Ring gear error elimination method based on BPNN

Abstract

The present invention discloses a ring gear error elimination method based on BPNN. First, the time difference Δt between two adjacent teeth is normalized and preprocessed. Then, a constructed BP neural network is used to calculate and eliminate the ring gear error. Finally, the speed information of Δt after the ring gear error is eliminated is restored by denormalization, thereby obtaining angular velocity information that can reflect the original vibration characteristics of the wheel and eliminate the ring gear error. By separating the correlation between the ring gear error calculation and the vehicle speed, the present invention can, on the one hand, avoid the interference of speed information on the calculation of the ring gear error. On the other hand, it can reduce the computational complexity of calculating the ring gear error using the BP neural network, thereby accurately calculating and eliminating the ring gear error while eliminating the vibration of the wheel rotation itself. The algorithm structure of the present invention is simple, has low computational complexity and storage space requirements, and can meet the real-time requirements of vehicle control.

Description

Technical Field

[0001] The present invention relates to the field of automobile technology, and in particular to a method for eliminating gear ring errors based on BPNN. Background Art

[0002] The wheel speed sensors used in today's vehicles detect wheel speed by detecting changes in the magnetic field generated by the rolling ring gear. A wheel typically has dozens of teeth. As the wheel rolls, adjacent teeth pass the sensor, generating changes in the magnetic field. By detecting the rate of change in the magnetic field, the speed at which the teeth pass the sensor—the wheel's rolling angular velocity (wheel speed)—can be determined. However, in actual production, circular wheels cannot be divided into dozens of equal parts, and production errors are almost inevitable. Consequently, the distances between different teeth on the same wheel may vary. This means that when the wheel rotates at a constant speed, the varying distances between adjacent teeth can cause speed jumps when the sensor is used for speed detection. Furthermore, since the number of teeth on the ring gear of a wheel is fixed, these speed jumps occur periodically during wheel rotation.

[0003] Data indicates that tire blowouts account for as much as 60% of traffic accidents, with 75% of these blowouts caused by insufficient tire pressure. Tire pressure monitoring systems can be divided into two types: direct and indirect. Indirect systems determine tire leaks based on tire rotational speed differences and tire vibration frequencies. Extracting tire vibration frequencies requires spectral analysis of the wheel speed signal. However, when analyzing the frequency domain characteristics of wheel angular velocity in the iTPMS system, the periodic signal generated by wheel ring gear errors can interfere with the wheel-related tire pressure signals. In severe cases, this can completely overwhelm the tire pressure-related frequency domain characteristics, preventing the iTPMS system from acquiring tire pressure signatures and, consequently, preventing alarms. Traditional filters, while suppressing ring gear errors, also suppress the wheel's vibration spectrum. This distorts the wheel vibration derived from the wheel speed signal, thus affecting tire pressure signature recognition.

[0004] BPNN is a method that has only been widely used for noise elimination in recent years. Machine learning is a better way to solve problems that cannot be accurately solved by mathematical analysis. The ring gear error cannot be accurately measured without removing the wheel and measuring it with a measuring instrument. Indirect measurement methods are also unable to accurately measure the ring gear error due to the vibration caused by the rotation of the wheel.

[0005] In the prior art, patent publication number CN 109991438 A discloses a method and device for eliminating the measurement error of the angular velocity of an automobile ring gear. The method comprises the following steps: 1) assuming that a ring gear with N teeth is mounted on a rotating shaft, reading the ring gear signal, and obtaining the time t at which the tooth k passes through k , let the error of tooth i be δi, and get the angular velocity that eliminates the error; 2) The time interval between teeth Δti(k) = t k -t k-1 , a complete cycle time is Ti(k)=t k+N / 2 –t k-N / 2 , then Δti / Ti = 1 / N + Δi(+ek), and the error of tooth i, δi = 2πΔi = 2π(Δti / Ti - 1 / N) + ek; 3) Use an adaptive filter to filter the error of tooth i; 4) Use the error of tooth i to correct the angular velocity of the gear teeth according to the corresponding relationship. However, while this solution uses a filter to suppress the ring gear error, it also suppresses the vibration spectrum of the wheel itself, causing distortion in the wheel vibration obtained from the wheel speed signal.

