Incremental fuzzy width learning electromechanical equipment fault diagnosis method based on stream data

Abstract

The invention relates to an incremental fuzzy width learning electromechanical equipment fault diagnosis method based on stream data. The method comprises the following steps: acquiring characteristic parameters and state data of an actuator at each operation moment; constructing a fault diagnosis model; obtaining a target data set; increasing nodes of the enhancement layer, and obtaining a target fault diagnosis model; and obtaining an electromechanical equipment fault diagnosis result by using the target fault diagnosis model. According to the method, through an incremental clustering method based on time weighting, a weight attenuation mechanism is introduced to adapt to data changes, the positions and the number of fuzzy width learning clustering center points are dynamically adjusted, so that the model has an online updating function in a real-time data stream scene, and meanwhile, through incremental updating of enhanced nodes, the diagnosis accuracy of the model is improved.

Description

Technical Field

[0001] The invention relates to the technical field of fault diagnosis, and in particular to a method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of stream data. Background Art

[0002] With the continuous development of modern technology, complex electromechanical equipment has been widely used in the field of national defense. Intelligent fault diagnosis of complex electromechanical equipment has important theoretical significance and application value for ensuring the safety and stability of the system, timely troubleshooting, and avoiding major casualties and property losses. The fault diagnosis method based on machine learning uses machine learning technology to establish an intelligent fault diagnosis model based on historical data. It does not require complex physical and mathematical models and has become an important part of the research on fault diagnosis technology of complex electromechanical systems.

[0003] In actual engineering applications, how to deal with the continuously collected new samples has become an important research topic. Through the current literature search, it is found that the use of incremental learning to perform real-time health monitoring and diagnostic model updates for complex electromechanical equipment can more efficiently diagnose faults. For example, Dai Jinling et al. combined a type of kernel extreme learning machine with data incremental learning in "Online Fault Detection Method Based on OCKELM and Incremental Learning" to achieve online fault diagnosis of sequential sample data; Wang Wenbo et al. proposed a rolling bearing fault prediction method based on deep learning and incremental update in "Research on Rolling Bearing Fault Prediction Method Based on Deep Learning and Incremental Update", which enables the model to have parameter self-learning and self-updating capabilities; Lu Songfeng et al. proposed a hierarchical federated incremental learning algorithm based on cloud, edge, and end architecture in "Incremental Federated Learning Algorithm for Cloud-Edge-End Collaboration".

[0004] The prior art proposes an efficient incremental learning system that does not require a deep architecture, namely, width learning. Compared with the "deep" network structure of deep learning, width learning is more inclined to the "width" development of the network, which makes it have the advantages of simple structure and fast training speed. At the same time, researchers have proposed fuzzy width learning, combining the Takagi-Sugeno fuzzy system with BLS, which improves the model performance while retaining the advantage of fast calculation speed of width learning. In the field of fault diagnosis, width learning provides a fast and efficient method for fault diagnosis. Jiang SB et al. applied incremental width learning to the fault diagnosis of three-phase asynchronous motors in "An efficient fault diagnostic method for three-phase induction motors based on incremental broad learning and non-negative matrix factorization", and optimized the network by using empirical mode decomposition, sample entropy and non-negative matrix factorization; and combined CNN with BLS in a deep and broad learning framework based on domain knowledge. However, the width learning method is mainly carried out in a batch update mode. Due to the variable operating conditions of equipment in actual industrial monitoring, the traditional fault diagnosis model cannot be updated once the training is completed, which affects the judgment accuracy of the model.

[0005] Therefore, it is necessary to provide an incremental fuzzy width learning electromechanical equipment fault diagnosis method based on flow data to solve the above problems. Summary of the invention

[0006] The present invention provides an incremental fuzzy width learning electromechanical equipment fault diagnosis method based on streaming data, which adopts the "data stream incremental learning" strategy to build a fault diagnosis model. At the same time, the "fuzzy width learning" strategy is used to improve the diagnostic accuracy of the fault diagnosis model to solve the problem that the traditional fault diagnosis model cannot be updated once the training is completed due to the changeable operating conditions of the equipment in actual industrial monitoring, thereby affecting the judgment accuracy of the model.

