A bearing slip monitoring method based on knowledge graph and distributed transfer learning

Abstract

The present invention discloses a bearing slip monitoring method based on knowledge graph and distributed transfer learning. First, typical slip tasks are divided according to expert experience and a bearing slip knowledge graph is constructed using Neo4j software; second, bearing vibration acceleration data, cage and inner ring speed data under different typical tasks are collected, and the obtained slip rate is divided into levels; then, a distributed transfer learning model is trained according to the collected bearing vibration and slip rate data under different tasks, and a slip diagnosis model set is constructed; according to the collected vibration data corresponding to known working parameters, the slip diagnosis model under the corresponding typical work is selected to monitor the bearing slip rate; finally, for the vibration data corresponding to the unknown working parameters, the constructed knowledge graph and the vibration data of the known working parameters are matched to obtain the cause of the bearing slip. The present invention makes full use of the bearing working information and vibration acceleration data to monitor the slip state, and has high practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing slip diagnosis, and mainly to a bearing slip monitoring method based on knowledge graph and distributed transfer learning. Background Art

[0002] Bearing slippage is a common phenomenon in rotating machinery, especially in bearings running in complex and harsh environments. According to statistics, bearing slippage failures account for about 37% of the total number of failures. Bearing slippage can cause equipment vibration, fatigue wear, motion instability and significantly reduce the service life of the bearing. In severe cases, it may cause equipment shutdown. Therefore, it is very necessary to monitor bearing slippage.

[0003] In high-end equipment, how to accurately monitor bearing slip under the conditions of limited sensor installation space and lubricating oil pollution is a very challenging problem. Related research on bearing slip monitoring is usually divided into two categories: non-destructive monitoring and destructive monitoring. The main strategies adopted by non-destructive monitoring methods include those based on optical sensors, acoustic signals, and electromagnetic effect signals. The relevant strategies are difficult to adapt to the harsh working environment of high-end bearings; the main strategies adopted by destructive monitoring methods include those based on bearing material structure modification and embedded chips. The relevant strategies require improvements to the bearing structure, which is not suitable for the non-destructive monitoring needs of high-end bearings and may reduce the bearing operating performance. In view of the shortcomings of existing bearing slip monitoring strategies, monitoring using vibration signals is the most feasible and practical method.

[0004] Research shows that the bearing slip rate is closely related to operating parameters, structural parameters, assembly parameters, foreign matter intrusion, and rotor characteristics. Therefore, using only bearing vibration acceleration data for bearing slip monitoring is difficult to meet the needs of complex and variable operating conditions. The query, matching, and reasoning functions based on the knowledge graph fully explore the relationship between operating parameters, structural parameters, assembly parameters, foreign matter intrusion, rotor characteristics, and bearing slip rate, and build a bearing slip diagnosis model under different typical operating conditions based on distributed transfer learning to achieve bearing slip monitoring under complex and variable operating conditions.

[0005] At the same time, determining the cause of bearing slip under unknown working conditions based only on vibration signals has high engineering requirements. Through data similarity measurement and knowledge graph query, accurate analysis of the cause of bearing slip can be achieved. Commonly used data similarity measurement methods include Euclidean distance, Mahalanobis distance, correlation coefficient, entropy, divergence, maximum mean difference and domain adaptation algorithm. Among them, the domain adaptation algorithm can comprehensively analyze the similarity between data from different scales. Summary of the invention

[0006] Purpose of the invention: Based on the problems existing in the above-mentioned background technology, the present invention provides a bearing slip monitoring method based on knowledge graph and distributed transfer learning, which can realize accurate measurement and monitoring of bearing slip rate based on flexible rotor deflection and bearing vibration acceleration data.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A bearing slip monitoring method based on knowledge graph and distributed transfer learning includes the following steps:

[0009] Step S1, collecting and measuring working parameters, wherein the measuring working parameters include measuring rotor characteristic parameters R is the type of rotor characteristics, r is the number of types of rotor characteristics; measurement assembly parameters A is the type of assembly parameter, a is the number of types of assembly parameters; Measuring the characteristics of intrusion foreign matter F is the type of intruding foreign body characteristics, f is the number of types of intruding foreign body characteristics; measure bearing structural parameters S is the type of bearing structure, s is the number of bearing structure types; measure bearing operating parameters W is the type of working condition, w is the number of working condition types; the measured working parameters are divided into levels, and the typical slipping conditions are further divided according to expert experience D is the type of typical slip condition, and d is the number of types of typical slip condition;

[0010] Step S2: Collect different typical slipping tasks Bearing vibration acceleration data A under acc , cage speed data S ca And the inner ring speed data S in , and divide the obtained slip rate data X into slip levels T is the type of slip level, t is the number of slip level types; combined with the typical slip work obtained in step S1 And the classification of slip rate Use Neo4j software to build a knowledge graph of bearing slip;

[0011] Step S3: different typical slipping operations collected in step S2 Bearing vibration acceleration data A under acc And the slip rate data K are used to train the distributed transfer learning model and build a slip diagnosis model set

[0012] Step S4: Calculate the vibration acceleration data A corresponding to the measured working parameters collected in step S1. y Select the typical slipping work D in step S3 iThe skid diagnosis model M under (i=1,2,…,d) i (i=1,2,...,m) monitor bearing slip rate;

[0013] Step S5: collect the vibration acceleration data A corresponding to the unknown measurement working parameters. w , according to the vibration acceleration data A of step S2 and step S4 acc The constructed knowledge graph performs slip matching based on the subdomain adaptive algorithm to obtain the bearing slip work and analyze the causes.

