A method for monitoring the health status of electric fluid machinery

Abstract

The present invention provides a method for monitoring the health status of an electric fluid machinery, including constructing a 2D model state tensor, stacking to generate a 3D model state tensor for building a model; during the monitoring period, constructing a 2D monitoring state tensor as a model input, reconstructing the model output through similarity calculation; comparing the monitoring input with the model output to obtain a monitoring deviation, comparing it with a warning threshold, and realizing health status monitoring. The present invention can effectively monitor the health status of an electric fluid machinery, and timely warn when there is a risk of failure in the monitoring, with high accuracy and sensitivity.

Description

Technical Field

[0001] The present invention relates to the field of health status monitoring of electric fluid machinery, and in particular to a health status monitoring method of electric fluid machinery based on transient operating condition data and tensor similarity analysis. Background Art

[0002] Traditional methods for health status monitoring of electric fluid machinery mainly rely on steady-state operating data of the equipment, such as vibration analysis, temperature over-limit monitoring, etc. Since the physical state and operating characteristics of the equipment will change significantly during transient conditions such as startup, shutdown or load changes, these changes often indicate potential failures and performance degradation. Therefore, most traditional methods have limitations when monitoring the health status of transient conditions of electric fluid machinery.

[0003] In recent years, with the development of sensor technology, data acquisition technology and data analysis technology, data-driven equipment health status monitoring methods have gradually been applied. In particular, using transient operating condition data for health status monitoring can more comprehensively reflect the performance changes of equipment under dynamic working conditions and provide richer information for health status assessment.

[0004] Therefore, developing a health status monitoring method for electric fluid machinery based on transient operating condition data and tensor similarity analysis is of great practical significance for improving the accuracy and sensitivity of monitoring and ensuring the safe and efficient operation of equipment. Summary of the invention

[0005] The purpose of the present invention is to provide a method for monitoring the health status of electric fluid machinery based on transient operating condition data and tensor similarity analysis, aiming to solve some limitations existing in the prior art and improve the accuracy and sensitivity of monitoring the health status of electric fluid machinery.

[0006] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0007] A method for monitoring the health status of an electric fluid machinery, characterized in that the method for monitoring the health status of an electric fluid machinery is based on transient operating condition data and tensor similarity analysis, and comprises the following steps:

[0008] S1. Collect transient operating condition data of the electric fluid machinery, mark the data, and clarify the safety status corresponding to the transient operating condition data at different times; perform data preprocessing, including data screening and normalization, and use the preprocessed data as a model database of the monitoring method;

[0009] S2, defining the data in the model database as a 1-dimensional model state tensor, using a step-size adaptive sliding time window operation to generate multiple 2-dimensional model state tensors containing time series information;

[0010] S3, remove similar 2D model state tensors, stack the retained 2D model state tensors, obtain the 3D model state tensor and its similarity matrix, and the model is established;

[0011] S4, taking the 2D monitoring state tensor corresponding to the electric fluid machinery in the monitoring period as the input of the model; calculating the similarity between the 2D monitoring state tensor and all the 2D model state tensors to form a similarity vector; obtaining a weight vector through similarity vector calculation, and performing the first modal multiplication with the 3D model state tensor to obtain the output of the model;

[0012] S5. Calculate the deviation between the model input and output, extract the latest data in the deviation and denormalize it, compare the denormalized deviation data with the set warning threshold, obtain the monitoring result, and realize the health status monitoring of the electric fluid machinery.

[0013] Preferably, in step S1, the following steps are specifically included:

[0014] S101. During the operation of the electric fluid machinery, transient operating condition data are collected, including parameter data under start-up, shutdown and variable load conditions, wherein the parameters include electrical parameters, mechanical parameters and temperature parameters. The electrical parameters include motor current, the mechanical parameters include bearing vibration and motor speed, and the temperature parameters include motor temperature and bearing temperature.

[0015] S102, labeling the data collected in step S101 with a safety status label, wherein the safety status label includes "safe" and "failure risk";

[0016] S103, preprocessing the data annotated in step S102, retaining the data with the safety status label "safe", and normalizing the data through the standard score normalization method; using the preprocessed data as a model database of the monitoring method for building the model.

[0017] To further explain, standard score standardization is a method of converting data into distributed data with a mean of 0 and a standard deviation of 1, so that the data can be compared and calculated on the same scale.