[0007] The present invention provides an incremental fuzzy width learning electromechanical equipment fault diagnosis method based on streaming data, which adopts the following technical scheme, including: Acquire the characteristic parameters and state data of the actuator at each operating moment, take the characteristic parameters as input data, take the state data corresponding to the actuator as output data, and construct a data set; wherein the characteristic parameters include: temperature data of the actuator and acceleration data of three axes; the state data include: normal operation, blocking fault, spalling fault and motor fault; Constructing a fault diagnosis model, the fault diagnosis model includes an enhancement layer and a mapping layer, the mapping layer is used to perform fuzzy mapping on the input data and output data in the data set, and output the defuzzified mapping result; the enhancement layer is used to perform nonlinear changes on the defuzzified mapping result and output the fault diagnosis result; A cluster is set for each characteristic parameter, the maturity of the cluster is set according to the time decay function, and the cluster of each characteristic parameter at each moment is obtained according to the maturity of the cluster, the linear weighted sum of all characteristic parameters in each characteristic parameter of the input data of the data set, and the weighted square sum of all characteristic parameters in each characteristic parameter of the input data of the data set; Obtain the cluster center and cluster radius of the cluster cluster, dynamically update the cluster center and cluster radius of the cluster cluster at the next moment through the cluster radius and radius threshold of the cluster cluster, obtain the target cluster center and target cluster radius of the cluster cluster, remove the data in the cluster cluster that does not belong to the range of the cluster center and the target cluster radius, and obtain the target cluster cluster for each feature parameter; obtain the target data set according to the target cluster cluster for each feature parameter; Adding nodes to the enhanced layer of the fault diagnosis model according to a preset number of enhanced nodes, and obtaining the fault diagnosis model after adding the enhanced nodes, obtaining the loss function value of the enhanced fault diagnosis model according to the target data set, and when the loss function value does not meet the preset loss function threshold, adding a preset number of enhanced nodes again until the loss function value of the enhanced fault diagnosis model meets the preset loss function threshold, and then taking the enhanced fault diagnosis model as the target fault diagnosis model; The input data at the current moment is input into the target fault diagnosis model, and the electromechanical equipment fault diagnosis result is output.

[0008] Preferably, the characteristic parameters are sequentially subjected to data cleaning and data standardization, and the standardized characteristic parameters are used as input data.

[0009] Preferably, the step of performing data cleaning on the characteristic parameters is: Traverse the collected characteristic parameters and mark the missing values ​​and outliers. The outliers are determined by the 6σ criterion. The missing values ​​and outliers are replaced by the neighboring value filling method to complete data cleaning.

[0010] Preferably, the step of performing data standardization on the characteristic parameters is: performing data standardization processing on the data set using a Z-score standardization method.

[0011] Preferably, the mapping layer includes: a fuzzification mapping layer and a defuzzification mapping layer; the fuzzification mapping layer is used to perform fuzzification mapping on the input data and output data in the data set, and the defuzzification mapping layer is used to defuzzify the fuzzified output; The fuzzy mapping layer model expression is:

[0012]

[0013]

[0014]

[0015] In the formula, Indicates The output value of the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The normalized weight of the kth rule in the i-th fuzzy subsystem is the characteristic parameter; represents the weighted sum function; Indicates The mth parameter among the characteristic parameters; Indicates The fuzzy coefficient of the tth parameter in the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The tth parameter among the characteristic parameters; Indicates The output of the i-th fuzzy subsystem corresponding to the input of the characteristic parameter; express The output of the i-th fuzzy subsystem among the fuzzy subsystems; The fuzzy mapping layer is represented as The output of the fuzzy subsystem; The defuzzification mapping layer model expression is:

[0016] ,......,

[0017]

[0018] In the formula, The fuzzy mapping layer is represented as The defuzzified output when there are fuzzy subsystems; express The defuzzified output of the i-th fuzzy subsystem among the fuzzy subsystems; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; is the mapping coefficient of the sth characteristic parameter under the kth rule of the i-th fuzzy subsystem; represents the output of the fuzzy subsystem at the mapping layer, ; represents the weight of the i-th fuzzy subsystem, .

[0019] Preferably, the expression of the enhancement layer is:

[0020]

[0021] In the formula, It represents the output when the enhancement layer has m enhancement nodes. represents the output of the mth enhancement node of the enhancement layer; The enhancement layer The output of the enhancement node; The fuzzy mapping layer is represented as The output of the fuzzy subsystem; represents the random weight of the jth enhancement node in the enhancement layer; represents the random drift of the jth enhancement node in the enhancement layer; Represents the random mapping value of the jth enhancement node of the enhancement layer.