[0014] Preferably, the implementation process of step S1 is:

[0015] Step S1.1, collect and measure working parameters, wherein the measuring working parameters include measuring rotor characteristic parameters Including eccentricity R1 and rotor resonance frequency R2; measuring assembly parameters Including assembly interference A1 and assembly preload A2; measuring the characteristics of intruding foreign matter Including foreign body material F1, foreign body structure F2 and foreign body quantity F3; measure bearing structure parameters Including gap ratio S1, guide mode S2, friction coefficient S3, number of rollers S4, surface roughness S5, axial clearance S6 and radial clearance S7; measure bearing operating parameters Including inner ring acceleration W1, axial load W2, radial load W3, lubricating oil viscosity W4, lubricating oil temperature W5, lubricating oil flow W6 and inner ring speed W7;

[0016] Step S1.2: classify the measurement working parameters collected in step S1.1 into different levels, specifically including:

[0017] The eccentricity R1 is divided into three levels The rotor resonance frequency R2 is divided into two levels The assembly interference A1 is divided into three levels The assembly preload force A2 is divided into three levels The foreign material F1 is divided into two levels {F1 1 ,F1 2}; Divide the foreign body structure F2 into two levels The number of foreign matter F3 is divided into two levels The gap ratio S1 is divided into three levels The boot mode S2 is divided into two levels The friction coefficient S3 is divided into three levels The number of rollers S4 is divided into three levels Surface roughness S5 is divided into three levels The axial clearance S6 is divided into three levels The radial clearance S7 is divided into three levels The inner ring acceleration W1 is divided into three levels {W1 1 ,W1 2 ,W1 3}; Divide the axial load W2 into three levels The radial load W3 is divided into three levels Lubricating oil viscosity W4 is divided into three grades The lubricating oil temperature W5 is divided into three levels The lubricating oil flow W6 is divided into three levels The inner ring speed W7 is divided into three levels Among them, the parameter level set ε that is most likely to cause an increase in slip rate max Satisfying formula (1):

[0018]

[0019] The parameter level set ε that leads to medium increase in slip rate mid Satisfying formula (2):

[0020]

[0021] The parameter level set ε that is less likely to cause the slip rate to increase min Satisfying formula (3):

[0022]

[0023] Step S1.3: Combine the levels divided in step S1.2 and divide the typical slipping tasks into D1, D2, D3, D4, and D5 according to expert experience; wherein the typical slipping task D1 satisfies formula (4):

[0024]

[0025] Among them, the typical slipping work D1 includes the following tasks: n1 is the number of sub-tasks included in the typical slipping task D1;

[0026] Typical slipping operation D2 satisfies formula (5):

[0027]

[0028] The typical slipping work D2 includes the following tasks: n2 is the number of sub-tasks included in the typical slipping task D2;

[0029] Typical slipping operation D3 satisfies formula (6):

[0030]

[0031] The typical slipping work D3 includes the following tasks: n3 is the number of sub-tasks included in the typical slipping task D3;

[0032] Typical slipping operation D4 satisfies formula (7):

[0033]

[0034] Typical slippage work D4 includes the following tasks: n4 is the number of sub-tasks included in the typical slipping task D4;

[0035] Typical slipping operation D5 satisfies formula (8):

[0036]

[0037] Typical slippage work D5 includes the following tasks: n5 is the number of sub-tasks included in the typical slipping task D5;

[0038] Among them, the value of parameter i is: 1, 2, 3.

[0039] Preferably, the implementation process of step S2 is:

[0040] Step 2.1: Collect different typical slipping tasks Bearing vibration acceleration data A under acc , cage speed data S ca And the inner ring speed data S in ; Among them, the bearing vibration acceleration data A acc Satisfying formula (9):

[0041]

[0042] Among them, A acc1 The bearing vibration acceleration data corresponding to the typical slipping operation D1 is: are the bearing vibration acceleration data corresponding to the n1 sub-tasks contained in the typical slipping task D1; A acc2 The bearing vibration acceleration data corresponding to the typical slipping operation D2 is: are the bearing vibration acceleration data corresponding to the n2 sub-tasks contained in the typical slipping task D2; A acc3 The bearing vibration acceleration data corresponding to the typical slipping operation D3 is: They are the bearing vibration acceleration data corresponding to the n3 sub-tasks contained in the typical slipping task D3; A acc4The bearing vibration acceleration data corresponding to the typical slipping operation D4 is shown in Figure 2. They are the bearing vibration acceleration data corresponding to the n4 sub-tasks contained in the typical slipping task D4; A acc5 The bearing vibration acceleration data corresponding to the typical slipping operation D5 is: They are the bearing vibration acceleration data corresponding to the n5 sub-tasks included in the typical slipping task D5;

[0043] Cage speed data S ca Satisfying formula (10):

[0044]

[0045] Among them, S ca1 The cage speed data corresponding to the typical slipping operation D1 is shown below. are the cage speed data corresponding to the n1 sub-tasks included in the typical slipping task D1; S ca2 The cage speed data corresponding to the typical slipping operation D2 is: are the cage speed data corresponding to the n2 sub-tasks included in the typical slipping task D2; S ca3 The cage speed data corresponding to the typical slipping operation D3 is shown below. are the cage speed data corresponding to the n3 sub-tasks included in the typical slipping task D3; S ca4 The cage speed data corresponding to the typical slipping operation D4 is shown below. are the cage speed data corresponding to the n4 sub-tasks included in the typical slipping task D4; S ca5 The cage speed data corresponding to the typical slipping operation D5 is shown below. They are the cage speed data corresponding to the n5 sub-tasks included in the typical slipping task D5;

