Intelligent oil and gas diagnosis method based on voiceprint recognition

Abstract

The invention relates to the technical field of voiceprint recognition, in particular to an intelligent oil and gas diagnosis method based on voiceprint recognition, and the method comprises the steps: carrying out the fault diagnosis of all equipment based on an operation voiceprint signal, and obtaining an equipment fault diagnosis result; under the condition that the equipment fault diagnosis result represents that the target equipment breaks down, a target node corresponding to the target equipment is determined in a fault propagation map, and a fault propagation path of the target equipment is predicted based on the weight of an edge from the target node to a next node in the fault propagation map, a fault risk value of the target node and a fault degree factor; and outputting a cluster fault diagnosis result corresponding to the oil and gas equipment cluster at least based on the equipment fault diagnosis result of the target equipment and the fault propagation path. According to the invention, the fault of the single-point equipment can be found in time, and the conduction path of the fault between the equipment can be quantified, so that the cascading fault reaction possibly caused by the fault of the single-point equipment can be found in time, and system-level oil gas fault diagnosis is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of voiceprint recognition, and particularly relates to an intelligent oil and gas diagnosis method based on voiceprint recognition. BACKGROUND

[0002] With the rapid development of the oil and gas industry, stable operation of oil and gas equipment is crucial to ensure energy supply. Traditional oil and gas equipment fault diagnosis methods mainly rely on manual inspection and simple instrument monitoring, which have the following shortcomings: manual inspection is inefficient and it is difficult to find potential faults in time; instrument monitoring data is limited and cannot fully reflect the equipment operating state. SUMMARY

[0003] To overcome the problems in the related art, the embodiments of the present application provide an intelligent oil and gas diagnosis method based on voiceprint recognition to solve the defects in the related art.

[0004] According to a first aspect of the embodiments of the present application, an intelligent oil and gas diagnosis method based on voiceprint recognition is provided, comprising:

[0005] Step S101, collecting operating voiceprint signals of each device in an oil and gas equipment cluster;

[0006] Step S102, performing fault diagnosis on the devices based on the operating voiceprint signals to obtain device fault diagnosis results;

[0007] Step S103, in the case that the device fault diagnosis results represent that a target device has a fault, determining a target node corresponding to the target device in a fault propagation graph, and predicting a fault propagation path of the target device based on a weight of an edge from the target node to a next node in the fault propagation graph, a fault risk value of the target node, and a fault degree factor, wherein the nodes of the fault propagation graph represent the devices in the oil and gas equipment cluster, the weight of the edge is obtained by dividing the number of associated faults from the target device to the device corresponding to the next node by the total number of faults of the target device, the fault risk value of the target node is 1, and the fault degree factor is obtained based on the average downtime of the target device;

[0008] Step S104, outputting a cluster fault diagnosis result corresponding to the oil and gas equipment cluster based at least on the device fault diagnosis result of the target device and the fault propagation path.

[0009] In one embodiment, the operating voiceprint signals are collected by sound sensors arranged at the devices, the fault diagnosis on the devices based on the operating voiceprint signals to obtain the device fault diagnosis results comprises:

[0010] The operation soundprint signals collected by each acoustic sensor are matrix processed from the time dimension to obtain a mixed signal matrix, and a spatial topology matrix is constructed based on the typical fault soundprint frequencies of each device and the relative position information of each acoustic sensor and each device. Signal separation processing is performed based on the mixed signal matrix and the spatial topology matrix by a fast independent component analysis algorithm to obtain single soundprint signals corresponding to each device respectively, wherein the objective function of the fast independent component analysis algorithm includes a noise suppression term for minimizing noise energy.

[0011] Fault diagnosis is performed based on the single soundprint signals corresponding to each device respectively to obtain a device fault diagnosis result.

[0012] In one embodiment, the matrix elements of the spatial topology matrix are obtained by the following calculation formula:

[0013]

[0014] wherein a xy represents the matrix element corresponding to acoustic sensor x and device y in the spatial topology matrix, d xy represents the straight-line distance from device y to acoustic sensor x, θ xy represents the angle between the main axis direction of device y and the line connecting acoustic sensor x, a represents a preset distance attenuation index, and β represents a preset frequency attenuation index, f k represents the typical fault soundprint frequency of device y.

[0015] In one embodiment, fault diagnosis is performed based on the single soundprint signals corresponding to each device respectively to obtain a device fault diagnosis result, including:

[0016] For each of the devices, Hilbert transform is performed on the single soundprint signal corresponding to the device to obtain a transformed soundprint signal, arctangent function calculation is performed based on the single soundprint signal and the transformed soundprint signal to obtain an instantaneous phase, phase difference is calculated based on the corresponding instantaneous phase within a preset time window to obtain a phase jump angle, and a fault diagnosis result representing that the device has failed is generated in the case where the phase jump angle exceeds a preset angle threshold.