[0022] Preferably, the expression of the fault diagnosis model is: make ,but:

[0023] In the formula, represents the output of the fault diagnosis model; The weight of the fault diagnosis model; represents the weight of the enhancement layer; It represents the output when the enhancement layer has m enhancement nodes; Indicates that The preceding value is defined as the symbol The following value; Represents the weight of the mapping layer; Represents the rule output of the fuzzy subsystem of the mapping layer; represents the weight of the i-th fuzzy subsystem in the mapping layer; Represents the mapping coefficient of the i-th fuzzy subsystem in the mapping layer.

[0024] Preferably, clusters are grouped using triples It indicates that, A vector representing the linear weighted sum of all feature parameters of the input data of the dataset; A vector representing the weighted sum of squares of all feature parameters in the input data of the dataset; Represents maturity; the cluster center and cluster radius of the cluster are:

[0025]

[0026] In the formula, ; Represents the cluster radius of the cluster; Indicates the number of types of feature parameters in the input data; Represents the cluster center of the cluster.

[0027] Preferably, the steps of dynamically updating the cluster center point and cluster radius of the cluster cluster at each moment through the cluster radius and radius threshold of the cluster cluster are: The j-th characteristic parameter at the corresponding moment when the cluster radius of the cluster is less than the radius threshold is merged into the cluster at each moment to obtain a new cluster; Get the cluster center and cluster radius of the new cluster.

[0028] Preferably, the expression of the fault diagnosis model after each enhancement node is added is:

[0029]

[0030]

[0031] In the formula, It represents the fault diagnosis model after the first enhancement node is added; Represents the network structure of the fault diagnosis model before adding enhanced nodes; It represents the network structure of the fault diagnosis model after the first enhancement node is added; represents the target weight of the fault diagnosis model after the first enhancement node is added; represents the weight of the fault diagnosis model before adding the enhanced node; represents the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before the enhanced node; A first intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the node enhancement; A second intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the enhanced nodes.

[0032] The beneficial effects of the present invention are: When processing real-time data streams, the time-weighted incremental clustering method introduces a weight decay mechanism to adapt to data changes, dynamically adjusts the position and number of fuzzy width learning clustering center points, and enables the model to have online update capabilities in real-time data stream scenarios. At the same time, the diagnostic accuracy of the model is improved by incrementally updating the enhanced nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A flow chart of a method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of stream data according to the present invention; Figure 2 It is a flow chart of incremental clustering steps based on time weighting in an embodiment of the present invention; Figure 3 It is a schematic diagram comparing the diagnostic accuracy of the method of the present invention and FBLS in the embodiment.

[0035] Figure 4 It is a schematic diagram comparing the training time of the method of the present invention and FBLS in the embodiment.

[0036] Figure 5 is the method parameter of the present invention in the embodiment Schematic diagram of the impact on the diagnostic accuracy of RFBLS. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0038] An embodiment of a method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data of the present invention is as follows: Figure 1 As shown, including: S1, obtaining characteristic parameters and status data of the actuator at each operating moment; Specifically, characteristic parameters and status data of the actuator at each operating moment are obtained, and the characteristic parameters are used as input data, and the status data corresponding to the actuator is used as output data to construct a data set; wherein the characteristic parameters include: temperature data of the actuator and acceleration data in three axes; the status data include: normal operation, blocking fault, spalling fault and motor fault.

[0039] In this embodiment, the characteristic parameters are sequentially cleaned and standardized, and the standardized characteristic parameters are used as input data, wherein the step of data cleaning of the characteristic parameters is: traversing the collected characteristic parameters, marking missing values ​​and outliers, wherein the determination of outliers adopts the 6σ criterion, that is, the number greater than the sum of the mean and 6 times the standard deviation is regarded as an outlier; the missing values ​​and outliers are replaced by the neighboring value filling method, that is, the average value of the 5 points before and after the missing value or outlier is used to fill and replace, and the data cleaning is completed. The Z-score standardization method is used to process the data set, eliminate the difference between the dimensions, complete the data standardization, and the characteristic parameters that have finally completed the data standardization are used as input data.

[0040] S2, build a fault diagnosis model; Specifically, a fault diagnosis model is constructed, which includes an enhancement layer and a mapping layer. The mapping layer is used to perform fuzzy mapping on the input data and output data in the data set and output the defuzzified mapping result; the enhancement layer is used to perform nonlinear changes on the defuzzified mapping result and output the fault diagnosis result.

