Fault diagnosis system and method for magneto-rheological semi-active suspension actuating mechanism

Abstract

The invention belongs to the technical field of fault diagnosis, and discloses a magneto-rheological semi-active suspension actuator fault diagnosis system and method, and the method comprises the steps: constructing a training set based on historical execution data; constructing a fault classification model based on a machine learning algorithm, and training the fault classification model by using the training set; and collecting real-time execution data, inputting the real-time execution data into the trained fault classification model, and outputting to obtain a fault type related to the real-time execution data. The fault classification model comprises a feature extraction layer, a feature fusion layer and a classification layer; the feature extraction layer extracts time domain fault features and dimension reduction frequency domain fault features through a convolutional neural network and a fast Fourier transform dual channel; and the feature fusion layer performs superposition fusion on the time domain fault features and the dimensionality reduction frequency domain fault features, and performs dimensionality reduction again on the combined fault features obtained through superposition fusion through kernel principal component analysis. The feature information integrity is ensured, and the diagnosis confidence is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and in particular relates to a fault diagnosis system and method for a magnetorheological semi-active suspension actuator. Background Art

[0002] As a key component of the vehicle chassis, the suspension system plays a vital role in improving vehicle safety and comfort. Semi-active suspension, with its low energy consumption, simple structure, and ability to effectively improve vehicle comfort and handling, is a key development direction in suspension technology. The magnetorheological damper (MRD), a semi-active intelligent vibration reduction device with continuously adjustable damping, can achieve real-time dynamic adjustment of the damping force by varying the input current to control the applied magnetic field. With advantages such as simple structure, wide adjustable damping force range, rapid response, and low energy consumption, it has become a new generation of semi-active controllers in automotive semi-active suspension.

[0003] However, the strong hysteretic nonlinearity of MRDs not only poses challenges to their modeling and control but also significantly increases the complexity of fault diagnosis. Traditional fault diagnosis techniques typically rely on linear or non-weakly linear assumptions, making them difficult to directly apply to highly nonlinear systems like MRDs. Furthermore, MRDs are actuators that are constantly exposed to complex vibration, temperature fluctuations, and electromagnetic interference, making them prone to a variety of typical faults.

[0004] In addition, MRD fault diagnosis technology mostly focuses on a single fault type, such as only detecting coil current abnormalities, and lacks comprehensive diagnostic capabilities for multi-fault coupling scenarios, resulting in an increased false alarm rate. Therefore, there is an urgent need for a highly robust, low-dependency online fault diagnosis system that can realize rapid identification of multi-modal faults of magnetorheological suspension actuators under limited sample conditions, providing support for fault-tolerant control and maintenance decisions. Summary of the Invention

[0005] In view of this, in order to solve the problems raised in the above background technology, an object of the present invention is to provide a magnetorheological semi-active suspension actuator fault diagnosis system and method.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A magnetorheological semi-active suspension actuator fault diagnosis method comprising:

[0008] Build a training set based on historical execution data;

[0009] Building a fault classification model based on a machine learning algorithm, and training the fault classification model using the training set;

[0010] Real-time execution data is collected, the real-time execution data is input into a trained fault classification model, and the fault type related to the real-time execution data is output.

[0011] Preferably, the fault diagnosis method further includes preprocessing the historical execution data and the real-time execution data.

[0012] Preferably, the preprocessing includes filtering and denoising:

[0013] ;

[0014] Among them, the cutoff frequency It means filtering out high-frequency noise above 500Hz in the data. This represents a 4th order filter with a roll-off rate of -80dB / decade.

[0015] Preferably, the preprocessing includes normalization:

[0016] ;

[0017] in, is the minimum value in the data set, is the maximum value in the data set, The data are normalized.

[0018] Preferably, the preprocessing includes segmented windowing:

[0019] ;

[0020] Where N=512 indicates that the window length is consistent with the frame length, and n is the position number of each data point in a covered frame signal.

[0021] Preferably, the fault classification model includes a feature extraction layer and a feature fusion layer.

