Personalized fault diagnosis customization method and system

Abstract

The invention relates to the field of fault diagnosis and health management, and discloses a personalized fault diagnosis customization method and system. According to the method, a transfer learning strategy is used in an initial stage, and a fault diagnosis model of each device is efficiently trained with fewer labeled samples. For a subsequent personalized customization process, personalized customization triggering rules are designed by considering factors which may influence the performance of the fault diagnosis model by state change of equipment and environment change in vehicle-line network information interaction, and a trigger is designed according to the rules. The trigger can quickly respond to changes of internal and external conditions and trigger layered fine adjustment of an equipment fault diagnosis model with performance changes, so that the influence of the conditions on the diagnosis performance is dealt with, and the reliability and safety of equipment are improved. The method effectively improves the fault diagnosis accuracy of different types and different numbers of equipment of the rail transit vehicle, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis and health management, and in particular to a personalized fault diagnosis customization method and system. Technical Background

[0002] With the rapid development of rail transit, there are many types of rail transit vehicles, and the number of vehicles of different types, the number of on-board equipment, and the complexity are increasing. The traditional fault diagnosis method for a single type of equipment and not considering the factors affecting the performance of the equipment under environmental changes can no longer meet the needs of diversified and complex changes in vehicle-line network conditions. Traditional fault diagnosis is usually based on offline data training to build a fixed parameter model, which lacks the ability to adapt to changes in the equipment's own state, such as device aging, component defects, equipment performance degradation, etc., and changes in the actual working environment of the vehicle-line network (rail transit network, lines, and vehicles actually running on the lines) that cause equipment performance degradation or failure. Different models, different equipment, and running on different lines, even if the same model and the same model of equipment are running on the same line, different environmental changes (such as sunny, windy, rainy, and snowy) may cause different changes in the equipment's own state, resulting in a significant decrease in the diagnostic generalization ability of the traditional fault diagnosis method based on the fixed parameter model. Therefore, there is an urgent need for a personalized fault diagnosis method and system that can be customized to consider the factors affecting the performance of the fault diagnosis model caused by changes in the equipment's own state and environmental changes in the vehicle-line network information interaction. Summary of the invention

[0003] The present invention aims to disclose a personalized fault diagnosis customization method and system to realize personalized customization of a fault diagnosis model.

[0004] To achieve the above objectives, in a first aspect, the present invention discloses a personalized fault diagnosis customization method, comprising:

[0005] S1: Collect sensor data of different models and numbered devices of rail transit vehicles to obtain number sets and model sets, and construct a mapping relationship from number sets to model sets;

[0006] S2: constructing a related set of source domain data sets of sensor data based on the mapping relationship, and constructing a related set of target source domain data sets of sensor data;

[0007] S3: Initialize a fault diagnosis model for each device, and train the fault diagnosis models of all devices based on the relevant set of the source domain data set and the relevant set of the target source domain data set to obtain the basic fault diagnosis model with optimal parameters for all devices;

[0008] S4: obtaining a sensor data set newly collected by the device in the vehicle-line-network information interaction and its corresponding fault category attribute label set, and collecting target semantic information, and constructing a personalized customized trigger rule encoding matrix based on the target semantic information;

[0009] S5: constructing an optimal decision tree model of the personalized customized trigger rule encoding matrix with respect to an optimal sub-matrix set;

[0010] S6: constructing a hierarchical freezing model based on the optimal basic fault diagnosis model of all the equipment parameters;

[0011] S7: Using the personalized customized trigger rule encoding matrix, the newly collected sensor data set and its corresponding fault category attribute label set, the hierarchical frozen model is fine-tuned to construct a personalized customized model.

[0012] To achieve the above-mentioned purpose, in a second aspect, the present invention further discloses a personalized fault diagnosis customization method and system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the computer program.

[0013] The present invention has the following beneficial effects

[0014] The personalized fault diagnosis customization method provided in this application uses a transfer learning strategy in the initial stage to efficiently train the fault diagnosis model of each device with fewer labeled samples. For the subsequent personalized customization process, by considering the factors that may affect the performance of the fault diagnosis model due to changes in the device's own state and environmental changes in the vehicle-line-network information interaction, personalized customization trigger rules are designed, and triggers are designed based on these rules. The trigger can quickly respond to changes in internal and external conditions, triggering hierarchical fine-tuning of the fault diagnosis model of the equipment with performance changes, thereby coping with the impact of conditions on diagnostic performance, and can effectively adapt to specific application scenarios and user needs, improve the accuracy and generalization ability of fault diagnosis of equipment of different models and numbers of rail transit vehicles, and thus improve the reliability and safety of rail transit vehicle operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a personalized fault diagnosis customization method according to a preferred embodiment of the present invention;

[0016] Figure 2 A personalized fault diagnosis customization method according to a preferred embodiment of the present invention uses a transfer learning strategy in the initial stage, and a visualization result diagram of source domain and target domain data features extracted by a feature extractor unit;

[0017] Figure 3A personalized fault diagnosis customization method according to a preferred embodiment of the present invention constructs a personalized customization model for the k-1th time Layered freezing model Schematic diagram;

[0018] Figure 4 A personalized fault diagnosis customization method according to a preferred embodiment of the present invention constructs a personalized customization model for the k-1th time Confusion matrix of model evaluation results before stratified fine-tuning;

[0019] Figure 5 A personalized fault diagnosis customization method according to a preferred embodiment of the present invention constructs a personalized customization model for the k-1th time Confusion matrix of model evaluation results after stratified fine-tuning; DETAILED DESCRIPTION

[0020] The technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the usual meanings understood by persons with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate quantity restrictions, but indicate the existence of at least one. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship also changes accordingly.

[0022] It is worth noting that the personalized fault diagnosis customization method provided by this application is mainly aimed at the field of rail transit and can be applied to rail transit vehicles including high-speed rail / EMU, ordinary speed trains, subway trains, light rail trains, monorail trains, maglev trains, heavy-load locomotives, etc. The maintenance, inspection and repair stations of these vehicles, such as the inspection and repair stations of high-speed rail / EMU, are EMU depots and EMU repair bases.

