Wheel hub motor electric automobile drive system commonality and fault feature collaborative extraction method

Abstract

The application discloses a kind of wheel hub motor electric vehicle drive system commonality and fault feature collaborative extraction method in the field of automobile and fault detection, respectively collect the motor bearing vibration signal of each subsystem and carry out time-frequency analysis, calculate all amplitude as element to form the time-frequency matrix of each subsystem, the time-frequency element of the feature of each subsystem under the same system internal working state or the same external coupling excitation is classified into a class, the element in the data set after classification is classified into common feature element set and individual feature element set using hierarchical clustering algorithm, set empty common feature matrix and empty individual feature matrix with the same row and column of time-frequency matrix, all elements in common feature element set and individual feature element set are respectively placed in the same element position with its time-frequency, construct common feature matrix and individual feature matrix, get common feature parameter and individual feature parameter by inverse transformation, improve the accuracy and timeliness of feature extraction.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of automobile and fault detection, and relates to the extraction of fault features of a wheel hub motor electric vehicle drive system, in particular to an extraction method of common features and fault features of a wheel hub motor electric vehicle drive system. BACKGROUND

[0002] The wheel hub motor is a hot topic in the new energy vehicle industry at present. For a vehicle, the wheel hub motor can make the structure of the vehicle simpler, and the drive system composed of the wheel hub motor is more flexible, which can realize various complex driving modes. Compared with other drive systems, the wheel hub motor is also more suitable for use in new energy vehicles. However, due to the special working position and complex road conditions of the wheel hub motor, the wheel hub motor is easily affected by the external environment when working normally, such as being attacked by water, dust and other factors, which leads to the failure of the wheel hub motor. Therefore, the research on the extraction of wheel hub motor fault features is a key link for the operation and maintenance of the drive system. At present, for the fault feature extraction of the automobile drive system with multiple wheel hub motors, whether it is for the local fault feature of the subsystem or for the global common feature of the drive system, it is mostly based on the state extraction of the single subsystem itself, for example:

[0003] (1) In 2019, XUE et al. published a wheel hub motor bearing fault feature extraction method based on AHN in the journal "Journal of Huazhong University of Science and Technology (Natural Science Edition)", which uses the characteristics of organic compounds and their molecules to encapsulate information, separates high and low frequency signals in the signal, filters out interference signals and extracts fault feature signals under strong intermittent interference, and then realizes the fault feature extraction of the wheel hub motor bearing. However, this method only uses the vibration signal of a single motor bearing as the input of the feature extraction method to extract the fault feature signal, and in the automobile drive system with multiple wheel hub motors, only the single fault feature of each subsystem can be extracted, which has limitations.

[0004] (2) In 2020, LEE et al. published "Optimised approach of feature selection based on genetic and binary state transitionalgorithm in the classification of bearing fault in BLDC motor" in IET Electric Power Applications, which proposed a feature extraction algorithm based on genetic and binary state transition algorithm, which can effectively extract motor bearing fault information, and has high extraction accuracy and strong pertinence under certain conditions. However, this method is difficult to maintain relatively stable fault feature extraction effect in the full working domain of the hub direct drive system, which has wide range, frequent changes and strong randomness. SUMMARY

[0005] Based on the above problems, the present application provides a method for collaborative extraction of common features and fault features of a hub motor electric vehicle drive system based on time-frequency analysis theory and hierarchical clustering algorithm, to improve the accuracy and timeliness of fault features.