[0018] Preferably, in step S2, the following steps are specifically included:

[0019] S201, extracting data at a certain moment from the model database, arranging the extracted data and defining it as a 1-dimensional model state tensor as follows:

[0020] ,

[0021] in, Represents the time index, Indicates the number of measurement points, express The 1-dimensional model state tensor at time, Indicates the index number of the measuring point. Indicates The measuring points are Normalized data at the moment, represents the set of positive integers, Indicated by all n A set of real vectors of dimension ;

[0022] S202, apply a time window operation to the data, repeat step S201 at consecutive moments, obtain multiple 1D model state tensors and incorporate them into the time window, and synthesize the 2D model state tensor as follows:

[0023] ,

[0024] in, Indicates the number of times step S201 is repeated, represents the 2D model state tensor corresponding to the time window, Indicated by all n OK L A collection of real matrices with columns;

[0025] S203, sliding on the time axis to execute step S202, and calculating the standard deviation mean of the data in the time window, and adaptively selecting the size of the sliding step according to the standard deviation mean, the standard deviation mean is defined as follows:

[0026] ,

[0027] in, express Time to The standard deviation mean of the data in the time window at time , represents the standard deviation function, Represents a 2D model state tensor No. row vectors; the 2D model state tensors obtained by the adaptive step-size sliding time window operation are expressed as ,in Represents the number of resulting 2D model state tensors.

[0028] Further explanation: In step S203, the sliding step size is a mapping of the mean standard deviation of the data in the time window, that is, ,in express Time to The sliding step size of the time window at the moment, Represents the step length mapping function; in order not to miss data during the sliding process of the time window, the sliding step length needs to satisfy .

[0029] Preferably, in step S3, the following steps are specifically included:

[0030] S301, define a similarity operator, and the similarity calculation rule of two 2D model state tensors is as follows:

[0031] ,

[0032] in, Denotes a similarity operator whose result is a scalar, with the subscript , , Respectively , The 2D model state tensor at time, express No. Row vector and No. Row vector The similarity between Indicates The weight of each measurement point data in calculating the similarity;

[0033] S302, search and The most similar 2D model state tensor , where the subscript , , let the similarity ;

[0034] S303, yes Repeat step S302 to obtain the subscript set , similarity set , set the similarity threshold ,get Greater than threshold The subscript set corresponding to the elements of , similar 2D model state tensors are eliminated, and the set of retained 2D model state tensors is ,in Indicates the subscript set corresponding to the moment index of the retained 2D model state tensor, subscript Representing a collection Elements in

[0035] S304: The set obtained in step S303 The number of state tensors of the two-dimensional model is , the 2D model state tensors are stacked to construct the 3D model state tensor as follows:

[0036] ,

[0037] in, represents the 3D model state tensor, Represents the stacking operation of the 2D model state tensor, Represents all shapes The set of three-dimensional real tensors of ;

[0038] S305. Define index , and Defined as The similarity matrix of p Line q The column elements satisfy:

[0039] ,

[0040] in, represents a similarity operator whose result is a matrix, and Represents the 3D model state tensor respectively No. p The first q slices, i.e., 2D model state tensor.

[0041] Further explanation: In step S301, the row vector and The similarity can be calculated by using Euclidean distance, cosine similarity and other calculation methods; similarity weight The value of should refer to the importance of the corresponding measuring point in health status monitoring.

[0042] Further explanation, in step S302, The most similar 2D model state tensor The corresponding subscript The selection should satisfy: .

[0043] Further explanation: In step S303, the similarity threshold The setting of can refer to the similarity set ,For example Set to on Number of digits , it indicates that there is A number of 2D model state tensors are considered to be similar and need to be removed.

[0044] Further explanation: In step S305, the similarity matrix is a symmetric matrix with all diagonal elements equal to zero.

