AMT gearbox fault diagnosis method and system based on BP neural network and wavelet packet features

Abstract

The invention discloses an AMT gearbox fault diagnosis method and system based on a BP neural network and wavelet packet features, and the method comprises the steps: symmetrically arranging measurement points at the two sides of a to-be-detected AMT gearbox, collecting vibration signals, carrying out the wavelet packet energy feature extraction, obtaining an energy vector, inputting the energy vector into a pre-trained BP neural network, and carrying out the fault diagnosis of the to-be-detected AMT gearbox. Performing fault diagnosis; the pre-training of the BP neural network comprises the following steps: assembling a normal AMT gearbox and a gearbox containing a fault AMT gearbox to obtain a to-be-tested AMT gearbox; measuring points are symmetrically arranged on the two sides of the AMT gearbox to be measured, and vibration signals of the measuring points are collected to obtain time signals; performing wavelet packet energy feature extraction on the time signal to obtain an energy vector; training the neural network until the error is smaller than a preset value; the neural network is evaluated, and if the accuracy reaches a set value, the neural network is used for fault diagnosis; the gearbox automatically triggers detection through built-in online diagnosis of the AMT controller, so that the accuracy of fault recognition is improved, and personal errors are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of gearbox fault diagnosis, and relates to a method and system for diagnosing AMT gearbox faults based on BP neural network and wavelet packet features. Background Art

[0002] With the rapid development of the automotive industry, automated transmission technology has become a key factor in improving vehicle performance, driving comfort, and fuel economy. An automated mechanical transmission (AMT) combines the high transmission efficiency of a manual transmission with the convenience of an automatic transmission. By utilizing an electronic control system to automatically shift gears, it significantly reduces the driver's operational burden, demonstrating a significant advantage in traffic congestion or long-distance driving scenarios. AMT not only retains the core advantage of manual transmissions' high transmission efficiency, but also utilizes intelligent control algorithms to optimize shifting logic, enabling precise shift timing and effectively reducing power interruptions, thereby improving fuel economy and meeting the dual demands of high performance and low emissions for modern vehicles.

[0003] However, in practical applications, AMT technology still faces many challenges. Especially under heavy-load and high-load operating conditions, such as commercial vehicles and engineering vehicles, the total vehicle load is large and the engine torque output is high, resulting in a significant increase in the mechanical stress and thermal load on the AMT transmission. This harsh operating environment accelerates problems such as component wear, hydraulic system failure and electronic control unit failure, making the failure probability of the AMT transmission relatively high. Once a failure occurs, it not only affects the normal operation of the vehicle, but may also cause safety hazards, increase maintenance costs and downtime. Therefore, fault diagnosis technology for AMT transmissions has become a key issue that needs to be solved urgently in the field of automotive maintenance, and is also an important research direction to ensure vehicle reliability and economy.

[0004] Currently, fault diagnosis for AMT transmissions primarily relies on traditional manual experience. Maintenance personnel make preliminary judgments through intuitive methods such as "listening" (for abnormal sounds), "looking" (for instrument indications, leaks, etc.), and "touching" (for temperature, vibration). While simple and cost-effective, diagnostic results are highly dependent on the maintenance personnel's experience and technical skills. Differences in fault signature recognition among personnel, coupled with the uncertainty of subjective judgment, make it difficult to ensure the accuracy and reliability of diagnostic results. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention aims to provide a method and system for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a method for diagnosing AMT gearbox faults based on BP neural network and wavelet packet features, comprising the following steps: arranging measuring points symmetrically on both sides of the AMT gearbox to be tested, collecting vibration signals, and obtaining time series signals; performing wavelet packet energy feature extraction on the time series signals to obtain energy vectors. ; The energy vector The input is input into a pre-trained BP neural network for fault diagnosis; the pre-trained BP neural network is obtained by the following training steps: assembling a normal AMT gearbox and n faulty AMT gearboxes to obtain an AMT gearbox to be tested, each faulty AMT gearbox contains a typical fault; symmetrically arranging measuring points on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, collecting vibration signals of the measuring points to obtain a time series signal; performing wavelet packet energy feature extraction on the time signal to obtain an energy vector ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