[0006] To achieve the above purpose, the technical scheme adopted by the present application comprises the following steps:

[0007] Step 1): According to the operation scene and external excitation condition of the hub motor electric vehicle drive system, the motor bearing vibration signals of each subsystem of the drive system are collected respectively, the motor bearing vibration signals are subjected to time-frequency analysis, the amplitude of each motor bearing vibration signal of each subsystem at each time and each frequency is calculated, and all amplitudes are used as elements to form a time-frequency matrix of each subsystem;

[0008] Step 2): The time-frequency elements of the features of each subsystem under the same system internal working state or the same external coupling excitation are classified into a class, and the classified data set is obtained;

[0009] Step 3): The elements in the classified data set are classified by using a hierarchical clustering algorithm, and the corresponding blocks are obtained after merging the blocks, and according to the distribution of the elements in the corresponding blocks, the elements are classified into a common feature element set and an individual feature element set;

[0010] Step 4): Set an empty common feature matrix and an empty individual feature matrix with the same row and column as the time-frequency matrix described in step 1), and put all elements in the common feature element set described in step 3) into the empty common feature matrix at the same time-frequency element position, respectively, and put all elements in the individual feature element set into the empty individual feature matrix at the same time-frequency element position, respectively, and fill the positions of the elements not stored with zeros to construct the common feature matrix and the individual feature matrix of each subsystem;

[0011] Step 5): Inverse transform the common feature matrix and the individual feature matrix of all subsystems to obtain the common feature parameters representing the common feature of the system and the individual feature parameters representing the fault feature of the system in the time domain.

[0012] Further, in step 2), when the road condition is simple, the elements with the same frequency and the same time in the time-frequency matrix are selected to be classified into one class; when the road condition is complex, the corresponding elements at the 1st time point in the 1st time-frequency matrix, the corresponding elements at the 2nd time point in the 2nd time-frequency matrix, and the corresponding elements at the n / 2th time point in the n / 2th time-frequency matrix are selected to be classified into one class.

[0013] Further, the hierarchical clustering algorithm in step 3) is:

[0014] Step A: Define each element in the sample set as an independent initial block, and take the mean value of all elements in each initial block as the block center value,

[0015] Step B: Calculate the block distance between all pairs of blocks, and when the minimum block distance d min ≤α, α is a set threshold, merge several blocks with the minimum block distance d min into a new block, and calculate the mean value of the new block to obtain the block center value of the new block;

[0016] Step B: Repeat step B until d min >α, to obtain the corresponding block after the merged block.

[0017] Further, the distribution of the elements is:

[0018] Case 1: There is a block with more than p / 2 elements in β blocks with β>1, and p is the total number of elements in the data set, then all elements in the block are common feature elements and are classified into the common feature element set, and the elements in the remaining blocks are classified into the individual feature element set:

[0019] Case 2: in the block of beta = 1, increase the threshold alpha once, and reclassify the elements of the data set based on the hierarchical clustering algorithm, then execute the case 1, if beta = 1, then all the elements in the data set are included in the common feature element set;

[0020] Case 3: in the beta block of beta > 1, there is a block containing the most elements, but the number of elements in the block is less than or equal to q / 2, reduce the size of the threshold alpha once, and reclassify the elements of the data set based on the hierarchical clustering algorithm, then execute the case 1, if the number of elements in the block containing the most elements is still less than or equal to p / 2, then the common feature element set and the individual feature element set cannot be obtained;

[0021] Case 4: if there are several blocks containing the same number of elements after classification, the common feature element set and the individual feature element set cannot be obtained.

[0022] The beneficial effects of the present application are:

[0023] 1. The present application faces the characteristics of wide range, frequent change and strong randomness of the hub direct drive system full working domain, based on the hierarchical clustering algorithm, the common features and fault features of the motor bearing vibration signals of each subsystem of the hub direct drive system can be extracted at the same time, and the accuracy and timeliness of feature extraction are improved.

[0024] 2. The present application processes the motor bearing vibration signals of each subsystem of the hub direct drive system based on the time-frequency analysis method, through the method of constructing time-frequency matrix, it is beneficial to classify and process the information of each subsystem at the same frequency, thereby establishing a synchronous motor bearing vibration signal time-frequency matrix collaborative decomposition mechanism of each subsystem facing the system common feature filtering and local fault feature extraction, which can take the real-time system common feature as the benchmark of signal processing, and is beneficial to find the local fault feature of the subsystem under the current state, and maintains the fault feature extraction effect in the full working domain.

