New energy power assembly abnormal sound fault positioning method and device
Starting from the three fault sources of motor, motor controller and speed reduction device, real-time data of the powertrain is collected and analyzed, and fault diagnosis is used by BP neural network, the problem of difficult to quickly locate the abnormal noise fault of the new energy powertrain in the existing technology is solved, and efficient and accurate fault positioning and maintenance suggestions are achieved.
Patent Information
- Application Number
- CN202510189619.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-20
- Publication Date
- 2025-06-24
AI Technical Summary
It is difficult for the existing technology to quickly locate the causes of abnormal noise failure of new energy powertrains and judge its severity. Especially in complex working conditions and changing environments, there is a lack of comprehensive diagnostic analysis of electrical signals such as voltage, current, speed, and torque.
By starting from the three fault sources of the powertrain abnormal noise failure, the motor, motor controller, and speed reduction device, basic parameters and real-time data are collected, data preprocessing and feature extraction are performed, and BP neural network is used for training and analysis, fault causes are sorted and repair suggestions are generated.
It realizes rapid positioning and fault cause sorting of powertrain abnormal noise faults, provides data support and refined decision-making, has the characteristics of fast calculation speed and high accuracy, and is suitable for after-sales fault positioning and factory routine inspection.
Smart Images
Figure CN120198097A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of powertrain fault location, and more specifically, to a method and device for locating abnormal noise faults of a new energy powertrain. Background Art
[0002] The new energy powertrain is the core component of a vehicle, including a motor, a motor controller, and a reduction device (gearbox or axle). Its performance indicators and fault problems have attracted much attention in the market. In the middle and late stages of operation, the abnormal noise problem of the powertrain is a typical fault type, often causing customer complaints. It is a sign of irreversible mechanical damage and may even lead to serious vehicle faults.
[0003] When an abnormal noise fault occurs in the powertrain, the current common solutions are to subjectively judge by the human ear or to collect and analyze the vibration noise spectrum. Subjective judgment requires rich on-site experience and is subjective, and the results lack data support. Collecting the vibration noise spectrum can only analyze typical fault types and lacks comprehensive diagnostic analysis of electrical signals such as voltage, current, speed, and torque in complex working conditions and changing environments.
[0004] Therefore, how to quickly locate the fault cause and evaluate the severity of the abnormal noise fault of the powertrain has become an urgent problem to be solved. Summary of the Invention
[0005] In view of the deficiencies of the prior art, the present invention provides a method and device for locating abnormal noise faults of a new energy powertrain. The method and device for locating abnormal noise faults of the new energy powertrain start from three fault sources: the motor, the motor controller, and the reduction device, sort and locate the detailed causes of the abnormal noise fault according to the contribution amount, provide data support and at the same time provide refined decision-making for after-sales maintenance. At the same time, with the help of a big data training model, through training with enough samples, it has the characteristics of fast calculation speed and high accuracy.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] A method for locating abnormal noise faults of a new energy powertrain, comprising the following steps:
[0008] S1. Parameter input and signal acquisition: Collect and input the basic parameters of the powertrain, and collect CAN messages and vibration noise data in real time;
[0009] S2. Data preprocessing: Perform filtering processing on the collected vibration noise signals and CAN message data, and extract the inherent characteristic frequencies, time domain and frequency domain data of the detailed components;
[0010] S3. Complete the training of the typical fault database: Collect various typical abnormal sound fault sample data, extract eigenvalue as input, and train through a BP neural network until the error of the output result converges;
[0011] S4. Fault analysis: Input the preprocessed data into the data analysis module of the trained BP neural network, extract relevant eigenvalues, obtain the eigenvalues of the hidden layer after multiple transmissions through the completed training large database, obtain the result at the output layer, and sort the detailed causes of the abnormal sound fault according to the contribution amount;
[0012] S5. Fault location and decision-making: Determine the top three main fault causes according to the contribution amount sorting result, and generate corresponding maintenance suggestions.
[0013] Further, the input of the powertrain parameters includes at least one of the following parameters: parameters of the motor bearing, rotor dynamic balance, number of motor rotor poles, number of motor stator slots; switching frequency and response time of the motor controller; gear data and reduction ratio of each stage of the reduction device; modal frequencies of the motor, motor control, and reduction device.