[0022] Preferably, the feature extraction layer is used to extract time domain fault features and dimension-reduced frequency domain fault features through a convolutional neural network and fast Fourier transform dual channel.

[0023] Preferably, the feature fusion layer is used to superimpose and fuse the time domain fault features and the dimension-reduced frequency domain fault features, and is also used to re-dimension-reduce the joint fault features obtained by superposition and fusion through kernel principal component analysis.

[0024] Preferably, the fault classification model uses the new joint features obtained after re-dimensionality reduction as input and the fault type as output, and the fault type includes normal execution, damper failure, magnetorheological fluid performance degradation and elastic element fatigue.

[0025] As a general inventive concept, the present invention also provides the following technical solutions:

[0026] A magnetorheological semi-active suspension actuator fault diagnosis system, comprising:

[0027] Data classification module; the data classification module is used to classify and label historical execution data according to fault type, and use the historical execution data with classification labels to construct a training set;

[0028] Model building module; the model building module builds a fault classification model based on a machine learning algorithm;

[0029] Training module; the training module trains the fault classification model based on the training set;

[0030] Data acquisition module; the data acquisition module is used to collect real-time execution data;

[0031] Control module; the control module is used to control the trained fault classification model to take the real-time execution data as input and the fault type as output.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention realizes the extraction of magnetorheological damper fault characteristics, and extracts the time domain characteristics and frequency domain characteristics of the magnetorheological damper separately. Compared with the traditional single time domain feature extraction and frequency domain feature extraction, it covers more complete fault information, avoids the lack of information integrity in single time domain feature and single frequency domain feature extraction, and combines convolutional neural network and fast Fourier transform to improve diagnostic confidence. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flow chart of the fault diagnosis method of the present invention;

[0035] Figure 2 A flow chart showing fault classification using the fault diagnosis method of the present invention;

[0036] Figure 3 This is a structural block diagram of the fault diagnosis system of the present invention. DETAILED DESCRIPTION

[0037] To further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the content disclosed in the specification, so that people familiar with the technology can understand and read them. They are not used to limit the limitations of the implementation of the present invention and therefore have no technical significance. Any structural modifications, changes in proportional relationships, or adjustments in size should still fall within the scope of the technical content disclosed in the present invention without affecting the efficacy and objectives that can be achieved by the present invention. At the same time, terms such as "upper", "lower", "left", "right", and "middle" used in this specification are only for ease of description and are not used to limit the scope of implementation. Changes or adjustments in their relative relationships should also be considered as the scope of implementation of the present invention without substantially changing the technical content. It should be noted that the terms "first", "second", etc. in the specification and claims of this application 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 are interchangeable where appropriate for the embodiments of the present application described herein.

[0038] Example 1

[0039] like Figure 1 As shown, a magnetorheological semi-active suspension actuator fault diagnosis method provided by an embodiment of the present invention includes the following steps:

[0040] S1. Build a training set based on historical execution data;

[0041] Specifically, historical execution data such as vibration pressure, piston displacement and control current signals are collected by pressure sensors, displacement sensors and current sensors installed on the magnetorheological semi-active suspension actuator. The historical execution data are classified and marked according to the fault type. In the present invention, the fault types are divided into four categories: normal execution, damper failure, magnetorheological fluid performance degradation and elastic element fatigue.

[0042] During bench testing:

[0043] Collect normal data of the vehicle when it is driving at a normal constant speed of 60 km / h on a Class B road (ISO 8608:2016 standard), including 2000 sets of vibration pressure, piston displacement and control current signals;

[0044] Through physical intervention, the system simulates damper failure and collects abnormal data while the vehicle is traveling at 60 km / h on a Class B road. This includes 800 sets of vibration pressure, piston displacement, and control current signals.

[0045] Physical intervention was used to simulate the performance degradation of magnetorheological fluid (MRF). Abnormal data was collected while the vehicle was traveling at 60 km / h on a Class B road. The data included 600 sets of vibration pressure, piston displacement, and control current signals.