[0046] Inner ring speed data S in Satisfying formula (11):

[0047]

[0048] Among them, S in1 The inner ring speed data corresponding to the typical slipping operation D1 is shown below. are the inner ring speed data corresponding to the n1 sub-tasks included in the typical slipping task D1; S in2 The inner ring speed data corresponding to the typical slipping operation D2 is: They are the inner ring speed data corresponding to the n2 sub-tasks included in the typical slipping task D2; S in3 The inner ring speed data corresponding to the typical slipping operation D3 is shown below. They are the inner ring speed data corresponding to the n3 sub-tasks included in the typical slipping task D3; S in4 The inner ring speed data corresponding to the typical slipping operation D4 is shown below. They are the inner ring speed data corresponding to the n4 sub-tasks included in the typical slipping task D4; S in5 The inner ring speed data corresponding to the typical slipping operation D5 is shown below. They are the inner ring speed data corresponding to the n5 sub-tasks included in the typical slipping task D5;

[0049] The calculation formula of bearing slip rate satisfies formula (12):

[0050]

[0051] Among them, S ca is the bearing cage speed data; X is the bearing cage slip rate; S cm is the theoretical speed data of the bearing cage, satisfying:

[0052]

[0053] Among them, R w is the bearing roller radius; R m is the bearing pitch radius; S in is the inner ring speed data; the bearing slip rate X satisfies formula (14):

[0054]

[0055] Among them, X1 is the bearing slip rate corresponding to the typical slipping work D1, are the bearing slip rates corresponding to the n1 sub-tasks contained in the typical slipping task D1; X2 is the bearing slip rate corresponding to the typical slipping task D2, are the bearing slip rates corresponding to the n2 sub-tasks included in the typical slipping task D2; X3 is the bearing slip rate corresponding to the typical slipping task D3, are the bearing slip rates corresponding to the n3 sub-tasks included in the typical slipping task D3; X4 is the bearing slip rate corresponding to the typical slipping task D4, are the bearing slip rates corresponding to the n4 sub-tasks included in the typical slipping task D4; X5 is the bearing slip rate corresponding to the typical slipping task D5, They are the bearing slip rates corresponding to the n5 sub-tasks included in the typical slip task D5;

[0056] According to the typical slipping work D1, D2, D3, D4, and D5 divided in step S1.3, the bearing slip levels in formula (14) are respectively divided into severe slip level T1, relatively severe slip level T2, medium slip level T3, relatively slight slip level T4, and slight slip level T5;

[0057] Step 2.2: respectively calculate the measured working parameters collected in step S1.1 and the typical slip working parameters obtained in step S1.3. and the slip rate grade classified in step S2.1 Neo4j software is used as a node to build a bearing slip knowledge graph.

[0058] Preferably, the implementation process of step S3 is:

[0059] Step S3.1, divide the typical slipping work, vibration acceleration data and bearing slip rate data collected in step S2.1 into five groups, specifically including: the vibration acceleration data A corresponding to the typical slipping work D1 acc1 The vibration acceleration data A corresponding to the bearing slip rate X1 and typical slip work D2 acc2 The vibration acceleration data A corresponding to the bearing slip rate X2 and typical slip work D3 acc3 The vibration acceleration data A corresponding to the bearing slip rate X3 and typical slip work D4 acc4 The vibration acceleration data A corresponding to the bearing slip rate X4 and typical slip work D5 acc5 and bearing slip rate X5;

[0060] Step S3.2: Based on the transfer learning algorithm and the five groups of data divided according to step S3.1, the distributed transfer learning models M1, M2, M3, M4, and M5 are trained respectively, and a skid diagnosis model set is constructed. Specifically include:

[0061] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M1 separately The parameters of the local model are

[0062] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M2 separately The parameters of the local model are

[0063] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M3 separately The local model parameters are

[0064] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M4 separately The local model parameters are

[0065] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M5 separately The local model parameters are

[0066] Step S3.3: Integrate the local models established in step S3.2 respectively, specifically:

[0067] Finding local model parameters The average is Use it as the parameter of the distributed transfer learning model M1; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M2; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M3; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M4; obtain the local model parameters The average is It is used as a parameter of the distributed transfer learning model M5.

[0068] Preferably, the implementation process of step S4 is:

[0069] The typical slipping work established in step S1.3 is searched according to the collected measurement working parameters, and the slip rate is monitored using the distributed transfer learning model established in step S3.3 corresponding to the typical slipping work.