[0041] The method for establishing the fault diagnosis model in this embodiment is consistent with the model establishment method in "Fuzzy broad learning system: Anovel neuro-fuzzy model for regression and classification". Specifically, the mapping layer includes: a fuzzification mapping layer and a defuzzification mapping layer; the fuzzification mapping layer is used to fuzzify the input data and output data in the data set, and the defuzzification mapping layer is used to defuzzify the fuzzified output; Step 21, establish the fuzzy mapping layer model: In the Takagi-Sugeno fuzzy system, let the characteristic parameters , , , assuming that the first Fuzzy subsystem exists rules, then there are Rules, that is … , , , then we have: (1) In the formula, Indicates The output value of the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The fuzzy coefficient of the tth parameter in the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The tth parameter among the characteristic parameters; represents the weighted sum function; Indicates The mth parameter among the characteristic parameters.

[0042] Then, Characteristic parameters The corresponding The expression of the fuzzy subsystem is: (2) In the formula, Indicates The output of the i-th fuzzy subsystem corresponding to the input of the characteristic parameter; Indicates The characteristic parameters of the fuzzy subsystem under the kth rule of the i-th fuzzy subsystem; Indicates The normalized weight of the kth rule in the i-th fuzzy subsystem is the characteristic parameter; , , , here, Indicates The characteristic parameter is the weight of the kth rule in the i-th fuzzy subsystem; Indicates The fuzziness of the tth parameter of the characteristic parameter in the kth rule of the i-th fuzzy subsystem is , the standard deviation is Gaussian distribution, for the convenience of calculation, Take 1, mean Depend on The cluster center value directly obtained by the algorithm; The output of the fuzzy subsystem for all input parameters in the data set is: (3) Then, the fuzzy mapping layer is The output of the fuzzy subsystem is: (4) Step 22, establish a defuzzification mapping layer: In order to preserve the input characteristics, the i-th fuzzy subsystem has the s-th characteristic parameter The defuzzified output for: (5) In the formula, Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; is the mapping coefficient of the sth characteristic parameter under the kth rule of the ith fuzzy subsystem.

[0043] Then The defuzzified output of the fuzzy subsystem is formula (6), and the fuzzy mapping layer is formula (7): (6) (7) In the formula, The fuzzy mapping layer is represented as The defuzzified output when there are fuzzy subsystems; express The defuzzified output of the i-th fuzzy subsystem among the fuzzy subsystems; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; is the mapping coefficient of the sth characteristic parameter under the kth rule of the i-th fuzzy subsystem; represents the output of the fuzzy subsystem at the mapping layer, ; represents the weight of the i-th fuzzy subsystem, .

[0044] Step 23: Create an enhancement layer: The outputs of the fuzzy rules generated by each fuzzy subsystem are not immediately summarized into one value, but are all sent to the enhancement layer for further nonlinear transformation to obtain the first The output of the enhancement layer node is: (8) Then, the output of the enhancement layer when there are m enhancement nodes, that is, the enhancement layer is: (9) In the formula, It represents the output when the enhancement layer has m enhancement nodes. represents the output of the mth enhancement node of the enhancement layer; The enhancement layer The output of the enhancement node; The fuzzy mapping layer is represented as The output of the fuzzy subsystem; represents the random weight of the jth enhancement node in the enhancement layer; represents the random drift of the jth enhancement node in the enhancement layer; Represents the random mapping value of the jth enhancement node of the enhancement layer.

[0045] Step 24: Establish a fault diagnosis model: Combining the mapping layer and the enhancement layer, the final fault diagnosis model is obtained as follows: (10) In the formula, is the weight of the enhancement layer, represents the output of the fault diagnosis model; make ,but: (11) In the formula, represents the output of the fault diagnosis model; The weight of the fault diagnosis model; represents the weight of the enhancement layer; It represents the output when the enhancement layer has m enhancement nodes; Indicates that The preceding value is defined as the symbol The following value; Represents the weight of the mapping layer; Represents the rule output of the fuzzy subsystem of the mapping layer; represents the weight of the i-th fuzzy subsystem in the mapping layer; Represents the mapping coefficient of the i-th fuzzy subsystem in the mapping layer.

[0046] Using the pseudo-inverse solution technology of ridge regression, the weight of the fault diagnosis model is obtained as follows: (12) in, .

[0047] S3, obtain the target data set; Specifically, a cluster is set for each characteristic parameter, the maturity of the cluster is set according to the time decay function, and the cluster of each characteristic parameter at each moment is obtained according to the maturity of the cluster, the linear weighted sum of all characteristic parameters in each characteristic parameter of the input data of the data set, and the weighted square sum of all characteristic parameters in each characteristic parameter of the input data of the data set; The cluster center and cluster radius of the cluster cluster are obtained, and the cluster center and cluster radius of the cluster cluster at the next moment are dynamically updated through the cluster radius and radius threshold of the cluster cluster to obtain the target cluster center and target cluster radius of the cluster cluster, and the data in the cluster cluster that does not belong to the range of the cluster center and the target cluster radius are removed to obtain the target cluster cluster for each feature parameter; the target data set is obtained according to the target cluster cluster for each feature parameter.