[0014] Further, during signal acquisition in step S1, use the same clock to ensure the synchronization of the powertrain data. Collect the time-domain signals of vibration and noise of the powertrain body through vibration sensors and microphones, collect messages through the CAN communication of the motor controller, and parse out the given speed, feedback speed, given torque, feedback torque, three-phase current, three-phase voltage, and temperature time-domain signals of the motor.
[0015] Further, the data preprocessing includes:
[0016] S2.1. Calculate the inherent characteristic frequency values of the causes of each component;
[0017] S2.2. Filter the collected data, then conduct time-domain analysis, extract the root mean square value, square root amplitude, peak-to-peak value, waveform index, margin index, skewness, kurtosis, pulse index of the signal, and the difference follow of the given speed and feedback speed, given torque and feedback torque;
[0018] S2.3. Through short-time Fourier transform of the collected voltage, current, vibration, and noise time-domain signals, obtain the frequency-domain signals, extract power spectrum, peak value, root mean square value, and power spectrum information, and extract eigenvalues.
[0019] The typical abnormal noise faults of the further powertrain sub-components can be further divided into: rotor dynamic balance failure, bearing failure, motor electromagnetic noise, motor resonance, motor controller modulation, motor controller imbalance, motor controller Tip-in-out, gear meshing of the reduction gear, and reduction gear resonance. Collect data on each typical fault and normal condition as training samples, with each sample size greater than 100 and the number of training times greater than 1000 or the satisfaction of the output result greater than 98% to end the big data training. The original signals and basic parameters corresponding to the typical abnormal noise faults are marked with the characteristic values corresponding in the time domain and frequency domain, so as to accurately complete the identification and judgment of the abnormal noise faults.
[0020] The further training of the typical fault database includes the following training steps:
[0021] S3.1. Collect data on each fault and non-fault. Through data preprocessing, extract the inherent characteristic frequencies, time-domain and frequency-domain data of the sub-components as the input layer of the BP neural network, and the input vector: X = (x1, x2,..., x i ,..., x m );
[0022] S3.2. The vector of the L-th hidden layer:
[0023]
[0024] Let be the connection weight between the i-th neuron of the (L - 1)-th layer and the j-th neuron of the L-th layer, and be the bias of the j-th neuron of the L-th layer;
[0025]
[0026] where where is the input of the j-th neuron of the L-th layer, and take the Sigmoid function as the activation function;
[0027] S3.3. The output vector of the output layer: Y = (y1, y2,..., y k ,..., y n ). Calculate the error between the output result and the expected result, and use the gradient descent method to backpropagate the error to update the weights and biases of the upper-layer nodes until the output result error converges.
[0028] S3.4. Input the existing typical fault sample data and observe whether the output result meets the requirements to complete the training of the typical fault database.
[0029] In the further fault analysis, for the satisfactory data, add the collected data to the sample library of the typical fault database to improve the calculation accuracy of the model.
[0030] A device for locating abnormal noise faults of a new energy powertrain, comprising:
[0031] Typical fault database, used to store trained BP neural network models and typical fault feature data;
[0032] Signal acquisition and preprocessing module, used to collect basic parameters of the powertrain, collect CAN messages and vibration noise data in real time, and perform data preprocessing;
[0033] The data analysis module based on BP neural network is used to extract relevant eigenvalues. The eigenvalues of the hidden layer that have been passed multiple times are obtained through the large database that has completed the training. The results are obtained in the output layer. The causes of abnormal noise faults are sorted according to the contribution amount.
[0034] The fault location decision module is used to sort the factors according to their contribution. The top three are regarded as the main contributors to the abnormal noise of the powertrain, and corresponding suggestions are given for the main sub-factors.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] The present invention is provided with a typical fault database, a signal acquisition and preprocessing module, a data analysis module based on BP neural network, and a fault location decision module;
[0037] Through the abnormal noise fault location, workflow and cause analysis of the powertrain, starting from the three fault sources of the powertrain abnormal noise fault, namely the motor, motor controller and reduction device, by inputting the basic parameters of the powertrain, collecting the motor controller CAN message and the vibration and noise data of the powertrain in real time;
[0038] Then, time domain analysis, frequency analysis and time-frequency graph analysis are performed and input into the data analysis module based on BP neural network that has completed the training of typical fault database. The relevant eigenvalues are extracted and the eigenvalues of multiple hidden layers are confirmed. The subdivided causes of abnormal noise faults are sorted according to the size of their contribution. Finally, corresponding processing suggestions are given according to the fault graph.