[0045] Preferably, in step S4, the following steps are specifically included:

[0046] S401, collect and normalize the transient operating condition data of the electric fluid machinery used for monitoring at a certain moment in the monitoring period, and obtain a 1-dimensional monitoring state tensor:

[0047] ,

[0048] in, Indicates the monitoring time index, express The 1-dimensional monitoring state tensor at the moment, Indicates The measuring points are Normalized data at the moment;

[0049] S402, using time window operation, in the monitoring period L Repeat step S401 for each continuous monitoring moment to obtain L The 1-dimensional monitoring state tensor is incorporated into the time window to synthesize the 2-dimensional monitoring state tensor as the input of the model:

[0050] ,

[0051] in, Indicated in Time to The 2D monitoring state tensor corresponding to the time window at the moment;

[0052] S403, calculation model input With the 3D model state tensor The similarity of all slices in constitutes the similarity vector , The similarity operator represents the result of a vector. p Elements satisfy:

[0053] ,

[0054] S404, through similarity vector The original weight vector is calculated as follows:

[0055] ,

[0056] in, represents the original weight vector obtained by the model, Represents the similarity matrix The inverse matrix of ; normalize the original weight vector so that the sum of its elements is 1, and the normalized weight vector is as follows:

[0057] ,

[0058] in, represents the normalized weight vector obtained by the model, represents the normalization function, express No. p elements;

[0059] S405, the 3D model state tensor With the normalized weight vector Performing the first modal multiplication, the output of the model is as follows:

[0060] ,

[0061] in, represents the reconstructed output of the model, Represents a 3D model state tensor With vector Multiply along the first mode.

[0062] It is further explained that the normalization parameters in step S401 should be the same as the normalization parameters in step S103.

[0063] Further explanation: In step S404, the similarity matrix The reversibility of is a necessary condition for the feasibility of the monitoring method. Assuming that the 3D model state tensor The two slices are similar, that is, , then we have the similarity matrix No. p Row (column) and q The rows (columns) are similar, at this time is an ill-conditioned matrix or even a singular matrix, resulting in The numerical stability of is poor, which makes the method infeasible. Therefore, step S3 eliminates similar 2D model state tensors to ensure the feasibility of the monitoring method.

[0064] Preferably, in step S5, the following steps are specifically included:

[0065] S501. Calculation model input With model output Deviation, extract the latest monitoring time The deviation data is normalized and the monitoring deviation vector is obtained:

[0066] ,

[0067] in, express n The data of the measurement points are The monitoring deviation vector at time, Indicates Monitoring deviation value of each measuring point data;

[0068] S502, yes n Set monitoring and warning thresholds for each measuring point, including high warning thresholds , low warning threshold , when an element in the monitoring deviation vector exceeds the warning threshold, the monitoring warning is triggered, otherwise it is considered The data at the time of monitoring meets the conditions of normal working conditions;

[0069] S503, repeating steps S4, S501, and S502 for the remaining monitoring moments in the monitoring period, obtaining monitoring results for the remaining monitoring moments, and realizing health status monitoring of the electric fluid machinery.

[0070] Further explanation, in step S502, if or , indicating a 1-dimensional monitoring state tensor No. If the data of a measuring point is abnormal, a monitoring warning will be triggered. , Respectively represent the warning threshold , No. elements.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] The present invention utilizes the transient operating condition information of electric fluid machinery to construct a state tensor, designs a tensor similarity calculation method and realizes health status monitoring of the electric fluid machinery. By converting transient operating condition data into a high-dimensional tensor, the present invention can better capture the intrinsic connections and similarity patterns between data, improve the ability to characterize the state, and realize accurate similarity analysis, so as to improve the accuracy and sensitivity of health status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A working flow chart of the method for monitoring the health status of an electric fluid machinery provided by the present invention;

[0074] Figure 2 A schematic diagram of the principle of constructing a 2D model state tensor in step S2;

[0075] Figure 3The flowchart for removing similar 2D model state tensors in step S3;

[0076] Figure 4 A schematic diagram of the principle of constructing a 3D model state tensor in step S3;

[0077] Figure 5 A schematic diagram of the monitoring input and model output of the motor temperature in a state monitoring task under a safe working condition according to an embodiment of the present invention;

[0078] Figure 6 A schematic diagram of a monitoring deviation of a motor temperature in a state monitoring task under a safe working condition according to an embodiment of the present invention;

[0079] Figure 7 A schematic diagram of the monitoring input and model output of the motor temperature in a state monitoring task under a fault risk condition according to an embodiment of the present invention;

[0080] Figure 8 The diagram is a schematic diagram of monitoring deviation of motor temperature and triggering of fault warning in a state monitoring task under a fault risk condition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0082] The electric fluid machinery selected in this embodiment is a large industrial electric water pump, such as Figure 1 As shown, a workflow diagram of the health status monitoring method of an electric fluid machinery based on transient operating condition data and tensor similarity analysis in this embodiment is shown. The steps in the embodiment are described below.