[0006] In view of this, it is necessary to design a gear ring error elimination method based on BPNN to solve the above problems. Summary of the Invention

[0007] The object of the present invention is to provide a BPNN-based ring gear error elimination method for eliminating the periodic interference signal generated by the ring gear error without affecting the original vibration characteristics of the wheel.

[0008] To achieve the above-mentioned object of the invention, the present invention provides a method for eliminating ring gear errors based on BPNN, comprising the following steps:

[0009] S1. Perform normalization preprocessing on the ring gear data to eliminate the influence of vehicle speed information on the time difference Δt between two adjacent teeth;

[0010] S2. Constructing a BPNN architecture to calculate and eliminate the ring gear error;

[0011] S3. Using an inverse normalization method, restore the velocity information of the time difference between the two adjacent teeth after the ring gear error is eliminated, thereby obtaining the angular velocity information after the ring gear error is eliminated that can reflect the original vibration characteristics of the wheel.

[0012] As a further improvement of the present invention, in step S1, the normalization preprocessing is to normalize the speed, and the normalization calculation formula is as follows:

[0013] △t i =t i+1 -t i

[0014] A timer is used to record the time it takes for each tooth on the ring gear to pass through the sensor, where ti represents the time it takes for the i-th tooth on the ring gear to pass through the sensor.

[0015]

[0016] in, is the normalized time difference between two adjacent teeth, i is the subscript of the number of teeth, n is the total number of teeth on the gear ring, k is the normalization constant, △t i is the difference between the arrival time of the i+1th tooth and the ith tooth.

[0017] As a further improvement of the present invention, the coefficient of variation c is used v To evaluate the distribution characteristics of Δt before and after normalization, the coefficient of variation c v The calculation formula is as follows:

[0018]

[0019] Among them, σ is the standard deviation of the sample and μ is the sample mean.

[0020] As a further improvement of the present invention, in step S2, the BPNN architecture has three layers: an input layer, a hidden layer, and an output layer. The input layer is an input matrix with n dimensions. The input matrix is an n-by-N matrix, where N is the number of sample sets. The calculation formula of the input matrix is as follows:

[0021] For the mth sample x m :

[0022]

[0023] According to the calculation formula of the input matrix, calculation is performed on n samples respectively.

[0024] As a further improvement of the present invention, the hidden layer has an adjustable hyperparameter, namely the number of units in the hidden layer; preferably, the number of units in the hidden layer is 15.

[0025] As a further improvement of the present invention, the activation function used in the hidden layer is a tanh function, the gradient of which ranges from 0 to 1 during network back propagation. The expression of the tanh function is:

[0026]

[0027] As a further improvement of the present invention, the value range of the normalization constant k is consistent with the range of the gradient.

[0028] As a further improvement of the present invention, the output layer converts the output of the hidden layer into an output matrix with n dimensions through a fully connected layer. The calculation formula of the output matrix is as follows:

[0029]

[0030] The calculation result of the mth sample represents the correlation between the current tooth and all remaining teeth.

[0031] As a further improvement of the present invention, after the learning of the network converges, the corresponding relationship between each tooth and the other teeth is obtained, so as to eliminate the ring gear error; suppose the learned network is used to predict the sample x, and the output value obtained is Right now:

[0032]

[0033] The correction of the ring gear error is:

[0034]

[0035] in, Represents the time difference between two adjacent teeth after the ring gear error is eliminated.

[0036] As a further improvement of the present invention, in step S3, the denormalization calculation formula is as follows:

[0037]

[0038] in, The time difference between two adjacent teeth is used to restore the speed information.

[0039] The beneficial effects of the present invention are:

[0040] 1. The present invention separates the relationship between ring gear error calculation and vehicle speed by preprocessing the normalized time difference Δt between two adjacent teeth. This facilitates the subsequent construction of a BP neural network architecture for ring gear error calculation. This prevents speed information from interfering with the ring gear error calculation and reduces the computational complexity of the BP neural network. This allows accurate calculation and elimination of the ring gear error while eliminating the vibration of the wheel rotation itself.

[0041] 2. This invention combines data normalization preprocessing before calculating the ring gear error with denormalization to restore Δt velocity information after eliminating the ring gear error. This method obtains ring gear error-free angular velocity information that reflects the original vibration characteristics of the wheel. This provides the iTPMS system with input information with a higher signal-to-noise ratio for frequency domain analysis, facilitating the iTPMS's identification of tire pressure changes.

