Equipment fault diagnosis method combining digital twinning and machine learning

Abstract

The invention belongs to the technical field of fault diagnosis of mining mechanical equipment, and provides an equipment fault diagnosis method combining digital twinning and machine learning in order to timely find a fault occurrence part during equipment operation, which comprises the following steps: acquiring monitoring data; constructing a digital twinborn model based on the physical parameters of the equipment entity, and performing analog simulation on the digital twinborn model based on real-time monitoring data; combining a principal component analysis algorithm and a locality preserving projection algorithm to construct a locality-global embedding learning algorithm model; monitoring data in real time based on a local-global embedded learning algorithm model and history, and obtaining a diagnosis result of the model; based on a diagnosis result of a local-global embedded learning algorithm model and a simulation result of a digital twinborn model, equipment fault information is determined, and fault signs are found in time, so that fault diagnosis of mining machinery equipment is carried out in time, workers are helped to know the running state and performance of the equipment more clearly, and the working efficiency is improved. And more reasonable maintenance plans and strategies are formulated.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis of mining machinery and equipment, and specifically relates to an equipment fault diagnosis method combining digital twin and machine learning. Background Art

[0002] As an important equipment for realizing smart mines, mining machinery and equipment play an indispensable role in coal mining, and their working performance directly affects the efficiency and economic returns of coal mining. However, due to the large number of parts and components of mining machinery and equipment and the extremely harsh working environment, mining machinery and equipment frequently fail during operation. These failures often bring huge economic losses and even threaten the personal safety of underground personnel. In order to facilitate maintenance and resume production, it is particularly important to find out the fault location in time, and fault diagnosis of mining machinery and equipment is particularly important.

[0003] Traditional fault diagnosis usually requires manual observation and analysis, while digital twin technology remotely monitors the status of equipment by collecting and transmitting data from physical entities in real time. This method can not only detect signs of potential faults in a timely manner, but also accurately locate and diagnose faults that occur. Technicians can monitor the status of equipment in real time through remote devices without having to go to the site in person, which greatly saves time and human resources and ensures the personal safety of staff to a certain extent.

[0004] Combining the high-fidelity behavior simulation characteristics of digital twins and the powerful data mining capabilities of machine learning can provide rich data analysis and visualization. This combination enables the diagnostic system to monitor the health status of equipment in real time and accurately, and detect signs of failure in a timely manner, thereby better realizing the fault diagnosis of mining machinery and equipment, helping staff to understand the operating status and performance of the equipment more clearly, and then formulate more reasonable maintenance plans and strategies. Therefore, in order to find out the location of the failure of mining machinery and equipment during operation in a timely manner, it is necessary to study a method for equipment fault diagnosis that can combine digital twins and machine learning. Summary of the invention

[0005] In order to solve at least one of the above-mentioned technical problems existing in the prior art, the present invention provides a device fault diagnosis method combining digital twins and machine learning.

[0006] The present invention is implemented by the following technical solution: a device fault diagnosis method combining digital twins and machine learning, comprising the following steps:

[0007] Acquire monitoring data, wherein the monitoring data includes historical monitoring data and real-time monitoring data, and the monitoring data includes at least equipment operation data and environmental parameters;

[0008] Building a digital twin model based on physical parameters of the device entity, and simulating the digital twin model based on the real-time monitoring data;

[0009] Combining principal component analysis algorithm and local preservation projection algorithm, a local-global embedding learning algorithm model is constructed;

[0010] Based on the local-global embedding learning algorithm model and historical and real-time monitoring data, the diagnostic results of the model are obtained;

[0011] Based on the diagnosis results of the local-global embedding learning algorithm model and the simulation results of the digital twin model, the equipment fault information is determined.

[0012] Preferably, the principal component analysis algorithm and the local preserving projection algorithm are combined to construct a local-global embedding learning algorithm model, which specifically includes:

[0013] Processing the monitoring data to obtain a standardized data set;

[0014] Processing the standardized data set using a principal component analysis algorithm and a locality preserving projection algorithm to obtain an objective function;

[0015] Construct a local-global embedding learning algorithm model based on the objective function.

[0016] Preferably, it also includes:

[0017] Perform feature extraction based on the objective function to determine the principal component;

[0018] Calculating a statistic associated with the historical monitoring data based on the determined principal components, and determining a control limit of the statistic using a kernel density estimation method;

[0019] A fault sample is determined according to the control limit, and a fault contribution graph is calculated according to the fault sample.