[0083] Step S1, collect transient operating data of the electric water supply pump, mark the data, and clarify the safety status corresponding to the transient data at different times; pre-process the data, including data screening and normalization, and use the pre-processed data as the model database of the monitoring method. Specifically, the following steps are included:

[0084] 1.1) During the operation of the electric water supply pump, transient operating data are collected, including parameter data under pump startup, shutdown and variable load conditions. The parameter data include electrical parameters, mechanical parameters and temperature parameters. The electrical parameters include motor current, the mechanical parameters include bearing vibration, motor speed, etc., and the temperature parameters include motor temperature and bearing temperature.

[0085] In this embodiment, a total of 26 key parameters are obtained by screening according to step 1, as shown in Table 1.

[0086] Table 1 Key parameter set

[0087] Parameter name Parameter Type quantity Motor speed Mechanical parameters 1 Motor current Electrical parameters 1 Motor temperature Temperature parameters 1 Bearing vibration displacement Mechanical parameters 4 Bearing vibration speed Mechanical parameters 4 Bearing vibration acceleration Mechanical parameters 4 Bearing temperature Temperature parameters 8 Pump inlet pressure Mechanical parameters 1 Pump outlet pressure Mechanical parameters 1 Feed tank water level Mechanical parameters 1

[0088] 1.2) Label the collected data with safety labels, label normal operating data as “safe”, and label operating data with failure risks as “failure risk”.

[0089] 1.3) Preprocess the data, retain the data with the safety status label of "safe", and normalize the data through the standard score normalization method; use the preprocessed data as the model database of the monitoring method for building the model.

[0090] Step S2: define the data in the model database as a 1-dimensional model state tensor, and use a step-size adaptive sliding time window operation to generate multiple 2-dimensional model state tensors containing time series information. Step 2 specifically includes the following steps:

[0091] 2.1) Extract the data at a certain moment from the model database, arrange the extracted data and define it as a 1-dimensional model state tensor as follows:

[0092] ,

[0093] in, Represents the time index, express The 1-dimensional model state tensor at time, Indicates the index number of the measuring point. Indicates The measuring points are Normalized data at the moment.

[0094] 2.2) The principle of constructing a 2D model state tensor is as follows Figure 2 As shown, Figure 2 In this embodiment, the parameters are . Figure 2 It indicates that a time window operation is applied to the data. Step 2.1 is repeated at 10 consecutive moments to obtain 10 1D model state tensors and incorporate them into the time window. The synthesized 2D model state tensor is as follows:

[0095] ,

[0096] in, Represents the 2D model state tensor corresponding to the time window.

[0097] 2.3) Slide and execute step 2.2 on the time axis, calculate the standard deviation mean of the data in the time window, and adaptively select the size of the sliding step. The standard deviation mean is defined as follows:

[0098] ,

[0099] in, express Time to The standard deviation mean of the data in the time window at time , represents the standard deviation function, Represents a 2D model state tensor No. A row vector.

[0100] In this embodiment, the standard deviation mean The sliding step size of the corresponding time window satisfy ,in is the negative correlation linear mapping and then rounded, and the maximum step size is 10 and the minimum step size is 1; 2000 2D model state tensors are obtained through adaptive step size sliding time window operation, which are respectively expressed as .

[0101] Step S3, remove similar 2D model state tensors, stack the remaining 2D model state tensors, obtain the 3D model state tensor and its similarity matrix, and the model is established. Step 3 specifically includes the following steps:

[0102] 3.1) Define the similarity operator, and the similarity calculation rule of two 2D model state tensors is as follows:

[0103] ,

[0104] in, Denotes a similarity operator whose result is a scalar, with the subscript , , Respectively , The 2D model state tensor at time, express No. Row vector and No. Row vector The similarity between Indicates The data of each measurement point is used to calculate the weight of similarity.

[0105] In this embodiment, the similarity between row vectors is calculated using the Euclidean distance method. The optional values ​​are 0.25, 0.5, 0.75 and 1, which respectively indicate that the parameter is less important, generally important, important and very important when used for health status monitoring tasks. The weight coefficients of motor speed, motor temperature and bearing temperature parameters are set to 1; the weight coefficients of motor current and bearing vibration parameters are set to 0.75; the weight coefficients of pump inlet and outlet pressure parameters are set to 0.5; and the weight parameter of water level parameter of water tank is set to 0.25.

[0106] 3.2) Find and The most similar 2D model state tensor , where the subscript , , let the similarity .