[0025] 3. The present application can classify the features under the same internal working state or external coupling excitation of each subsystem of the hub direct drive system. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is the collaborative extraction flowchart of the common features and fault features of each subsystem of the hub direct drive system of the present application;

[0027] Figure 2 is Figure 1 the flowchart of extracting the common feature elements and fault feature elements of the hub direct drive system based on the hierarchical clustering algorithm in the present application;

[0028] Figure 3 is the classification flow chart of the data set under different classification conditions based on the hierarchical clustering algorithm. DETAILED DESCRIPTION

[0029] In the hub motor electric vehicle driving system, each subsystem is subsystem 1, subsystem 2, …, subsystem n / 2, …, subsystem n, and n is the total number of subsystems. Among them, subsystem 1, subsystem 2, …, subsystem n / 2 are the subsystems on the same side of the hub motor electric vehicle, and subsystem n / 2+1, subsystem n / 2+2, …, subsystem n are the subsystems on the other side; Subsystem 1 and subsystem n / 2+1 are subsystems belonging to the same axle, subsystem 2 and subsystem n / 2+2 are subsystems belonging to the same axle, and subsystem n / 2 and subsystem n are subsystems belonging to the same axle, and so on.

[0030] In the real running scene of the hub motor electric vehicle driving system, the hub motor bearing vibration signals of each subsystem are collected, and the bearing vibration signals corresponding to each subsystem are respectively: x (1) , x (2) , …, x (n) Such signals have the characteristics of fast instantaneous frequency change and no periodicity, and belong to non-stationary signals. Therefore, the present application needs to use a time-frequency analysis method suitable for non-stationary signals, such as short-time Fourier transform, wavelet transform with high, medium and low frequency analysis capability, Hilbert-Huang transform which can perform empirical mode decomposition on signals, and other time-frequency analysis methods. The vibration signals of each subsystem are analyzed by time-frequency analysis to construct a time-frequency matrix. The specific steps are as follows:

[0031] According to the running scene of the hub motor electric vehicle driving system and the change of external excitation conditions, the motor bearing vibration signals x (1) , x (2) , …, x (n) of subsystem 1, subsystem 2, …, subsystem n are collected at t (0) , t (1) , t (2) , …, t (M) in the period of 0 to t M , the starting sampling time point t (0) =0, the sampling time interval is T k , T k is the time difference between the k+1 time point t (k) and the k time point t (k-1) , k=1,2,...,M, M is the number of time points, and M is greater than n. The size of T k is the ratio of the wheelbase to the instantaneous speed of the electric vehicle at that moment.

[0032] Vibration signals x of the motor bearings in subsystems 1, 2, ..., n are collected at the same frequency interval Δf. (1) x (2) , ..., x (n) Y frequency points f (1) f (2) , ..., f (Y) Frequency interval

[0033] This invention selects short-time Fourier transform to analyze the vibration signal x of the motor bearing. (1) x (2) , ..., x (n) For time-frequency analysis, the formula for the short-time Fourier transform is as follows:

[0034]

[0035] in, The vibration signal x of the motor bearing in subsystem i (i) At the corresponding time point t (m) Corresponding frequency point f (y) The magnitude of the amplitude.

[0036] See Figure 1 As shown, based on the time-frequency analysis transformation formula, the vibration signal x of each hub motor bearing in each subsystem can be calculated. (1) x (2) , ..., x (n) At each time t (1) , t (2) , ..., t (M) With each frequency f (1) f (2) , ..., f (Y) The amplitude below All amplitudes The time-frequency matrix S, which is composed of elements of each subsystem, (1) S (2) S (n) :

[0037]