[0023] like Figure 1 As shown, this embodiment provides a personalized fault diagnosis customization method, including:

[0024] S1: Collect sensor data of different models and numbered devices in rail transit vehicles to obtain number sets and model sets, and construct a mapping relationship from number sets to model sets;

[0025] S2: constructing a related set of source domain data sets of sensor data based on the mapping relationship, and constructing a related set of target source domain data sets of sensor data;

[0026] S3: Initialize a fault diagnosis model for each device, and train the fault diagnosis models of all devices based on the relevant set of the source domain data set and the relevant set of the target source domain data set to obtain the basic fault diagnosis model with optimal parameters for all devices;

[0027] S4: obtaining a sensor data set newly collected by the device in the vehicle-line-network information interaction and its corresponding fault category attribute label set, and collecting target semantic information, and constructing a personalized customized trigger rule encoding matrix based on the target semantic information;

[0028] S5: constructing an optimal decision tree model of the personalized customized trigger rule encoding matrix with respect to an optimal sub-matrix set;

[0029] S6: constructing a hierarchical freezing model based on the optimal basic fault diagnosis model of all the equipment parameters;

[0030] S7: Using the personalized customized trigger rule encoding matrix, the optimal decision tree model, the newly collected sensor data set and its corresponding fault category attribute label set, the hierarchical frozen model is fine-tuned to construct a personalized customized model.

[0031] In this embodiment, the device is a device in a rail transit vehicle, such as a traction motor. Taking x vehicles under the jurisdiction of a motor vehicle depot as an example, there are y traction motors in each vehicle, and the total number of traction motors is B=xy, that is, there are B numbered devices (traction motors), and these B numbered traction motors belong to different models (different models of vehicles). For example, devices numbered 1 to 3 belong to model 1 (vehicle 1), devices numbered 4 to 6 belong to model 2 (vehicle 2), devices numbered 7 to 9 belong to model 3 (vehicle 3), devices numbered 10 to 12 belong to model 3 (vehicle 4), and so on; therefore, the device number b uniquely identifies a certain device of a certain vehicle, such as b=10 identifies the first device of model 3 of vehicle 4.

[0032] The personalized customization in this application is to customize a fault diagnosis model for each device on each vehicle. The optimal basic fault diagnosis model for all B devices is constructed based on historical data. For example, if the states of devices 1 and 5 change during the k-th vehicle-line network information interaction, and the environmental change affects the failure of device 10, the k-th personalized customization is performed using the data collected for the k-th time, that is, only the fault diagnosis model parameters of devices 1 (belonging to model 1), 5 (belonging to model 2), and 10 (belonging to model 3) are fine-tuned (different parameters of each device are fine-tuned according to different changes), and the model parameters of other devices 2, 3, 4, 6, 7, 8, 9, 11, and 12 are frozen, and the fault diagnosis model customized for the k-1th time is still used.

[0033] Among them, the sensor data is mainly time series such as the stator current sensor signal of the traction motor. The source domain dataset refers to the labeled dataset, and the target domain dataset refers to the unlabeled dataset. Specifically, both the source domain dataset and the target domain dataset are constructed from the original three-phase current data collected by the sensor and are subsets of the original dataset.

[0034] The above-mentioned personalized fault diagnosis customization method uses a transfer learning strategy in the initial stage to efficiently train the fault diagnosis model of each device with fewer labeled samples. For the subsequent personalized customization process, by considering the factors that may affect the performance of the fault diagnosis model due to changes in the device's own state and environmental changes in the vehicle-line-network information interaction, personalized customization trigger rules are designed, and triggers are designed based on these rules. The trigger can quickly respond to changes in internal and external conditions, triggering hierarchical fine-tuning of the fault diagnosis model of the device with performance changes, thereby coping with the impact of conditions on diagnostic performance. It can effectively adapt to specific application scenarios and user needs, improve the accuracy and generalization ability of fault diagnosis of equipment of different models and numbers of rail transit vehicles, and thus improve the reliability and safety of rail transit vehicle operation.

[0035] It should be noted that the device described in this embodiment is a device in a rail transit vehicle such as a traction motor, and the sensor data collected is a stator current signal of the traction motor. The main hyperparameters for fault diagnosis model training in this embodiment are shown in Table 1.

[0036] Table 1 Basic fault diagnosis model hyperparameters

[0037] Learning Rate Iterations Batch size Optimizer Learning rate adjustment strategy 0.01 200 64 SGD Cosine annealing

[0038] Further, based on the above parameters, the steps of the personalized fault diagnosis customization method provided by the present application are described in detail as follows:

[0039] S11: Collect sensor data of B numbered devices of A different models of rail transit vehicles to form an original data set D = {D1,...,Db ,…,D B}, Represents the number of sampling points of the sensor signal, that is, the data set D b The total number of elements in, b is the device number, b = 1, 2, ..., B, B is the total number of numbers; all device numbers constitute the number set C B ={1,2,…,b,…B}; all device models constitute the model set C A ={1,2,…,a,…A}, a is the device model, a=1,2,…,A, A is the total number of models;

[0040] Construct a mapping relationship from a number set to a model set, described as:

[0041] φ:C B →C A ; (1)

[0042] Among them, φ:C B →C A Represents the number set C B To model set C A The mapping of

[0043] If the device numbered b belongs to the ath type, it is recorded as:

[0044] φ(b)=a; a=1,2,…,A; (2)

[0045] Construct an index set of all device numbers belonging to the ath model, denoted as:

[0046] μ a ={b|φ(b)=a,b∈{1,2,…,B}}; (3)

[0047] In the formula, μ a Represents the index set of all device numbers belonging to the a-th model;

[0048] Take μ a The first index in represents the number of the first device belonging to the a-th type of equipment, min() represents the operation of taking the minimum number of all numbers in the brackets; the corresponding Dataset a=1,2,…,A; construct source domain dataset Its index set is In the formula, Indicates taking μ A The first index in Indicates the number of the first device belonging to type A of equipment;