[0007] Furthermore, the measuring points are symmetrically arranged on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, including: two measuring points are symmetrically arranged on both sides of the faulty shaft of each faulty AMT gearbox, and 2n measuring points corresponding to n types of faults are arranged on both sides of the normal AMT gearbox, and the total number of measuring points is 4n.

[0008] Furthermore, the energy vector is obtained by extracting the wavelet packet energy feature of the time signal. The method also includes: time signal preprocessing and data set construction; the time signal preprocessing is used to remove background noise; the data set construction is to perform multi-label annotation on the filtered time signal according to the measurement point location and fault type, intercept samples according to the preset length, and divide each sample into a first training set and a first test set; the first training set is subjected to wavelet packet energy feature extraction to obtain an energy vector , perform wavelet packet energy feature extraction on the first test set to obtain the energy vector .

[0009] Furthermore, the energy vector is obtained by extracting the wavelet packet energy feature of the time signal. , including: converting the time signal Discretized into the coefficient sequence of the 0th layer and the 0th frequency band , among which Layer, the coefficient of the kth frequency band is ; Layer, the coefficient of the kth frequency band is Perform low-pass filtering and high-pass filtering to generate the j+1th layer sub-band coefficients; calculate the energy value of each layer of each band to obtain the energy vector .

[0010] Furthermore, the energy vector for:

[0011] in, For the Layer, The wavelet coefficient energy in the frequency band is is the number of decomposition layers, = .

[0012] Furthermore, the Layer, The energy of wavelet coefficients in a frequency band is:

[0013] in, For the Layer, -1 wavelet coefficient energy in the frequency band, For the Layer, The frequency band wavelet coefficients, is the index variable.

[0014] Furthermore, the Layer, The coefficients of the frequency bands are Perform low-pass filtering, including: and location The next wavelet coefficients, passed through a low-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to obtain.

[0015] Furthermore, the Layer, The coefficients of the frequency bands are Perform high-pass filtering, including: and location The next wavelet coefficients, passed through a high-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to obtain.

[0016] Furthermore, it also includes online diagnosis, which uses a classifier to identify time series signals, and performs maintenance processing when the identification failure rate is above 80%.

[0017] The present invention also provides an AMT gearbox fault diagnosis system based on BP neural network and wavelet packet features, comprising: a signal acquisition module for symmetrically arranging measuring points on both sides of the AMT gearbox to be tested, collecting vibration signals, and obtaining time series signals; a feature extraction module for performing wavelet packet energy feature extraction on the time series signals, and obtaining energy vectors. ; Fault diagnosis module: used to convert the energy vector Input into the pre-trained BP neural network for fault diagnosis; The pre-trained BP neural network is obtained through the following training steps: Assemble a normal AMT gearbox and n faulty AMT gearboxes to obtain the AMT gearbox to be tested, and each faulty AMT gearbox contains a typical fault; symmetrically arrange measurement points on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, collect the vibration signals of the measurement points, and obtain a time series signal; perform wavelet packet energy feature extraction on the time signal to obtain an energy vector ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

[0018] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention provides a method for diagnosing AMT transmission faults based on BP neural network and wavelet packet features. The transmission can automatically trigger detection according to a preset mileage, such as every 10,000 kilometers, through the built-in online diagnosis of the AMT controller. Real-time operating data such as speed, torque, temperature, etc. are used for precise analysis, which significantly improves the accuracy of fault identification and reduces human error. Traditional manual identification of transmission faults relies on the driver's experience, is time-consuming and labor-intensive, has a high misjudgment rate, and requires entering a service station for unpacking and inspection, resulting in a cumbersome process.