[0042] 3. The present invention builds a BPNN architecture to calculate the gear ring error. By designing an activation function in the hidden layer, it solves the gradient vanishing problem of the network during the learning process to a certain extent, thereby avoiding the problem of learning stagnation leading to the inability to obtain the optimal solution.

[0043] 4. The BPNN-based calculation method designed in the present invention is simple. During the calculation process, it is only necessary to perform unified calculations on all the ring gears when the wheel runs one circle, without having to calculate each ring gear separately, thereby significantly reducing the amount of calculation. At the same time, the algorithm structure of the present invention is simple, with low computational complexity and storage space requirements, and can meet the real-time requirements of vehicle control. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is the time domain distribution diagram of the ring gear error of the present invention.

[0045] Figure 2 This is the gear ring error spectrum diagram of the present invention.

[0046] Figure 3 Schematic diagram of the time difference change between two adjacent teeth of the present invention.

[0047] Figure 4 Schematic diagram of the change of the time difference between the first tooth and the second tooth of the present invention.

[0048] Figure 5 Schematic diagram of the change of the time difference between the first tooth and the second tooth after normalization processing of the present invention.

[0049] Figure 6 This is the BPNN structure diagram of the present invention.

[0050] Figure 7 This is the time domain distribution diagram before and after the ring gear error is eliminated in the present invention.

[0051] Figure 8 This is a signal spectrum diagram after the ring gear error of the present invention is eliminated. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] It should also be noted here that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions of the present invention are shown in the drawings, while other details that are not closely related to the present invention are omitted.

[0054] In addition, it should be noted that the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0055] like Figures 1 to 8 As shown, the present invention provides a method for eliminating gear ring errors based on BPNN, comprising the following steps:

[0056] S1. Perform normalization preprocessing on the ring gear data to eliminate the influence of vehicle speed information on the time difference Δt between two adjacent teeth;

[0057] S2. Constructing a BPNN architecture to calculate and eliminate the ring gear error;

[0058] S3. Use the inverse normalization method to restore the velocity information of the time difference between two adjacent teeth with the ring gear error eliminated, so as to obtain the angular velocity information with the ring gear error eliminated that can reflect the original vibration characteristics of the wheel.

[0059] In the processing of wheel speed signals, a timer is usually used to record the time it takes for each tooth on the ring gear to pass through the sensor, which is recorded as t i , represents the arrival time of the i-th tooth, and the characteristic of the ring gear error can be represented by the arrival time difference △t of two adjacent teeth. The difference between the arrival time of the i+1-th tooth and the i-th tooth is:

[0060] △t i =t i+1 -t i

[0061] In this embodiment, the wheel rim has 40 teeth, such as Figure 1 As shown in the figure, it can be clearly seen from the peaks and troughs of the signal that Δt has a distribution characteristic with a tooth number of 40 as a period. This means that as the vehicle moves, there will be periodic signals in the rolling angular velocity of the wheel. Figure 2 The error spectrum of the ring gear shows that there are fixed-frequency harmonics with extremely large peak values. These harmonics will affect the spectrum analysis of wheel vibration.

[0062] Since the faster the vehicle speed, the shorter the interval between two adjacent teeth will be, and conversely, the slower the vehicle speed, the longer the interval between two adjacent teeth will be, so Δt in the time domain will also change accordingly, and the results are as follows: Figure 3As shown in the figure, this change with vehicle speed will cause the Δt between adjacent teeth to change due to the ring gear error. Taking the difference between the Δt between the first and second teeth and the difference between all Δt in a circle as an example, the trend of the time difference between the first and second teeth changing with time is shown in the figure. Figure 4 As shown. Figure 3 and Figure 4 It can be seen that the time difference between the same two teeth is consistent with the changing trend of the vehicle speed. Therefore, Δt contains not only the information of the ring gear error, but also the speed information. Therefore, before using BPNN to eliminate the ring gear error, the speed information in Δt needs to be eliminated.

[0063] Specifically, in step S1, the calculation formula for processing the speed using the normalization method is:

[0064]

[0065] in, is the normalized time difference between two adjacent teeth, i is the tooth number subscript, k is the normalization constant, the value range of the normalization constant k is consistent with the range of the gradient, in this embodiment, k is 1, △t i is the difference between the arrival time of the i+1th tooth and the ith tooth.

