtained from the above formula. The sensor collects the correlation of roller parameter signals at all times, reducing the random error generated when the sensor measures the roller parameter signals. Therefore, the estimated value of the comprehensive characteristic true value is expressed as: ； The total mean square error expression is as follows: ； When passing When implementing true value estimation, It is proportional to the number of collected roller parameter signals. When the number of collected roller parameter signals increases, Increase, at this time will decrease accordingly; the value of the balancing factor affects the efficiency of solving the optimal estimate and the signal fusion effect. The above method is used to improve the roller parameter signal processing capability.

Images