A centrifugal pump fault diagnosis method and system based on multi-source data fusion

Abstract

The application is suitable for the technical field of fault diagnosis, and specifically provides a centrifugal pump fault diagnosis method and system based on multi-source data fusion. In the diagnosis method, multi-source heterogeneous data of the centrifugal pump is collected, associated features are extracted to construct multi-source fusion feature units corresponding to each operation stage, and a state feature chain capable of continuously representing working condition evolution is formed. A nonlinear coupling mapping relationship between each feature is established to form a multi-source coupling response network reflecting the fault evolution direction, a fault risk index sequence corresponding to the current operation state is generated, and a fault feature mode library established in advance is matched to determine fault attribute information and output corresponding diagnosis information. The rationality, accuracy and foresight of the centrifugal pump fault diagnosis are greatly improved, the maintenance personnel can scientifically formulate a maintenance plan, and the precise control of operation and maintenance cost and the effective extension of the service life of the centrifugal pump equipment are realized.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, and in particular relates to a method and system for diagnosing centrifugal pump faults based on multi-source data fusion. Background Technology

[0002] In the field of centrifugal pump fault diagnosis, multi-source data fusion methods are widely used. These methods integrate multi-dimensional monitoring data such as pressure, flow rate, temperature, and vibration during centrifugal pump operation to identify faults. Traditional multi-source data fusion methods mostly treat each dimension of data as evidence generated independently at the same time, neglecting the inherent correlation between them. In reality, early faults in centrifugal pumps often begin with a single physical quantity and then gradually propagate to other parameters. For example, localized cavitation first disturbs the pressure, then affects the flow rate, subsequently triggering vibration and temperature changes—a sequential and causal transmission process.

[0003] However, traditional methods fragment this transmission process, failing to fully explore and reflect the centrifugal pump failure evolution process. This results in insufficient rationality and accuracy of data fusion. When diagnosing faults, existing methods often rely on the threshold exceeding of a single characteristic parameter. They are insensitive to subtle shifts in the coupling relationships between various characteristic parameters in the early stages of a fault, such as transmission delay and response gain. This often fails to capture early fault signals in a timely manner, leading to delayed fault warnings and missed optimal maintenance opportunities. Furthermore, when outputting results, only the fault type can be output, without accurately determining the specific evolution stage of the fault. In other words, it is impossible to determine which stage the fault has developed to. Maintenance personnel cannot arrange maintenance plans accordingly, which is not conducive to achieving precise control of maintenance costs and extending equipment lifespan. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for diagnosing centrifugal pump faults based on multi-source data fusion, in order to solve the technical problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution.

[0006] According to an embodiment of the present invention, a method for diagnosing centrifugal pump faults based on multi-source data fusion is provided, comprising the following steps:

[0007] Based on the multi-source synchronous state sequence, the associated features representing load change, energy transfer, thermal response and mechanical disturbance are extracted, and multi-source fusion feature units corresponding to each operating stage are constructed to form a state feature chain that can continuously represent the evolution of the operating condition.

[0008] Based on the state feature chain, a nonlinear coupling mapping relationship between each feature is established, the causal transmission path of disturbance between features is identified, and a multi-source coupled response network reflecting the direction of fault evolution is formed.

[0009] Based on the multi-source coupled response network, the characteristic offset amplitude, inter-feature transmission delay and response gain at continuous time are jointly analyzed to extract abnormal evolution components that deviate from the normal coupling law and generate a fault risk index sequence corresponding to the current operating state.

[0010] Based on the fault risk index sequence, the abnormal evolution component is matched with a pre-established fault feature pattern library to determine fault attribute information.

[0011] Based on the fault attribute information, corresponding diagnostic information is output, including warning level and / or trend prediction results.

[0012] Furthermore, by collecting multi-source heterogeneous data from centrifugal pumps and performing preprocessing, a time-aligned multi-source synchronous state sequence is formed.

[0013] in:

[0014] Multi-source heterogeneous data includes at least two of the following: vibration data, noise data, current data, temperature data, pressure data, and flow rate data; the preprocessing includes:

[0015] Unified noise reduction processing based on variational mode decomposition is performed on the signals of each channel;

[0016] Using the vibration signal as a reference sequence, a dynamic time warping algorithm is used to align the noise-reduced signals of each channel in time, forming a multi-source synchronous state sequence that corresponds to the sampling time and retains the fault propagation delay information.