[0048] In this embodiment, an incremental clustering step based on time weighting is proposed. By setting cluster clusters, the cluster center points are dynamically adjusted to adapt to the model update in the data stream scenario. Specifically, the incremental clustering step is: Suppose a cluster is , using triples Represents clusters as , when the time is , the characteristic parameters within the cluster are , is the characteristic parameter at the nth moment, where , is the dth characteristic parameter at the i-th moment, and the timestamp of each moment is , then the expression of maturity is: (13) in , The maturity threshold indicating whether a cluster is mature, Greater than or equal to When Less than When it is immature. is the weight decay function, as time goes by Exponential decay, through which the weight of each sample is gradually reduced over time, that is, the attention paid to the new feature parameters is greater than the old feature parameters in the data set. The model training method based on incremental learning needs to continuously learn and update the model based on the new data collected after deployment. Therefore, it is necessary to make the model pay more attention to new samples (new feature parameters) so as to learn new knowledge. In general, , To control the attenuation rate, for this embodiment The value is 0.02.

[0049] (14) in For the period 1-n The linear weighted sum of the characteristic parameters, A vector representing the linearly weighted sum of all feature parameters of the input data set.

[0050] (15) in For the period 1-n The weighted sum of squares of the characteristic parameters; A vector representing the weighted sum of squares of all feature parameters in the input data of the dataset.

[0051] Let cluster The center point of (in order to be consistent with other clustering algorithms, cluster The position vector of the cluster center is uniformly called the cluster center) c, and the radius is , then: (16) In the formula, ; c is the location vector of the cluster center, j=1,2,3,4; (17) In the formula, Represents the cluster radius of the cluster; Indicates the number of types of feature parameters in the input data; Represents the cluster center of the cluster.

[0052] Next, Update the process over time, as long as, , is the maximum threshold radius of a cluster corresponding to each characteristic parameter, then At this moment, the jth parameter of this characteristic parameter is merged into the cluster corresponding to the characteristic parameter. At this time, we have: (18) exist After a while, three The decay is: (19) Here we assume that the cluster exist No more data that meets the conditions (i.e., obtained through new input data) is received within the time period ) new data.

[0053] The updated cluster center and cluster radius of the cluster cluster corresponding to each characteristic parameter are: (20) (twenty one) The data that does not belong to the cluster center point and the target cluster radius range is removed from the cluster cluster to obtain the target cluster cluster of each characteristic parameter; the target data set is obtained according to the target cluster cluster of each characteristic parameter.

[0054] In this embodiment, clusters evolve over time, and a weight decay mechanism is used to give newer samples more attention than older samples. Since the weight of each cluster decays exponentially over time, the model can adapt to concept drift. In subsequent use, when the cluster is a mature cluster, the mature cluster is used as the real cluster center, and the immature cluster is recorded. When the immature cluster matures, it is used as the cluster center again. At the same time, as time goes by, when the weight of the mature cluster decreases and becomes an immature cluster, the cluster center is removed in time, such as Figure 2 As shown, , maturity threshold , , for the maturation process of clusters, when When , let the input data be , a new cluster is formed According to the above formulas (13) to (21), we can get ,at this time is an immature cluster; When the new input data is ,at this time , ,at this time , Still an immature cluster; when When the new input data is , , , ,at this time , Become a mature cluster. Here, by default, in all cases, the radius of the mature cluster Must be smaller than the maximum threshold radius For the process of clustering from mature to immature, When the new input data is ,make The radius to the cluster center is greater than , a new cluster is formed , , is an immature cluster, and , , , , Still a mature cluster; when When the new input data is , and belongs to the immature cluster ,at this time , , , Still an immature cluster, , , , , From mature clusters to immature clusters.