[0017] In one embodiment, fault diagnosis is performed based on the single soundprint signals corresponding to each device respectively to obtain a device fault diagnosis result, including:

[0018] The single voiceprint signal corresponding to each of the devices is subjected to wavelet packet decomposition to obtain decomposition signals of different frequencies, and a plurality of target characteristic frequency bands are determined according to the type of the device and a preset correspondence relationship between characteristic frequency bands and device types, the decomposition signals belonging to each target characteristic frequency band are filtered from the decomposition signals, the in-band signal energy of each target characteristic frequency band is calculated, the in-band signal energies of the plurality of target characteristic frequency bands are weighted and summed to obtain a target frequency band energy value, and a fault diagnosis result indicating that the device has failed is generated if the target frequency band energy value is less than a preset energy threshold, wherein the weight of each target characteristic frequency band in the weighted sum is determined according to the importance, repair cost and signal-to-noise ratio of the corresponding target characteristic frequency band of the device.

[0019] In one embodiment, the weight of each target characteristic frequency band in the weighted sum is determined by:

[0020] The target importance coefficient of the device is obtained based on the type of the device and a preset correspondence relationship between importance coefficients and device types, and the repair cost coefficient of the device is obtained by dividing the historical average repair cost of the device by the total value of the device.

[0021] The detection sensitivity coefficient of the target characteristic frequency band is determined based on the signal-to-noise ratio of the target characteristic frequency band.

[0022] For each target characteristic frequency band, the target importance coefficient, the repair cost coefficient and the detection sensitivity coefficient of the target characteristic frequency band are multiplied and then normalized to obtain the weight of the target characteristic frequency band.

[0023] In one embodiment, the fault propagation path of the target device is predicted based on the weight of the edge from the target node to the next node in the fault propagation graph, the fault risk value of the target node and the fault degree factor, including:

[0024] The target node is taken as an initial fault node, and the following process is repeatedly executed:

[0025] The fault risk value of the device corresponding to the next node is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node and the fault degree factor, and if a preset stopping condition is not reached, the next node is taken as a new fault node, otherwise the repetition is stopped, wherein the preset stopping condition includes that the next node is a tail node or the number of repetitions reaches a preset number.

[0026] The risk nodes whose fault risk values are higher than a preset risk threshold are determined, and the fault propagation path of the target device is obtained based on the connection relationship between the risk nodes.

[0027] In an embodiment, the fault risk value of the next node corresponding device is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node, and a fault degree factor, including:

[0028] The fault risk value of the next node corresponding device is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node, and a fault degree factor according to the following calculation formula:

[0029]

[0030] wherein PR(v i ) represents the fault risk value of the next node v i corresponding device, γ represents a preset damping coefficient, PR(v j ) represents the fault risk value of the fault node v j , w ji represents the weight of the edge from the fault node v j to the next node v i , and L(v j ) represents the fault degree factor of the fault node v j .

[0031] In an embodiment, the cluster fault diagnosis result corresponding to the oil and gas equipment cluster is output based on at least the device fault diagnosis result of the target device and the fault propagation path, including:

[0032] The device fault diagnosis result of the target device, the fault propagation path, and the fault propagation graph are input into a large language model to obtain the cluster fault diagnosis result of the oil and gas equipment cluster output by the large language model, wherein the cluster fault diagnosis result includes the fault propagation path, the fault risk value of each node in the fault propagation path, and fault maintenance suggestions.

[0033] The technical scheme provided by the embodiments of the present application can include the following beneficial effects:

[0034] The intelligent oil and gas diagnosis method based on voiceprint recognition provided in the embodiments of the present application can collect the running voiceprint signals of each device in the oil and gas device cluster, then perform fault diagnosis on each device based on the running voiceprint signals to obtain a device fault diagnosis result, and in the case that the device fault diagnosis result represents that the target device has a fault, the fault propagation path of the target device can be predicted based on a fault propagation graph, so as to output a cluster fault diagnosis result corresponding to the oil and gas device cluster based on at least the target device and the fault propagation path. In this way, the automatic oil and gas fault diagnosis can be realized based on voiceprint recognition, the human participation in the oil and gas fault diagnosis process is reduced, and thus the efficiency of the oil and gas fault diagnosis is improved, and potential faults can be found in time. Moreover, the fault propagation path is predicted based on the fault propagation graph, not only the single-point device fault can be found in time, but also the conduction path of the fault among devices can be quantified, so that the chain fault reaction possibly caused by the single-point device fault can be found in time, and the system-level oil and gas fault diagnosis is realized. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0036] Figure 1 is a flowchart of an intelligent oil and gas diagnosis method based on voiceprint recognition proposed in the present application. DETAILED DESCRIPTION

[0037] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The following description is only one of the exemplary embodiments consistent with the present application. Therefore, it is not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of devices and methods consistent with some aspects of the present application.

