Rotating equipment fault early warning and intelligent diagnosis method
By building a fault warning model and mechanism model, combining historical and real-time data analysis, fault warning and intelligent diagnosis of rotating equipment are achieved, problems that cannot be early warning in the existing technology are solved, and maintenance efficiency and equipment safety are improved.
Patent Information
- Application Number
- CN202410952567.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art cannot provide early warning of rotating equipment, resulting in low maintenance efficiency.
By collecting historical monitoring data of rotating equipment, a fault warning model is built, and analyses are combined with real-time monitoring data to generate fault warning results. When the fault warning result is a specific type, the scheduling mechanism model performs fault diagnosis and generates an intelligent diagnosis log.
It realizes timely detection and early warning of potential fault risks of rotating equipment, improves maintenance efficiency, and ensures the safe operation of equipment.
Smart Images

Figure CN119005940A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of rotating equipment monitoring, and in particular relates to a rotating equipment fault early warning and intelligent diagnosis method. Background Art
[0002] Rotating equipment refers to mechanical equipment that achieves its working purpose through rotational motion. This type of equipment is widely used in various industries, especially in the fields of manufacturing, transportation, energy production, etc. Rotating equipment usually contains one or more rotating parts, which are driven by rotating equipment, engines or other power sources to achieve various physical actions or work processes. In the prior art, only the sensor data corresponding to the rotating equipment is often collected, and an alarm is issued when a certain sensor data exceeds the limit. Although the abnormality of the equipment can be detected in time, it is impossible to give early warning of the fault, resulting in low maintenance efficiency. Summary of the invention
[0003] The present invention provides a rotating equipment fault early warning and intelligent diagnosis method, which is used to solve the problem that the prior art cannot provide early warning of faults, thereby resulting in low maintenance efficiency.
[0004] A method for early warning and intelligent diagnosis of rotating equipment faults, comprising:
[0005] Collect historical monitoring data corresponding to rotating equipment and historical fault types corresponding to the historical monitoring data;
[0006] Constructing a rotating equipment fault warning model, and training the rotating equipment fault warning model using historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data, to obtain a trained rotating equipment fault warning model;
[0007] The rotating equipment fault warning model is deployed on a platform designated by the staff, and real-time monitoring data is collected, and the real-time monitoring data is analyzed by the deployed rotating equipment fault warning model to obtain a fault warning result; wherein the fault warning result includes a specific type of fault or the absence of a fault;
[0008] When the fault warning result is a specific type of fault, the pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain the fault diagnosis result;
[0009] According to the fault diagnosis results and real-time monitoring data, an intelligent fault diagnosis log is generated, and the intelligent fault diagnosis log is transmitted to the equipment pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
[0010] Furthermore, historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data are collected, including:
[0011] Based on the preset data length N, at least the voltage data, current data, temperature data, speed data, vibration data and corresponding historical fault types of the rotating equipment at N consecutive time points are collected, and the collected voltage data, current data, temperature data, speed data and vibration data are formed into a data matrix to obtain historical monitoring data and historical fault types corresponding to the historical monitoring data;
[0012] By using the above method, multiple pairs of historical monitoring data and historical fault types corresponding to the historical monitoring data are obtained.
[0013] Furthermore, a rotating equipment fault warning model is constructed, and the rotating equipment fault warning model is trained using historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data, to obtain a trained rotating equipment fault warning model, including:
[0014] Use CNN to build a fault warning model for rotating equipment;
[0015] Determine the minimum data length n of a single type of data and set the counter k=n; where n <N;
[0016] According to the voltage data, current data, temperature data, speed data and vibration data in the historical monitoring data corresponding to the rotating equipment, the voltage data, current data, temperature data, speed data and vibration data at the most recent k time points are taken out to obtain sample data, and the historical fault type corresponding to the historical monitoring data is used as the sample label corresponding to the sample data;
[0017] Each historical monitoring data is trimmed to obtain a number of sample data and sample labels corresponding to the sample data;
[0018] Using a number of sample data and sample labels corresponding to the sample data to form sample data pairs, and dividing the sample data pairs into a training set and a validation set;
[0019] The rotation equipment fault warning model is trained using a training set to obtain a rotation equipment fault warning model corresponding to a data length of k;
[0020] The validation set is used to obtain the accuracy of the rotating equipment fault warning model corresponding to the data length k;
[0021] Determine whether the count value of counter k is equal to N. If so, obtain Nn rotating equipment fault warning models. Otherwise, increase the count value of k by one and return to the sample data acquisition step.