[0107] Among them, the subscript Should meet .

[0108] 3.3) The workflow for removing similar 2D model state tensors is as follows Figure 3 shown. Figure 3 In this embodiment, the parameters are Subscript Repeat step 3.2 to get the subscript set , similarity set ; Set similarity threshold ,get Greater than threshold The subscript set corresponding to the elements of , similar 2D model state tensors are eliminated, and the set of retained 2D model state tensors is ,in Indicates the subscript set corresponding to the moment index of the retained 2D model state tensor, subscript Representing a collection The elements in .

[0109] In this embodiment, the similarity Set to the median of , indicating that there is A number of 2D model state tensors are considered to be similar and need to be eliminated.

[0110] 3.4) The principle of constructing the 3D model state tensor is as follows Figure 4 As shown, Figure 4 In this embodiment, the parameters are . Will gather The 1000 2D model state tensors in are stacked and the 3D model state tensor is constructed as follows:

[0111] ,

[0112] in, represents the 3D model state tensor, Represents a stacking operation on a 2D model state tensor.

[0113] 3.5) Define index , and Defined as The similarity matrix of p Line q The column elements satisfy:

[0114] ,

[0115] in, Represents a similarity operator whose calculation result is a matrix, and Represents the 3D model state tensor respectively No. p The first q slices, i.e., 2D model state tensor.

[0116] Step S4, using the 2D monitoring state tensor corresponding to the electric water supply pump in the monitoring period as the input of the model; calculating the similarity between the 2D monitoring state tensor and all 2D model state tensors to form a similarity vector; obtaining a weight vector through similarity vector calculation, and performing the first modal multiplication with the 3D model state tensor to obtain the output of the model. Step 4 specifically includes the following steps:

[0117] 4.1) Collect and normalize the transient operating data of the electric water supply pump used for monitoring at a certain time in the monitoring period to obtain a 1-dimensional monitoring state tensor:

[0118] ,

[0119] in, Indicates the monitoring time index, express The 1-dimensional monitoring state tensor at the moment, Indicates The measuring points are Normalized data at the moment.

[0120] In this example, the normalization parameters used should be the same as those in step 1.3.

[0121] 4.2) Using the time window operation, repeat step 4.1 at 10 consecutive monitoring moments in the monitoring period to obtain 10 1-dimensional monitoring state tensors and incorporate them into the time window to synthesize a 2-dimensional monitoring state tensor as the input of the model:

[0122] ,

[0123] in, Indicated in Time to The 2D monitoring state tensor corresponding to the time window of time.

[0124] 4.3) Calculation model input With the 3D model state tensor The similarity of all slices in constitutes the similarity vector , The similarity operator represents the result of a vector. p Elements satisfy:

[0125] .

[0126] 4.4) Through the similarity vector The original weight vector is calculated as follows:

[0127] ,

[0128] in, represents the original weight vector obtained by the model, Represents the similarity matrix The inverse matrix of ; normalize the original weight vector so that the sum of its elements is 1, and the normalized weight vector is as follows:

[0129] ,

[0130] in, represents the normalized weight vector obtained by the model, represents the normalization function, express No. p elements.

[0131] 4.5) Convert the 3D model state tensor With the normalized weight vector Performing the first modal multiplication, the output of the model is as follows:

[0132] ,

[0133] in, represents the reconstructed output of the model, Represents a 3D model state tensor With vector Multiply along the first mode.

[0134] Step S5, calculate the deviation between the model input and output, extract the latest data in the deviation and denormalize it, compare the denormalized deviation data with the set warning threshold, obtain the monitoring result, and realize the health status monitoring of the electric water supply pump. Step 5 specifically includes the following steps:

[0135] 5.1) Calculation model input With model output Deviation, extract the latest monitoring time The deviation data is normalized and the monitoring deviation vector is obtained:

[0136] ,

[0137] in, Indicates that the data of 26 measuring points are The monitoring deviation vector at time, Indicates The monitoring deviation value of the data at each measuring point.

[0138] 5.2) Set monitoring and early warning thresholds for 26 measurement points, including high warning thresholds , low warning threshold , when an element in the monitoring deviation vector exceeds the warning threshold, the monitoring warning is triggered, otherwise it is considered The data at the monitoring time meets the conditions of normal working conditions.