[0038] The terms used in the present application are merely for the purpose of describing particular embodiments and are not intended to limit the present application. The singular forms "a," "an," and "the" used in the present application are also intended to include the plural forms unless the context clearly indicates otherwise.

[0039] As described in the background, the conventional oil and gas device fault diagnosis means mainly relies on manual inspection and simple instrument monitoring, which has the following disadvantages: the manual inspection is inefficient and it is difficult to find potential faults in time; the instrument monitoring data are limited and cannot fully reflect the device running state.

[0040] In addition, the existing voiceprint recognition diagnosis method usually focuses on single-point device failure. However, in the oil and gas scene, single-point device failure can cause a chain failure reaction. For example, centrifugal pump failure can trigger valve cavitation risk, and valve cavitation risk can further trigger pipeline rupture risk. According to the existing voiceprint recognition diagnosis method, only centrifugal pump failure can be identified, and the possible failure of the valve and the pipeline cannot be found in advance, that is, the chain failure reaction caused by single-point device failure cannot be found in time, thereby affecting the efficiency of oil and gas failure diagnosis.

[0041] Based on this, at least one embodiment of the present application provides an intelligent oil and gas diagnosis method based on voiceprint recognition, please refer to the accompanying Figure 1 which shows the flow of the method, including steps S101 to S104.

[0042] In step S101, the running voiceprint signals of each device in the oil and gas device cluster are collected.

[0043] In step S102, the failure of each device is diagnosed based on the running voiceprint signals, and the device failure diagnosis result is obtained.

[0044] In step S103, in the case that the device failure diagnosis result represents that the target device fails, the target node corresponding to the target device is determined in the failure propagation graph, and the failure propagation path of the target device is predicted based on the weight of the edge from the target node to the next node in the failure propagation graph, the failure risk value of the target node and the failure degree factor.

[0045] Among them, the nodes of the failure propagation graph represent the devices in the oil and gas device cluster, the weight of the edge is obtained by dividing the number of associated failures from the target device to the corresponding device of the next node by the total number of failures of the target device, the failure risk value is 1, and the failure degree factor is obtained based on the average downtime of the target device.

[0046] In step S104, at least based on the device failure diagnosis result and the failure propagation path of the target device, the cluster failure diagnosis result corresponding to the oil and gas device cluster is output.

[0047] Thus, automatic oil and gas failure diagnosis can be realized based on voiceprint recognition, reducing human participation in the oil and gas failure diagnosis process, thereby improving the efficiency of oil and gas failure diagnosis and discovering potential failures in time. Moreover, the failure propagation path is predicted based on the failure propagation graph, which not only can discover single-point device failure in time, but also can quantify the conduction path of failure between devices, thereby discovering the chain failure reaction caused by single-point device failure in time, and realizing system-level oil and gas failure diagnosis.

[0048] For example, in step S101, the running voiceprint signal is a voiceprint signal collected during the running of the device. Anti-electromagnetic interference MEMS (Micro-Electro-Mechanical Systems) acoustic sensor arrays can be deployed at key nodes of pumps, compressors and other devices in the oil and gas device cluster. Each sensor is built-in with a mechanical vibration filter to filter out pipeline fluid impact noise.

[0049] In some embodiments, in step S102, if the running voiceprint signal is collected by the acoustic sensors arranged at the devices, the running voiceprint signals collected by the acoustic sensors can be matrix processed from the time dimension to obtain a mixed signal matrix, and a spatial topology matrix can be constructed based on the typical fault voiceprint frequencies of the devices and the relative position information of the acoustic sensors and the devices. The signal separation processing can be performed based on the mixed signal matrix and the spatial topology matrix by a fast independent component analysis algorithm to obtain single voiceprint signals corresponding to the devices respectively. The fault diagnosis can be performed based on the single voiceprint signals corresponding to the devices respectively to obtain the device fault diagnosis result. The objective function of the fast independent component analysis algorithm includes a noise suppression term for minimizing the noise energy.