[0020] Preferably, determining the fault samples according to the control limit and calculating the fault contribution graph according to the fault samples specifically includes:

[0021] Calculating new statistics related to the standardized data set based on the local-global embedding learning algorithm model;

[0022] Whether the system fails is determined based on the new statistic and the control limit. When the new statistic exceeds the control limit, the system fails, a fault sample is determined, and a fault contribution graph is obtained based on the fault sample.

[0023] Preferably, the objective function is:

[0024]

[0025] Among them, J LPP (P) is the local preservation projection objective function, J PCA (P) is the principal component analysis objective function, U is the local matrix of the measured data set, Σ is the global covariance matrix defined by the standardized data set, P is the projection matrix, and T is the matrix transpose.

[0026] Preferably, the standardized data set is processed using a principal component analysis algorithm and a locality preserving projection algorithm to obtain an objective function, which specifically includes:

[0027] The principal component analysis algorithm and the standardized data set are used for calculation to obtain the principal component analysis objective function;

[0028] The local preservation projection algorithm and the local area information of the standardized data set are used for calculation to obtain the local preservation projection objective function;

[0029] The objective function is obtained by minimizing the local scattering local preserving projection objective function and maximizing the global scattering principal component analysis objective function.

[0030] Preferably, a digital twin model is constructed based on the physical parameters of the equipment, specifically including:

[0031] Build a three-dimensional model of the virtual space based on the physical parameters of the device;

[0032] According to the actual working status of the equipment, the parent-child relationship between the components is set in the virtual space three-dimensional model, and components are added to build a high-fidelity model;

[0033] A simulation environment is built in the software based on environmental parameters, and the behavior of the high-fidelity model is simulated in the simulation environment to obtain a digital twin model.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention combines the high-fidelity behavioral simulation characteristics of digital twins with the powerful data mining capabilities of machine learning to provide rich data analysis and visualization. This combination enables the diagnostic system to monitor the health status of equipment in real time and accurately, and promptly detect signs of faults, thereby better realizing fault diagnosis of mining machinery and equipment, helping staff to more clearly understand the operating status and performance of the equipment, and then formulate more reasonable maintenance plans and strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a flowchart of a device fault diagnosis method combining digital twin and machine learning according to an embodiment of the present invention;

[0038] Figure 2 It is a schematic diagram of a high-fidelity behavior simulation construction process of a scraper conveyor virtual space provided by an embodiment of the present invention;

[0039] Figure 3 It is a schematic diagram of the construction process of the local-global embedding algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In conjunction with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size should fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention. It should be noted that in this specification, relational terms such as first and second are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0042] The present invention provides an embodiment, such as Figure 1 As shown, a device fault diagnosis method combining digital twins and machine learning includes the following steps:

[0043] S1: Acquire monitoring data, wherein the monitoring data includes historical monitoring data and real-time monitoring data, and the monitoring data includes at least equipment operation data and environmental parameters.

[0044] In this embodiment, sensors are arranged at reasonable locations of the monitored mining machinery and equipment to obtain real-time operation data, environmental parameters and other monitoring data of the mining machinery and equipment, and store them to obtain historical monitoring data such as historical operation data and historical environmental parameters.

[0045] S2: Construct a digital twin model based on the physical parameters of the device entity, and simulate the digital twin model based on the real-time monitoring data.

[0046] Optionally, a virtual space three-dimensional model is constructed based on the physical parameters of the device; according to the actual working status of the device, the parent-child relationship between the components is set in the virtual space three-dimensional model, and components are added to construct a high-fidelity model; a simulation environment is constructed in the software based on the environmental parameters, and the behavior of the high-fidelity model is simulated in the simulation environment to obtain a digital twin model.

[0047] In this embodiment, based on the physical parameters of the mining machinery and equipment entity, SolidWorks software is used to build a high-fidelity model of the mining machinery and equipment in the virtual space; a hyper-realistic simulation environment is built in Unity 3D software according to the environmental parameters in the monitoring parameters. At the same time, the high-fidelity model built in SolidWorks is imported into Unity 3D, and the high-fidelity model is driven by the real-time equipment operation data in the monitoring data to perform high-fidelity behavior simulation in the hyper-realistic simulation environment, thereby realizing the visualization of the real-time status of the mining machinery and equipment in the physical space, and providing a qualitative basis for the fault diagnosis of the mining machinery and equipment.

[0048] In this embodiment, taking the scraper conveyor as an example, Figure 2 As shown, based on the physical parameters of the mining machinery and equipment entity, the three-dimensional model of the mining machinery and equipment is constructed in the virtual space using SolidWorks software. The physical parameters include: geometric parameters, material properties; the geometric parameters include: length, width, height, hole depth, etc., and the material properties include: material type, stiffness, hardness, elastic modulus, etc.