[0038] Based on the actual operating scenarios of in-wheel motor electric vehicle drive systems, all time-frequency matrices S (1) S (2) S (n) amplitude in The correspondence of elements will be able to reflect the characteristics of each subsystem under the same internal operating state (e.g., the hub motors of each subsystem operate under the same internal load) or the same external coupling excitation (e.g., each subsystem is subjected to the same road surface excitation) in terms of time-frequency elements (amplitude). They are classified into one category. Based on the actual operating scenarios of hub motor electric vehicle drive systems, the classification criteria are as follows:

[0039] 1. When a hub motor electric vehicle operates under simple road conditions (i.e., from 0 to t) M within the time period t (1) , t (2) , ..., t (M) At any given moment, when the road conditions experienced by the drive system are approximately the same, the time-frequency matrix S is selected. (1) S (2) S (n) Elements of the same frequency and time can characterize the system characteristics of each subsystem at the same frequency and time, and can be classified into one category.

[0040] 2. When a hub motor electric vehicle is driving on complex road conditions, the first time-frequency matrix S is selected at the same frequency. (1) The first time point t in (1) The corresponding elements, the second time-frequency matrix S (2) The second time point t in (2) The corresponding elements, ..., up to the n / 2th time-frequency matrix S (n / 2) The n / 2th time point t (n / 2) The corresponding elements can characterize the system features of each subsystem under the same real-time road conditions, and therefore can be classified into one category.

[0041] By analyzing different scenarios during the operation of the hub motor electric vehicle drive system, and selecting appropriate classification conditions, the time-frequency matrix S is... (1) S (2) S (n) all elements Classification is performed. This invention covers the period from 0 to t. M Within the time period, the time-frequency matrix S (1) S (2) S (n) The elements in the dataset are divided into R classes based on the different states of the system. Each class constitutes a dataset, resulting in R categorized datasets. The elements in each dataset represent the system characteristics driving the system in that state, denoted as dataset Q. (1) Q (2) ,...,Q (R) This dataset serves as the input sample set for subsequent hierarchical clustering algorithms.

[0042] Based on the hierarchical clustering algorithm in the theory of machine learning, the elements in the data set Q (r) are classified respectively, r = 1, 2,..., R. As shown in Figure 2 , the specific process is as follows:

[0043] Step A: In the initial stage, the total number of elements in the input sample set Q (r) is p, and each of the p elements is defined as an independent initial block In the initial block , the mean value of all elements in each block is taken as the block center value, and each block center value is denoted as Based on the block center value, the block distance between two blocks is defined as And g ≠ h.

[0044] Step B: Calculate the block distance between all pairs of blocks, and compare the sizes of all block distances, and record the minimum block distance as d min . If d min ≤ α, α is a threshold value set by the present application according to the size of the elements in the data set, which can be flexibly adjusted when performing hierarchical clustering algorithm classification on different data sets, then a number of blocks with the minimum block distance d min are merged into a new block, and the mean value of the new block is calculated to obtain the block center value of the new block. If d min > α, stop merging blocks.

[0045] Step C: Repeat step B until d min > α, i.e. stop. Finally, the corresponding block (r) of the input sample set Q (r) after merging blocks is obtained, and β is the number of blocks.

[0046] Based on the classification principle of hierarchical clustering algorithm, the elements in the same block have high similarity. Therefore, for the distribution of elements in each block in the block , the elements are classified into common feature element set z (r) and individual feature element set g (r) . As shown in Figure 3 , the specific element distribution and classification are as follows:

[0047] Classification case 1: if β > 1 after the hierarchical clustering algorithm ends, and there is a block D r (λ) in the β blocks, the number of elements in D r (λ) is greater than p / 2, p is the sample set Q (r)The total number of elements in the block D (r) , the elements in the block D r (λ) , the elements in the block D r (λ) , and the elements in the block D (r) represent the common characteristics of the system in the working state. At the same time, the remaining elements in the data set Q (r) are classified as the individual characteristic element set g (r) , representing the fault characteristics of the system in the working state.