[0049] For source domain dataset All elements in the source domain are labeled with fault category labels, and the fault category label set corresponding to the source domain dataset is constructed, which is described as:

[0050]

[0051] In the formula, Indicates the source domain dataset belonging to the ath model The corresponding fault category label set, a=1,2,…,A, represents the field of real numbers, Indicates the source domain dataset belonging to the ath model Middle The fault category label value of the element, γ z is the fault category attribute label set The zth label in, Z is the total number of labels, z = 1, 2, ..., Z,

[0052] Construct the index set of all device numbers belonging to the ath model, excluding the first index, and record it as:

[0053]

[0054] In the formula Indicates that all device number indexes belonging to the ath model are excluded from the first index A collection of other indexes arranged in ascending order, x a =1,2,…,A a , A a express The total number of elements in; take out the corresponding All indexes in x a =1,2,…,A a A dataset, a=1,2,…,A, construct the target domain dataset:

[0055]

[0056] Indicates the number The data set corresponding to the device;

[0057] Its index collection is:

[0058]

[0059] In the formula, Indicates the Ath type of equipment.A The device number;

[0060] The original dataset consists of the source domain dataset and the target domain dataset. Its index set is Optionally, the S3 includes:

[0061] S31: Construct a fault diagnosis model for device number b, which is described as:

[0062]

[0063] in Indicates the fault diagnosis model of device numbered b, symbol Represents a function composite operation, They represent the feature extractor model and classifier model of the equipment fault diagnosis model numbered b, and the formulas are:

[0064]

[0065] in, They respectively represent the feature extractor basic unit and the classifier basic unit of the equipment fault diagnosis model numbered b, θ=1,2,…,β, ω=1,2,…,α, β and α respectively represent the number of feature extractor layers and the number of classifier layers.

[0066] As a preferred implementation of this embodiment, the feature extractor model of the fault diagnosis model It is an 18-layer one-dimensional residual neural network, which includes 5 basic units, β = 5, among which the feature extractor basic unit Includes 1 one-dimensional convolution, 1 batch normalization layer and 1 maximum pooling layer, the basic unit of the feature extractor ~Basic unit In the example, each basic unit includes only one residual block, and the basic unit of the feature extractor It includes 1 residual block, 1 average pooling layer and 1 flattening layer; the classifier model of the fault diagnosis model It is a multi-layer perceptron network, including 1 classifier basic unit α=1 The basic unit of the classifier includes 1 linear layer.

[0067] S32: Using the source domain and target domain data sets, the fault diagnosis model of all devices is trained to obtain the basic fault diagnosis model with optimal parameters for all devices.

[0068] For the a-th model, the number is the index set The device pointed to by the index in the source domain is b = 1, 2, ..., B, a = 1, 2, ..., A. Index Collection All indexes in Dataset referred to Training a fault diagnosis model The gradient descent method is used to optimize and update, and the index set numbered as the ath model is obtained. The optimal basic fault diagnosis model for the equipment pointed to by the index The model parameter optimization function is expressed as:

[0069]

[0070] in, Indicates that it belongs to the ath model and is numbered Optimal basic fault diagnosis model for equipment parameters The optimal model parameters are Indicates that it belongs to the ath model and is numbered Fault diagnosis model of equipment Model parameters of Indicates that in all parameters Take the parameter that satisfies the minimum condition of the operation result in the brackets L ce (·) represents the cross entropy loss function; The model parameters are Input is the source domain dataset Fault Diagnosis Model The output of γ a (·) represents the label mapping function of device type a.

[0071] Construct an index set belonging to the ath model and numbered as The optimal basic fault diagnosis model set for all parameters of the device pointed to by the index The number is Optimal basic fault diagnosis model for equipment parameters It is expressed as:

[0072] (10)

[0073] In the formula They represent the optimal basic fault diagnosis model The optimal feature extractor model and classifier model are as follows:

[0074]

[0075] in, They represent the optimal feature extractor basic unit and the classifier basic unit respectively.

[0076] For the a-th model, the number is the index set The device pointed to by the index in is b=1,2,…,B, a=1,2,…,A, using the target domain dataset Index Collection All indexes in x a =1,2,…,A a , refers to the target domain dataset Training a fault diagnosis model Get the index set belonging to the ath model and numbered The optimal basic fault diagnosis model for the equipment pointed to by the index in The model parameter optimization function is expressed as:

[0077]

[0078] in, Indicates that it belongs to the ath model and the number is an index set Index Parameter-optimal basic fault diagnosis model for the indicated equipment The optimal model parameters are Indicates that it belongs to the ath model and the number is an index set Index Fault diagnosis model of the indicated equipment Model parameters of Indicates that in all parameters Take the parameter that satisfies the minimum condition of the operation result in the brackets L adapt (·) represents the domain adaptation loss function, η∈[0,1] represents the weight parameter of the domain adaptation loss; The model parameters are Input is the source domain dataset Fault Diagnosis Model The output, Respectively represent the input as the target domain dataset Source domain dataset Fault Diagnosis Model Feature extractor model Output; In this embodiment, two groups of asynchronous AC motors with the same model and different working conditions are selected for transfer learning, where the domain adaptive loss function selects the maximum mean difference loss, η is 0.5, and the visualization result of the source domain and target domain data features extracted by the feature extractor unit is shown in the figure Figure 2 As shown;

[0079] Construct an index set belonging to the ath model and numbered as The optimal basic fault diagnosis model for all parameters of the device pointed to by the index Will It is expressed as:

[0080]

[0081] In the formula They represent the optimal basic fault diagnosis model The optimal feature extractor model and classifier model are as follows:

[0082]

[0083] in, They represent the optimal basic fault diagnosis model The optimal feature extractor basic unit and classifier basic unit, x a =1,2,…,A a , a=1,2,…,A.