[0070] Preferably, the implementation process of step S5 is:

[0071] Step S5.1: Perform slippage matching based on subdomain adaptive algorithm:

[0072] The vibration acceleration data A collected in step S2 acc As the source domain sample, the vibration acceleration data A corresponding to the unknown measurement working parameters is collected w As the target domain sample; the loss function of the subdomain adaptive algorithm satisfies:

[0073]

[0074] Among them, i and j are parameters; N s is the number of samples in the source domain; N t is the number of samples in the target domain; C is the number of subdomain categories; and are the weights of the sample belonging to category c; and are activation factors respectively; κ(·) is the multi-core local maximum mean difference and satisfies:

[0075]

[0076] Among them, β u is the weight of different kernel functions; m is the number of kernel functions, u is the parameter; k u is the u-th kernel function; k is the multi-kernel function;

[0077] Step S5.2, obtain the bearing slippage work and analyze the cause:

[0078] When the loss function (15) of the subdomain adaptive algorithm proposed in step S5.1 obtains the minimum value, the work of the source domain sample is the work of the target domain sample; the work of the target domain sample obtained by matching is combined to search the knowledge graph established in step S2.2, and the parameter level set ε in the work of the target domain sample that is most likely to cause an increase in the slip rate as described in step S1.2 is analyzed. max parameters and regard them as the cause of bearing slip.

[0079] Beneficial effects:

[0080] (1) Since the bearing slip rate is closely related to operating parameters, structural parameters, assembly parameters, foreign matter intrusion and rotor characteristics, it is difficult to use a single determined deep learning network to accurately detect the bearing slip rate under complex and changeable operating conditions. The present invention sorts out the relationship between the bearing slip rate and operating parameters, structural parameters, assembly parameters, foreign matter intrusion and rotor characteristics based on the knowledge graph, and realizes accurate detection of the bearing slip rate based on distributed transfer learning; the algorithm is designed closely in combination with the specific ideas of engineering implementation and has high engineering application value;

[0081] (2) By making full use of the query, matching and reasoning functions of the knowledge graph, the analysis of the causes of bearing slip based only on bearing vibration acceleration data was realized based on the subdomain adaptive algorithm and the knowledge graph. At the same time, the many tasks that affect bearing slip were summarized into five typical tasks, which can realize a more convenient, fast and accurate analysis of the causes of bearing slip. The relevant technology has a high promotion and application value.

[0082] (3) Based on the distributed transfer learning algorithm, a set of skid diagnosis models were established for different typical tasks. Through the integration of local models, the effect was better than the simple combination of single local diagnosis models.

[0083] (4) The present invention combines expert experience and engineering practice to propose a parameter classification standard that affects the size of the bearing slip rate. This is the first standard clearly proposed in the relevant field, and provides guidance for actual bearing slip analysis and the formulation of bearing slip suppression measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A flow chart of a bearing slip monitoring method based on knowledge graph and distributed transfer learning provided by the present invention;

[0085] Figure 2 A structural diagram of a bearing skidding test bench for a certain type of flexible rotor system provided by the present invention;

[0086] Figure 3 This is a knowledge graph of bearing slippage in an embodiment of the present invention;

[0087] Figure 4 is a bearing slip curve collected in an embodiment of the present invention;

[0088] Figure 5 The structure of the bearing slippage distributed transfer learning diagnosis model in the embodiment of the present invention;

[0089] Figure 6 This is a flow chart of distributed transfer learning training in an embodiment of the present invention;

[0090] Figure 7 It is a local transfer learning model in an embodiment of the present invention;

[0091] Figure 8 This is the bearing slip monitoring result in the embodiment of the present invention. DETAILED DESCRIPTION

[0092] The present invention will be further described below in conjunction with the accompanying drawings.

[0093] See also Figure 1-Figure 8 This embodiment provides a bearing slip monitoring method based on knowledge graph and distributed transfer learning. The process of the method is as follows: Figure 1 As shown in FIG. 1 , this embodiment uses a certain type of flexible rotor system bearing slip test bench as an example to verify the algorithm performance. The structure of a certain type of flexible rotor system bearing slip test bench is as follows: Figure 2 The strategy specifically includes the following steps:

[0094] Step S1, measuring working parameters; including measuring rotor characteristic parameters R is the type of rotor characteristics, r is the number of types of rotor characteristics; measurement assembly parameters A is the type of assembly parameter, a is the number of types of assembly parameters; Measuring the characteristics of intrusion foreign matter F is the type of intruding foreign body characteristics, f is the number of types of intruding foreign body characteristics; measure bearing structural parameters S is the type of bearing structure, s is the number of types of bearing structure; measurement parameters W is the type of working condition, w is the number of working condition types; the measured parameters are divided into grades, and typical slipping work is further divided according to expert experience. D is the type of typical slipping work, and d is the number of types of typical slipping conditions;

[0095] Specifically, in this embodiment, step S1 includes:

[0096] Step S1.1: Measure rotor characteristics Including eccentricity R1 and rotor resonance frequency R2; measuring assembly parameters Including assembly interference A1 and assembly preload A2; measuring the characteristics of intruding foreign matter Including foreign body material F1, foreign body structure F2 and foreign body quantity F3; measure bearing structure parameters Including gap ratio S1, guide mode S2, friction coefficient S3, number of rollers S4, surface roughness S5, axial clearance S6 and radial clearance S7; measure bearing operating parameters Including inner ring acceleration W1, axial load W2, radial load W3, lubricating oil viscosity W4, lubricating oil temperature W5, lubricating oil flow W6 and inner ring speed W7;

[0097] Step S1.2: Classify the parameters measured in step S1.1 into different levels, specifically including:

[0098] The eccentricity R1 is divided into three levels The rotor resonance frequency R2 is divided into two levels The assembly interference A1 is divided into three levels The assembly preload force A2 is divided into three levels The foreign material F1 is divided into two levels {F1 1 ,F1 2}; Divide the foreign body structure F2 into two levels The number of foreign matter F3 is divided into two levels The gap ratio S1 is divided into three levels The boot mode S2 is divided into two levels The friction coefficient S3 is divided into three levels The number of rollers S4 is divided into three levels Surface roughness S5 is divided into three levels The axial clearance S6 is divided into three levels The radial clearance S7 is divided into three levels The inner ring acceleration W1 is divided into three levels {W1 1 ,W1 2 ,W1 3}; Divide the axial load W2 into three levels The radial load W3 is divided into three levels Lubricating oil viscosity W4 is divided into three grades The lubricating oil temperature W5 is divided into three levels The lubricating oil flow W6 is divided into three levels The inner ring speed W7 is divided into three levels Among them, the parameter level set ε that is most likely to cause an increase in slip rate max Satisfying formula (1):

[0099]

[0100] Medium parameter level that leads to increased slip rate ε mid The set satisfies formula (2):

[0101]

[0102] The parameter level set ε that is less likely to cause the slip rate to increase min Satisfying formula (3):

[0103]

[0104] In this embodiment, the parameter level classification results are shown in Table 1:

[0105] Table 1 Parameter level classification

[0106]

[0107]

[0108] Step S1.3: Combine the parameter levels divided in step S1.2 and divide the typical slipping work into D1, D2, D3, D4, and D5 according to expert experience; wherein the typical slipping work D1 satisfies formula (4):

[0109]

[0110] Among them, the typical slipping work D1 includes the following tasks: n1 is the number of sub-tasks included in the typical slipping task D1;

[0111] Typical slipping operation D2 satisfies formula (5):

[0112]

[0113] The typical slipping work D2 includes the following tasks: n2 is the number of sub-tasks included in the typical slipping task D2;

[0114] Typical slipping operation D3 satisfies formula (6):

[0115]

[0116] The typical slipping work D3 includes the following tasks: n3 is the number of sub-tasks included in the typical slipping task D3;

[0117] Typical slipping operation D4 satisfies formula (7):

[0118]

[0119] Typical slippage work D4 includes the following tasks: n4 is the number of sub-tasks included in the typical slipping task D4;

[0120] Typical slipping operation D5 satisfies formula (8):

[0121]

[0122] Typical slippage work D5 includes the following tasks: n5 is the number of sub-tasks included in the typical slipping task D5;

[0123] Step S2: Collect different typical slipping tasks Bearing vibration acceleration data A underacc , cage speed data S ca And the inner ring speed data S in , and divide the obtained slip rate data X into slip levels t is the number of types of slip levels; combined with the typical slip work obtained in step S1 And the classification of slip rate Use Neo4j software to build a bearing slip knowledge graph; the bearing slip knowledge graph is as follows Figure 3 shown.

[0124] Specifically, in this embodiment, step S2 includes:

[0125] Step 2.1: Collect different typical slipping tasks Bearing vibration acceleration data under Bearing vibration acceleration data A acc , cage speed data S ca And the inner ring speed data S in ; Among them, the bearing vibration acceleration data A acc Satisfying formula (9):

[0126]

[0127] Cage speed data S ca Satisfying formula (10):

[0128]

[0129] Inner ring speed data S in Satisfying formula (11):

[0130]

[0131] The calculation formula of bearing slip rate satisfies formula (12):

[0132]

[0133] Among them, S ca is the actual speed of the bearing cage; X is the slip rate of the bearing cage; S cm is the theoretical speed of the bearing cage, satisfying:

[0134]

[0135] Among them, R w =5mm is the bearing roller radius; R m =60mm is the bearing pitch radius; S in is the speed of the inner ring of the bearing; the bearing slip rate X satisfies formula (14):

[0136]

[0137] According to the typical slipping work D1, D2, D3, D4, and D5 divided in step S1.3, the bearing slip levels in formula (14) are respectively divided into severe slip level T1, relatively severe slip level T2, medium slip level T3, relatively slight slip level T4, and slight slip level T5; the collected bearing slip curve is as follows: Figure 4 In this embodiment, the results of the bearing slip level classification are shown in Table 2:

[0138] Table 2 Bearing slip level classification results

[0139] Slip level <![CDATA[Severe T1]]> <![CDATA[Severe T2]]> <![CDATA[Medium T3]]> <![CDATA[Relatively mild T4]]> <![CDATA[Minor T5]]> Slip rate More than 50% 20％-50％ 10％-20％ 2％-10％ Below 2%

[0140] Step 2.2: Combine the parameters measured in step S1.1 and the typical slipping work obtained in step S1.3 and the slip rate grade classified in step S2.1 They are respectively used as nodes to construct the bearing slip knowledge graph using Neo4j software.

[0141] Step S3: different typical slipping operations collected in step S2 Bearing vibration acceleration data A under acc And the slip rate data K are used to train the distributed transfer learning model and build a slip diagnosis model set

[0142] Specifically, in this embodiment, step S3 includes:

[0143] Step S3.1, divide the data collected in step S2.1 into five groups, including: vibration acceleration data A corresponding to the typical slipping work D1 acc1 The vibration acceleration data A corresponding to the bearing slip rate data X1 and the typical slip work D2 acc2 The vibration acceleration data A corresponding to the bearing slip rate data X2 and the typical slip work D3 acc3 The vibration acceleration data A corresponding to the bearing slip rate data X3 and the typical slip work D4 acc4 The vibration acceleration data A corresponding to the bearing slip rate data X4 and the typical slip work D5 acc5 and bearing slip rate data X5;

[0144] Step S3.2: Based on the transfer learning algorithm and the five groups of data divided according to step S3.1, the distributed transfer learning models M1, M2, M3, M4, and M5 are trained respectively, and a skid diagnosis model set is constructed. The structure of the bearing slip distributed transfer learning diagnosis model is as follows Figure 5 As shown in Figure 2, the distributed transfer learning training process is as follows: Figure 6 Specifically including:

[0145] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip data Train the local models of model M1 separately The model parameters are The local model structure is as follows Figure 7 shown.