[0050] It should be understood that in the oil and gas station, the device cluster (such as pump group, compressor, valve) is densely distributed, and the voiceprint signals generated during the running of the devices are mixed in space to form a complex mixed acoustic field. For example, the signal collected by a single acoustic sensor can be composed of pump voiceprint, valve voiceprint and environmental noise. Therefore, the voiceprint collected by a single sensor is the superposition of signals of multiple devices, and if it is directly used for fault diagnosis, the accuracy of the fault diagnosis result will be affected.

[0051] To solve this problem, the embodiments of the present application are based on the fast independent component analysis (FastICA) algorithm, and physical constraints are added through a spatial topology matrix, so that the voiceprint is decoupled in real time, laying a foundation for subsequent accurate fault diagnosis.

[0052] For example, the mixed signal matrix is an m x 1 matrix, where m represents the number of acoustic sensors, i.e. each row of the mixed signal matrix represents a running voiceprint signal collected by an acoustic sensor.

[0053] For example, the spatial topology matrix is an m x n matrix, where n represents the number of devices. The rows of the spatial topology matrix represent the acoustic sensor dimension, for example, the xth row describes the mixed influence of all device voiceprints on the xth sensor; the columns of the spatial topology matrix represent the device dimension, for example, the yth column describes the transfer function of the voiceprint of device y to all acoustic sensors.

[0054] In some embodiments, the matrix elements of the spatial topology matrix are obtained by the following calculation formula:

[0055]

[0056] wherein a xy represents the matrix element in the spatial topology matrix corresponding to the sound sensor x and the device y, d xy represents the straight-line distance from the device y to the sound sensor x, θ xy represents the angle between the main axis direction of the device y and the line connecting the sound sensor x, a represents the preset distance attenuation index, β represents the preset frequency attenuation index, f k represents the typical fault acoustic fingerprint frequency of the device y.

[0057] For example, in the oil and gas pipeline environment, the preset distance attenuation index a can be defaulted to 1.8, and the preset frequency attenuation index β can be defaulted to 0.002. The typical fault acoustic fingerprint frequency of the device can be obtained based on historical fault data analysis.

[0058] For example, the main axis direction of the device is the main direction of the acoustic wave radiation generated by the device, which is jointly determined by the device type, mechanical structure and installation method. For example, the main axis direction of the centrifugal pump can be the axial direction along the driving shaft, the main axis direction of the regulating valve can be the medium flow direction, and the main axis direction of the oil pipeline can be the normal direction of the pipe wall.

[0059] It should be understood that the row of the spatial topology matrix can realize the long-term state tracking of a single device, support progressive fault diagnosis, the column of the spatial topology matrix can capture the transient correlation event of multiple devices, support burst fault positioning, and the overall spatial topology matrix can provide the basis for spatio-temporal joint analysis, empower device cluster collaborative diagnosis, realize real-time decoupling of acoustic fingerprints, thereby solving the acoustic fingerprint aliasing problem specific to the oil and gas scene, and laying a foundation for subsequent accurate fault diagnosis.

[0060] For example, the core calculation formula of the fast independent component analysis algorithm in the embodiments of the present application is:

[0061] X(t) = A·S(t) + N(t) (2)

[0062] wherein X(t) represents the mixed signal matrix, A represents the spatial topology matrix, S(t) is the single acoustic fingerprint matrix corresponding to the n devices, and N(t) represents the environmental noise vector.

[0063] Thus, the pure environmental noise can be measured when the device is shut down, then the noise statistics are calculated, and a noise suppression term -λ||N|| 2(λ is a preset noise suppression coefficient) to improve robustness by minimizing noise energy. Then, a constrained pseudo-inverse is constructed using the calibration noise to obtain a rough separated voiceprint signal. Finally, adaptive noise reduction, such as time-varying threshold filtering, can also be performed on the rough separated voiceprint signal to obtain a single voiceprint signal.

[0064] In some embodiments, based on the single voiceprint signals corresponding to each device respectively, device fault diagnosis is performed to obtain a device fault diagnosis result, including: for each of the devices, performing Hilbert transform on the single voiceprint signal corresponding to the device to obtain a transformed voiceprint signal, performing arctangent function calculation based on the single voiceprint signal and the transformed voiceprint signal to obtain an instantaneous phase, calculating a phase difference within a preset time window based on the corresponding instantaneous phase to obtain a phase jump angle, and in a case where the phase jump angle exceeds a preset angle threshold, generating a fault diagnosis result representing that the device has a fault.