[0019] The present invention discloses a method for diagnosing AMT transmission faults based on BP neural network and wavelet packet features. Through automatic online diagnosis, potential faults can be identified in advance without disassembly, avoiding unnecessary repairs in service stations, reducing manpower input in service stations, shortening vehicle downtime, and improving operational efficiency. Traditional unpacking inspections require drivers to actively enter the service station, resulting in increased vehicle downtime and additional labor costs.

[0020] The present invention provides a method for diagnosing AMT transmission faults based on BP neural network and wavelet packet features. The method monitors the transmission status in real time, can provide early warning of potential risks, and shift maintenance from "passive repair" to "active prevention", thereby extending the service life of the transmission and reducing the incidence of major failures.

[0021] The present invention provides a method for diagnosing AMT transmission faults based on BP neural network and wavelet packet features. Online diagnosis can timely discover hidden faults and avoid the risk of traffic accidents caused by transmission failure. At the same time, through continuous data accumulation and model optimization, the diagnostic capability can be iteratively upgraded over time, further ensuring the stability and safety of the vehicle's long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a method for diagnosing AMT transmission faults based on BP neural network and wavelet packet features of the present invention; Figure 2 This is a flowchart of online diagnosis in an embodiment of the present invention; Figure 3 This is a flowchart of neural network training in an embodiment of the present invention; Figure 4 This is a signal frequency band energy distribution diagram of different fault forms in an embodiment of the present invention; Figure 5 This is a comparison chart of the predicted results and the actual results in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] Example 1 The present invention provides a method for diagnosing AMT gearbox faults based on BP neural network and wavelet packet features. Figure 1 The method comprises arranging measuring points symmetrically on both sides of the AMT gearbox to be tested, collecting vibration signals, and obtaining time series signals; performing wavelet packet energy feature extraction on the time series signals to obtain energy vectors ; The energy vector The input is input into a pre-trained BP neural network for fault diagnosis; the pre-trained BP neural network is obtained by the following training steps: assembling a normal AMT gearbox and n faulty AMT gearboxes to obtain an AMT gearbox to be tested, each faulty AMT gearbox contains a typical fault; symmetrically arranging measuring points on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, collecting vibration signals of the measuring points to obtain a time series signal; performing wavelet packet energy feature extraction on the time signal to obtain an energy vector ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

[0025] Existing fault diagnosis for AMT transmissions relies primarily on traditional manual experience-based methods. The results are highly dependent on the maintenance personnel's experience and technical skills, making it difficult to guarantee the accuracy and reliability of the diagnostic results. This invention uses automatic online diagnosis to proactively identify potential faults without disassembly, promptly uncovering hidden faults and avoiding the risk of accidents caused by transmission failure.

[0026] Assemble normal AMT gearboxes and n faulty AMT gearboxes. Specifically, assemble a normal AMT gearbox without faults as a control group, and assemble three AMT gearboxes containing three typical faults. It should be noted that each faulty gearbox contains only one typical fault.

[0027] Measuring points are arranged symmetrically on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, including two measuring points symmetrically arranged on both sides of the faulty shaft of each faulty AMT gearbox, and 2n measuring points corresponding to n types of faults of n AMT gearboxes are arranged on both sides of the normal AMT gearbox, with a total of 4n measuring points.

[0028] Specifically, two measuring points were placed symmetrically on either side of the axis where the fault occurred for each fault box. Therefore, the faults of the three fault boxes were located on three different axes, for a total of six measuring points. A normal AMT transmission, serving as the control group, required six measuring points, for a total of 4n measuring points, or 12.

[0029] Place the faulty AMT gearbox on the test bench. The speed modulation n1 of the first faulty AMT gearbox, the speed modulation n2 of the second faulty AMT gearbox, the speed modulation n3 of the third faulty AMT gearbox, and the normal AMT gearbox are used as a control. After the engine speed stabilizes, the vibration signals of the faulty AMT gearboxes are collected to obtain 12 time signals.