[0139] 5.3) Repeat steps 4, 5.1, and 5.2 for the remaining monitoring moments in the monitoring period to obtain the monitoring results of the remaining monitoring moments, thereby realizing the health status monitoring of the electric water supply pump.

[0140] Taking the monitoring of motor temperature as an example, Figure 5 This is a schematic diagram of monitoring input and model output under safe working conditions. Figure 6 It is a schematic diagram of monitoring deviation under safe working conditions; Figure 7 This is a schematic diagram of monitoring input and model output under fault risk conditions. Figure 8 Schematic diagram of monitoring deviation under failure risk conditions.

[0141] In this embodiment, the monitoring results are statistically analyzed: for the health status monitoring of safe working conditions, after the monitoring deviation vector is normalized again, the root mean square error is 0.74%, and the fault false alarm rate is 0.00%; for the health status monitoring with fault risk, the warning time of this method is 362s earlier than the experience-based warning on average, and the fault missed alarm rate is 0.00%. It can be seen that the present invention uses transient working condition data and tensor similarity analysis to accurately analyze the multi-parameter change information in transient working conditions, monitor the health status of the machinery, and can promptly issue fault warnings, with high accuracy and sensitivity.

[0142] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for monitoring the health status of an electric fluid machinery, characterized in that: The electric fluid machinery health status monitoring method is based on transient operating condition data and tensor similarity analysis, and includes the following steps: S1. Collect transient operating data of electric fluid machinery, mark the data, and clarify the safety status corresponding to the transient operating data at different times; perform data preprocessing, including data screening and normalization, and use the preprocessed data as a model database of the monitoring method; S2. define the data in the model database as a 1-dimensional model state tensor, and use a step-size adaptive sliding time window operation to generate multiple 2-dimensional model state tensors containing time series information; S3. remove similar 2D model state tensors, stack the retained 2D model state tensors, obtain the 3D model state tensor and its similarity matrix, and the model is established; S4. taking the 2D monitoring state tensor corresponding to the electric fluid machinery in the monitoring period as the input of the model; Calculate the similarity between the 2D monitoring state tensor and all 2D model state tensors to form a similarity vector; obtain a weight vector through similarity vector calculation, and perform first mode multiplication with the 3D model state tensor to obtain the output of the model; S5. Calculate the deviation between the model input and output, extract the latest data in the deviation and denormalize it, compare the denormalized deviation data with the set warning threshold, obtain the monitoring result, and realize the health status monitoring of the electric fluid machinery; In step S3, the following steps are specifically included: S301, define a similarity operator, and the similarity calculation rule of two 2D model state tensors is as follows: in, Denotes a similarity operator whose result is a scalar, with the subscript X(i a ,L)、X(i b , L) represent i a 、i b The 2D model state tensor at the moment, s(r k (i a ,L),r k (i a ,L)) represents X(i a ,L)’s k-th row vector r k (i a ,L) and X(i b ,L)’s k-th row vector r k (i b ,L) similarity, β k Indicates the weight of the kth measurement point data in calculating similarity; S302, search for a′ ,L) the most similar 2D model state tensor X(i θ(a′) ,L), where the subscript Let similarity S303, yes Repeat step S302 to obtain the subscript set Similarity Set Set the similarity threshold γ e , we get the value greater than the threshold value γ in Γ e The subscript set corresponding to the elements of And γ(a′)>γ e }, similar 2D model state tensors are eliminated, and the set of retained 2D model state tensors is {X(i c ,L)|c∈Θ r }, where Θ r =Θ e ∪{1} represents the subscript set corresponding to the time index of the retained 2D model state tensor, and the subscript c represents the set Θ r Elements in S304, the set {X(i c ,L)|c∈Θ r The number of two-dimensional model state tensors in} is m2∈ The 2D model state tensors are stacked to construct a 3D model state tensor as follows: in, represents the 3D model state tensor, Represents the stacking operation of the 2D model state tensor, Represents the set of all three-dimensional real tensors of shape m2×n×L; S305. Define index and will Defined as a similarity matrix of m2×m2, the elements of the pth row and qth column of the similarity matrix satisfy: in, The result is a matrix similarity operator, T p With T q They respectively represent the p-th and q-th slices of the 3D model state tensor T, that is, the 2D model state tensor.