[0065] It should be understood that the existing oil and gas diagnosis scheme of voiceprint recognition only focuses on the frequency domain amplitude and does not fully utilize the phase information of the voiceprint. However, research has found that the phase spectrum is more sensitive to early oil and gas faults. Therefore, the dynamic phase analysis performed by the embodiments of the present application can capture phase abnormalities of early faults, thereby solving the problem of difficult detection of early micro-faults and improving the sensitivity of oil and gas fault diagnosis.

[0066] For example, the arctangent function calculation based on the single voiceprint signal and the transformed voiceprint signal to obtain the instantaneous phase can be performed according to the following calculation formula:

[0067]

[0068] wherein φ(t) represents the instantaneous phase, represents the transformed voiceprint signal, and s(t) represents the single voiceprint signal.

[0069] It should be understood that the instantaneous phase obtained according to the calculation formula (3) is in radian measure, and it can also be converted into degree measure, thereby providing two kinds of outputs in radian measure and degree measure.

[0070] For example, the preset time window and the preset angle threshold can be set according to requirements, such as setting the preset time window to 10 ms and the preset angle threshold to 45°. Correspondingly, the phase jump angle is the phase difference between two instantaneous phases with an interval of 10 ms. If the phase jump angle is greater than 45°, a fault diagnosis result representing that the device has a fault is generated. If the phase jump angle is less than or equal to 45°, a fault diagnosis result representing that the device is operating normally is generated. In this way, the instantaneous mechanical impact can be captured through the phase jump, early micro-faults can be discovered in time, and the sensitivity of oil and gas fault diagnosis can be improved.

[0071] In some embodiments, the fault diagnosis is performed based on the single voiceprint signal corresponding to each device respectively, and a device fault diagnosis result is obtained, including: for each device in the devices, performing wavelet packet decomposition on the single voiceprint signal corresponding to the device to obtain a decomposition signal of different frequencies, and determining a plurality of target characteristic frequency bands according to the type of the device and a preset correspondence relationship between the characteristic frequency bands and the device types, screening the decomposition signal belonging to each target characteristic frequency band in the decomposition signal, calculating the in-band signal energy of each target characteristic frequency band, and performing weighted summation on the in-band signal energy of the plurality of target characteristic frequency bands to obtain a target frequency band energy value, and in a case where the target frequency band energy value is less than a preset energy threshold, generating a fault diagnosis result representing that the device has a fault, wherein the weight of the weighted summation is determined according to the importance of the device, the repair cost, and the signal-to-noise ratio of the corresponding target characteristic frequency band.

[0072] For example, the preset correspondence relationship between the characteristic frequency bands and the device types can be pre-configured based on historical data. For example, the characteristic frequency bands of a centrifugal pump include a base frequency band, a bearing fault band, a cavitation band, and a seal leakage band. The frequency range of the base frequency band is 0-1 kHz, the frequency range of the bearing fault band is 2-8 kHz, the frequency range of the cavitation band is 8-12 kHz, and the frequency range of the seal leakage band is 12-16 kHz. The characteristic frequency bands of a regulating valve include a flow noise band, a cavitation band, a micro-leakage band, and a mechanical vibration band. The frequency range of the flow noise band is 0-2 kHz, the frequency range of the cavitation band is 2-6 kHz, the frequency range of the micro-leakage band is 8-12 kHz, and the frequency range of the mechanical vibration band is greater than 12 kHz.

[0073] Thus, after determining the device type, the plurality of target characteristic frequency bands can be determined based on the preset correspondence relationship between the characteristic frequency bands and the device types. Then, the decomposition signal belonging to each target characteristic frequency band is screened in the decomposition signal. Next, for each target characteristic frequency band, the in-band signal energy of the target characteristic frequency band is calculated based on the decomposition signal of the target characteristic frequency band. Finally, the in-band signal energy of the plurality of target characteristic frequency bands is weighted and summed to obtain a target frequency band energy value. In a case where the target frequency band energy value is less than a preset energy threshold, a fault diagnosis result representing that the device has a fault is generated. In a case where the target frequency band energy value is greater than or equal to the preset energy threshold, a fault diagnosis result representing that the device is running normally is generated. The preset energy threshold can be set according to actual conditions, for example, it can be set to 1.2.

[0074] In some embodiments, the weight of each target characteristic frequency band in the weighted summation is determined as follows: based on the type of the device and a preset correspondence between the importance coefficient and the type of the device, a target importance coefficient of the device is obtained, and a historical average repair cost of the device is divided by a total value of the device to obtain a repair cost coefficient of the device; based on a signal-to-noise ratio of the target characteristic frequency band, a detection sensitivity coefficient is determined; for each target characteristic frequency band, the target importance coefficient, the repair cost coefficient and the detection sensitivity coefficient of the target characteristic frequency band are multiplied and then normalized to obtain the weight of the target characteristic frequency band.