[0022] According to the accuracies corresponding to the Nn rotating equipment fault warning models, the rotating equipment fault warning model with the highest accuracy is determined, and the trained rotating equipment fault warning model is obtained.
[0023] Furthermore, the rotation equipment fault warning model is trained using a training set to obtain a rotation equipment fault warning model corresponding to a data length of k, including:
[0024] Randomly initializing the model parameters of the rotating equipment fault warning model between the upper limit and the lower limit of the parameters of the rotating equipment fault warning model to obtain the model parameter code;
[0025] Based on the above method, the model parameter codes are obtained repeatedly to obtain multiple model parameter codes;
[0026] For any model parameter code, apply the model parameter code to the rotating equipment fault warning model, and obtain the error function value of the rotating equipment fault warning model based on the training set, and obtain the fitness value corresponding to the model parameter code according to the error function value;
[0027] According to the fitness values corresponding to the model parameter codes, determining the code with the largest fitness value as the optimal parameter code;
[0028] Based on the optimal parameter encoding, a first search strategy of weighted and perturbation learning is used to perform global optimization on each model parameter encoding to obtain a model parameter encoding after global optimization;
[0029] The second search strategy of mutual attraction within the neighborhood is used to locally optimize the model parameter encoding after global optimization, and the model parameter encoding after local optimization is obtained;
[0030] The third search strategy of worst position exclusion learning is used to perform extended optimization on the model parameter encoding after local optimization to obtain the model parameter encoding after extended optimization;
[0031] Repeat the above three optimization steps until the training times reach the maximum training times, re-obtain the optimal parameter encoding, and use the re-obtained optimal parameter encoding as the final parameter of the rotating equipment fault warning model to obtain the rotating equipment fault warning model corresponding to the data length k.
[0032] Further, obtaining the fitness value corresponding to the model parameter encoding according to the error function value includes:
[0033] Adding the error function value corresponding to the model parameter encoding to a preset noise value to obtain a first added term;
[0034] Taking the reciprocal of the first added term to obtain a reciprocal term;
[0035] After performing an exponential operation on the reciprocal term using a power function between (0,1), it is multiplied by a preset adjustment factor to obtain the fitness value corresponding to the model parameter encoding.
[0036] Furthermore, based on the optimal parameter encoding, a first search strategy of weighted and perturbation learning is used to perform global optimization on each model parameter encoding to obtain a model parameter encoding after global optimization, including:
[0037] The current number of training times is divided by the maximum number of training times to obtain the first division term;
[0038] Subtract the first division term from the preset constant term to obtain an inertia weight that gradually decreases as the number of training times increases;
[0039] The model parameter code is weighted by using an inertia weight to obtain a weighted model parameter code;
[0040] Generate a first random number between [0, 1], and weight the optimal parameter code by squaring the first random number to obtain the weighted optimal parameter code;
[0041] The weighted optimal parameter encoding is subtracted from the model parameter encoding, and then multiplied by the jump search step size and the fitness value corresponding to the optimal parameter encoding to obtain the first update item for global optimization;
[0042] The weighted model parameter encoding is added to the first update term to obtain the model parameter encoding after global optimization.
[0043] Furthermore, a second search strategy of mutual attraction within a neighborhood is used to locally optimize the model parameter encoding after global optimization, and the model parameter encoding after local optimization is obtained, including:
[0044] For the model parameter encoding after global optimization, a random neighborhood search direction is determined between (-1,1);
[0045] Determine the target model parameter encoding that is closest to the model parameter encoding after global optimization and has a larger fitness value;
[0046] The target model parameter code is subtracted from the model parameter code after global optimization to obtain the first subtraction term;
[0047] Generate a second random number between [0, 1], and weight the first subtraction term by the second random number, and then multiply the weight by the fitness value corresponding to the target model parameter encoding to obtain a neighborhood search step length;
[0048] Using a neighborhood search direction and a neighborhood search step size, determine a second update item for local optimization;
[0049] The model parameter encoding after global optimization is added to the second update term to obtain the model parameter encoding after local optimization.