[0146] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip data Train the local models of model M2 separately The model parameters are

[0147] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip data Train the local models of model M3 separately The model parameters are

[0148] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip data Train the local models of model M4 separately The model parameters are

[0149] Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip data Train the local models of model M5 separately The model parameters are

[0150] Step S3.3: Integrate the local models established in step S3.2 respectively, specifically:

[0151] Finding local model parameters The average is Take it as the parameter of model M1; obtain the local model parameters The average is Take it as the parameter of model M2; obtain the local model parameters The average is Take it as the parameter of model M3; obtain the local model parameters The average is Take it as the parameter of model M4; obtain the local model parameters The average is Use it as the parameter of model M5;

[0152] Step S4: According to the vibration acceleration data A corresponding to the known working parameters collected y Select the typical slipping work D in step S3 i The skid diagnosis model M under (i=1,2,...,d) i (i=1,2,…,m) monitor bearing slip rate;

[0153] Specifically, in this embodiment, step S4 includes:

[0154] According to the data of the known working parameters collected, the typical slipping work established in step S1.3 is found, and the distributed transfer learning model established in step S3.3 corresponding to the typical slipping work is used to monitor the slip rate; the bearing slip monitoring results are as follows: Figure 8 shown.

[0155] Step S5: collect the vibration acceleration data A corresponding to the unknown working parameters. w , according to the vibration acceleration data A of the known working parameters collected in step S2 acc The constructed knowledge graph performs slip matching based on the subdomain adaptive algorithm, and then obtains the bearing slip work and analyzes the causes;

[0156] Specifically, in this embodiment, step S5 includes:

[0157] Step S5.1: Slip matching based on subdomain adaptive algorithm

[0158] The vibration acceleration data A of the known working parameters collected in step S2 acc As the source domain sample, the vibration acceleration data A corresponding to the unknown working parameters collected is w As the target domain sample; the loss function of the subdomain adaptive algorithm satisfies:

[0159]

[0160] Among them, i and j are parameters; N s is the number of samples in the source domain; N t is the number of samples in the target domain; C is the number of subdomain categories; and are the weights of the sample belonging to category c; and are activation factors respectively; κ(·) is the multi-core local maximum mean difference and satisfies:

[0161]

[0162] Among them, β u is the weight of different kernel functions; m is the number of kernel functions, u is the parameter; k u is the u-th kernel function; k is the multi-kernel function;

[0163] Step S5.2: Obtain bearing slippage and analyze the cause

[0164] When the loss function (15) of the subdomain adaptive algorithm proposed in step S5.1 reaches the minimum value, the work of the source domain sample is the work of the target domain sample; the knowledge graph established in step S2.2 is searched in combination with the work of the target domain sample obtained by matching.

[0165] The parameter level set ε that is most likely to cause the slip rate to increase in step S1.2 is analyzed in the target domain sample. max parameters and regard them as the cause of bearing slip.