[0048] Classification case 2: if β = 1 after the hierarchical clustering algorithm ends, that is, all the elements in the data set Q (r) are in the block D r (1) , it is indicated that the common characteristics and the fault characteristics of the system in the state cannot be distinguished at the threshold α. Therefore, in the classification case 1, the size of the threshold α is appropriately increased once (the value of the threshold α in the present application is changed at most once), and the increase amplitude is controlled at about 5% of the threshold α, which can amplify the fault characteristics of the system in the state and reclassify the elements of the data set Q (r) based on the hierarchical clustering algorithm. If β > 1 after the second classification ends, and there is a block D r (λ) , the number of elements in the block D r (λ) is greater than p / 2, the elements in the block D r (λ) are defined as the common characteristic elements and are classified in the common characteristic element set z (r) , representing the common characteristics of the system in the state; and all the elements in the remaining blocks are classified as the individual characteristic element set g (r) , representing the fault characteristics of the system in the state. If β = 1 after the second classification ends, it is indicated that the subsystems of the system in the state run smoothly and no fault elements appear, and therefore, all the elements in the data set Q (r) are defined as the common characteristic elements and are classified in the common characteristic element set z (r) .

[0049] Classification case 3: if β > 1 after the hierarchical clustering algorithm ends, and there is a block D r containing the most elements among the β blocks.(λ) But D r (λ) The number of elements in the block is less than or equal to q / 2. In classification case 3, on the one hand, it indicates that block D... r (λ) The number of elements in D is limited, which cannot effectively characterize the common features of the system in this state, and because D r (λ) The elements in the block have high similarity and are still relatively numerous compared to the elements in other blocks, and cannot characterize the fault features of the system in this state; on the other hand, it may also indicate that the common features and fault features of the system in this state cannot be distinguished at the threshold α.

[0050] Therefore, in classification case 3, appropriately reducing the threshold α once (in this invention, the value of threshold α is changed at most once), with the reduction controlled to around 5% of threshold α, can amplify the common features of the system in this state, and then the dataset Q... (i) The elements are reclassified based on a hierarchical clustering algorithm. If, after the second classification, β > 1, and there exists a block D among the β blocks... r (λ) D r (λ) If the number of elements in block D is greater than p / 2, then block D... r (λ) The elements in the set are defined as common feature elements, and they are categorized into the common feature element set z. (r) In this context, the common characteristics of the system under this state are represented; simultaneously, all elements in the remaining blocks are categorized into a set of individual characteristic elements g under this external stimulus. (r) This represents the fault characteristics of the system under this state. If, after the second classification, the number of elements in the block containing the most elements is still less than p / 2, it indicates that the common characteristics of the system under the internal working state or external coupling excitation conditions are not obvious, and the set of common characteristic elements cannot be obtained; correspondingly, the set of individual characteristic elements representing the internal working state or external coupling excitation conditions of the system also cannot be obtained.

[0051] Classification Case 4: If, after classification, there are several blocks containing the same number of elements, it indicates that the common characteristics of the system under this external coupling stimulus are not obvious and cannot effectively represent the common characteristics of the system under its internal working state or external coupling stimulus. Therefore, dataset Q... (r) It is impossible to obtain a set of common feature elements; correspondingly, it is also impossible to obtain a set of individual feature elements that characterize the system under internal working conditions or external coupling excitation conditions.

[0052] Based on this, respectively, dataset Q (1) Q (2)..., Q (R) The hierarchical clustering algorithm is adopted to obtain the common characteristic element set z of the hub direct drive system under different internal working conditions or external coupling excitation conditions (1) ,z (2) ..., z (R) and the individual characteristic element set g (1) ,g (2) ..., g (R) , and the union of all common characteristic element sets is obtained to obtain the global common characteristic element set z of the system; correspondingly, the union of all individual characteristic element sets is obtained to obtain the global individual characteristic element set g of the system.