[0050] Furthermore, the third search strategy of worst position exclusion learning is used to perform extended optimization on the model parameter encoding after local optimization, and the model parameter encoding after extended optimization is obtained, including:
[0051] Generate a third random number between [0, 1], and weight the model parameter encoding after the local optimization by the third random number to obtain the weighted model parameter encoding;
[0052] The worst parameter encoding with the smallest fitness is subtracted from the weighted model parameter encoding, and the absolute value of each dimension parameter is taken to obtain the second subtraction term;
[0053] The fitness value corresponding to the model parameter encoding after local optimization is subtracted from the fitness value corresponding to the worst parameter encoding, and a preset noise value is added to obtain a second addition term;
[0054] After dividing the second subtraction term by the second addition term, and controlling it with a step size control coefficient, a third update term for expanding the optimization is determined;
[0055] The model parameter encoding after local optimization is added to the third update term to obtain the model parameter encoding after extended optimization.
[0056] Furthermore, when the fault warning result is a specific type of fault, the pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain the fault diagnosis result, including:
[0057] According to the working principle and physical characteristics of rotating equipment, establish the mechanism model of rotating equipment;
[0058] Input the real-time monitoring data of the rotating equipment into the mechanism model; wherein the real-time monitoring data includes voltage data, current data, temperature data, speed data and vibration data;
[0059] Based on the real-time monitoring data of the rotating equipment, the mechanism model of the rotating equipment is operated to determine the operation trajectory of each component of the rotating equipment;
[0060] The running tracks of the various components of the rotating equipment are analyzed to determine the fault type of the rotating equipment and obtain the fault diagnosis result.
[0061] Furthermore, according to the fault diagnosis result and the real-time monitoring data, a fault intelligent diagnosis log is generated, and the fault intelligent diagnosis log is transmitted to a device pre-set by the staff, including:
[0062] According to the fault diagnosis results and real-time monitoring data, a PDF version report is generated according to a preset template, and a fault intelligent diagnosis log is generated;
[0063] The fault intelligent diagnosis log is transmitted to the equipment pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
[0064] The present invention provides a method for early warning and intelligent diagnosis of rotating equipment faults. By real-time monitoring of the sensor parameters of rotating equipment and analyzing these sensor parameters using deep learning technology, potential fault risks can be discovered in a timely manner, and possible fault occurrence can be warned. Possible fault types can be warned based on abnormal conditions of the sheath parameters. After predicting a possible fault, the scheduling mechanism model determines the fault type, allowing staff to take corresponding protective measures to prevent the occurrence or expansion of the fault, thereby ensuring the safe operation of rotating equipment and having important significance for energy supply and socio-economic development. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0066] Figure 1 A flow chart of a method for fault warning and intelligent diagnosis of rotating equipment provided in an embodiment of the present invention.
[0067] Figure 2 A flow chart for collecting historical monitoring data corresponding to rotating equipment and historical fault types corresponding to the historical monitoring data provided by an embodiment of the present invention.
[0068] Figure 3 A flowchart of obtaining a trained rotating equipment fault warning model provided in an embodiment of the present invention.
[0069] Figure 4 A flow chart of using a training set to train the rotating equipment fault warning model is provided in an embodiment of the present invention.
[0070] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0071] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.
[0072] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.
[0073] like Figure 1 As shown, an embodiment of the present invention provides a method for early warning and intelligent diagnosis of rotating equipment faults, comprising:
[0074] S101. Collect historical monitoring data corresponding to rotating equipment and historical fault types corresponding to the historical monitoring data.
[0075] Optionally, data collection can be performed through human-computer interaction, or it can be directly pulled from a historical fault database, so that the changing patterns of the data can be better reflected, and these changing patterns can be learned, thereby achieving fault warning.
[0076] S102, constructing a rotating equipment fault warning model, and training the rotating equipment fault warning model using historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data, to obtain a trained rotating equipment fault warning model;
[0077] A deep learning model can be used to build a rotating equipment fault warning model. After the rotating equipment fault warning model is built, it cannot be used directly. Therefore, it is necessary to first learn the relationship between historical data so that the rotating equipment fault warning model has the ability to identify and analyze data.
[0078] S103, deploying the rotating equipment fault warning model on a platform designated by the staff, collecting real-time monitoring data, and analyzing the real-time monitoring data through the deployed rotating equipment fault warning model to obtain a fault warning result; wherein the fault warning result includes a specific type of fault or the absence of a fault;
[0079] For example, the rotating equipment fault warning model can be deployed on a cloud server or a private computer system designated by a staff member, so that the collected data can be analyzed.
[0080] S104. When the fault warning result is a specific type of fault, a pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain a fault diagnosis result.