[0046] Through physical intervention, elastic component fatigue is simulated, and abnormal data is collected when the vehicle is driving at 60km / h on a Class B road, including 600 sets of vibration pressure, piston displacement and control current signals.

[0047] Then the following training set is constructed:

[0048] Fault type Training data Verify data Test data Execution is normal 1400 300 300 Damper failure 560 120 120 Magnetorheological fluid performance degradation 420 90 90 elastic element fatigue 420 90 90

[0049] S2. Preprocessing the historical execution data contained in the training set;

[0050] Specifically, the preprocessing includes filtering denoising, normalization and segmented windowing;

[0051] (1) Filtering and denoising: ; Among them, the cutoff frequency It means filtering out high-frequency noise above 500Hz in the data. This represents a 4th order filter with a roll-off rate of -80dB / decade.

[0052] (2) Normalization: ;in, is the minimum value in the data set, is the maximum value in the data set, The data are normalized.

[0053] (3) Segmented windowing: 512 points / frame, frame shift 256 points, that is, each second of data is divided into 3 frames;

[0054] Hanning window function: Where N=512 means the window length is the same as the frame length, ensuring that each frame of data is completely windowed. n is the position number of each data point in a frame of signal covered by the window function.

[0055] S3. Constructing a fault classification model based on a machine learning algorithm and training the fault classification model using the training set;

[0056] Specifically, the fault classification model includes a feature extraction layer, a feature fusion layer and a classification layer;

[0057] As can be seen from the above, the training set includes training data, verification data and test data. First, the fault classification model is trained using the training data:

[0058] S31. The feature extraction layer extracts time domain fault features and dimension-reduced frequency domain fault features through a convolutional neural network and fast Fourier transform dual channels:

[0059] (1) Construct a convolutional neural network to extract time domain fault features:

[0060] A 5-layer convolutional neural network structure is used, including the input layer, the first convolutional layer, the second convolutional layer, the global pooling layer and the fully connected layer;

[0061] Input layer: accepts the preprocessed time domain signal sequence;

[0062] The first convolutional layer: 32 convolution kernels with a width of 5, ReLU as the activation function, and output feature maps;

[0063] The second convolutional layer: 64 convolution kernels with a width of 3, ReLU as the activation function, and output feature maps;

[0064] Global pooling layer: compresses the time domain dimension to generate a 64-dimensional feature vector;

[0065] Fully connected layer: 10-node hidden layer, selects tanh activation function, and finally outputs a 10-dimensional feature vector;

[0066] In summary, the time domain fault characteristics are obtained .

[0067] (2) Fast Fourier Transform:

[0068] ; Where X(k) is the fast Fourier transform output, including real and imaginary parts, representing the kth frequency component; x(n) is the value of the nth sampling point of the signal; is the weight of the Hanning window at point n; N=512 is the number of data points after framing;

[0069] Single frame spectrum amplitude: ; Where A(k) is the output amplitude;

[0070] Band energy ratio: ; Where A(k) is the amplitude of the output corresponding to frequency k, and is the starting point and end point of the frequency band;

[0071] Harmonic distortion: ; The fundamental frequency is defined as the natural frequency of the suspension system. is the amplitude corresponding to the fundamental frequency, is the amplitude corresponding to the nth harmonic;

[0072] In summary, the 5-dimensional frequency domain fault characteristics are extracted based on the amplitude spectrum .

[0073] In addition, for the four major types of faults provided by the present invention: frequency band division: low frequency (0-20Hz) reflects normal vibration of the suspension, medium frequency (20-100Hz) reflects damper failure, and high frequency (100-500Hz) reflects magnetorheological fluid performance degradation; if the harmonic distortion rate is high, it indicates that the significant nonlinearity in the signal is caused by fatigue of the elastic element.