[0084] According to formulas (9) to (16), the optimal basic fault diagnosis model set of all equipment parameters is obtained: It is expressed as:

[0085]

[0086] The S4 includes:

[0087] S41: Collect the sensor data set newly collected by the device and its corresponding label set; collect the semantic information required to build a personalized customized trigger rule encoding matrix;

[0088] Collect the sensor data set in the vehicle-line network information interaction collected by the device numbered b (b=1,2,3,…,B) for the kth time and its corresponding fault category attribute label set γ z is the fault category attribute label set The zth tag in the dataset, z = 1, 2, ..., Z, where Z is the total number of tags, k is the number of personalized customizations, k = 1, 2, 3, ..., K, where K is the total number of personalized customizations; collect the semantic information in the vehicle-line network information interaction required for constructing the personalized customization trigger rule encoding matrix for the kth time;

[0089] S42: Construct a personalized customized trigger rule coding matrix.

[0090] Encode the semantic information that affects the trigger result in the k-th personalized customization trigger rule coding matrix, and construct the k-th personalized customization trigger rule coding matrix. The formula is:

[0091]

[0092] in, The personalized trigger rule encoding matrix constructed for the kth time, Represents the personalized trigger rule encoding matrix P k The element at the i-th row and j-th column, i = 1, 2, 3, ..., n, j = 1, 2, 3, ..., m, n is the personalized customized trigger rule encoding matrix P constructed for the kth time k is the total number of rules (total number of rows), and m is the total number of semantic information encodings (total number of columns) of the personalized customization trigger rule encoding matrix constructed for the kth time.

[0093] The S5 includes:

[0094] Construct the optimal decision tree model of the optimal sub-matrix set of the personalized customized trigger rule encoding matrix.

[0095] S51: Calculate the entropy of the k-th construction of the personalized customized trigger rule encoding matrix.

[0096] Construct the row rule trigger result decision column vector corresponding to the k-th personalized customization trigger rule encoding matrix, the formula is:

[0097]

[0098] In the formula Represents the kth personalized customization trigger rule encoding matrix The corresponding row rule trigger result decision vector; o k；i P k The i-th row rule The corresponding trigger result, o k；i ∈{0,1}, i=1,2,3,…,n;

[0099] Calculate the entropy of the k-th personalized customization trigger rule encoding matrix, the formula is:

[0100]

[0101] Among them, H(P k ) is the kth personalized customization trigger rule encoding matrix The entropy of Respectively represent P k The i-th row rule The corresponding trigger result o k；i (i=1,2,3,…,n) The number of rows with 0 and 1 accounts for P k The ratio of the total number of rows is:

[0102]

[0103] Where 1(·) is an indicator function, which takes the value 1 when the condition in the brackets is true, and takes the value 0 otherwise.

[0104] S52: Calculate the entropy of the sub-matrix of the k-th personalized customization trigger rule encoding matrix.

[0105] Construct a sub-matrix of the k-th personalized customization trigger rule encoding matrix, the formula is:

[0106]

[0107] in Encoding matrix for the kth personalized trigger rule The j-th column element of The submatrix with value v, v∈{0,1} represents the personalized trigger rule encoding matrix P k The element at the i-th (i=1,2,3,…,n) row and j-th (j=1,2,3,…,m) column The value of P k The j-th element in the Submatrix with value v the number of rows, i∈{1,2,…,n}} is the row set of the submatrix, and size(·) represents the number of elements in the set, that is, the number of rows of the submatrix.

[0108] Specifically, the column elements take values When , the sub-matrices of the k-th personalized customization trigger rule encoding matrix are constructed as and Get values ​​for column elements respectively Time matrix the number of rows,

[0109] Construct the kth personalized customization trigger rule encoding matrix P k The j-th (j=1,2,3,…,m) column element All sub-matrices with values ​​v∈{0,1} are: The number of rows corresponding to all sub-matrices is:

[0110] Calculate the entropy of the submatrix of the k-th personalized customization trigger rule encoding matrix, the formula is:

[0111]

[0112] in, Encoding matrix for the kth personalized trigger rule The j-th column element of Submatrix with value v∈{0,1} The entropy of Represent submatrices BOC rule triggering results The number of rows with 0 and 1 occupies the submatrix The ratio of the total number of rows is:

[0113]

[0114] The kth personalized customization trigger rule encoding matrix P k The j-th (j=1,2,3,…,m) column element The entropy of all sub-matrices with values ​​v∈{0,1} is:

[0115] S53: Calculate the column where the column element with the largest information gain in the personalized customization trigger rule encoding matrix is ​​located;

[0116] Calculate the kth personalized customization trigger rule encoding matrix P k The j-th column element of The information gain is:

[0117]

[0118] In the formula Encoding matrix P for customized trigger rules k The j-th column element of Information gain of

[0119] The kth personalized customization trigger rule encoding matrix P k The j-th (j=1,2,3,…,m) column element All information gains are

[0120] Select the column element with the largest information gain Get the column it is in, the formula is:

[0121]

[0122] Where j* is the personalized trigger rule encoding matrix P k The column element with the largest information gain The column where j * ∈{1,2,3,…,m}, Indicates taking the column j (j=1,2,3,…,m) that satisfies the maximum condition of the operation result in the brackets;

[0123] S54: Construct the kth personalized customization trigger rule encoding matrix P according to the column where the column element with the maximum information gain is located k The Lth optimal submatrix set of ;

[0124] According to the personalized trigger rule encoding matrix P k The jth position where the information gain is the largest * Column Elements Take values ​​of v∈{0,1} to construct the first group of optimal submatrices Right now and Get values ​​for column elements respectively The optimal submatrix the number of rows, The number of optimal submatrices in the first group is N1=2 1 =2;

[0125] The personalized trigger rule encoding matrix P k The first optimal submatrix in According to the information gain of the largest Column Elements Take the value v∈{0,1} to construct the second group of optimal submatrices Right now and is the second group of optimal submatrices, The optimal submatrices are The number of rows, and The number of optimal submatrices in the second group is N2 = 2 2 =4; (the submatrix can be an empty set).