[0081] The mechanism model is a model that describes the behavior of the system through physical laws and empirical formulas based on a deep understanding of physical phenomena. This model usually contains detailed physical processes and mechanisms, and can reflect the changing patterns of various physical quantities and parameters within the system over time or space. Therefore, the mechanism model corresponding to the rotating equipment can be pre-built, and when a failure may occur (i.e. when a fault warning is generated), the mechanism model is dispatched for simulation analysis to determine the operating parameters of the rotating equipment, and then the fault analysis is performed based on the operating parameters.
[0082] S105. Generate a fault intelligent diagnosis log based on the fault diagnosis result and real-time monitoring data, and transmit the fault intelligent diagnosis log to a device pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
[0083] Optionally, in order to further improve the efficiency of the staff in handling exceptions, the handling method corresponding to the fault diagnosis result can also be scheduled in the database, and the handling method can be displayed in the fault intelligent diagnosis log.
[0084] like Figure 2 As shown, the historical monitoring data corresponding to the rotating equipment and the historical fault types corresponding to the historical monitoring data are collected, including:
[0085] S101.1. Based on a preset data length N, at least voltage data, current data, temperature data, rotation speed data, vibration data and corresponding historical fault types of the rotating equipment at N consecutive time points are collected, and the collected voltage data, current data, temperature data, rotation speed data and vibration data are formed into a data matrix to obtain historical monitoring data and historical fault types corresponding to the historical monitoring data;
[0086] It is worth noting that in addition to voltage data, current data, temperature data, rotation speed data and vibration data, other types of monitoring data may also be included to improve the accuracy of fault warning.
[0087] S101.2. Using the above method, obtain multiple pairs of historical monitoring data and historical fault types corresponding to the historical monitoring data.
[0088] In an embodiment of the present invention, constructing a rotating equipment fault warning model includes: using CNN (convolutional neural network) to construct a rotating equipment fault warning model.
[0089] It is worth noting that CNN is only a preferred selection method in the embodiment of the present invention, and other neural networks can also be used to construct a fault warning model for rotating equipment.
[0090] like Figure 3 As shown, the rotation equipment fault early warning model is trained using the historical monitoring data corresponding to the rotation equipment and the historical fault types corresponding to the historical monitoring data, to obtain a trained rotation equipment fault early warning model, including:
[0091] S102.1. Determine the minimum data length n of a single type of data and set a counter k=n; wherein n <N;
[0092] S102.2. According to the voltage data, current data, temperature data, speed data and vibration data in the historical monitoring data corresponding to the rotating equipment, the voltage data, current data, temperature data, speed data and vibration data at the most recent k time points are taken out to obtain sample data, and the historical fault type corresponding to the historical monitoring data is used as the sample label corresponding to the sample data;
[0093] In monitoring data, there are often sensor data measured by multiple sensors at multiple time points. For example, there may be multiple pressure sensors that measure pressure data at N consecutive time points. Therefore, the data of all sensors of the same type at the same time point can be constructed into a row of data, and the data at multiple time points can be arranged in sequence into multiple rows of data to obtain the data matrix corresponding to a certain sensor data at N time points; the data matrices corresponding to the voltage data, current data, temperature data, speed data and vibration data are connected in series from top to bottom to obtain sample data.
[0094] Optionally, when the number of sensors for each type of data is the same, the dimension of each row of the collected sample data is the same, while when the number of sensors is different, the row dimensions may be different. Therefore, zero padding can be performed at the end of the row data with fewer dimensions, so that the data matrix is more regular and convenient for subsequent data recognition. It is worth noting that the subsequent data to be recognized also needs to use the same method to construct the data matrix to ensure that the data can be accurately recognized.
[0095] The data matrix can also be normalized to reduce the complexity of the data and thus improve the efficiency of data recognition.
[0096] S102.3. Data cutting is performed on each historical monitoring data to obtain a number of sample data and sample labels corresponding to the sample data;
[0097] S102.4. Use a number of sample data and sample labels corresponding to the sample data to form sample data pairs, and divide the sample data pairs into a training set and a validation set;
[0098] S102.5. Use the training set to train the rotating equipment fault warning model to obtain the rotating equipment fault warning model corresponding to the data length k;
[0099] S102.6. Use the validation set to obtain the accuracy of the corresponding rotating equipment fault warning model when the data length is k;
[0100] S102.7. Determine whether the count value of counter k is equal to N. If so, obtain Nn rotating equipment fault warning models. Otherwise, increase the count value of k by one and return to the sample data acquisition step.
[0101] S102.8. According to the accuracies corresponding to the Nn rotating equipment fault warning models, determine the rotating equipment fault warning model with the highest accuracy, and obtain a trained rotating equipment fault warning model.