[0074] S32. The feature fusion layer superimposes and fuses the time domain fault features and the reduced-dimensional frequency domain fault features, and further re-reduces the dimension of the combined fault features obtained by superimposing and fusion through kernel principal component analysis;

[0075] (1) Feature fusion

[0076] The 10-dimensional time domain fault features and the 5-dimensional frequency domain fault features are combined to obtain a 15-dimensional feature vector (joint fault features). ; Input the 15-dimensional feature vector into the Gaussian kernel function to construct the kernel matrix: ;

[0077] Center the kernel matrix: Where is to find the mean of K columns, is to find the mean of K rows, is the global mean correction;

[0078] (2) Re-dimensionality reduction

[0079] Solve the principal component direction in the high-dimensional space corresponding to the joint fault feature: ;

[0080] Then perform projection dimensionality reduction: ;in k=5 is the dimension after dimensionality reduction;

[0081] In summary, the 5-dimensional joint fault features after dimensionality reduction are obtained.

[0082] S33. The classification layer performs nonlinear classification on the new joint features obtained after re-dimensionality reduction by a support vector machine (specifically, an RBF kernel function) to obtain the classified fault type;

[0083] The RBF kernel function is expressed as:

[0084] ; ;

[0085] The input training data is is the sample data, There are 4 types of faults.

[0086] Optimization constraints: ; Constraints are used to ensure that samples are classified normally and are located outside the interval );

[0087] In addition, it is allowed to adjust To soften the border: Where It is the implicit expression of the input feature map by the RBF kernel function, and b is the bias term used to control the hyperplane. is the normal vector of the hyperplane and is used to determine the direction of the boundary.

[0088] In summary, the five-dimensional joint fault features are classified into four major fault types according to the correlation and threshold range shown in the following table:

[0089] Table 1

[0090] principal component Main associated features Typical fault response Dimension 1 Low frequency energy and high frequency stability Normal dimension 1 is high, and it decreases significantly when tired Dimension 2 High frequency noise, THD nonlinearity Dimension 2 increases abnormally when magnetorheological fluid degrades Dimension 3 Medium frequency energy, damper impact Dimension 2 fluctuates violently when the damper fails Dimension 4 Waveform distortion, periodic attenuation Elastic fatigue dimension 4 decreases, degradation increases Dimension 5 Multi-band energy coupling Mixed Fault Dimension 5 Anomaly

[0091] Table 2

[0092] Fault type Dimension 1 Dimension 2 Dimension 3 Dimension 4 Dimension 5 normal [0.5,1.2] [-0.3,0.3] [-0.5,0.5] [-0.2,0.4] [-0.1,0.2] Damper failure [0.2,0.8] [-0.5,0.1] [1.0,2.0] [-0.5,0.0] [0.3,0.8] Magnetorheological fluid degradation [0.4,1.0] [1.5,2.5] [-0.3,0.5] [0.6,1.2] [-0.5,0.1] elastic element fatigue [0.0,0.4] [-0.2,0.4] [-0.8,0.0] [-0.8,-0.2] [-0.3,0.3]

[0093] After the training is completed, the validation data is used to verify the classification results, and the test data is used to test the accuracy of the classification results to ensure that the trained fault classification model has a classification accuracy of more than 90%.

[0094] S4 collects real-time execution data, inputs the real-time execution data into the trained fault classification model, and outputs the fault type of the real-time execution data;

[0095] Specifically, if Figure 2 As shown, this step includes:

[0096] S41. Collect real-time execution data; collect real-time execution data such as vibration pressure, piston displacement, and control current signals through the pressure sensor, displacement sensor, and current sensor installed on the magnetorheological semi-active suspension actuator.

[0097] S42. Preprocess the real-time execution data; specifically, the preprocessing method of this step is the same as the preprocessing method of step S2.

[0098] S43. The feature extraction layer extracts time domain fault features and dimensionality-reduced frequency domain fault features through a convolutional neural network and a fast Fourier transform dual channel;

[0099] S44. The feature fusion layer superimposes and fuses the time domain fault features and the reduced-dimensional frequency domain fault features, and further re-reduces the dimension of the combined fault features obtained by superimposing and fusion through kernel principal component analysis;

[0100] S45. The classification layer performs nonlinear classification on the new joint features obtained after re-dimensionality reduction through the RBF kernel function to obtain the classified fault type.