2. The method according to claim 1, characterized in that In step S1, the following steps are specifically included: S101. During the operation of the electric fluid machinery, transient operating condition data are collected, including parameter data under start-up, shutdown and variable load conditions, wherein the parameters include electrical parameters, mechanical parameters and temperature parameters. The electrical parameters include motor current, the mechanical parameters include bearing vibration and motor speed, and the temperature parameters include motor temperature and bearing temperature. S102, labeling the data collected in step S101 with a safety status label, wherein the safety status label includes "safe" and "failure risk"; S103, preprocessing the data annotated in step S102, retaining the data with the safety status label "safe", and normalizing the data through the standard score normalization method; using the preprocessed data as a model database of the monitoring method for building the model.

3. The method according to claim 1, characterized in that In step S2, the following steps are specifically included: S201, extracting data at a certain moment from the model database, arranging the extracted data and defining it as a 1-dimensional model state tensor as follows: in, Represents the time index, Indicates the number of measurement points, Represents the 1-dimensional model state tensor at time i, Indicates the index number of the measuring point, x k (i) represents the normalized data of the kth measurement point at time i, represents the set of positive integers, represents the set of all n-dimensional real vectors; S202, apply a time window operation to the data, repeat step S201 at consecutive moments, obtain multiple 1D model state tensors and incorporate them into the time window, and synthesize the 2D model state tensor as follows: X(i,L)=[x(i-L+1),x(i-L+2)…x(i)] in, Indicates the number of times step S201 is repeated, represents the 2D model state tensor corresponding to the time window, represents the set of all real matrices with n rows and L columns; S203, sliding on the time axis to execute step S202, and calculating the standard deviation mean of the data in the time window, and adaptively selecting the size of the sliding step according to the standard deviation mean, the standard deviation mean is defined as follows: in, represents the standard deviation mean of the data in the time window from time i-L+1 to time i, σ(·) represents the standard deviation function, and r k (i,L) represents the kth row vector of the 2D model state tensor X(i,L); the 2D model state tensors obtained by the adaptive step sliding time window operation are respectively expressed as where m1∈N * Represents the number of resulting 2D model state tensors.

4. The method according to claim 1, characterized in that: In step S4, the following steps are specifically included: S401, collect and normalize the transient operating condition data of the electric fluid machinery used for monitoring at a certain moment in the monitoring period, and obtain a 1-dimensional monitoring state tensor: in, Indicates the monitoring time index, Represents the 1-dimensional monitoring state tensor at time j, y k (j) represents the normalized data of the kth measuring point at time j; S402, using the time window operation, repeating step S401 at L consecutive monitoring moments in the monitoring period, obtaining L 1-dimensional monitoring state tensors and incorporating them into the time window, synthesizing a 2-dimensional monitoring state tensor as the input of the model: Y(j,L)=[y(j-L+1),y(j-L+2)...y(j)] in, Represents the 2D monitoring state tensor corresponding to the time window from time j-L+1 to time j; S403, calculate the similarity between the model input Y(j,L) and all slices in the 3D model state tensor T, and form a similarity vector Said The result is a vector similarity operator, and the p-th element of the similarity vector satisfies: S404, through similarity vector The original weight vector is calculated as follows: in, represents the original weight vector obtained by the model, Represents the similarity matrix The inverse matrix of ; normalize the original weight vector so that the sum of its elements is 1, and the normalized weight vector is as follows: Among them, w norm Represents the normalized weight vector obtained by the model, f norm (·) represents the normalization function, w p represents the pth element of w; S405, the 3D model state tensor T and the normalized weight vector w norm Performing the first modal multiplication, the output of the model is as follows: in, represents the reconstructed output of the model, T×1w norm Represents the 3D model state tensor T and vector w norm Multiply along the first mode.

5. The method according to claim 4, characterized in that In step S5, the following steps are specifically included: S501, calculate model input Y(j,L) and model output Y out (j, L), extract the deviation data of the latest monitoring time j and normalize it to obtain the monitoring deviation vector: in, represents the monitoring deviation vector of n measuring point data at time j, e k (j) represents the monitoring deviation value of the kth measuring point data; S502: Setting monitoring and early warning thresholds for n measuring points, including high early warning thresholds Low warning threshold When an element in the monitoring deviation vector exceeds the warning threshold, the monitoring warning is triggered, otherwise it is considered that the data at the j monitoring time meets the conditions of normal working conditions; S503, repeating steps S4, S501, and S502 for the remaining monitoring moments in the monitoring period, obtaining monitoring results for the remaining monitoring moments, and realizing health status monitoring of the electric fluid machinery.

Images