[0075] For example, the preset correspondence between the importance coefficient and the type of the device can be configured based on the importance of the device. For example, the importance coefficient of a centrifugal pump, which is a key device, can be configured as 1.5, and the importance coefficient of an adjusting valve, which is a general device, can be configured as 0.8.

[0076] For example, the detection sensitivity coefficient can be determined based on the signal-to-noise ratio of the target characteristic frequency band according to the following calculation formula:

[0077]

[0078] wherein D represents the detection sensitivity coefficient of the target characteristic frequency band, and SNR represents the signal-to-noise ratio of the target characteristic frequency band.

[0079] It should be understood that for different frequency bands of the same device, the importance coefficient and the repair cost coefficient are fixed, and the detection sensitivity coefficient dynamically changes with the frequency band. Therefore, for each target characteristic frequency band, the same importance coefficient, the same repair cost coefficient and the detection sensitivity coefficient of the target characteristic frequency band are multiplied and then normalized to obtain the weight of the target characteristic frequency band. Thus, the importance coefficient and the repair cost coefficient can focus on the high-risk failure of the key device, and the detection sensitivity coefficient can strengthen the high signal-to-noise ratio frequency band and suppress the low signal-to-noise ratio interference, so that the weight in the frequency band energy value calculation process is more in line with the actual situation, thereby improving the accuracy of fault diagnosis.

[0080] It should be understood that in the oil and gas fault diagnosis process, the above phase jump detection or frequency band energy value calculation can be performed separately, or the phase jump detection and the frequency band energy value calculation can be combined, which is not limited in the present application. Among them, the phase jump is sensitive to micron-level structure damage, and the frequency band energy value can quantify the energy aggregation degree of the fault characteristic frequency, so that early micro-failure can be found in time.

[0081] For example, in step S103, the target node can be taken as an initial fault node, and the following process can be repeatedly performed: based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node, and the fault degree factor, the fault risk value of the corresponding device of the next node is predicted, if the preset stopping condition is not reached, the next node is taken as a new fault node, otherwise the repeated execution is stopped, wherein the preset stopping condition includes that the next node is a tail node or the number of repetitions reaches a preset number; the risk nodes with fault risk values higher than a preset risk threshold are determined, and based on the connection relationship between the risk nodes, the fault propagation path of the target device is obtained.

[0082] For example, the tail node is a node without a next node, and the preset number can be set according to actual conditions, for example, set to 10.

[0083] In some embodiments, based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node, and the fault degree factor, the fault risk value of the corresponding device of the next node is predicted, including: based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node, and the fault degree factor, the fault risk value of the corresponding device of the next node is predicted according to the following calculation formula:

[0084]

[0085] wherein PR(v i ) represents the fault risk value of the corresponding device of the next node v i , γ represents a preset damping coefficient, PR(v j ) represents the fault risk value of the fault node v j , w ji represents the weight of the edge from the fault node v j to the next node v i , and L(v j ) represents the fault degree factor of the fault node v j .

[0086] For example, the preset damping coefficient γ can be set according to actual conditions, for example, can be set to 0.85.

[0087] For example, the fault risk value of the initial fault node in the fault propagation graph is 1, and the fault risk value of the subsequent fault node can be calculated according to the calculation formula (5).

[0088] For example, the historical maintenance record contains a fault device and an associated device that subsequently fails, so the weight of each edge in the fault propagation graph can be calculated based on the historical maintenance record. For example, when the centrifugal pump fails, there are 8 times that cause the outlet valve to fail within 24 hours (a total of 10 times of centrifugal pump failure), so the weight of the edge from the centrifugal pump node to the outlet valve node in the fault propagation graph is: 0.8 ÷ 10 = 0.8. When the outlet valve fails, there are 5 times that cause the oil pipeline to fail within 24 hours (a total of 10 times of outlet valve failure), so the weight of the edge from the outlet valve node to the oil pipeline node in the fault propagation graph is: 0.5 ÷ 10 = 0.5.

[0089] For example, the fault degree factor can be obtained based on the average downtime of the device failure. For example, the average downtime of the centrifugal pump failure is 12 hours, so the fault degree factor is: 12 ÷ 24 = 0.5, and the average downtime of the outlet valve failure is 8 hours, so the fault degree factor is: 8 ÷ 24 ≈ 0.33.