[0102] The data clipping and identification method provided in the embodiment of the present invention can determine the optimal data identification length, thereby ensuring that the fault warning task can be completed while avoiding the data being too long.
[0103] Optionally, in addition to screening based on accuracy, other neural network evaluation indicators can also be used for screening, and ultimately the rotating equipment fault warning model with the best recognition and analysis effect is determined, thereby ensuring the accuracy of the fault warning.
[0104] like Figure 4 As shown, the rotation equipment fault warning model is trained using a training set to obtain a corresponding rotation equipment fault warning model when the data length is k, including:
[0105] a1. Randomly initialize the model parameters (such as weights) of the rotating equipment fault warning model between the upper limit and the lower limit of the parameters of the rotating equipment fault warning model to obtain the model parameter code;
[0106] a2. Based on the above method, repeatedly obtain the model parameter codes multiple times to obtain multiple model parameter codes;
[0107] a3. For any model parameter code, apply the model parameter code to the rotating equipment fault warning model, and obtain the error function value of the rotating equipment fault warning model based on the training set, and obtain the fitness value corresponding to the model parameter code according to the error function value;
[0108] Optionally, the error function value can be set to a cross entropy error or a mean square error. By obtaining these errors, the quality of the model parameter encoding can be judged.
[0109] a4. According to the fitness values corresponding to the model parameter codes, determine the code with the largest fitness value as the optimal parameter code;
[0110] a5. Based on the optimal parameter encoding, a first search strategy of weighted and perturbation learning is used to perform global optimization on each model parameter encoding to obtain a model parameter encoding after global optimization;
[0111] a6. Use the second search strategy of mutual attraction within the neighborhood to perform local optimization on the model parameter encoding after global optimization to obtain the model parameter encoding after local optimization;
[0112] a7. Using the third search strategy of worst position exclusion learning to perform extended optimization on the model parameter encoding after local optimization, to obtain the model parameter encoding after extended optimization;
[0113] a8. Repeat the above three optimization steps until the number of training times reaches the maximum number of training times, re-obtain the optimal parameter code, and use the re-obtained optimal parameter code as the final parameter of the rotating equipment fault warning model to obtain the rotating equipment fault warning model corresponding to the data length k.
[0114] In the prior art, the optimization of the neural network is poor, and it is easy to fall into the local optimum and the convergence effect is poor. Therefore, the embodiment of the present invention improves the training process.
[0115] In the embodiment of the present invention, obtaining the fitness value corresponding to the model parameter encoding according to the error function value includes:
[0116] Add the error function value E corresponding to the model parameter encoding to the preset noise value ε to obtain a first addition term A1; wherein the noise value ε can be set to 0.0001;
[0117] Taking the reciprocal of the first added term A1, a reciprocal term I=1 / A1 is obtained;
[0118] After performing exponential operation on the reciprocal term using the power function a between (0,1), multiplying it by the preset adjustment factor c, the fitness value f corresponding to the model parameter encoding is obtained. a .
[0119] In the embodiment of the present invention, based on the optimal parameter encoding, a first search strategy of weighted and perturbation learning is adopted to perform global optimization on each model parameter encoding to obtain the model parameter encoding after global optimization, including:
[0120] Divide the current number of training times t by the maximum number of training times T to obtain the first division term t / T;
[0121] Subtract the first division term t / T from the preset constant term B (for example, set to 2) to obtain the inertia weight w that gradually decreases as the number of training times increases;
[0122] The model parameter encoding is weighted by using the inertia weight to obtain the weighted model parameter encoding in, represents the i-th model parameter encoding in the t-th training process, i = 1, 2, ..., M, M represents the total number of model parameter encodings;
[0123] Generate a first random number r1 between [0,1], and encode the optimal parameter by squaring the first random number r1 Weighted, get the optimal parameter encoding after weighting
[0124] The weighted optimal parameter encoding is subtracted from the model parameter encoding, and then multiplied by the jump search step size levy and the fitness value corresponding to the optimal parameter encoding. Get the first update item for global optimization Where levy represents the random step length generated by Levy flight;
[0125] Add the weighted model parameter encoding to the first update term to obtain the model parameter encoding after global optimization
[0126] The first search strategy of weighted and perturbation learning provided in the embodiment of the present invention can make the model parameter encoding have a certain random jump while quickly approaching the optimal position, thereby ensuring the search speed and global searchability, and can effectively avoid falling into the local optimum. In addition, an adaptive inertia weight is set, which can reduce the weight of the model parameter encoding itself as the training progresses, and can learn more optimal position information, which is conducive to the convergence of the algorithm in the later stage.