[0126] The personalized trigger rule encoding matrix P k The l-1th optimal submatrix in According to the information gain of the largest Column Elements Take the value v∈{0,1} to construct the optimal submatrix of group l The number of optimal submatrices in the lth group is N l =2 l ;

[0127] The personalized trigger rule encoding matrix P k The L-1th optimal submatrix in According to the information gain of the largest Column Elements Take the value v∈{0,1} to construct the Lth group of optimal submatrices The number of optimal submatrices in group L is N L =2 L; At this time, the personalized trigger rule encoding matrix P k The entropy of each optimal submatrix in the Lth optimal submatrix set is equal to 0; the personalized customized trigger rule encoding matrix P k The Lth optimal submatrix set in It is expressed as:

[0128]

[0129] in, Encoding matrix P for customized trigger rules k The Lth optimal submatrix set in The tth optimal submatrix of L =2 L ;in, The brackets represent another concise way of expressing the symbol before the brackets, and the meaning is the same; see the left and right sides of the first equal sign in Formula 27. Personalized customized trigger rule encoding matrix P k The Lth group of optimal submatrices The t-th sub-matrix in The i-th row rule The trigger result Equal to c t ∈{0,1}, i∈{1,2,…,n}} is the tth optimal submatrix The row set of is the tth optimal submatrix the number of rows, Personalized customized trigger rule encoding matrix P k The Lth optimal submatrix set in The trigger result set is C = {c1, c2, ..., c t ,…,c T}.

[0130] S55: Constructing an optimal decision tree model of the personalized customized trigger rule encoding matrix with respect to the Lth group of optimal submatrices, the formula is:

[0131]

[0132] in, Encoding matrix P for customized trigger rules k About the Lth Group of Optimal Submatrices The optimal decision tree model of Indicates that it is taken from P k Any i-th row of c is used as the input of the decision tree model, i = 1, 2, 3, ..., n; t is the trigger result of the tth optimal submatrix, Indicates that the indicator function is used when the condition in the brackets The value is 1 when established.

[0133] The S6 includes:

[0134] The optimal basic fault diagnosis model set M based on all equipment parameters b , construct a hierarchical freezing model, the formula is:

[0135]

[0136] in, Indicates the hierarchical freezing model of device number b. Indicates model M of device number b b The freezing operation function, They are A set of indicating functions of whether each layer is frozen.

[0137] They are The indicator function of whether the layer in is frozen, its value is 0 for freezing (in the frozen state, the input of the layer is directly passed to the output without updating the parameters of the layer), and its value is 1 for activation (in the activated state, the parameters of the layer are updated); They represent the feature extractor model and the classifier model of the hierarchical frozen model of device numbered b, respectively. They represent the feature extractor basic unit and classifier basic unit of the hierarchical freezing model of device numbered b, θ=1,2,…,β,ω=1,2,…,α,β、αrepresent the number of feature extractor layers and classifier layers respectively, They satisfy the following expressions respectively:

[0138]

[0139] Wherein, id(·) is the identity function of the basic unit in the brackets, which is used to activate the update of the parameters of the corresponding basic unit; They respectively represent the feature extractor basic unit and the classifier basic unit of the optimal basic fault diagnosis model of equipment parameters numbered b; b=1,2,3,…,B.

[0140] The S7 includes:

[0141] S71: Using the constructed k-th personalized customization trigger rule encoding matrix P k According to formula (28), the personalized trigger rule coding matrix P is calculated k About the Lth Group of Optimal Submatrices The optimal decision result is the kth personalized customization model. Row triggering rules i=1,2,3,…,n;

[0142] when When the k-1th personalized customization model is triggered Perform the k-th layer fine-tuning, using the sensor data set newly collected by the device numbered b (b = 1, 2, 3, ..., B) for the kth time and its corresponding fault category attribute label set Construct the k-th personalized customization model, the formula is:

[0143]

[0144] in, They represent the personalized customized models constructed for the kth and k-1th times of device b, respectively. represents the fine-tuning operation function of the personalized customized model built for the kth time by device numbered b, represents the fine-tuning learning rate of the personalized customized model built for the kth time by device numbered b; They represent the personalized customized model constructed for the kth time by device numbered b. Feature extractor model, classifier model; They represent the personalized customized model constructed for the kth time by device numbered b. The feature extractor basic unit and the classifier basic unit are as follows: θ=1,2,…,β,ω=1,2,…,α,β、αrepresent the number of feature extractor layers and classifier layers respectively; Represents the personalized customization model constructed for the k-1th time The freezing operation function satisfies the following relationship:

[0145]

[0146] In the formula, Indicates the personalized customized model built for the k-1th time by device number b The hierarchical freezing model, They represent the personalized customization model constructed for the k-1th time by device numbered b. Layered freezing model The feature extractor model and classifier model of They represent the personalized customization model constructed for the k-1th time by device numbered b. Layered freezing model The feature extractor basic unit and the classifier basic unit, They satisfy the following expressions respectively:

[0147]

[0148] Where id(·) is the identity function of the basic unit in the brackets, which is used to activate the update of the parameters of the corresponding basic unit; They represent the personalized customization model constructed for the k-1th time by device numbered b. Feature extractor model Classifier Model An indicator function of whether the layer in is frozen, where a value of 0 indicates freezing and a value of 1 indicates activation; They represent the personalized customization model constructed for the k-1th time by device numbered b. The feature extractor basic unit and the classifier basic unit;

[0149] Specifically, when k=1, it indicates that the personalized customized model U is constructed for the 0th time by the device numbered b. b 0 The hierarchical freezing model is Its freezing operation function is in That is to say Personalized custom model built by the 0th time Layered freezing model right After the first fine-tuning, the first personalized customization model is constructed as

[0150] when When , freeze (do not trigger) the k-1th personalized customization model Perform the k-th layered fine-tuning, and let the k-th constructed personalized customization model be

[0151] Let k=k+1, k=1,2,3,…,K, and go to S21 to execute until k>K and stop the operation.