[0055] S4, adding nodes of the enhancement layer and obtaining a target fault diagnosis model; Since step S2 establishes a fault diagnosis model, step 4 uses an incremental learning method to update the model established in step 2. Unlike traditional models, once trained and deployed, the model parameters are solidified and cannot be changed or new data features can be learned. This patent is based on incremental learning to enable the model to self-learn and self-update based on new data after deployment, which is more in line with industrial implementation needs. Therefore, in this embodiment, the enhanced layer of the fault diagnosis model is increased according to a preset number of enhanced nodes, and the fault diagnosis model after the enhanced nodes are added is obtained. The loss function value of the enhanced fault diagnosis model is obtained according to the target data set. When the loss function value does not meet the preset loss function threshold, the preset number of enhanced nodes is added again until the loss function value of the enhanced fault diagnosis model meets the preset loss function threshold, and the enhanced fault diagnosis model is used as the target fault diagnosis model.

[0056] In this embodiment, it is assumed that 20 enhanced nodes are added each time, and the network structure of the fault diagnosis model before adding enhanced nodes is and weights are , the network structure after adding enhanced nodes and weights are , The weights of 20 newly added enhanced nodes are as follows: (twenty two) in, , therefore, the weight of each enhanced node is: (twenty three) According to the network structure after each enhancement node is added and the weight after each enhancement node is added, the fault diagnosis model after each enhancement node is added can be obtained. The fault diagnosis model after each enhancement node is added is:

[0057] In the formula, It represents the fault diagnosis model after the first enhancement node is added; Represents the network structure of the fault diagnosis model before adding enhanced nodes; It represents the network structure of the fault diagnosis model after the first enhancement node is added; represents the target weight of the fault diagnosis model after the first enhancement node is added; represents the weight of the fault diagnosis model before adding the enhanced node; represents the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before the enhanced node; A first intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the node enhancement; A second intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the enhanced nodes.

[0058] The fault diagnosis model after each enhanced node is tested according to the target data set, and the loss function value of the fault diagnosis model after the enhanced node is added is compared with the preset loss function threshold (10 in this embodiment). -6 ), if the loss function value is greater than or equal to the loss function threshold, the enhanced node is added again until the loss function value is less than the loss function threshold, and the fault diagnosis model after adding the enhanced node is used as the target fault diagnosis model.

[0059] S5. Obtaining mechanical and electrical equipment fault diagnosis results using the target fault diagnosis model; The input data at the current moment is input into the target fault diagnosis model, and the electromechanical equipment fault diagnosis result is output.

[0060] The present invention will be described below in conjunction with the accompanying drawings: In this embodiment, an aircraft actuator is used as an example to illustrate, and four characteristic parameters of the aircraft actuator, namely, X-axis acceleration, Y-axis acceleration, Z-axis acceleration, and temperature, are selected for experiment, and four conditions, namely, normal operation, blocking fault, peeling fault, and motor fault, are used as state parameters. The selected experimental data set contains a total of 600 sets of original data sets.

[0061] Step 1: Verify the validity of the target data set obtained in step S3: Analyze the performance of RFBLS (real-time data stream model) after introducing time-weighted incremental clustering in step S3 and set the number of mapping layer nodes 14, the number of nodes added each time the enhancement layer is added is 60, let , , , the results are as follows Figure 3 , Figure 4 As shown, Figure 3 It represents the relationship between the number of cluster centers and RFBLS and FBLS. Figure 4 It shows the comparison of training time of RFBLS and FBLS (retrained model). The following conclusions can be drawn from the figure: 1) With lower time cost, the average fault diagnosis rate under real-time data stream reaches 99%, which is close to the fault diagnosis accuracy of cloud FBLS. However, compared with FBLS (retraining), the training time required for RFBLS to execute the model update process is shorter, about half of the former, which shows the feasibility of the method proposed in this invention and its advantages over the FBLS fault diagnosis method in real-time data stream scenarios.

[0062] 2) As the cluster centers change, the fault diagnosis accuracy of RFBLS and FBLS both show a process of first decreasing, then fluctuating, and finally gradually and steadily increasing. This is because there are fewer data at the beginning, the model fitting accuracy is higher (even overfitting may occur), and the fault diagnosis accuracy is higher. As the data increases, the diagnosis accuracy decreases, but at the same time, the increase in data will also increase the number of cluster centers. However, when the number of centers increases, the diagnosis accuracy will increase again, so it fluctuates. Finally, the number of cluster centers is large (the network structure is relatively complex), the amount of data is relatively large, and the single data has little effect on the diagnosis accuracy, so it shows a stable or even slightly increasing trend.

[0063] 3) As the cluster center changes, the diagnostic accuracy will fluctuate to a certain extent, and the diagnostic accuracy of RFBLS is smoother. Although the diagnostic accuracy of FBLS is higher, the actual comparison shows that the difference between the two does not exceed 0.2%. The reason why RFBLS fluctuates less than FBLS is that FBLS re-clusters each time it is trained, which may lead to large differences in the cluster centers of two adjacent times, so there will be a large gap in the diagnostic accuracy. RFBLS uses incremental clustering, and the changes in the cluster centers of two adjacent times are relatively small, so it is more stable.