[0053] In order to more clearly and intuitively reflect the common characteristics and fault characteristics of each subsystem under the corresponding time-frequency conditions, and obtain the characteristic parameters in system fault diagnosis, based on the obtained common characteristic element set z and individual characteristic element set g, the common characteristic matrix Z (1) ,Z (2) ..., Z (n) and the individual characteristic matrix G (1) ,G (2) ..., G (n) of each subsystem are constructed.

[0054] Each element in the common characteristic element set is placed in the empty common characteristic matrix of the corresponding subsystem before classification, and is placed at the corresponding time-frequency position of the element; correspondingly, each element in the individual characteristic element set is placed in the individual characteristic matrix of the corresponding subsystem before classification, and is placed at the corresponding time-frequency position of the element. The specific steps are as follows:

[0055] First, for subsystem 1, subsystem 2,..., and subsystem n, the common characteristic matrix Z (i) ,Z (1) ..., Z (2) in the same row and column as the time-frequency matrix S (n) (i=1, 2,..., n) is set, and the individual characteristic matrix G (i) ,G (1) ..., G (2) in the same row and column as the time-frequency matrix S (n) (i=1, 2,..., n) is set. At this time, Z (1) ,Z (2) ..., Z (n) and G (1) ,G (2) ..., G (n) are all empty matrices.

[0056] Further, all elements in the global common feature element set z are put into the empty common feature matrix of the original subsystem respectively at the same time-frequency position as itself. For example, assuming that element is classified into the common feature element set z, which is the element in the mth row and yth column of the original time-frequency matrix S (j) , then element is put into the mth row and yth column of the common feature matrix Z (j) . Based on this, all elements in the common feature element set z are put into the common feature matrix Z (1) , Z (2) ,..., Z (n) . If there is no common feature element in a time point and a frequency point of the common feature matrix of a subsystem, then zero is filled in the position of the time point and the frequency point.

[0057] Similarly, all elements in the global individual feature element set g are put into the empty individual feature matrix G (1) , G (2) ,..., G (n) . If there is no individual feature element in a time point and a frequency point of the individual feature matrix of a subsystem, then zero is filled in the position of the time point and the frequency point.

[0058] Based on this, the empty matrix Z (i) and G (i) are filled with all element positions to obtain the common feature matrix Z (1) , Z (2) ,..., Z (n) and the individual feature matrix G (1) , G (2) ,..., G (n) of each subsystem.

[0059] Based on the time-frequency analysis theory, inverse transformation is performed on the common feature matrix Z (1) , Z (2) ,..., Z (n) and the individual feature matrix G (1) , G (2) ,..., G (n) of all subsystems, such as short-time inverse Fourier transform method, to obtain the common feature matrix Z (1) , Z (2) ,..., Z (n) and the individual feature matrix G (1) , G (2) ,..., G (n)The elements in the matrix are calculated to obtain common characteristic parameters and individual characteristic parameters of each subsystem in the time domain, which represent common characteristics and fault characteristics of the system, and are used as inputs for system fault type diagnosis.

Claims

1. A wheel hub motor electric vehicle drive system commonality and fault feature collaborative extraction method, characterized by The method comprises the following steps: Step 1): According to the operation scene of the in-wheel motor electric vehicle driving system and external excitation conditions, the motor bearing vibration signals of each subsystem of the driving system are collected respectively, time-frequency analysis is performed on the motor bearing vibration signals, the amplitude of each motor bearing vibration signal of each subsystem at each time and each frequency is calculated, and all amplitudes are used as elements to form a time-frequency matrix of each subsystem; Step 2): The characteristic time-frequency elements of each subsystem under the same system internal working state or the same external coupling excitation are classified into a category to obtain a classified data set; Step 3): The elements in the classified data set are classified by using a hierarchical clustering algorithm, and corresponding blocks are obtained after merging the blocks, and the elements are classified into a common characteristic element set and an individual characteristic element set according to the distribution of the elements in the corresponding blocks; Step 4): A common characteristic matrix and an individual characteristic matrix are set, all elements in the common characteristic element set are respectively placed in the common characteristic matrix at the same time-frequency element position, all elements in the individual characteristic element set are respectively placed in the individual characteristic matrix at the same time-frequency element position, and zero is filled in the positions of the unplaced elements to construct the common characteristic matrix and the individual characteristic matrix of each subsystem; Step 5): The common characteristic matrix and the individual characteristic matrix of all subsystems are inversely transformed to obtain common characteristic parameters representing the common characteristics of the system and individual characteristic parameters representing the fault characteristics of the system in the time domain.

2. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method according to claim 1, characterized in that: In step 2), when the road condition is simple, the elements at the same frequency and the same time in the time-frequency matrix are classified into a category; when the road condition is complex, the corresponding elements at the first time point in the first time-frequency matrix, the corresponding elements at the second time point in the second time-frequency matrix, and the corresponding elements at the n / 2 time point in the n / 2 time-frequency matrix are classified into a category.

3. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 1, characterized in that: The hierarchical clustering algorithm in step 3) is as follows: Step A: Each element in the sample set is defined as an independent initial block, and the mean value of all elements in each initial block is used as the block center value, Step B: Calculate the block distance between all pairs of blocks, and when the minimum block distance d min ≤ α, α is a set threshold, merge several blocks with the minimum block distance d min into a new block, and calculate the mean value of the new block to obtain the block center value of the new block; Step B: Repeat Step B until d min > a, to obtain the corresponding block after the merging of the blocks.

4. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method according to claim 3, characterized in that: The distribution of the elements is as follows: Case 1: In the β blocks with β>1, there is a block with an element number greater than p / 2, p is the total number of elements in the data set, then all elements in the block are common characteristic elements and are classified into a common characteristic element set, and the elements in the remaining blocks are classified into an individual characteristic element set: Case 2: In the block with β=1, the threshold value α is increased by one, the elements in the data set are reclassified based on the hierarchical clustering algorithm, and then the case 1 is performed; if β is still 1, all elements in the data set are classified into a common characteristic element set. Case 3: there is one block containing the most elements in the beta blocks, but the number of elements in the block is less than or equal to q / 2, reduce the size of the threshold alpha once, and reclassify the elements of the data set based on the hierarchical clustering algorithm, then execute the case 1; if the number of elements in the block containing the most elements is still less than or equal to p / 2, the common feature element set and the individual feature element set cannot be obtained; Case 4: if there are several blocks containing the same number of elements after classification, the common feature element set and the individual feature element set cannot be obtained.

5. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method according to claim 1, characterized in that: Step 1) each subsystem is subsystem 1, 2, …, n / 2, …, n, n is the total number of subsystems, subsystem 1, 2, …, n / 2 are on the same side of the in-wheel motor electric vehicle, subsystem n / 2+1, n / 2+2, …, n are on the other side; Subsystem 1 and subsystem n / 2+1 belong to the same axle, subsystem 2 and subsystem n / 2+2 belong to the same axle, subsystem n / 2 and subsystem n are on the same axle, and so on.

6. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 1, characterized in that: In step 1), the motor bearing vibration signal x of subsystem i (i) At the corresponding time point t (m) , the amplitude at the corresponding frequency point f (y) is: Y is the number of frequency points, and M is the number of time points.

7. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 3, characterized in that: each block center value is the block distance between each pair of blocks and g≠h.

8. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 1, characterized in that: In step 1), the time interval for collecting the motor bearing vibration signal samples is the ratio of the wheelbase to the instantaneous speed of the electric vehicle at that moment.

9. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 1, characterized in that: The time-frequency matrix S (1) , S (2) ,..., S (n) is: is the amplitude.

10. The wheel motor electric vehicle drive system commonality and fault feature collaborative extraction method of claim 4, characterized in that: The amplitude of increasing or decreasing the threshold alpha is 5%.

Images