[0127] In the embodiment of the present invention, a second search strategy of mutual attraction within a neighborhood is used to perform local optimization on the model parameter encoding after global optimization, and the model parameter encoding after local optimization is obtained, including:
[0128] For the model parameter encoding after global optimization, a random neighborhood search direction γ is determined between (-1,1);
[0129] Determine the model parameter encoding after global optimization of distance The nearest target model with a larger fitness value is encoded in the parameters
[0130] Target model parameter encoding Model parameter encoding after subtracting global optimization Get the first subtraction term C1; represents the model parameter encoding after the mth global optimization, represents the target model parameter encoding corresponding to the model parameter encoding after the mth global optimization, m = 1, 2, ..., M;
[0131] Generate a second random number r2 between [0,1], and weight the first subtraction term C1 by the second random number r2, and then multiply it by the fitness value corresponding to the target model parameter encoding to obtain the neighborhood search step length. Among them, f g Indicates the fitness value corresponding to the target model parameter encoding;
[0132] Adopt neighborhood search direction γ and neighborhood search step size Determine the second update term for local optimization
[0133]
[0134] Add the model parameter encoding after global optimization to the second update term to obtain the model parameter encoding after local optimization
[0135]
[0136] The embodiment of the present invention provides a second search strategy of mutual attraction within the neighborhood, which can enable each model parameter encoding to quickly move to a better position in its neighborhood, explore more solution spaces, and thus effectively improve the possibility of a global optimal solution and the convergence speed of the algorithm.
[0137] In the embodiment of the present invention, the third search strategy of the worst position exclusion learning is adopted to perform extended optimization on the model parameter encoding after local optimization to obtain the model parameter encoding after extended optimization, including:
[0138] Generate a third random number r3 between [0,1], and encode the model parameters after the local optimization by the third random number r3 After weighting, the weighted model parameter encoding is obtained in, Represents the model parameter encoding after the lth local optimization, l = 1, 2, ..., M;
[0139] Using weighted model parameter encoding Subtract the worst parameter encoding with the smallest fitness And take the absolute value of each dimension parameter as The second subtraction term C2 is obtained;
[0140] The fitness value f corresponding to the model parameter encoding after local optimization l t Subtract the fitness value corresponding to the worst parameter encoding Then, a preset noise value ε is added to obtain a second added term A2;
[0141] After dividing the second subtraction term C2 by the second addition term A2, and controlling by the step size control coefficient, the third update term Δ3 = K*(C2 / A2) for extended optimization is determined;
[0142] Add the model parameter encoding after local optimization to the third update term to obtain the model parameter encoding after extended optimization
[0143]
[0144] The third search strategy of worst-position exclusion learning provided in the embodiment of the present invention can enable the model parameter encodings in poor positions to quickly move toward the model parameter encodings in better positions, and enable the model parameter encodings in better positions to search in a better direction, thereby achieving further optimization of the model parameters, allowing the algorithm to avoid searching in these poor positions, and effectively improving the search speed.
[0145] When the fault warning result is a specific type of fault, the pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain the fault diagnosis results, including:
[0146] According to the working principle and physical characteristics of rotating equipment, establish the mechanism model of rotating equipment;
[0147] Input the real-time monitoring data of the rotating equipment into the mechanism model; wherein the real-time monitoring data includes voltage data, current data, temperature data, speed data and vibration data;
[0148] Based on the real-time monitoring data of the rotating equipment, the mechanism model of the rotating equipment is operated to determine the operation trajectory of each component of the rotating equipment;
[0149] The running tracks of the various components of the rotating equipment are analyzed to determine the fault type of the rotating equipment and obtain the fault diagnosis result.
[0150] Under different fault conditions, the operating trajectory of rotating equipment will be different. Therefore, the actual operating trajectory can be compared with the corresponding operating trajectory under different faults to determine the fault type of the rotating equipment.
[0151] Generate a fault intelligent diagnosis log based on the fault diagnosis result and real-time monitoring data, and transmit the fault intelligent diagnosis log to a device pre-set by the staff, including:
[0152] According to the fault diagnosis results and real-time monitoring data, a PDF version report is generated according to a preset template, and a fault intelligent diagnosis log is generated;
[0153] The fault intelligent diagnosis log is transmitted to the equipment pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
[0154] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.