[0166] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A bearing slip monitoring method based on knowledge graph and distributed transfer learning, characterized in that: The following steps are involved: Step S1, collecting and measuring working parameters, wherein the measuring working parameters include measuring rotor characteristic parameters R is the type of rotor characteristics, r is the number of types of rotor characteristics; measurement assembly parameters A is the type of assembly parameter, a is the number of types of assembly parameters; Measuring the characteristics of intrusion foreign matter F is the type of intruding foreign body characteristics, f is the number of types of intruding foreign body characteristics; measure bearing structural parameters S is the type of bearing structure, s is the number of bearing structure types; measure bearing operating parameters W is the type of working condition, w is the number of working condition types; the measured working parameters are divided into levels, and the typical slipping conditions are further divided according to expert experience D is the type of typical slip condition, and d is the number of types of typical slip condition; Step S2: Collect different typical slipping tasks Bearing vibration acceleration data A under acc , cage speed data S ca And the inner ring speed data S in , and divide the obtained slip rate data X into slip levels T is the type of slip level, t is the number of slip level types; combined with the typical slip work obtained in step S1 And the classification of slip rate Use Neo4j software to build a knowledge graph of bearing slip; Step S3: different typical slipping operations collected in step S2 Bearing vibration acceleration data A under acc And the slip rate data K are used to train the distributed transfer learning model and build a slip diagnosis model set Step S4: Calculate the vibration acceleration data A corresponding to the measured working parameters collected in step S1. y Select the typical slipping work D in step S3 i The skid diagnosis model M under (i=1,2,...,d) i (i=1,2,...,m) monitor bearing slip rate; Step S5: collect the vibration acceleration data A corresponding to the unknown measurement working parameters. w , according to the vibration acceleration data A of step S2 and step S4 acc The constructed knowledge graph performs slip matching based on the subdomain adaptive algorithm, and then obtains the bearing slip work and analyzes the causes; The implementation process of step S1 is: Step S1.1, collect and measure working parameters, wherein the measuring working parameters include measuring rotor characteristic parameters Including eccentricity R1 and rotor resonance frequency R2; measuring assembly parameters Including assembly interference A1 and assembly preload A2; measuring the characteristics of intruding foreign matter Including foreign body material F1, foreign body structure F2 and foreign body quantity F3; measure bearing structure parameters Including gap ratio S1, guide mode S2, friction coefficient S3, number of rollers S4, surface roughness S5, axial clearance S6 and radial clearance S7; measure bearing operating parameters Including inner ring acceleration W1, axial load W2, radial load W3, lubricating oil viscosity W4, lubricating oil temperature W5, lubricating oil flow W6 and inner ring speed W7; Step S1.2: classify the measurement working parameters collected in step S1.1 into different levels, specifically including: The eccentricity R1 is divided into three levels The rotor resonance frequency R2 is divided into two levels The assembly interference A1 is divided into three levels The assembly preload force A2 is divided into three levels Foreign material F1 is divided into two levels The foreign body structure F2 is divided into two levels The number of foreign matter F3 is divided into two levels The gap ratio S1 is divided into three levels The boot mode S2 is divided into two levels The friction coefficient S3 is divided into three levels The number of rollers S4 is divided into three levels Surface roughness S5 is divided into three levels The axial clearance S6 is divided into three levels The radial clearance S7 is divided into three levels The inner ring acceleration W1 is divided into three levels The axial load W2 is divided into three levels The radial load W3 is divided into three levels Lubricating oil viscosity W4 is divided into three grades The lubricating oil temperature W5 is divided into three levels The lubricating oil flow W6 is divided into three levels The inner ring speed W7 is divided into three levels Among them, the parameter level set ε that is most likely to cause an increase in slip rate max Satisfying formula (1): The parameter level set ε that leads to medium increase in slip rate mid Satisfying formula (2): The parameter level set ε that is less likely to cause the slip rate to increase min Satisfying formula (3): Step S1.3: Combine the levels divided in step S1.2 and divide the typical slipping tasks into D1, D2, D3, D4, and D5 according to expert experience; wherein the typical slipping task D1 satisfies formula (4): Among them, the typical slipping work D1 includes the following tasks: n1 is the number of sub-tasks included in the typical slipping task D1; Typical slipping operation D2 satisfies formula (5): The typical slipping work D2 includes the following tasks: n2 is the number of sub-tasks included in the typical slipping task D2; Typical slipping operation D3 satisfies formula (6): Among them, the typical slipping work D3 includes the following tasks: n3 is the number of sub-tasks included in the typical slipping task D3; Typical slipping operation D4 satisfies formula (7): Typical slippage work D4 includes the following tasks: n4 is the number of sub-tasks included in the typical slipping task D4; Typical slipping operation D5 satisfies formula (8): Typical slippage work D5 includes the following tasks: n5 is the number of sub-tasks included in the typical slipping task D5; Among them, the value of parameter i is: 1, 2, 3; The implementation process of step S5 is: Step S5.1: Perform slippage matching based on subdomain adaptive algorithm: The vibration acceleration data A collected in step S2 acc As the source domain sample, the vibration acceleration data A corresponding to the unknown measurement working parameters is collected w As the target domain sample; the loss function of the subdomain adaptive algorithm satisfies: Among them, i and j are parameters; N s is the number of samples in the source domain; N t is the number of samples in the target domain; C is the number of subdomain categories; and are the weights of the sample belonging to category c; and are activation factors respectively; κ(·) is the multi-core local maximum mean difference and satisfies: Among them, β u is the weight of different kernel functions; m is the number of kernel functions, u is the parameter; k u is the u-th kernel function; k is the multi-kernel function; Step S5.2, obtain the bearing slippage work and analyze the cause: When the loss function (15) of the subdomain adaptive algorithm proposed in step S5.1 obtains the minimum value, the work of the source domain sample is the work of the target domain sample; the work of the target domain sample obtained by matching is combined to search the knowledge graph established in step S2, and the parameter level set ε in the work of the target domain sample that is most likely to cause an increase in the slip rate as described in step S1.2 is analyzed. max parameters and regard them as the cause of bearing slip.