[0101] Example 2

[0102] like Figure 3 As shown, a magnetorheological semi-active suspension actuator fault diagnosis system provided by an embodiment of the present invention includes the following structure:

[0103] Data classification module; the data classification module is used to classify and label historical execution data according to fault type, and use the historical execution data with classification labels to construct a training set;

[0104] Model building module; the model building module builds a fault classification model based on a machine learning algorithm;

[0105] Training module; the training module trains the fault classification model based on the training set;

[0106] Data acquisition module; the data acquisition module is used to collect real-time execution data;

[0107] Control module; the control module is used to control the trained fault classification model to take the real-time execution data as input and the fault type as output.

[0108] Example 3

[0109] An embodiment of the present invention provides an intelligent terminal including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the steps of the method described in the first embodiment.

[0110] Example 4

[0111] An embodiment of the present invention provides a computer-readable storage medium, on which a magnetorheological semi-active suspension actuator fault diagnosis program is stored. When the magnetorheological semi-active suspension actuator fault diagnosis program is executed by a processor, the steps of the method described in the above embodiment 1 are implemented.

[0112] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments, and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A magnetorheological semi-active suspension actuator fault diagnosis method, characterized in that: include: Build a training set based on historical execution data; Building a fault classification model based on a machine learning algorithm, and training the fault classification model using the training set; Real-time execution data is collected, the real-time execution data is input into a trained fault classification model, and the fault type related to the real-time execution data is output.

2. The magnetorheological semi-active suspension actuator fault diagnosis method according to claim 1, characterized in that: The method also includes pre-processing the historical execution data and the real-time execution data.

3. The magnetorheological semi-active suspension actuator fault diagnosis method according to claim 2, characterized in that: The preprocessing includes filtering and denoising: ； Among them, the cutoff frequency It means filtering out high-frequency noise above 500Hz in the data. This represents a 4th order filter with a roll-off rate of -80dB / decade.

4. The magnetorheological semi-active suspension actuator fault diagnosis method according to claim 2, characterized in that: The preprocessing includes normalization: ； in, is the minimum value in the data set, is the maximum value in the data set, The data are normalized.

5. The magnetorheological semi-active suspension actuator fault diagnosis method according to claim 2, characterized in that: The preprocessing includes segmented windowing: ； Where N=512 indicates that the window length is consistent with the frame length, and n is the position number of each data point in a covered frame signal.

6. The magnetorheological semi-active suspension actuator fault diagnosis and method according to claim 1, characterized in that: The fault classification model includes a feature extraction layer, a feature fusion layer and a classification layer.

7. The magnetorheological semi-active suspension actuator fault diagnosis and method according to claim 6, characterized in that: The feature extraction layer extracts time domain fault features and dimension-reduced frequency domain fault features through convolutional neural network and fast Fourier transform dual channels.

8. The magnetorheological semi-active suspension actuator fault diagnosis and method according to claim 7, characterized in that: The feature fusion layer superimposes and fuses the time domain fault features and the dimension-reduced frequency domain fault features, and further re-dimension-reduces the combined fault features obtained by superimposing and fusion through kernel principal component analysis.

9. The magnetorheological semi-active suspension actuator fault diagnosis and method according to claim 9, characterized in that: The classification layer performs nonlinear classification on the new joint features obtained after re-dimensionality reduction through the RBF kernel function to obtain classified fault types, and the fault types include normal execution, damper failure, magnetorheological fluid performance degradation and elastic element fatigue.

10. A magnetorheological semi-active suspension actuator fault diagnosis system, characterized in that: include: Data classification module; The data classification module is used to classify and label historical execution data according to fault types, and to construct a training set using the historical execution data with classification labels; Model building module; the model building module builds a fault classification model based on a machine learning algorithm; Training modules; The training module trains the fault classification model based on the training set; Data acquisition module; the data acquisition module is used to collect real-time execution data; Control module; The control module is used to control the trained fault classification model to take the real-time execution data as input and the fault type as output.

Images