[0090] Thus, the process connection between devices is converted into a fault propagation graph, and the weight is assigned based on historical data. When a fault of a certain device is detected, the fault propagation path can be predicted by combining the fault propagation graph and the above calculation formula (5), not only considering the fault propagation probability (i.e., the weight of the edge), but also considering the severity of the device failure (i.e., the fault degree factor), so that the fault risk value is more in line with the actual loss, thereby improving the accuracy of the fault propagation path and timely discovering the cascading failure reaction that may be caused by a single-point device failure, and realizing system-level oil and gas fault diagnosis.

[0091] In some embodiments, in step S104, the device fault diagnosis result and the fault propagation path of the target device can be output as a cluster fault diagnosis result corresponding to an oil and gas device cluster.

[0092] In some embodiments, in step S104, the device fault diagnosis result, the fault propagation path, and the fault propagation graph of the target device can be input into a large language model to obtain a cluster fault diagnosis result corresponding to an oil and gas device cluster output by the large language model, wherein the cluster fault diagnosis result includes the fault propagation path, the fault risk value of each node in the fault propagation path, and the fault maintenance suggestion.

[0093] For example, the large language model is used to generate a cluster fault diagnosis result corresponding to an oil and gas device cluster according to the input device fault information, the fault propagation path, and the fault propagation graph. The general large language model can be fine-tuned based on sample device fault diagnosis results, sample fault propagation paths, sample fault propagation graphs, and sample cluster fault diagnosis results to obtain the large language model in the present application, thereby improving the rationality of the cluster fault diagnosis result output by the large language model.

[0094] Therefore, not only the fault propagation path and the fault risk value of each node in the fault propagation path can be output, but also a maintenance strategy can be generated, so that the equipment with a higher fault risk value is preferentially maintained, and the efficiency of oil and gas fault processing is improved.

[0095] Through the intelligent oil and gas diagnosis method based on voiceprint recognition in the present application, dynamic modeling of a fault propagation graph, real-time decoupling of multiple device voiceprints, and detection of phase mutations of sound waves can be realized, so that the method is suitable for a high-density device cluster scene in the oil and gas industry, the efficiency of oil and gas fault diagnosis is improved, and potential faults can be found in time.

[0096] The preferred embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited to the specific details in the above-described embodiments. Within the technical concept of the present application, various simple modifications can be made to the technical solutions of the present application, and these simple modifications all belong to the protection scope of the present application.

[0097] In addition, it should be noted that each specific technical feature described in the above specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, various possible combinations are not described again in the present application.

[0098] Furthermore, any combination of the various different embodiments of the present application can also be made, as long as it does not deviate from the idea of the present application, and it should also be considered as disclosed in the present application.

Claims

1. A method for intelligent oil and gas diagnosis based on voiceprint recognition, characterized in that, The method comprises: Step S101, collecting operation voiceprint signals of each device in an oil and gas equipment cluster; Step S102, performing fault diagnosis on the devices based on the operation voiceprint signals to obtain device fault diagnosis results; Step S103, in the case that the device fault diagnosis results represent that a target device has failed, determining a target node corresponding to the target device in a fault propagation graph, and predicting a fault propagation path of the target device based on a weight of an edge from the target node to a next node in the fault propagation graph, a fault risk value of the target node, and a fault degree factor, wherein the nodes of the fault propagation graph represent the devices in the oil and gas equipment cluster, the weight of the edge is obtained by dividing the number of associated faults from the target device to the device corresponding to the next node by the total number of faults of the target device, the fault risk value of the target node is 1, and the fault degree factor is obtained based on the average downtime of the target device; Step S104, outputting a cluster fault diagnosis result corresponding to the oil and gas equipment cluster based on at least the device fault diagnosis result of the target device and the fault propagation path.

2. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 1, wherein, The operation voiceprint signals are collected by acoustic sensors arranged at the devices, and the device fault diagnosis results are obtained by performing fault diagnosis on the devices based on the operation voiceprint signals, which comprises: performing matrix processing on the operation voiceprint signals collected by the acoustic sensors in the time dimension to obtain a mixed signal matrix, constructing a spatial topology matrix based on the typical fault voiceprint frequencies of the devices and the relative position information between the acoustic sensors and the devices, and performing signal separation processing on the mixed signal matrix and the spatial topology matrix based on a fast independent component analysis algorithm to obtain single voiceprint signals corresponding to the devices respectively, wherein the objective function of the fast independent component analysis algorithm comprises a noise suppression term for minimizing noise energy; performing fault diagnosis on the single voiceprint signals corresponding to the devices respectively to obtain the device fault diagnosis results.

3. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 2, wherein, The matrix elements of the spatial topology matrix are obtained by the following calculation formula: wherein a xy represents the matrix element corresponding to the sound sensor x and the device y in the spatial topology matrix, d xy represents the straight-line distance from the device y to the sound sensor x, θ xy represents the angle between the main axis direction of the device y and the line connecting the device y and the sound sensor x, α represents the preset distance attenuation index, β represents the preset frequency attenuation index, f k represents the typical fault soundprint frequency of the device y.

4. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 2, wherein, performing fault diagnosis on the single voiceprint signals corresponding to the devices respectively to obtain the device fault diagnosis results, which comprises: for each of the devices, performing Hilbert transform on the single voiceprint signal corresponding to the device to obtain a transformed voiceprint signal, performing arctangent function calculation based on the single voiceprint signal and the transformed voiceprint signal to obtain an instantaneous phase, calculating a phase difference based on the corresponding instantaneous phase within a preset time window to obtain a phase jump angle, and in the case that the phase jump angle exceeds a preset angle threshold, generating a fault diagnosis result representing that the device has failed.

5. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 2, wherein, performing fault diagnosis on the single voiceprint signals corresponding to the devices respectively to obtain the device fault diagnosis results, which comprises: The single voiceprint signal corresponding to each of the devices is subjected to wavelet packet decomposition to obtain decomposition signals of different frequencies, and a plurality of target characteristic frequency bands are determined according to the type of the device and a preset correspondence relationship between characteristic frequency bands and device types, the decomposition signals belonging to each target characteristic frequency band are filtered from the decomposition signals, the in-band signal energy of each target characteristic frequency band is calculated, the in-band signal energies of the plurality of target characteristic frequency bands are weighted and summed to obtain a target frequency band energy value, and a fault diagnosis result indicating that the device has failed is generated if the target frequency band energy value is less than a preset energy threshold, wherein the weight of the weighted sum is determined according to the importance, repair cost and signal-to-noise ratio of the corresponding target characteristic frequency band of the device.

6. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 5, wherein, The weight of each target characteristic frequency band in the weighted sum is determined in the following manner: A target importance coefficient of the device is obtained based on the type of the device and a preset correspondence relationship between importance coefficients and device types, and a repair cost coefficient of the device is obtained by dividing the historical average repair cost of the device by the total value of the device; A detection sensitivity coefficient is determined based on the signal-to-noise ratio of the target characteristic frequency band; For each target characteristic frequency band, the target importance coefficient, the repair cost coefficient and the detection sensitivity coefficient of the target characteristic frequency band are multiplied and then normalized to obtain the weight of the target characteristic frequency band.

7. The voiceprint recognition based intelligent oil and gas diagnostic method according to any one of claims 1-6, characterized in that, The fault propagation path of the target device is predicted based on the weight of the edge from the target node to the next node in the fault propagation graph, the fault risk value of the target node and the fault degree factor, including: The target node is taken as an initial fault node, and the following process is repeatedly executed: The fault risk value of the device corresponding to the next node is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node and the fault degree factor, and if a preset stopping condition is not reached, the next node is taken as a new fault node, otherwise the repetition is stopped, wherein the preset stopping condition includes that the next node is a tail node or the number of repetitions reaches a preset number; The risk nodes with fault risk values higher than a preset risk threshold are determined, and the fault propagation path of the target device is obtained based on the connection relationship between the risk nodes.

8. The voiceprint recognition based intelligent oil and gas diagnostic method of claim 7, wherein, The fault risk value of the device corresponding to the next node is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node and the fault degree factor, including: The fault risk value of the device corresponding to the next node is predicted based on the weight of the edge from the fault node to the next node in the fault propagation graph, the fault risk value of the fault node and the fault degree factor according to the following calculation formula: wherein PR(v i ) represents a failure risk value of a next node v i , γ represents a preset damping coefficient, PR(v j ) represents a failure risk value of a failure node v j , w ji represents a weight of an edge from the failure node v j to a next node v i , and L(v j ) represents a failure degree factor of the failure node v j .

9. The voiceprint recognition based intelligent oil and gas diagnostic method according to any one of claims 1-6, characterized in that, At least based on the device fault diagnosis result of the target device and the fault propagation path, a cluster fault diagnosis result corresponding to the oil and gas equipment cluster is output, including: The device fault diagnosis result of the target device, the fault propagation path, and the fault propagation graph are input into a large language model to obtain a cluster fault diagnosis result corresponding to the oil and gas equipment cluster output by the large language model, wherein the cluster fault diagnosis result includes the fault propagation path, and a fault risk value and a fault maintenance suggestion of each node in the fault propagation path.

Images