[0030] The sliding average is used to filter the time signal to remove the influence of background noise on feature extraction.

[0031] Then, the filtered time signal is annotated with multiple labels according to the measurement point location and fault type, and samples are cut according to the preset length. Each sample is divided into the first training set and the first test set. The wavelet packet energy feature extraction of the first training set is performed to obtain the energy vector , perform wavelet packet energy feature extraction on the first test set to obtain the energy vector .

[0032] Specifically, the filtered time signal is labeled according to measurement point 1 or measurement point 2, and also according to the first fault, second fault, third fault, or control 1, control 2, or control 3, resulting in 12 labels. The time signal is then truncated, with each 1024 data points as a sample. 1000 samples are obtained under each label, for a total of 12 × 1000 samples. The 1000 samples under each label are randomly divided into two groups in a 1:1 ratio: the first training set and the first test set. That is, the first training set has 12 × 500 samples, and the first test set has 12 × 500 samples.

[0033] Wavelet packet energy feature extraction is performed on both the first training set and the first test set to obtain the energy vector Specifically, the time signal Discretized into the coefficient sequence of the 0th layer and the 0th frequency band , among which Layer, the coefficient of the kth frequency band is ; Layer, the coefficient of the kth frequency band is Perform low-pass filtering and high-pass filtering to generate the j+1th layer sub-band coefficients; calculate the energy value of each layer of each band to obtain the energy vector It should be noted that the time signal It is the general term for the first training set and the first test set after filtering.

[0034] Energy Vector for:

[0035] in, For the Layer, The wavelet coefficient energy in the frequency band is is the number of decomposition layers, = The number of decomposition layers of the present invention 3

[0036] in, For the Layer, -1 wavelet coefficient energy in the frequency band, For the Layer, The frequency band wavelet coefficients, is the index variable.

[0037] Specifically, the time signal Discretized coefficient sequence , the first level of decomposition: Decomposition Generation (low frequency) and (high frequency), and Decompose them separately and get 、 、 、 , repeat until layer, generate frequency bands, calculate the frequency band energy of each layer, and obtain the energy vector .

[0038] It should be noted that the Layer, The coefficients of the frequency bands are , generating child node coefficients through filtering. Filters include low-pass filters and high-pass filters. Low-pass filters are often used to remove high-frequency noise from signals. High-pass filters, in contrast to low-pass filters, allow high-frequency signals to pass while suppressing low-frequency signals.

[0039] In scale and location The next wavelet coefficients, passed through a low-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to get it. That is

[0040] In scale and location The next wavelet coefficients, passed through a high-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to get it. That is .

[0041] The energy vector The system is divided into six groups, each consisting of a fault group and a control group. Within a group, if a certain energy subvector in the fault group shows a significant increase while the control group remains relatively stable, it can be preliminarily determined that the energy frequency band corresponding to that subvector is strongly correlated with the fault, thus providing guidance for accurately locating the root cause of the fault.

[0042] Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The test set in is used to evaluate the neural network. If the accuracy reaches the set value, it is used for fault diagnosis.

[0043] Specifically, the neural network is first constructed, and then the neural network is trained, including calculating the hidden layer unit activation value, output layer response value, actual output and expected output. When the preset value E is less than 0.0001, the iterative process is terminated and the current network parameters are output; when the preset value E is greater than or equal to 0.0001, the hidden layer unit error is calculated in turn and the gradient descent method is used to adjust the weight matrix of each layer, such as Figure 3 After stopping the training, the trained neural network is tested on the respective test sets.