[0152] Specifically, in this embodiment, the number displayed is the personalized customization model built by device b for the k-1th time. Layered freezing model Schematic diagram as Figure 3 As shown in the figure, the feature extractor basic units 1, 2, and 3 are frozen, the feature extractor basic units 4, 5, and the classifier basic unit are activated, and the k-th vehicle-line network interaction information is received. According to the decision results of the decision tree model with personalized customized trigger rules, the confusion matrices of the model evaluation results before and after the k-th hierarchical fine-tuning are shown as follows: Figure 4 , 5 shown.

[0153] In summary, the personalized fault diagnosis customization method provided by the present application uses a transfer learning strategy in the initial stage to efficiently train the fault diagnosis model of each device with fewer labeled samples. For the subsequent personalized customization process, personalized customization trigger rules can be designed by considering the factors that may affect the performance of the fault diagnosis model due to changes in the device's own state and environmental changes in the vehicle-line-network information interaction, and triggers can be designed based on these rules. Through personalized rule triggers and external interactions, adjustments to the fault diagnosis model of the device with performance changes are triggered, thereby responding to the impact of conditions on diagnostic performance. The implementation of this method can be targeted at various existing devices without changing existing hardware devices, which is of great significance to improving the reliability of equipment operation and maintaining system security.

[0154] The present application also provides a personalized fault diagnosis customization system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program. The personalized fault diagnosis customization system can implement various embodiments of the above personalized fault diagnosis customization method and achieve the same effect, which will not be described in detail here.

[0155] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A personalized fault diagnosis customization method, characterized in that: include: S1: Collect sensor data of different models and numbered devices in rail transit vehicles to obtain number sets and model sets, and construct a mapping relationship from number sets to model sets; S2: constructing a related set of source domain data sets of sensor data based on the mapping relationship, and constructing a related set of target source domain data sets of sensor data; S3: Initialize a fault diagnosis model for each device, and train the fault diagnosis models of all devices based on the relevant set of the source domain data set and the relevant set of the target source domain data set to obtain the basic fault diagnosis model with optimal parameters for all devices; S4: obtaining a sensor data set newly collected by the device in the vehicle-line-network information interaction and its corresponding fault category attribute label set, and collecting target semantic information, and constructing a personalized customized trigger rule encoding matrix based on the target semantic information; S5: constructing an optimal decision tree model of the personalized customized trigger rule encoding matrix with respect to an optimal sub-matrix set; S6: constructing a hierarchical freezing model based on the optimal basic fault diagnosis model of all the equipment parameters; S7: Using the personalized customized trigger rule encoding matrix, the optimal decision tree model, the newly collected sensor data set and its corresponding fault category attribute label set, the hierarchical frozen model is fine-tuned to construct a personalized customized model.

2. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S1 includes: S11: Collect sensor data of B numbered devices of A different models of rail transit vehicles to form an original data set D = {D1,…,D b ,…,D B }, represents the number of sampling points of the sensor signal, b is the device number, b = 1, 2, ..., B, B is the total number; all device numbers constitute the number set C B ={1,2,…,b,…B}; all device models constitute the model set C A ={1,2,…,a,…A}, a is the device model, a=1,2,…,A, A is the total number of models; Construct a mapping relationship from a number set to a model set, as described below: f:C B →C A (1) Among them, φ:C B →C A Represents the number set C B To model set C A 's mapping.

3. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S2 includes: If the device numbered b belongs to the ath type, it is recorded as: φ(b)=a; a=1,2,…,A; (2) In the formula, A is the total number of models; Construct an index set of all device numbers belonging to the ath model, denoted as: m a ={b|φ(b)=a,b∈{1,2,…,B}}; (3) In the formula, μ a represents the index set of all device numbers belonging to the ath model, and B is the total number of numbers; Take μ a The first index in represents the number of the first device belonging to the a-th type of equipment, min() represents the operation of taking the minimum number of all numbers in the brackets; the corresponding Dataset a=1,2,…,A; construct source domain dataset The index set is as follows: In the formula, Indicates μ A The first index in Indicates the number of the first device belonging to type A of equipment; The source domain dataset All elements in the source domain are labeled with fault category labels, and the fault category label set corresponding to the source domain dataset is constructed, which is described as follows: In the formula, Indicates the source domain dataset belonging to the ath model The corresponding fault category label set, a=1,2,…,A, represents the field of real numbers, Indicates the source domain dataset belonging to the ath model Middle The fault category label value of the element, γ z is the fault category attribute label set The zth label in, Z is the total number of labels, z = 1, 2, ..., Z, Construct the index set of all device numbers belonging to the ath model, excluding the first index, and record it as: In the formula, Indicates that all device number indexes belonging to the ath model are excluded from the first index A collection of other indexes arranged in ascending order, x a =1,2,…,A a , A a express The total number of elements in; Take the corresponding All indexes in x a =1,2,…,A a The target domain dataset is constructed as follows: In the formula, Indicates the number The data set corresponding to the device; The index set of the target domain dataset is as follows: In the formula, Indicates the Ath type of equipment. A The device number; The original dataset consists of the source domain dataset and the target domain dataset. Its index set is 4. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S3 includes: S31: Construct a fault diagnosis model for device number b, which is described as: in represents the fault diagnosis model of the device numbered b, and the symbol ο represents the function composite operation. They represent the feature extractor model and classifier model of the equipment fault diagnosis model numbered b, and the formulas are: in, They represent the feature extractor basic unit and the classifier basic unit of the fault diagnosis model of the device numbered b, θ=1,2,…,β, ω=1,2,…,α, β and α represent the number of feature extractor layers and the number of classifier layers respectively; S32: using the relevant set of the source domain data set and the relevant set of the target source domain data set to train the fault diagnosis model of all devices, to obtain the basic fault diagnosis model with optimal parameters for all devices: For the a-th model, the number is the index set The device pointed to by the index in the source domain is b = 1, 2, ..., B, a = 1, 2, ..., A. Index Collection All indexes in Dataset referred to Training a fault diagnosis model The gradient descent method is used to optimize and update, and the index set numbered as the ath model is obtained. The optimal basic fault diagnosis model for the equipment pointed to by the index The model parameter optimization function satisfies the following relationship: in, Indicates that it belongs to the ath model and is numbered Parameter Optimization Basic Fault Diagnosis Model for Equipment The optimal model parameters are Indicates that it belongs to the ath model and is numbered Fault diagnosis model of equipment Model parameters of Indicates that in all parameters Take the parameter that satisfies the minimum condition of the operation result in the brackets L ce (·) represents the cross entropy loss function; The model parameters are Input is the source domain dataset Fault Diagnosis Model The output of γ a (·) represents the label mapping function of device type a; Construct an index set belonging to the ath model and numbered as The optimal basic fault diagnosis model set for all parameters of the device pointed to by the index The number is Optimal basic fault diagnosis model for equipment parameters It is expressed as: In the formula, They represent the optimal basic fault diagnosis model The optimal feature extractor model and the optimal classifier model are as follows: in, They represent the optimal feature extractor basic unit and the classifier basic unit respectively; For the a-th model, the number is the index set The device pointed to by the index in is b=1,2,…,B, a=1,2,…,A, using the target domain dataset Index Collection All indexes in x a =1,2,…,A a , refers to the target domain dataset Training a fault diagnosis model Get the index set belonging to the ath model and numbered The optimal basic fault diagnosis model for the equipment pointed to by the index in The model parameter optimization function is expressed as follows: in, Indicates that it belongs to the ath model and the number is an index set Index Parameter-optimal basic fault diagnosis model for the indicated equipment The optimal model parameters are Indicates that it belongs to the ath model and the number is an index set Index Fault diagnosis model of the indicated equipment Model parameters of Indicates that in all parameters Take the parameter that satisfies the minimum condition of the operation result in the brackets L adapt (·) represents the domain adaptation loss function, η∈[0,1] represents the weight parameter of the domain adaptation loss; The model parameters are Input is the source domain dataset Fault Diagnosis Model The output, Respectively represent the input as the target domain dataset Source domain dataset Fault Diagnosis Model Feature extractor model Output: Construct an index set belonging to the ath model and numbered as The optimal basic fault diagnosis model for all parameters of the device pointed to by the index Will It is expressed as follows: In the formula, They represent the optimal basic fault diagnosis model The optimal feature extractor model and classifier model are as follows: in, They represent the optimal basic fault diagnosis model The optimal feature extractor basic unit and classifier basic unit, x a =1,2,…,A a , a=1,2,…,A; According to formulas (9) to (16), the optimal basic fault diagnosis model set of all equipment parameters is obtained: It is expressed as:

5. A personalized fault diagnosis customization method according to claim 1, characterized in that: The S4 includes: S41: Collect the sensor data set of the kth vehicle-line network information interaction collected by the device numbered b, b = 1, 2, 3, ..., B and its corresponding fault category attribute label set γ z is the fault category attribute label set The zth tag in the dataset, z = 1, 2, ..., Z, where Z is the total number of tags, k is the number of personalized customizations, k = 1, 2, 3, ..., K, where K is the total number of personalized customizations; collect the semantic information in the vehicle-line network information interaction required for constructing the personalized customization trigger rule encoding matrix for the kth time; S42: Encode the semantic information that affects the trigger result in the k-th personalized customization trigger rule coding matrix, and construct the k-th personalized customization trigger rule coding matrix. The formula is as follows: in, The personalized trigger rule encoding matrix constructed for the kth time, Represents the personalized trigger rule encoding matrix P k The element at the i-th row and j-th column, i = 1, 2, 3, ..., n, j = 1, 2, 3, ..., m, n is the personalized customized trigger rule encoding matrix P constructed for the kth time k , m is the total number of rows of the personalized customized trigger rule encoding matrix constructed for the kth time, and m is the total number of columns of semantic information encoding.

6. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S5 includes: S51: Construct the row rule trigger result decision column vector corresponding to the k-th personalized customization trigger rule encoding matrix, the formula is as follows: The k =[o k；1 the k；2 … the k；i … the k；n ] T (19) In the formula, Represents the kth personalized customization trigger rule encoding matrix The corresponding row rule trigger result decision vector; o k；i P k The i-th row rule The corresponding trigger result, o k；i ∈{0,1}, i=1,2,3,…,n; Calculate the entropy of the k-th personalized customization trigger rule encoding matrix, the formula is as follows: Among them, H(P k ) is the kth personalized customization trigger rule encoding matrix The entropy of Respectively represent P k The i-th row rule The corresponding trigger result o k；i The number of rows with values ​​of 0 and 1 accounts for P k The ratio of the total number of rows, i = 1, 2, 3, ..., n, is as follows: Where 1(·) is an indicator function, which takes the value 1 when the condition in the brackets is true, and takes the value 0 otherwise; S52: construct a sub-matrix of the k-th personalized customization trigger rule encoding matrix, the formula is as follows: in, Encoding matrix for the kth personalized trigger rule The j-th column element of The submatrix with value v, v∈{0,1} represents the personalized trigger rule encoding matrix P k The element at row i and column j The value of i=1,2,3,…,n,j=1,2,3,…,m, P k The j-th element in the Submatrix with value v the number of rows, i∈{1,2,…,n}} is the row set of the submatrix, and size(·) represents the number of elements in the set; Construct the kth personalized customization trigger rule encoding matrix P k The j-th column element of All sub-matrices with values ​​v∈{0,1} are: The number of rows corresponding to all sub-matrices is: Calculate the entropy of the submatrix of the k-th personalized customization trigger rule encoding matrix, the formula is: in, Encoding matrix for the kth personalized trigger rule The j-th column element of Submatrix with value v∈{0,1} The entropy of Represent submatrices BOC rule triggering results k；i The number of rows with 0 and 1 occupies the submatrix The ratio of the total number of rows, The formula is as follows: The kth personalized customization trigger rule encoding matrix P k The j-th column element of The entropy of all sub-matrices with values ​​v∈{0,1}, j=1,2,3,…,m, is as follows: S53: Calculate the kth personalized customization trigger rule coding matrix P k The j-th column element of The information gain is as follows: In the formula, Encoding matrix P for customized trigger rules k The j-th column element of Information gain of The kth personalized customization trigger rule encoding matrix P k The j-th column element of All information gains are Select the column element with the largest information gain Get the column it is in, the formula is as follows: Where j* is the personalized trigger rule encoding matrix P k The column element with the largest information gain The column where j * ∈{1,2,3,…,m}, Indicates taking the column j that satisfies the maximum condition of the calculation result in the brackets; S234: Construct the kth personalized customization trigger rule encoding matrix P according to the column where the column element with the maximum information gain is located k The Lth optimal submatrix set of ; The personalized trigger rule encoding matrix P k The optimal submatrix of the l-1 group in According to the information gain of the largest Column Elements Take the value v∈{0,1} to construct the optimal submatrix of group l The number of optimal submatrices in the lth group is N l =2 l ; The personalized trigger rule encoding matrix P k The L-1th optimal submatrix in According to the information gain of the largest Column Elements Take the value v∈{0,1} to construct the Lth group of optimal submatrices The number of optimal submatrices in group L is N L =2 L ; At this time, the personalized trigger rule encoding matrix P k The entropy of each optimal submatrix in the Lth optimal submatrix set is equal to 0; the personalized customized trigger rule encoding matrix P k The Lth optimal submatrix set in It is expressed as follows: in, Encoding matrix P for customized trigger rules k The