[0017] Furthermore, the extraction of associated features and the construction of state feature chains specifically include:

[0018] Based on the dynamic time-series slice window and sliding step size, the multi-source synchronization state sequence is divided into multiple consecutive time-series state segments;

[0019] For each temporal state segment, an associated feature group is extracted. These associated feature groups are then arranged chronologically to form a state feature chain consisting of N multi-source fused feature units. The associated feature group includes:

[0020] Flow combined characteristics are used to characterize the pressure of load changes, including the position coordinates of the pump operating point on the HQ performance curve and the instantaneous velocity of movement;

[0021] Vibration energy coupling characteristics are used to characterize pressure pulsations that transmit energy, including the ratio coefficient of the frequency band energy of the pressure pulsation signal to the corresponding frequency band energy of the vibration signal.

[0022] Flow dynamics, used to characterize the temperature of the thermal response, including the hysteresis correlation coefficient between the rate of temperature change and the rate of flow change;

[0023] Noise coherence characteristics are used to characterize the vibration of mechanical disturbances, including the peak values ​​of the coherence function of the vibration and noise signals in the characteristic frequency band.

[0024] Furthermore, the construction of the multi-source coupled response network includes:

[0025] An initial topology network is constructed using each associated feature in the state feature chain as a network node and the physical causal relationship between the associated features as a directed edge. The direction of the directed edge represents the direction of transmission of fault disturbance between physical quantities, including at least the first transmission path from the pressure feature node to the vibration feature node and the second transmission path from the flow feature node to the temperature feature node.

[0026] Calculate the normalized propagation delay and dynamic response gain of each directed edge, and label them as attributes of the directed edges in the multi-source coupled response network.

[0027] Furthermore, the calculation of the normalized transmission delay specifically includes:

[0028] Cross-correlation analysis is performed on the two associated feature sequences connected by the directed edge. The time shift corresponding to the peak value of the cross-correlation function is taken as the current transmission delay of the directed edge. The ratio of the current transmission delay to the reference transmission delay under the device health baseline state is taken as the normalized transmission delay.

[0029] Furthermore, the calculation of the dynamic response gain specifically includes:

[0030] For two associated feature sequences connected by a directed edge, a transfer function is established, and the average amplitude gain within a preset fault feature frequency band is calculated as the current response gain. The ratio of the current response gain to the reference response gain under the device health reference state is used as the dynamic response gain.

[0031] Furthermore, the extraction of the anomalous evolutionary components specifically includes:

[0032] For each directed edge in the multi-source coupled response network, a coupling offset vector is defined, the elements of which include the difference between the normalized propagation delay and 1, and the difference between the dynamic response gain and 1.

[0033] The weighted summation of each coupling offset vector yields the overall network anomaly evolution component at the current moment.

[0034] The network's overall abnormal evolution components, arranged in chronological order across multiple consecutive moments, constitute the fault risk indicator sequence; wherein, the morphological characteristics of the fault risk indicator sequence include growth slope, acceleration, and current amplitude.

[0035] Furthermore, the fault feature pattern library stores typical topological change patterns of multi-source coupled response networks and typical growth curve shapes of fault risk index sequences under different evolution stages for different fault types.

[0036] Furthermore, the determination of fault attribute information includes:

[0037] The fault type is identified based on the location and orientation of the directed edges that have undergone significant shifts in the multi-source coupled response network;

[0038] Based on the growth slope, acceleration, and current amplitude of the fault risk indicator sequence, the typical growth curve shape in the fault characteristic pattern library is matched to determine the current evolution stage of the fault; wherein, the evolution stage includes at least the initial stage, the development stage, and the near-failure stage.

[0039] According to another embodiment of the present invention, a centrifugal pump fault diagnosis system based on multi-source data fusion is provided, comprising:

[0040] The state feature chain construction module is used to extract related features representing load changes, energy transfer, thermal response, and mechanical disturbances based on multi-source synchronous state sequences, construct multi-source fusion feature units corresponding to each operating stage, and form a state feature chain that can continuously represent the evolution of the operating conditions.

[0041] The coupled response network construction module is used to establish a nonlinear coupling mapping relationship between each associated feature based on the state feature chain, identify the causal transmission path of disturbances between features, and form a multi-source coupled response network that reflects the direction of fault evolution.

[0042] The abnormal evolution analysis module is used to perform joint analysis on the feature offset amplitude, inter-feature transmission delay and response gain at continuous time based on the multi-source coupled response network, extract abnormal evolution components that deviate from the normal coupling law, and generate a fault risk index sequence corresponding to the current operating state.