[0049] Save the 3D model created in SolidWorks as .sld format, and use 3D Max software to convert the .sld format to .fbx format to facilitate importing the model into Unity 3D software.

[0050] According to the actual working posture of mining machinery and equipment, the parent-child relationship between the components is set in Unity 3D, and components such as joints, rigid bodies, and collision bodies are added to complete the construction of a high-fidelity model of mining machinery and equipment.

[0051] A hyper-realistic simulation environment is constructed in Unity 3D software based on the environmental parameters in the monitoring parameters. At the same time, the real-time operating parameters in the monitoring data drive the high-fidelity model to perform high-fidelity behavior simulation in the hyper-realistic simulation environment, and the posture state of the physical space of the mining machinery and equipment is reflected in the virtual space in real time and dynamically, realizing the visualization of the real-time state of the physical space mining machinery and equipment. The environmental parameters include: temperature, humidity, etc.; the real-time operating parameters include: operating data and position information; the operating data include: speed, amplitude, operating time, etc.; the position information includes: moving distance, tilt angle, etc.

[0052] S3: Combine the principal component analysis algorithm and the local preserving projection algorithm to build a local-global embedding learning algorithm model.

[0053] Optionally, the monitoring data is processed to obtain a standardized data set; the standardized data set is processed using a principal component analysis algorithm and a local preserving projection algorithm to obtain an objective function; and a local-global embedding learning algorithm model is constructed based on the objective function.

[0054] Optionally, a local-global embedding learning algorithm model is constructed by combining a principal component analysis algorithm and a local preserving projection algorithm, specifically including: using the principal component analysis algorithm and a standardized data set to perform calculations to obtain a principal component analysis objective function; using the local preserving projection algorithm and local domain information of a standardized data set to perform calculations to obtain a local preserving projection objective function; and obtaining an objective function by minimizing the local scattering local preserving projection objective function and maximizing the global scattering principal component analysis objective function.

[0055] Optionally, feature extraction is performed based on the objective function to determine the principal components; statistics associated with the historical monitoring data are calculated based on the determined principal components, and control limits of the statistics are determined using a kernel density estimation method; fault samples are determined based on the control limits, and a fault contribution graph is calculated based on the fault samples.

[0056] Optionally, fault samples are determined according to the control limits, and a fault contribution graph is calculated based on the fault samples, specifically including: calculating new statistics related to the standardized data set based on the local-global embedding learning algorithm model; judging whether the system is faulty according to the new statistics and the control limits, when the new statistics exceed the control limits, the system fails, determining fault samples, and calculating a fault contribution graph based on the fault samples.

[0057] In this embodiment, x(i)=(x1(i),x2(i),…,x m (i)) is a sample of the monitoring data set X, where i = 1, 2, ..., n. The monitoring data is processed to obtain a standardized data set, the standardized data set X∈Rn×m , its standardized formula is:

[0058]

[0059] in, and are the mean and standard deviation of the jth variable in the data set. Its mean and variance are calculated by the following formula:

[0060]

[0061] In this embodiment, the Local-Global Embedding Algorithm (LGEA) model projects high-dimensional data into low-dimensional space by minimizing local scattering while maximizing global scattering projections, which can reveal the intrinsic geometric structure of imprecise data and obtain more useful low-dimensional information hidden in high-dimensional data. LGEA is a popular learning algorithm that incorporates the Locality Preserving Projections (LPP) algorithm into the Principal Component Analysis (PCA) algorithm to retain local information of the data. In this way, LGEA can discover local manifold structures in the data and maximize the preservation of local neighborhoods and global information in the data.

[0062] In this embodiment, for the standardized data set, the purpose of PCA is to find a projection axis P∈R m×l The Euclidean distance Y between all data projections is (y 1j ,y 2j ,…,y nj )∈R n×l (where j = 1, 2, ..., l) reaches the maximum. According to the monitoring data set X, y i =x(i)P, where i=1,2,…,n. Therefore, the principal component analysis objective function is:

[0063]

[0064] in, Σ is the global covariance matrix defined by the standardized data set, that is, Σ = 1 / n(X T X); eigenvector P = [P1, P2, ..., P l ](l<m), represents principal component (PCs), l is the number of selected PCs, and m is the number of all PCs.