Lth optimal submatrix set in The tth optimal submatrix of L =2 L ; Personalized customized trigger rule encoding matrix P k The Lth group of optimal submatrices The t-th sub-matrix in The i-th row rule The trigger result Equal to c t ∈{0,1}, i∈{1,2,…,n}} is the tth optimal submatrix The row set of is the tth optimal submatrix the number of rows, Personalized customized trigger rule encoding matrix P k The Lth optimal submatrix set in The trigger result set is C = {c1, c2, ..., c t ,…,c T }; S55: Constructing an optimal decision tree model of the personalized customized trigger rule encoding matrix with respect to the Lth group of optimal submatrices, the formula is as follows: in, Encoding matrix P for customized trigger rules k About the Lth Group of Optimal Submatrices The optimal decision tree model for Indicates that it is taken from P k Any i-th row of c is used as the input of the decision tree model, i = 1, 2, 3, ..., n; t is the trigger result of the tth optimal submatrix, Indicates that the indicator function is used when the condition in the brackets The value is 1 when established.

7. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S6 includes: The optimal basic fault diagnosis model set M based on all equipment parameters b , construct a hierarchical freezing model, the formula is as follows: in, Indicates the hierarchical freezing model of device number b. Indicates model M of device number b b The freezing operation function, They are A set of indicating functions of whether each layer is frozen. They are An indicator function of whether the layer in is frozen, where a value of 0 indicates freezing and a value of 1 indicates activation; They represent the feature extractor model and the classifier model of the hierarchical frozen model of device numbered b, respectively. They represent the feature extractor basic unit and classifier basic unit of the hierarchical freezing model of device numbered b, θ=1,2,…,β,ω=1,2,…,α,β、αrepresent the number of feature extractor layers and classifier layers respectively, They satisfy the following expressions respectively: Wherein, id(·) is the identity function of the basic unit in the brackets, which is used to activate the update of the parameters of the corresponding basic unit; They respectively represent the feature extractor basic unit and the classifier basic unit of the optimal basic fault diagnosis model of equipment parameters numbered b; b=1,2,3,…,B.

8. The personalized fault diagnosis customization method according to claim 1, characterized in that: The S7 includes: S71: Using the constructed k-th personalized customization trigger rule encoding matrix P k According to the optimal decision tree model of the personalized trigger rule encoding matrix about the Lth group of optimal sub-matrices, the personalized trigger rule encoding matrix P is calculated. k About the Lth Group of Optimal Submatrices The optimal decision result is the kth personalized customization model. Row triggering rules when When the k-1th personalized customization model is triggered Perform the k-th layered fine-tuning, using the sensor data set numbered b, b = 1, 2, 3, ..., B, which is newly collected by the device for the kth time and its corresponding fault category attribute label set Construct the k-th personalized customization model, the formula is as follows: in, They represent the personalized customized models constructed for the kth and k-1th times of device b, respectively. represents the fine-tuning operation function of the personalized customized model built for the kth time by device numbered b, represents the fine-tuning learning rate of the personalized customized model built for the kth time by device numbered b; They represent the personalized customized model constructed for the kth time by device numbered b. Feature extractor model, classifier model; They represent the personalized customized model constructed for the kth time by device numbered b. The feature extractor basic unit and the classifier basic unit are as follows: θ=1,2,…,β,ω=1,2,…,α,β、αrepresent the number of feature extractor layers and classifier layers respectively; Represents the personalized customization model constructed for the k-1th time The freezing operation function satisfies the following relationship: In the formula, Indicates the personalized customized model built for the k-1th time by device number b The hierarchical freezing model, They represent the personalized customization model constructed for the k-1th time by device numbered b. Layered freezing model The feature extractor model and classifier model of They represent the personalized customization model constructed for the k-1th time by device numbered b. Layered freezing model The basic unit of feature extractor and classifier, They satisfy the following expressions respectively: Where id(·) is the identity function of the basic unit in the brackets, which is used to activate the update of the parameters of the corresponding basic unit; They represent the personalized customization model constructed for the k-1th time by device numbered b. Feature extractor model Classifier Model An indicator function of whether the layer in is frozen, where a value of 0 indicates freezing and a value of 1 indicates activation; They represent the personalized customization model constructed for the k-1th time by device numbered b. The feature extractor basic unit and the classifier basic unit; when When the k-1th personalized customization model is frozen Perform the k-th layered fine-tuning, and let the k-th constructed personalized customization model be Let k=k+1, k=1,2,3,…,K, and go to S21 to execute until k>K and stop the operation.

9. A personalized fault diagnosis customization system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 8 are implemented.

Images