[0044] Online diagnosis uses a classifier to identify time signals, and the identification fault rate is over 80%, requiring maintenance. Specifically, if the results of both classifiers identify more than 8 faults, it is determined that a fault has occurred. Once the fault is confirmed, the controller in the system will act quickly. It will carefully package the detailed fault information into a CAN message, and accurately broadcast it to the instrument through the CAN bus, a high-speed communication channel. As an information display terminal directly facing the driver, the instrument will prompt the driver in an intuitive and eye-catching manner to enter the nearest service station for maintenance as soon as possible. From fault identification to information transmission to driver awareness, the entire process is efficient and smooth, which maximizes the protection of the safe operation of the equipment and driving safety, and builds a solid line of defense for the continued stable operation of the equipment and the safe travel of personnel. Figure 2 shown.

[0045] Three types of faults are selected: bearing inner race fault, ball bearing fault, and outer race fault. The speed of the three types of faults is 1750rpm, and the number of measurement points is 1. 1000 samples are selected and divided into the first training set and the first test set in a 1:1 ratio. The length of each sample is 1024. The first training set and the first test set are filtered, and then wavelet packet energy feature extraction is performed. The energy vectors of different fault forms are obtained. like Figure 4 As shown, it can be seen that there is a significant difference in the distribution, and the node coefficient is 4. When the preset value is less than 0.0001, the training is stopped and verified on the test set. The test results are as follows Figure 5As shown in Figure 1, the labels output by the network are compared with the known labels, and the accuracy reaches 99.5%.

[0046] It should be noted that during driving, an online diagnosis will be performed passively every 10,000 kilometers. The triggering condition for the online diagnosis is that the vehicle is driving or the engine reaches a fault-sensitive speed.

[0047] To summarize, the transmission can automatically trigger detection at a preset mileage, such as every 10,000 kilometers, through the built-in online diagnosis of the AMT controller. It uses real-time operating data such as speed, torque, and temperature for precise analysis, significantly improving the accuracy of fault identification and reducing human error. Traditional manual identification of transmission faults relies on the driver's experience, which is time-consuming and labor-intensive, with a high misjudgment rate, and requires entering the service station for unpacking and inspection, which is a cumbersome process.

[0048] Example 2 The present invention provides an AMT gearbox fault diagnosis system based on BP neural network and wavelet packet features, comprising a signal acquisition module, a feature extraction module and a fault diagnosis module.

[0049] Signal acquisition module: used to arrange measurement points symmetrically on both sides of the AMT gearbox to be tested, collect vibration signals, and obtain time series signals; Feature extraction module: used to extract the wavelet packet energy feature of the time series signal to obtain the energy vector ; Fault diagnosis module: used to convert the energy vector Input into the pre-trained BP neural network for fault diagnosis; The pre-trained BP neural network is obtained through the following training steps: Assemble a normal AMT gearbox and n faulty AMT gearboxes to obtain the AMT gearbox to be tested, and each faulty AMT gearbox contains a typical fault; symmetrically arrange measurement points on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, collect the vibration signals of the measurement points, and obtain a time series signal; perform wavelet packet energy feature extraction on the time signal to obtain an energy vector ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

[0050] The present invention provides an AMT transmission fault diagnosis system based on BP neural network and wavelet packet features, which can implement the same method steps as the above method, and therefore will not be described in detail.

[0051] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

Claims

1. A method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features, characterized in that: The following steps are involved: The measuring points are arranged symmetrically on both sides of the AMT gearbox to be tested, and the vibration signals are collected to obtain time series signals. The time series signal is subjected to wavelet packet energy feature extraction to obtain the energy vector ; The energy vector Input into the pre-trained BP neural network for fault diagnosis; The pre-trained BP neural network is obtained through the following training steps: Assemble a normal AMT transmission and n faulty AMT transmissions to obtain the AMT transmission to be tested, where each faulty AMT transmission contains a typical fault; Measuring points are arranged symmetrically on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of a normal AMT gearbox. Vibration signals of the measuring points are collected to obtain time series signals. The energy vector is obtained by extracting the wavelet packet energy feature of the time signal ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

2. The method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features according to claim 1 is characterized in that: The measuring points are arranged symmetrically on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of the normal AMT gearbox, including: Two measuring points are arranged symmetrically on both sides of the faulty axis of each faulty AMT gearbox, and 2n measuring points corresponding to n types of faults are arranged on both sides of the normal AMT gearbox, with a total of 4n measuring points.