[0064] 4) As the amount of data increases, the time cost required for RFBLS and FBLS increases. The reason for the increase in training time required for the analysis model to receive data and complete one update is that in FBLS, the amount of data required to be processed for each clustering under real-time data flow increases, so the training time increases. In RFBLS, although historical data does not need to be processed, as the data increases, the number of clusters will also increase, and the amount of data required to calculate each model update will increase, so the training time will also increase.

[0065] Step 2: Verify the validity of the node increment: Setting hyperparameters of the fault diagnosis model: number of network layers 2, the number of nodes in the mapping layer 6, the number of nodes added each time the enhancement layer is added The number of enhanced nodes is 20, and the number of enhanced nodes is dynamically increased by 20 each time, and finally reaches 100. The results are shown in Table 1. It can be concluded that with the increase of enhanced nodes, the model accuracy has been improved to a certain extent. This proves that the enhanced node increment method of RFBLS is effective. When the model cannot achieve the ideal performance after training, the model diagnosis accuracy can be improved by adding new enhanced nodes.

[0066] Table 1

[0067] This embodiment mainly discusses The impact on the accuracy of diagnosis, setting is 14, is 60, For the parameters of the weight decay function, set , , exist The value is in the range of Figure 5 As shown. Figure 5 It can be concluded that as As the size of the cluster increases, the number of cluster centers shows an overall downward trend. This is mainly because the mature clusters decay faster, resulting in an increase in the number of immature clusters. This shows that the number of cluster centers does have an impact on the accuracy of fault diagnosis. In summary. The incremental fuzzy width learning electromechanical equipment fault diagnosis method based on streaming data of the present invention performs excellently in the NASA actuator data set fault diagnosis problem, can effectively process real-time data streams, while retaining the learning speed of the BLS method, and can easily complete model training on an ordinary PC using a shorter practice, with a high fault diagnosis accuracy.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for fault diagnosis of electromechanical equipment based on incremental fuzzy width learning of streaming data, characterized in that: include: Acquire the characteristic parameters and state data of the actuator at each operating moment, take the characteristic parameters as input data, take the state data corresponding to the actuator as output data, and construct a data set; wherein the characteristic parameters include: temperature data of the actuator and acceleration data of three axes; the state data include: normal operation, blocking fault, spalling fault and motor fault; Constructing a fault diagnosis model, the fault diagnosis model includes an enhancement layer and a mapping layer, the mapping layer is used to perform fuzzy mapping on the input data and output data in the data set, and output the defuzzified mapping result; the enhancement layer is used to perform nonlinear changes on the defuzzified mapping result and output the fault diagnosis result; A cluster is set for each characteristic parameter, the maturity of the cluster is set according to the time decay function, and the cluster of each characteristic parameter at each moment is obtained according to the maturity of the cluster, the linear weighted sum of all characteristic parameters in each characteristic parameter of the input data of the data set, and the weighted square sum of all characteristic parameters in each characteristic parameter of the input data of the data set; Obtain the cluster center and cluster radius of the cluster cluster, dynamically update the cluster center and cluster radius of the cluster cluster at the next moment through the cluster radius and radius threshold of the cluster cluster, obtain the target cluster center and target cluster radius of the cluster cluster, remove the data in the cluster cluster that does not belong to the range of the cluster center and the target cluster radius, and obtain the target cluster cluster for each feature parameter; obtain the target data set according to the target cluster cluster for each feature parameter; Adding nodes to the enhanced layer of the fault diagnosis model according to a preset number of enhanced nodes, and obtaining the fault diagnosis model after adding the enhanced nodes, obtaining the loss function value of the enhanced fault diagnosis model according to the target data set, and when the loss function value does not meet the preset loss function threshold, adding a preset number of enhanced nodes again until the loss function value of the enhanced fault diagnosis model meets the preset loss function threshold, and then taking the enhanced fault diagnosis model as the target fault diagnosis model; The input data at the current moment is input into the target fault diagnosis model, and the electromechanical equipment fault diagnosis result is output.

2. According to claim 1, a method for fault diagnosis of electromechanical equipment based on incremental fuzzy width learning of streaming data is characterized in that: The feature parameters are cleaned and standardized in turn, and the standardized feature parameters are used as input data.