[0039] In addition, the present invention uses a big data training model and has the characteristics of fast calculation speed and high accuracy through sufficient sample training. A typical fault database is set up to train the original signals and basic parameters corresponding to typical abnormal sound faults, and mark the corresponding characteristic values in the time domain and frequency domain, so as to accurately complete the identification and judgment of abnormal sound faults.
[0040] Moreover, the present invention has the characteristics of being portable, platform-based and intelligent, and is suitable for after-sales fault location and routine factory inspection work scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings, where:
[0042] Figure 1 It is a schematic diagram of a device for locating abnormal noise faults in a new energy powertrain;
[0043] Figure 2 It is a schematic diagram of the training steps for training a typical fault database. Specific embodiments
[0044] In the description of the present invention, it should be noted that for orientation terms, if there are terms such as "center", "horizontal (X)", "longitudinal (Y)", "vertical (Z)", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., the orientation and position relationships indicated are based on the orientation or position relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of the present invention.
[0045] In addition, if there are terms "first" and "second", they are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, the meaning of "several" and "a number of" is two or more, unless otherwise clearly and specifically defined.
[0046] A method for locating abnormal noise faults in a new energy powertrain includes the following steps:
[0047] S1. Parameter input and signal acquisition: Collect and input the basic parameters of the powertrain, and collect CAN messages and vibration and noise data in real time;
[0048] S2. Data preprocessing: Filter the collected vibration and noise signals and CAN message data, and extract the inherent characteristic frequencies, time-domain and frequency-domain data of the sub-components;
[0049] S3. Complete the training of the typical fault database: Collect various typical abnormal noise fault sample data, extract the characteristic values as inputs, and train through a BP neural network until the error of the output result converges;
[0050] S4, Fault analysis: Input the preprocessed data into the data analysis module of the trained BP neural network, extract relevant eigenvalue, obtain the eigenvalue of the hidden layer transmitted multiple times through the trained large database, get the result at the output layer, and sort the detailed causes of abnormal noise faults according to the contribution amount;
[0051] S5, Fault location and decision-making: Determine the top three main fault causes according to the sorting result of the contribution amount, and generate corresponding maintenance suggestions.
[0052] Preferably, the input of the powertrain parameters includes at least one of the following parameters: the parameters of the motor bearing, rotor dynamic balance, number of motor rotor poles, number of motor stator slots; the switching frequency and response time of the motor controller; the gear data and reduction ratio of each stage of the reduction gear; the modal frequencies of the motor, motor control, and reduction gear.
[0053] Preferably, during signal acquisition in step S1, the same clock is used to ensure the synchronization of the powertrain data. The time-domain signals of the vibration and noise of the powertrain body are collected through vibration sensors and microphones, and the messages are collected through the CAN communication of the motor controller, and the given speed, feedback speed, given torque, feedback torque, three-phase current, three-phase voltage, and temperature time-domain signals of the motor are parsed.
[0054] Preferably, the data preprocessing includes:
[0055] S2.1, Calculate the inherent characteristic frequency values of the causes of each component;
[0056] S2.2, Filter the collected data, then conduct time-domain analysis, extract the root mean square value, peak-to-peak value, peak-to-peak value, waveform index, margin index, skewness, kurtosis, pulse index of the signal, and the difference follow of the given speed and feedback speed, given torque and feedback torque;
[0057] S2.3, Through short-time Fourier transform of the collected voltage, current, vibration, and noise time-domain signals, obtain the frequency-domain signals, extract the power spectrum, peak value, root mean square value, and power spectrum information, and extract the eigenvalue.
[0058] Preferably, the typical abnormal noise faults of the powertrain sub-components can be further divided into: rotor dynamic balance failure, bearing failure, motor electromagnetic noise, motor resonance, motor controller modulation, motor controller imbalance, motor controller Tip-in-out, gear meshing of the reduction gear, and reduction gear resonance. Data of each typical fault and normal condition are collected as training samples, with the number of each sample being greater than 100. The big data training can be ended only when the number of training times is greater than 1000 or the satisfaction degree of the output result is greater than 98%. The original signals and basic parameters corresponding to the typical abnormal noise faults are marked with the characteristic values corresponding in the time domain and frequency domain, so as to accurately complete the identification and judgment of the abnormal noise faults.