[0066] The essence of this normalization method is to normalize the time difference between two adjacent teeth with the time required for the wheel to roll one circle. The time required for the wheel to roll one circle includes the vehicle speed information, so the influence of speed is naturally eliminated.

[0067] After normalization, Δt changes with time as follows Figure 5 By comparison Figure 4 and Figure 5 It can be seen that the normalized Δt no longer changes with the change of speed. Due to the huge difference in sample size before and after normalization, the coefficient of variation c is used v To evaluate the Δt distribution characteristics before and after normalization, the coefficient of variation c v The calculation formula is as follows:

[0068]

[0069] Among them, σ is the standard deviation of the sample and μ is the sample mean.

[0070] The coefficient of variation can eliminate the size differences between different samples and only consider the degree of dispersion of the sample distribution.

[0071] The coefficient of variation of the sample before normalization is:

[0072] c v1 =0.1319

[0073] The coefficient of variation of the samples after normalization is:

[0074] c v2 =0.0026

[0075] c v1 and c v2 The difference of 51 times indicates that the normalized sample distribution is more compact and tends to a fixed value, which is the ring gear error. After normalization, BPNN can be used to eliminate the ring gear error.

[0076] Specifically, in step S2, the BPNN architecture has three layers: input layer, hidden layer, and output layer. The input layer is an input matrix with 40 dimensions. The input matrix X is a 40-by-N matrix. 40 is because the total number of teeth is 40, that is, there are 40 features. N is the number of sample sets. In this embodiment, N is 1000. For the first sample x 1 ,have:

[0077]

[0078] For the second sample x 2 have:

[0079]

[0080] The same applies to subsequent samples.

[0081] Specifically, the hidden layer has a hyperparameter that can be adjusted, namely the number of hidden units. This parameter determines the number of units with the same structure in the hidden layer. By traversing this parameter between 1 and 30, in the learning results of BPNN under different parameter settings, the fitting accuracy becomes higher and higher as the parameter changes from 1 to 15. When the parameter exceeds 15, the neural network will overfit, so 15 is used as the final number of units in the hidden layer. The activation function used in the hidden layer is the tanh function, and the expression of the tanh function is:

[0082]

[0083] The tanh function is symmetric around the origin, so its mean is 0. The gradient ranges from 0 to 1 during network backpropagation. The larger gradient range solves the gradient vanishing problem of the network during learning to a certain extent.

[0084] Specifically, the output layer converts the output of the hidden layer into an output matrix with 40 dimensions through a fully connected layer. The calculation formula of the output matrix is as follows:

[0085]

[0086] That is, the target matrix represents the correlation between the current tooth and all remaining teeth.

[0087] After the network converges, the corresponding relationship between each tooth and the other teeth can be obtained, so that the gear ring error can be eliminated. Suppose the learned network is used to predict the sample x, and the output value is Right now:

[0088]

[0089] The correction of the ring gear error is:

[0090]

[0091] in, Represents the time difference between two adjacent teeth after the ring gear error is eliminated.

[0092] Since Δt also plays a role in calculating vehicle speed in iTPMS, and frequency domain analysis still requires Δt to have its original dimension, and the normalized Δt removes speed information and changes its dimension, Δt needs to be restored after eliminating the ring gear error.

[0093] Specifically, the calculation formula for restoring the speed information of the time difference Δt between the two adjacent teeth after eliminating the ring gear error using the inverse normalization method is as follows:

[0094]

[0095] in, The time difference between two adjacent teeth is used to restore the speed information.

[0096] After the denormalization method is used, the time domain results after the ring gear error is eliminated are compared. Figure 7 As shown, from Figure 7 It can be seen that after eliminating the ring gear error, the time difference between two adjacent teeth no longer has periodic fluctuations, only the fluctuations caused by wheel vibration. The signal spectrum after eliminating the ring gear error is as follows Figure 8 As shown. Figure 8 It can be seen that the harmonic components in the signal have been greatly suppressed. Figure 2 The high energy components present in the have been eliminated.