[0065] In this embodiment, a projection transformation P is found through LPP, and the local neighborhood information of the monitoring data set X is retained by minimizing the local scattering local preservation projection objective function. The local preservation projection objective function is:

[0066]

[0067] Where U = X T LX is the local matrix of the monitoring data set X, L = CW is called the Laplace matrix, C is the diagonal matrix, C ii =∑ j w ij , w ij is the local similarity matrix, which is defined as follows:

[0068]

[0069] Among them, σ is an empirical constant, Ω k (x(i)) represents the k samples with the minimum Euclidean distance from sample x(j), where Ω k (x(i))={x|||x(i)-x(j)|| 2 ≤γ}, γ is a key parameter of LGEA. It can be used to select the appropriate neighborhood size by estimating the “quality” of the corresponding mapping, or it can be measured by residual variance, Bayesian information criterion, etc.

[0070] In this embodiment, in order to effectively extract information and reveal the intrinsic geometric structure of the data, the local neighborhood and global information of the data can be preserved by minimizing the local scattering local preservation projection objective function J LPP (P) and maximize the global scattering principal component analysis objective function J PCA The projection of (P) establishes the objective function, and its formula is as follows:

[0071]

[0072] In this embodiment, according to the above objective function, the minimization problem is transformed into a generalized eigenvalue problem to determine the principal components in the following form:

[0073] UP=λΣP

[0074] Where λ represents the eigenvalue of the eigenvector P.

[0075] Calculate the eigenvalue and eigenvector of the generalized eigenvalue problem. Let the eigenvector P = (P1, P2, ..., P m ) According to the eigenvalues ​​λ1,λ2,…,λ mSort from large to small, determine the principal components by the cumulative percentage variance standard, set the first l as the principal components, retain the first l column vectors according to the sorted eigenvalues ​​to form a transformation matrix The projection score matrix Y and residual matrix E of the data are expressed as follows:

[0076]

[0077] In this embodiment, the statistic T 2 is a measure of the spatial variation of the latent variable and the sum of normalized squared scores. Given a normalized dataset X∈R n×m , the i-th projection score matrix y i ∈R 1×l T 2 The indicator is given by:

[0078]

[0079] Among them, S -1 is the inverse matrix of the covariance matrix of the projection score matrix Y under normal operating conditions.

[0080] The statistic SPE is used to monitor the variation measures not captured by LGEA, which represents the Euclidean distance between the measured and estimated values, measured by the projection of the sample vector on the residual matrix E. For the residual feature e of the i-th data set of E i , its SPE index is given by:

[0081]

[0082] The kernel density estimation method is used to determine the upper control limits of the above two statistics, which can be calculated by the upper quantile of the significance level β:

[0083]

[0084] P(SPE≤SPE UCL )=1-β

[0085] in, and SPE UCL T 2 and the upper control limit of SPE, that is, the threshold between normal values ​​and abnormal values. If the new statistic T 2 Or if the SPE exceeds the threshold, the sample is abnormal.

[0086] In this embodiment, the statistic T 2 Although SPE can detect uncertain process faults, it cannot reveal the relevant process variables that cause the faults. Therefore, it is necessary to calculate the fault contribution diagram of each variable and find out the variables that have a high contribution to the fault statistics, that is, the fault variables.

[0087] The score vector and load vector of the LGEA model are non-orthogonal. Therefore, it is assumed that the new data after normalization of the real-time monitoring operation data is z new , calculate new statistics based on new data and SPE new , statistics is defined as follows:

[0088]

[0089] Therefore, the new input data z new The variable z i,new right The contributions of the statistics are as follows:

[0090]

[0091] Statistics SPE new They are defined as data z new The residual space e i,new The sum of squares:

[0092]

[0093] in, is the new sample at the i-th position variable SPE new The contribution of statistics.

[0094] S4: Based on the local-global embedding learning algorithm model and historical and real-time monitoring data, the diagnostic results of the model are obtained.

[0095] In this embodiment, Figure 3 As shown, historical monitoring data during the continuous normal operation of the equipment is obtained, the historical monitoring data is standardized to obtain a standardized data set, and then the global covariance matrix and the local matrix are calculated based on the standardized data set, the global covariance matrix is ​​maximized and scattered projection is performed, and the local matrix is ​​minimized and scattered projection is performed at the same time, and the objective function is established in combination with the above processing results, and then the principal component is determined, and then the historical data statistics are calculated, and the control upper limit of the historical data statistics is determined; the real-time monitoring operation data of the equipment operation is obtained, and the real-time monitoring operation data is standardized to obtain standardized new data, and then the standardized new data is processed to obtain new statistics, and the new statistics are compared with the control upper limit. When the new statistics is less than or equal to the control upper limit, the equipment is normal, then it returns to continuously obtain the real-time monitoring operation data, and performs subsequent calculations. When the new statistics is greater than the control upper limit, it indicates that the equipment is faulty and the current fault variable is identified.