[0059] Preferably, the training of the typical fault database includes the following training steps:
[0060] S3.1. Collect data of each fault and non-fault. Through data preprocessing, extract the inherent characteristic frequencies, time domain and frequency domain data of the sub-components as the input layer of the BP neural network, and the input vector: X = (x1, x2,..., x i ,..., x m );
[0061] S3.2. The vector of the L-th hidden layer:
[0062]
[0063] Let be the connection weight between the i-th neuron of the (L - 1)-th layer and the j-th neuron of the L-th layer, and be the bias of the j-th neuron of the L-th layer;
[0064]
[0065] where where is the input of the j-th neuron of the L-th layer, and the Sigmoid function is taken as the activation function;
[0066] S3.3. The output vector of the output layer: Y = (y1, y2,..., y k ,..., y n ). Calculate the error between the output result and the expected result, and use the gradient descent method to backpropagate the error to update the weights and biases of the upper-layer nodes until the output result error converges.
[0067] S3.4. Input the existing typical fault sample data and observe whether the output result meets the requirements to complete the training of the typical fault database.
[0068] Preferably, in the fault analysis, for the satisfactory data, add the collected data to the sample library of the typical fault database to improve the calculation accuracy of the model.
[0069] A device for locating abnormal noise faults of a new energy powertrain, comprising:
[0070] Typical fault database, used to store trained BP neural network models and typical fault feature data;
[0071] Signal acquisition and preprocessing module, used to collect basic parameters of the powertrain, collect CAN messages and vibration noise data in real time, and perform data preprocessing;
[0072] The data analysis module based on BP neural network is used to extract relevant eigenvalues. The eigenvalues of the hidden layer that have been passed multiple times are obtained through the large database that has completed the training. The results are obtained in the output layer. The causes of abnormal noise faults are sorted according to the contribution amount.
[0073] The fault location decision module is used to sort the factors according to their contribution. The top three are regarded as the main contributors to the abnormal noise of the powertrain, and corresponding suggestions are given for the main sub-factors.
[0074] advantage:
[0075] The present invention is provided with a typical fault database, a signal acquisition and preprocessing module, a data analysis module based on BP neural network, and a fault location decision module;
[0076] Through the abnormal noise fault location, workflow and cause analysis of the powertrain, starting from the three fault sources of the powertrain abnormal noise fault, namely the motor, motor controller and reduction device, by inputting the basic parameters of the powertrain, collecting the motor controller CAN message and the vibration and noise data of the powertrain in real time;
[0077] Then, time domain analysis, frequency analysis and time-frequency graph analysis are performed and input into the data analysis module based on BP neural network that has completed the training of typical fault database. The relevant eigenvalues are extracted and the eigenvalues of multiple hidden layers are confirmed. The subdivided causes of abnormal noise faults are sorted according to the size of their contribution. Finally, corresponding processing suggestions are given according to the fault graph.
[0078] In addition, the present invention uses a big data training model and has the characteristics of fast calculation speed and high accuracy through sufficient sample training. A typical fault database is set up to train the original signals and basic parameters corresponding to typical abnormal sound faults, and mark the corresponding characteristic values in the time domain and frequency domain, so as to accurately complete the identification and judgment of abnormal sound faults.
[0079] Moreover, the present invention has the characteristics of being portable, platform-based and intelligent, and is suitable for after-sales fault location and routine factory inspection work scenarios.
[0080] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.
Claims
1. A method for locating abnormal noise faults of new energy powertrains, characterized in that: The steps include: S1. Parameter input and signal acquisition: Collect and input basic parameters of the powertrain, and collect CAN messages and vibration noise data in real time; S2. Data preprocessing: filtering the collected vibration noise signals and CAN message data to extract the inherent characteristic frequency, time domain and frequency domain data of the subdivided components; S3. Complete typical fault database training: collect a variety of typical abnormal noise fault sample data, extract characteristic values as input, and train through BP neural network until the output result error converges; S4, fault analysis: input the preprocessed data into the data analysis module of the trained BP neural network, extract the relevant eigenvalues, obtain the eigenvalues of the hidden layer that have been passed multiple times through the trained large database, obtain the results in the output layer, and sort the subdivided causes of the abnormal noise fault according to the contribution amount; S5. Fault location and decision-making: According to the contribution ranking results, the top three main fault causes are determined and corresponding maintenance suggestions are generated.
2. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: The input of the powertrain parameters includes at least one of the following parameters: parameters of motor bearings, rotor dynamic balance, number of motor rotor poles, number of motor stator slots; switching frequency and response time of the motor controller; gear data and reduction ratio of each level of the reduction device; modal frequencies of the motor, motor control, and reduction device.
3. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: During signal collection in step S1, the same clock is used to ensure the synchronization of powertrain data. The vibration and noise time domain signals of the powertrain body are collected through vibration sensors and microphones. The messages are collected through CAN communication of the motor controller, and the given speed, feedback speed, given torque, feedback torque, three-phase current, three-phase voltage and temperature time domain signals of the motor are parsed.
4. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: The data preprocessing includes: S2.
1. Calculate and obtain the inherent characteristic frequency values of the inducements of each level of components; S2.
2. Filter the collected data and then conduct time domain analysis to extract the signal's RMS value, RMS amplitude, peak-to-peak value, waveform index, margin index, skewness, kurtosis, pulse index, and the difference between the given speed and feedback speed, and the given torque and feedback torque; S2.
3. The collected voltage, current, vibration and noise time domain signals are transformed into frequency domain signals through short-time Fourier transform, and the power spectrum, peak value, root mean square value, capacity spectrum information are extracted, and the characteristic value is extracted.
5. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: Typical abnormal noise failures of powertrain subdivision components can be divided into: rotor dynamic balance failure, bearing failure, motor electromagnetic noise, motor resonance, motor controller modulation, motor controller imbalance, motor controller Tip-in-out, reduction gear meshing, reduction gear resonance. Data of each typical fault and normal situation are collected as training samples. The big data training can be terminated only when the number of samples of each type is greater than 100, the number of training times is greater than 1000 times, or the output result satisfaction rate is greater than 98%. The original signals and basic parameters corresponding to the typical abnormal noise failures are marked with the corresponding eigenvalues in the time domain and frequency domain, so as to accurately complete the identification and judgment of the abnormal noise failure.
6. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: The typical fault database training includes the following training steps: S3.
1. Collect data of each fault and non-fault, extract the inherent characteristic frequency, time domain and frequency domain data of the subdivided components through data preprocessing as the input layer of the BP neural network, input vector: X = (x1, x2, ..., x i , ..., x m ); S3.2, vector of the Lth hidden layer: set up is the connection weight between the i-th neuron in the L-1 layer and the j-th neuron in the L-th layer, is the bias of the jth neuron in the Lth layer; in Where is the input of the jth neuron in the Lth layer, and the Sigmoid function is taken as the activation function; S3.3, output layer output vector: Y = (y1, y2, ..., y k , ..., y n ), find the error between the output result and the expected result, use the gradient descent method to back propagate the error, and update the weights and biases of the upper nodes until the output result error converges. S3.
4. Input the existing typical fault sample data, observe whether the output results meet the requirements, and complete the typical fault database training.
7. The method for locating abnormal noise fault of a new energy powertrain according to claim 1, characterized in that: In fault analysis, for satisfactory data, the collected data is added to the sample library of the typical fault database to improve the calculation accuracy of the model.
8. A device for locating abnormal noise faults of new energy powertrains, characterized in that: include: Typical fault database, used to store trained BP neural network models and typical fault feature data; Signal acquisition and preprocessing module, used to collect basic parameters of the powertrain, collect CAN messages and vibration noise data in real time, and perform data preprocessing; The data analysis module based on BP neural network is used to extract relevant eigenvalues. The eigenvalues of the hidden layer that have been passed multiple times are obtained through the large database that has completed the training. The results are obtained in the output layer. The causes of abnormal noise faults are sorted according to the contribution amount. The fault location decision module is used to sort the factors according to their contribution. The top three are regarded as the main contributors to the abnormal noise of the powertrain, and corresponding suggestions are given for the main sub-factors.