2. A bearing slip monitoring method based on knowledge graph and distributed transfer learning according to claim 1, characterized in that: The implementation process of step S2 is: Step 2.1: Collect different typical slipping tasks Bearing vibration acceleration data A under acc , cage speed data S ca And the inner ring speed data S in ; Among them, the bearing vibration acceleration data A acc Satisfying formula (9): Among them, A acc1 The bearing vibration acceleration data corresponding to the typical slipping operation D1 is: are the bearing vibration acceleration data corresponding to the n1 sub-tasks contained in the typical slipping task D1; A acc2 The bearing vibration acceleration data corresponding to the typical slipping operation D2 is: are the bearing vibration acceleration data corresponding to the n2 sub-tasks contained in the typical slipping task D2; A acc3 The bearing vibration acceleration data corresponding to the typical slipping operation D3 is: They are the bearing vibration acceleration data corresponding to the n3 sub-tasks contained in the typical slipping task D3; A acc4 The bearing vibration acceleration data corresponding to the typical slipping operation D4 is shown in Figure 2. They are the bearing vibration acceleration data corresponding to the n4 sub-tasks contained in the typical slipping task D4; A acc5 The bearing vibration acceleration data corresponding to the typical slipping operation D5 is: They are the bearing vibration acceleration data corresponding to the n5 sub-tasks included in the typical slipping task D5; Cage speed data S ca Satisfying formula (10): Among them, S ca1 The cage speed data corresponding to the typical slipping operation D1 is shown below. are the cage speed data corresponding to the n1 sub-tasks included in the typical slipping task D1; S ca2 The cage speed data corresponding to the typical slipping operation D2 is: are the cage speed data corresponding to the n2 sub-tasks included in the typical slipping task D2; S ca3 The cage speed data corresponding to the typical slipping operation D3 is shown below. are the cage speed data corresponding to the n3 sub-tasks included in the typical slipping task D3; S ca4 The cage speed data corresponding to the typical slipping operation D4 is shown below. are the cage speed data corresponding to the n4 sub-tasks included in the typical slipping task D4; S ca5 The cage speed data corresponding to the typical slipping operation D5 is shown below. They are the cage speed data corresponding to the n5 sub-tasks included in the typical slipping task D5; Inner ring speed data S in Satisfying formula (11): Among them, S in1 The inner ring speed data corresponding to the typical slipping operation D1 is shown below. are the inner ring speed data corresponding to the n1 sub-tasks included in the typical slipping task D1; S in2 The inner ring speed data corresponding to the typical slipping operation D2 is: They are the inner ring speed data corresponding to the n2 sub-tasks included in the typical slipping task D2; S in3 The inner ring speed data corresponding to the typical slipping operation D3 is shown below. They are the inner ring speed data corresponding to the n3 sub-tasks included in the typical slipping task D3; S in4 The inner ring speed data corresponding to the typical slipping operation D4 is shown below. They are the inner ring speed data corresponding to the n4 sub-tasks included in the typical slipping task D4; S in5 The inner ring speed data corresponding to the typical slipping operation D5 is shown below. They are the inner ring speed data corresponding to the n5 sub-tasks included in the typical slipping task D5; The calculation formula of bearing slip rate satisfies formula (12): Among them, S ca is the bearing cage speed data; X is the bearing cage slip rate; S cm is the theoretical speed data of the bearing cage, satisfying: Among them, R w is the bearing roller radius; R m is the bearing pitch radius; S in is the inner ring speed data; the bearing slip rate X satisfies formula (14): Among them, X1 is the bearing slip rate corresponding to the typical slipping work D1, are the bearing slip rates corresponding to the n1 sub-tasks contained in the typical slipping task D1; X2 is the bearing slip rate corresponding to the typical slipping task D2, are the bearing slip rates corresponding to the n2 sub-tasks included in the typical slipping task D2; X3 is the bearing slip rate corresponding to the typical slipping task D3, are the bearing slip rates corresponding to the n3 sub-tasks included in the typical slipping task D3; X4 is the bearing slip rate corresponding to the typical slipping task D4, are the bearing slip rates corresponding to the n4 sub-tasks included in the typical slipping task D4; X5 is the bearing slip rate corresponding to the typical slipping task D5, They are the bearing slip rates corresponding to the n5 sub-tasks included in the typical slip task D5; According to the typical slipping work D1, D2, D3, D4, and D5 divided in step S1.3, the bearing slip levels in formula (14) are respectively divided into severe slip level T1, relatively severe slip level T2, medium slip level T3, relatively slight slip level T4, and slight slip level T5; Step 2.2: respectively calculate the measured working parameters collected in step S1.1 and the typical slip working parameters obtained in step S1.

3. and the slip rate grade classified in step S2.1 Neo4j software is used as a node to build a knowledge graph of bearing slip.

3. A bearing slip monitoring method based on knowledge graph and distributed transfer learning according to claim 2, characterized in that: The implementation process of step S3 is: Step S3.1, divide the typical slipping work, vibration acceleration data and bearing slip rate data collected in step S2.1 into five groups, specifically including: the vibration acceleration data A corresponding to the typical slipping work D1 acc1 The vibration acceleration data A corresponding to the bearing slip rate X1 and typical slip work D2 acc2 The vibration acceleration data A corresponding to the bearing slip rate X2 and typical slip work D3 acc3 The vibration acceleration data A corresponding to the bearing slip rate X3 and typical slip work D4 acc4 The vibration acceleration data A corresponding to the bearing slip rate X4 and typical slip work D5 acc5 and bearing slip rate X5; Step S3.2: Based on the transfer learning algorithm and the five groups of data divided according to step S3.1, the distributed transfer learning models M1, M2, M3, M4, and M5 are trained respectively, and a skid diagnosis model set is constructed. Specifically include: Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M1 separately The parameters of the local model are Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M2 separately The parameters of the local model are Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M3 separately The local model parameters are Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M4 separately The local model parameters are Typical skidding work divided according to step S3.1 Corresponding vibration acceleration data and bearing slip rate Train the local models of the distributed transfer learning model M5 separately The local model parameters are Step S3.3: Integrate the local models established in step S3.2 respectively, specifically: Finding local model parameters The average is Use it as the parameter of the distributed transfer learning model M1; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M2; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M3; obtain the local model parameters The average is Use it as the parameter of the distributed transfer learning model M4; obtain the local model parameters The average is It is used as a parameter of the distributed transfer learning model M5.

4. A bearing slip monitoring method based on knowledge graph and distributed transfer learning according to claim 3, characterized in that: The implementation process of step S4 is: The typical slipping work established in step S1.3 is searched according to the collected measurement working parameters, and the slip rate is monitored using the distributed transfer learning model established in step S3.3 corresponding to the typical slipping work.

Images