3. The method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features according to claim 1 is characterized in that: The energy vector is obtained by performing wavelet packet energy feature extraction on the time signal The front also includes: time signal preprocessing and data set construction; The time signal preprocessing is used to remove background noise; The data set is constructed by performing multi-label annotation on the filtered time signal according to the measurement point location and the fault type, cutting samples according to a preset length, and dividing each sample into a first training set and a first test set; The energy vector is obtained by extracting the wavelet packet energy feature of the first training set , perform wavelet packet energy feature extraction on the first test set to obtain the energy vector .

4. The method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features according to claim 1, characterized in that: The energy vector is obtained by performing wavelet packet energy feature extraction on the time signal ,include: The time signal Discretized into the coefficient sequence of the 0th layer and the 0th frequency band , among which Layer, the coefficient of the kth frequency band is ; For the first Layer, the coefficient of the kth frequency band is Perform low-pass filtering and high-pass filtering to generate the j+1th layer sub-band coefficients; Calculate the energy value of each frequency band in each layer and obtain the energy vector .

5. The method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features according to claim 4 is characterized in that: The energy vector for: in, For the Layer, The wavelet coefficient energy in the frequency band is is the number of decomposition layers, = .

6. The method for AMT transmission fault diagnosis based on BP neural network and wavelet packet features according to claim 5, characterized in that: The said Layer, The energy of wavelet coefficients in a frequency band is: in, For the Layer, -1 wavelet coefficient energy in the frequency band, For the Layer, The frequency band wavelet coefficients, is the index variable.

7. The method for diagnosing AMT gearbox faults based on BP neural network and wavelet packet features according to claim 4, characterized in that: The said Layer, The coefficients of the frequency bands are Perform low-pass filtering, including: In scale and location The next wavelet coefficients, passed through a low-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to obtain.

8. The method for AMT gearbox fault diagnosis based on BP neural network and wavelet packet features according to claim 4 is characterized in that: The said Layer, The coefficients of the frequency bands are Perform high-pass filtering, including: In scale and location The next wavelet coefficients, passed through a high-pass filter On the previous scale and the wavelet packet coefficients at position k Perform convolution operation to obtain.

9. The method for AMT transmission fault diagnosis based on BP neural network and wavelet packet features according to claim 1, characterized in that: It also includes online diagnosis, which uses a classifier to identify time series signals, identifies a fault rate of more than 80%, and performs maintenance processing.

10. A fault diagnosis system for AMT transmission based on BP neural network and wavelet packet features, characterized in that: include: Signal acquisition module: used to arrange measurement points symmetrically on both sides of the AMT gearbox to be tested, collect vibration signals, and obtain time series signals; Feature extraction module: used to extract the wavelet packet energy feature of the time series signal to obtain the energy vector ; Fault diagnosis module: used to convert the energy vector Input into the pre-trained BP neural network for fault diagnosis; The pre-trained BP neural network is obtained through the following training steps: Assemble a normal AMT transmission and n faulty AMT transmissions to obtain the AMT transmission to be tested, where each faulty AMT transmission contains a typical fault; Measuring points are arranged symmetrically on both sides of the faulty shaft of each faulty AMT gearbox and on both sides of each shaft of a normal AMT gearbox. Vibration signals of the measuring points are collected to obtain time series signals. The energy vector is obtained by extracting the wavelet packet energy feature of the time signal ; Based on energy vector The neural network is trained with the training set in until the error is less than the preset value. The neural network is evaluated using the test set in . If the accuracy reaches the set value, the pre-trained BP neural network is obtained.

Images