[0096] S5: Determine equipment fault information based on the diagnosis results of the local-global embedding learning algorithm model and the simulation results of the digital twin model.

[0097] In this embodiment, digital twin technology is used to realize the real-time status visualization of mining machinery and equipment in physical space based on monitoring data; at the same time, online data drives the local-global embedded model to perform fault monitoring and fault variable identification; the visualized real-time status and fault diagnosis are comprehensively analyzed to realize the integration of digital twin and machine learning, and complete the qualitative and quantitative analysis of data; finally, the analysis results are visualized through the human-computer interaction system.

[0098] The embodiment of the present invention combines the digital twin and machine learning method for fault diagnosis of mining machinery and equipment. The digital twin is used to establish a high-fidelity model of the mining machinery and equipment in a virtual space. Real-time monitoring data drives the high-fidelity model to perform high-fidelity behavior simulation in the constructed hyper-realistic simulation environment. The real-time state of the physical space of the mining machinery and equipment is mirrored to the virtual space, thereby realizing the visualization of the real-time state of the mining machinery and equipment, and providing a qualitative basis for the fault diagnosis of the mining machinery and equipment. At the same time, the real-time monitoring operation data and the constructed machine learning algorithm are used to provide a quantitative basis for the fault diagnosis of the mining machinery and equipment. The diagnostic results of the constructed machine learning algorithm and the high-fidelity behavior simulation results in the virtual space are combined to perform quantitative and qualitative analysis to realize fault diagnosis of the mining machinery and equipment.

[0099] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A device fault diagnosis method combining digital twins and machine learning, characterized in that: The steps include: Acquire monitoring data, wherein the monitoring data includes historical monitoring data and real-time monitoring data, and the monitoring data includes at least equipment operation data and environmental parameters; Building a digital twin model based on physical parameters of the device entity, and simulating the digital twin model based on the real-time monitoring data; Combining principal component analysis algorithm and local preservation projection algorithm, a local-global embedding learning algorithm model is constructed; Based on the local-global embedding learning algorithm model and historical and real-time monitoring data, the diagnostic results of the model are obtained; Based on the diagnosis results of the local-global embedding learning algorithm model and the simulation results of the digital twin model, the equipment fault information is determined.

2. The method according to claim 1, characterized in that Combining the principal component analysis algorithm and the local preservation projection algorithm, a local-global embedding learning algorithm model is constructed, which includes: Processing the monitoring data to obtain a standardized data set; Processing the standardized data set using a principal component analysis algorithm and a locality preserving projection algorithm to obtain an objective function; Construct a local-global embedding learning algorithm model based on the objective function.

3. The method according to claim 2, characterized in that Also includes: Perform feature extraction based on the objective function to determine the principal component; Calculating a statistic associated with the historical monitoring data based on the determined principal components, and determining a control limit of the statistic using a kernel density estimation method; A fault sample is determined according to the control limit, and a fault contribution graph is calculated according to the fault sample.

4. The method according to claim 3, characterized in that The determining of the fault sample according to the control limit and calculating the fault contribution graph according to the fault sample specifically includes: Calculating new statistics related to the standardized data set based on the local-global embedding learning algorithm model; Whether the system fails is determined based on the new statistic and the control limit. When the new statistic exceeds the control limit, the system fails, a fault sample is determined, and a fault contribution graph is obtained based on the fault sample.

5. The method according to claim 3, characterized in that: The objective function is: Among them, J LPP (P) is the local preservation projection objective function, J PCA (P) is the principal component analysis objective function, U is the local matrix of the measured data set, Σ is the global covariance matrix defined by the standardized data set, P is the projection matrix, and T is the matrix transpose.

6. The method according to claim 2, characterized in that The method of processing the standardized data set by using the principal component analysis algorithm and the local preservation projection algorithm to obtain the objective function specifically includes: The principal component analysis algorithm and the standardized data set are used for calculation to obtain the principal component analysis objective function; The local preservation projection algorithm and the local area information of the standardized data set are used for calculation to obtain the local preservation projection objective function; The objective function is obtained by minimizing the local scattering local preserving projection objective function and maximizing the global scattering principal component analysis objective function.

7. The method according to claim 1, characterized in that The digital twin model is built based on the physical parameters of the equipment entity, including: Build a three-dimensional model of the virtual space based on the physical parameters of the device; According to the actual working status of the equipment, the parent-child relationship between the components is set in the virtual space three-dimensional model, and components are added to build a high-fidelity model; A simulation environment is built in the software based on environmental parameters, and the behavior of the high-fidelity model is simulated in the simulation environment to obtain a digital twin model.

Images