Claims
1. A method for early warning and intelligent diagnosis of rotating equipment faults, characterized in that: include: Collect historical monitoring data corresponding to rotating equipment and historical fault types corresponding to the historical monitoring data; Constructing a rotating equipment fault warning model, and training the rotating equipment fault warning model using historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data, to obtain a trained rotating equipment fault warning model; The rotating equipment fault warning model is deployed on a platform designated by the staff, and real-time monitoring data is collected, and the real-time monitoring data is analyzed by the deployed rotating equipment fault warning model to obtain a fault warning result; wherein the fault warning result includes a specific type of fault or the absence of a fault; When the fault warning result is a specific type of fault, the pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain the fault diagnosis result; According to the fault diagnosis results and real-time monitoring data, an intelligent fault diagnosis log is generated, and the intelligent fault diagnosis log is transmitted to the equipment pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
2. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 1 is characterized in that: Collect historical monitoring data corresponding to rotating equipment and historical fault types corresponding to the historical monitoring data, including: Based on the preset data length N, at least the voltage data, current data, temperature data, speed data, vibration data and corresponding historical fault types of the rotating equipment at N consecutive time points are collected, and the collected voltage data, current data, temperature data, speed data and vibration data are formed into a data matrix to obtain historical monitoring data and historical fault types corresponding to the historical monitoring data; By using the above method, multiple pairs of historical monitoring data and historical fault types corresponding to the historical monitoring data are obtained.
3. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 1, characterized in that: Constructing a rotating equipment fault warning model, and using historical monitoring data corresponding to the rotating equipment and historical fault types corresponding to the historical monitoring data to train the rotating equipment fault warning model, to obtain a trained rotating equipment fault warning model, including: Use CNN to build a fault warning model for rotating equipment; Determine the minimum data length n of a single type of data and set the counter k=n; where n <N; According to the voltage data, current data, temperature data, speed data and vibration data in the historical monitoring data corresponding to the rotating equipment, the voltage data, current data, temperature data, speed data and vibration data at the most recent k time points are taken out to obtain sample data, and the historical fault type corresponding to the historical monitoring data is used as the sample label corresponding to the sample data; Each historical monitoring data is trimmed to obtain a number of sample data and sample labels corresponding to the sample data; Using a number of sample data and sample labels corresponding to the sample data to form sample data pairs, and dividing the sample data pairs into a training set and a validation set; The rotation equipment fault warning model is trained using a training set to obtain a rotation equipment fault warning model corresponding to a data length of k; The validation set is used to obtain the accuracy of the rotating equipment fault warning model corresponding to the data length k; Determine whether the count value of counter k is equal to N. If so, obtain Nn rotating equipment fault warning models. Otherwise, increase the count value of k by one and return to the sample data acquisition step. According to the accuracies corresponding to the Nn rotating equipment fault warning models, the rotating equipment fault warning model with the highest accuracy is determined, and the trained rotating equipment fault warning model is obtained.
4. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 3 is characterized in that: The rotation equipment fault warning model is trained using a training set to obtain a rotation equipment fault warning model corresponding to a data length of k, including: Randomly initializing the model parameters of the rotating equipment fault warning model between the upper limit and the lower limit of the parameters of the rotating equipment fault warning model to obtain the model parameter code; Based on the above method, the model parameter codes are obtained repeatedly to obtain multiple model parameter codes; For any model parameter code, apply the model parameter code to the rotating equipment fault warning model, and obtain the error function value of the rotating equipment fault warning model based on the training set, and obtain the fitness value corresponding to the model parameter code according to the error function value; According to the fitness values corresponding to the model parameter codes, determining the code with the largest fitness value as the optimal parameter code; Based on the optimal parameter encoding, a first search strategy of weighted and perturbation learning is used to perform global optimization on each model parameter encoding to obtain a model parameter encoding after global optimization; The second search strategy of mutual attraction within the neighborhood is used to locally optimize the model parameter encoding after global optimization, and the model parameter encoding after local optimization is obtained; The third search strategy of worst position exclusion learning is used to perform extended optimization on the model parameter encoding after local optimization to obtain the model parameter encoding after extended optimization; Repeat the above three optimization steps until the training times reach the maximum training times, re-obtain the optimal parameter encoding, and use the re-obtained optimal parameter encoding as the final parameter of the rotating equipment fault warning model to obtain the rotating equipment fault warning model corresponding to the data length k.
5. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 4 is characterized in that: Obtaining a fitness value corresponding to the model parameter encoding according to the error function value includes: Adding the error function value corresponding to the model parameter encoding to a preset noise value to obtain a first added term; Taking the reciprocal of the first added term to obtain a reciprocal term; After performing an exponential operation on the reciprocal term using a power function between (0,1), it is multiplied by a preset adjustment factor to obtain the fitness value corresponding to the model parameter encoding.
6. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 5, characterized in that: Based on the optimal parameter encoding, the first search strategy of weighted and perturbation learning is used to perform global optimization on each model parameter encoding to obtain the model parameter encoding after global optimization, including: The current number of training times is divided by the maximum number of training times to obtain the first division term; Subtract the first division term from the preset constant term to obtain an inertia weight that gradually decreases as the number of training times increases; The model parameter code is weighted by using an inertia weight to obtain a weighted model parameter code; Generate a first random number between [0, 1], and weight the optimal parameter code by squaring the first random number to obtain the weighted optimal parameter code; The weighted optimal parameter encoding is subtracted from the model parameter encoding, and then multiplied by the jump search step size and the fitness value corresponding to the optimal parameter encoding to obtain the first update item for global optimization; The weighted model parameter encoding is added to the first update term to obtain the model parameter encoding after global optimization.
7. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 5, characterized in that: The second search strategy of mutual attraction within the neighborhood is used to locally optimize the model parameter encoding after global optimization, and the model parameter encoding after local optimization is obtained, including: For the model parameter encoding after global optimization, a random neighborhood search direction is determined between (-1,1); Determine the target model parameter encoding that is closest to the model parameter encoding after global optimization and has a larger fitness value; The target model parameter code is subtracted from the model parameter code after global optimization to obtain the first subtraction term; Generate a second random number between [0, 1], and weight the first subtraction term by the second random number, and then multiply the weight by the fitness value corresponding to the target model parameter encoding to obtain a neighborhood search step length; Using a neighborhood search direction and a neighborhood search step size, determine a second update item for local optimization; The model parameter encoding after global optimization is added to the second update term to obtain the model parameter encoding after local optimization.
8. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 5, characterized in that: The third search strategy of worst position exclusion learning is used to expand and optimize the model parameter encoding after local optimization, and the model parameter encoding after expanded optimization is obtained, including: Generate a third random number between [0, 1], and weight the model parameter encoding after the local optimization by the third random number to obtain the weighted model parameter encoding; The worst parameter encoding with the smallest fitness is subtracted from the weighted model parameter encoding, and the absolute value of each dimension parameter is taken to obtain the second subtraction term; The fitness value corresponding to the model parameter encoding after local optimization is subtracted from the fitness value corresponding to the worst parameter encoding, and a preset noise value is added to obtain a second addition term; After dividing the second subtraction term by the second addition term, and controlling it with a step size control coefficient, a third update term for expanding the optimization is determined; The model parameter encoding after local optimization is added to the third update term to obtain the model parameter encoding after extended optimization.
9. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 1, characterized in that: When the fault warning result is a specific type of fault, the pre-built mechanism model corresponding to the rotating equipment is dispatched, and the mechanism model is used to perform fault diagnosis to obtain the fault diagnosis results, including: According to the working principle and physical characteristics of rotating equipment, establish the mechanism model of rotating equipment; Input the real-time monitoring data of the rotating equipment into the mechanism model; wherein the real-time monitoring data includes voltage data, current data, temperature data, speed data and vibration data; Based on the real-time monitoring data of the rotating equipment, the mechanism model of the rotating equipment is operated to determine the operation trajectory of each component of the rotating equipment; The running tracks of the various components of the rotating equipment are analyzed to determine the fault type of the rotating equipment and obtain the fault diagnosis result.
10. The method for early warning and intelligent diagnosis of rotating equipment faults according to claim 1, characterized in that: Generate a fault intelligent diagnosis log based on the fault diagnosis result and real-time monitoring data, and transmit the fault intelligent diagnosis log to a device pre-set by the staff, including: According to the fault diagnosis results and real-time monitoring data, a PDF version report is generated according to a preset template, and a fault intelligent diagnosis log is generated; The fault intelligent diagnosis log is transmitted to the equipment pre-set by the staff to complete the fault early warning and intelligent diagnosis of the rotating equipment.
Citation Information
Patent Citations
Rail transit fault diagnosis method and system based on time series analysis
CN104091070A
Machine tool equipment fault early warning method and system based on digital twinning
CN117407814A
Frequency converter fault identification method and monitoring device based on XGBoost fault diagnosis model
CN117421677A
Chemical process fault diagnosis method and system based on improved space-time model
CN118094314A
Self-diagnosis device, programmable system, and computer-readable storage medium
WO2024027781A1