[0097] In summary, the present invention discloses a ring gear error elimination method based on BPNN. By pre-processing the normalized time difference Δt between two adjacent teeth, the correlation between the ring gear error calculation and the vehicle speed is separated, so as to facilitate the subsequent construction of a BP neural network architecture for ring gear error calculation. On the one hand, it avoids the interference of speed information on the ring gear error calculation. On the other hand, it can reduce the computational complexity of the ring gear error calculation using the BP neural network, thereby accurately calculating and eliminating the ring gear error while eliminating the vibration of the wheel rotation itself. At the same time, combined with the restoration of the speed information of Δt after the ring gear error is eliminated by denormalization, it can obtain the original vibration characteristics of the wheel and eliminate the periodic noise in the wheel angular velocity signal, thereby providing the iTPMS system with input information with a higher signal-to-noise ratio for frequency domain analysis, facilitating the iTPMS system to identify tire pressure changes. In addition, the present invention builds a BPNN architecture for ring gear error calculation. By designing an activation function in the hidden layer, it solves the gradient vanishing problem of the network during the learning process to a certain extent, thereby avoiding the problem of learning stagnation that leads to the inability to obtain the optimal solution. In addition, the BPNN-based calculation method designed in the present invention is simple. During the calculation process, it is only necessary to perform unified calculations on all the ring gears when the wheel runs one circle, without having to calculate each ring gear separately, thereby significantly reducing the amount of calculation. At the same time, the algorithm structure of the present invention is simple, with low computational complexity and storage space requirements, and can meet the real-time requirements of vehicle control.

[0098] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A gear ring error elimination method based on BPNN, characterized in that: The steps include: S1. Perform normalization preprocessing on the ring gear data to eliminate the influence of vehicle speed information on the time difference Δt between two adjacent teeth; S2. Constructing a BPNN architecture to calculate and eliminate the ring gear error; S3. Using an inverse normalization method, restore the velocity information of the time difference between the two adjacent teeth after the ring gear error is eliminated, thereby obtaining the angular velocity information after the ring gear error is eliminated that can reflect the original vibration characteristics of the wheel.

2. The ring gear error elimination method based on BPNN according to claim 1, characterized in that: In step S1, the normalization preprocessing is to normalize the speed, and the normalization calculation formula is as follows: △t i =t i+1 -t i A timer is used to record the time it takes for each tooth on the ring gear to pass through the sensor, where ti represents the time it takes for the i-th tooth on the ring gear to pass through the sensor. in, is the normalized time difference between two adjacent teeth, is the subscript of the number of teeth, n is the total number of teeth on the gear ring, k is the normalization constant, △t i For the The tooth and The difference in the arrival time of the teeth.

3. The ring gear error elimination method based on BPNN according to claim 2, characterized in that: Using the coefficient of variation To evaluate the normalization before and after The distribution characteristics, the coefficient of variation The calculation formula is as follows: in, is the standard deviation of the sample, is the sample mean.

4. The ring gear error elimination method based on BPNN according to claim 2, characterized in that: In step S2, the BPNN architecture has three layers: input layer, hidden layer and output layer. The input layer is an input matrix with n dimensions. The input matrix is an n-dimensional matrix. The matrix of is the number of sample sets, and the calculation formula of the input matrix is as follows: For the mth sample : According to the calculation formula of the input matrix, calculation is performed on n samples respectively.

5. The ring gear error elimination method based on BPNN according to claim 4, characterized in that: The hidden layer has a hyperparameter that can be adjusted, namely the number of units in the hidden layer.

6. The ring gear error elimination method based on BPNN according to claim 5, characterized in that: The activation function used in the hidden layer is the tanh function. The gradient of the tanh function ranges from 0 to 1 during network back propagation. The expression of the tanh function is: 。 7. The ring gear error elimination method based on BPNN according to claim 6, characterized in that: The value range of the normalization constant k is consistent with the range of the gradient.

8. The ring gear error elimination method based on BPNN according to claim 4, characterized in that: The output layer converts the output of the hidden layer into an output matrix with n dimensions through a fully connected layer. The calculation formula of the output matrix is as follows: The calculation result of the mth sample represents the correlation between the current tooth and all remaining teeth.

9. The method for eliminating ring gear errors based on BPNN according to claim 8, characterized in that: After the network converges, the corresponding relationship between each tooth and the other teeth is obtained to eliminate the gear ring error. Make a prediction and the output value is ,Right now: The correction of the ring gear error is: in, Represents the time difference between two adjacent teeth after the ring gear error is eliminated.

10. The ring gear error elimination method based on BPNN according to claim 1, characterized in that: In step S3, the denormalization calculation formula is as follows: in, The time difference between two adjacent teeth is used to restore the speed information.

Images