3. The method for fault diagnosis of electromechanical equipment based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The steps for data cleaning of feature parameters are as follows: Traverse the collected characteristic parameters and mark the missing values ​​and outliers. The outliers are determined by the 6σ criterion. The missing values ​​and outliers are replaced by the neighboring value filling method to complete data cleaning.

4. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The steps for data standardization of feature parameters are: using the Z-score standardization method to perform data standardization on the data set.

5. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The mapping layer includes: a fuzzification mapping layer and a defuzzification mapping layer; the fuzzification mapping layer is used to perform fuzzification mapping on the input data and output data in the data set, and the defuzzification mapping layer is used to defuzzify the fuzzified output; The fuzzy mapping layer model expression is: In the formula, Indicates The output value of the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The normalized weight of the kth rule in the i-th fuzzy subsystem is the characteristic parameter; represents the weighted sum function; Indicates The mth parameter among the characteristic parameters; Indicates The fuzzy coefficient of the tth parameter in the characteristic parameter under the kth rule of the i-th fuzzy subsystem; Indicates The tth parameter among the characteristic parameters; Indicates The output of the i-th fuzzy subsystem corresponding to the input of the characteristic parameter; express The output of the i-th fuzzy subsystem among the fuzzy subsystems; The fuzzy mapping layer is represented as The output of the fuzzy subsystem; The defuzzification mapping layer model expression is: ,......, In the formula, The fuzzy mapping layer is represented as The defuzzified output when there are fuzzy subsystems; express The defuzzified output of the i-th fuzzy subsystem among the fuzzy subsystems; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; Indicates The defuzzified output of the i-th fuzzy subsystem corresponding to the characteristic parameter; is the mapping coefficient of the sth characteristic parameter under the kth rule of the i-th fuzzy subsystem; represents the output of the fuzzy subsystem at the mapping layer, ; represents the weight of the i-th fuzzy subsystem, .

6. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The expression of the enhancement layer is: In the formula, It represents the output when the enhancement layer has m enhancement nodes. represents the output of the mth enhancement node of the enhancement layer; The enhancement layer The output of the enhancement node; The fuzzy mapping layer is represented as The output of the fuzzy subsystem; represents the random weight of the jth enhancement node in the enhancement layer; represents the random drift of the jth enhancement node in the enhancement layer; Represents the random mapping value of the jth enhancement node of the enhancement layer.

7. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The expression of the fault diagnosis model is: make ,but: In the formula, represents the output of the fault diagnosis model; represents the weight of the fault diagnosis model; represents the weight of the enhancement layer; It represents the output when the enhancement layer has m enhancement nodes; Indicates that The preceding value is defined as the symbol The following value; Represents the weight of the mapping layer; Represents the rule output of the fuzzy subsystem of the mapping layer; represents the weight of the i-th fuzzy subsystem in the mapping layer; Represents the mapping coefficient of the i-th fuzzy subsystem in the mapping layer.

8. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: Clustering with triples It indicates that, A vector representing the linear weighted sum of all feature parameters of the input data of the dataset; A vector representing the weighted sum of squares of all feature parameters in the input data of the dataset; Represents maturity; the cluster center and cluster radius of the cluster are: In the formula, ; Represents the cluster radius of the cluster; Indicates the number of types of feature parameters in the input data; Represents the cluster center of the cluster.

9. The method for diagnosing electromechanical equipment faults based on incremental fuzzy width learning of streaming data according to claim 1 is characterized in that: The steps of dynamically updating the cluster center point and cluster radius of the cluster at each moment through the cluster radius and radius threshold of the cluster are: The j-th characteristic parameter at the corresponding moment when the cluster radius of the cluster is less than the radius threshold is merged into the cluster at each moment to obtain a new cluster; Get the cluster center and cluster radius of the new cluster.

10. The method for fault diagnosis of electromechanical equipment based on incremental fuzzy width learning of streaming data according to claim 1, characterized in that: The expression of the fault diagnosis model after each enhancement node is added is: In the formula, It represents the fault diagnosis model after the first enhancement node is added; Represents the network structure of the fault diagnosis model before adding enhanced nodes; It represents the network structure of the fault diagnosis model after the first enhancement node is added; represents the target weight of the fault diagnosis model after the first enhancement node is added; represents the weight of the fault diagnosis model before adding the enhanced node; Represents the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before the enhanced node; A first intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the node enhancement; A second intermediate quantity representing the "pseudo-inverse" of the network structure matrix of the fault diagnosis model before and after the enhanced nodes.

Images