[0043] The fault mode matching module is used to match the abnormal evolution component with a pre-established fault feature pattern library based on the fault risk index sequence to determine fault attribute information; the fault attribute information includes fault type and fault evolution stage;

[0044] The diagnostic output module is used to output corresponding diagnostic information based on the fault attribute information, including warning level and / or trend prediction results.

[0045] Compared with existing technologies, the beneficial effects of the centrifugal pump fault diagnosis method and system based on multi-source data fusion of this invention are:

[0046] This invention effectively solves the problem that traditional multi-source data fusion methods process data from different dimensions independently and cannot reflect the causal transmission and evolution process of centrifugal pump failures by constructing time-aligned multi-source synchronous state sequences and forming a continuous state feature chain that characterizes the evolution of operating conditions. It relies on a multi-source coupled response network to achieve accurate perception of subtle shifts in coupling relationships such as transmission delay and response gain between features, and breaks through the limitations of single feature threshold criteria being insensitive to early faults and having delayed early warnings, so as to capture early fault signals in a timely manner.

[0047] This invention, by matching fault risk index sequences with a fault feature pattern library, can accurately determine the specific evolution stage of a fault based on the identification of fault types. Combined with the output of early warning levels and trend prediction results, it significantly improves the rationality, accuracy, and foresight of centrifugal pump fault diagnosis, making it easier for maintenance personnel to scientifically formulate maintenance plans, achieve precise control of maintenance costs, and effectively extend the service life of centrifugal pump equipment. Attached Figure Description

[0048] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0049] In the attached diagram:

[0050] Figure 1 This is a flowchart illustrating the implementation of a centrifugal pump fault diagnosis method based on multi-source data fusion according to the present invention.

[0051] Figure 2 This is a sub-flowchart of a centrifugal pump fault diagnosis method based on multi-source data fusion according to the present invention;

[0052] Figure 3 This is another sub-flowchart of a centrifugal pump fault diagnosis method based on multi-source data fusion according to the present invention;

[0053] Figure 4 This is a structural block diagram of a centrifugal pump fault diagnosis system based on multi-source data fusion according to the present invention. Detailed Implementation

[0054] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0055] In one embodiment of the present invention, a centrifugal pump fault diagnosis method based on multi-source data fusion is provided. The method aims to identify the causal transmission and dynamic coupling evolution relationship of centrifugal pump faults among multiple physical parameters, improve the rationality and accuracy of data fusion, achieve accurate identification of the specific evolution stage of the fault, and improve the timeliness of early fault warning.

[0056] For details, please refer to Figure 1 The centrifugal pump fault diagnosis method based on multi-source data fusion provided in this embodiment of the invention includes the following steps:

[0057] Step S101: Extract the associated features representing load change, energy transfer, thermal response and mechanical disturbance based on the multi-source synchronous state sequence, construct the multi-source fusion feature units corresponding to each operating stage, and form a state feature chain that can continuously represent the evolution of the working condition.

[0058] The multi-source synchronization state sequence of this invention is obtained by: collecting multi-source heterogeneous data from centrifugal pumps and preprocessing it to form a time-aligned multi-source synchronization state sequence; wherein:

[0059] The multi-source heterogeneous data in this embodiment includes at least two of the following: vibration data, noise data, current data, temperature data, pressure data, and flow rate data. Specifically, various types of sensors are deployed at key locations on the centrifugal pump to collect multi-source heterogeneous data, including vibration, noise, current, temperature, pressure, and flow rate. The collected raw data requires preprocessing to improve the accuracy and reliability of subsequent analysis.

[0060] Furthermore, in one implementation of the present invention, the preprocessing of multi-source heterogeneous data specifically includes: performing unified noise reduction processing on the signals of each channel based on variational mode decomposition; using the vibration signal as a reference sequence, using a dynamic time warping algorithm to perform time-series alignment on the noise-reduced signals of each channel to form a multi-source synchronous state sequence that corresponds to the sampling time and retains the fault propagation delay information.

[0061] In the unified noise reduction processing of each channel signal based on variational mode decomposition in this invention, each channel signal is adaptively decomposed into multiple modal components, the permutation entropy of each modal component is calculated, noise components with permutation entropy higher than a preset threshold are removed, and the remaining modal components are reconstructed into the noise-reduced signal.

[0062] Specifically, to preserve the temporal correlation characteristics between the signals in each channel, this embodiment employs Variational Mode Decomposition (VMD) for unified noise reduction of all channel signals. VMD adaptively decomposes the original signal into multiple mode components with different center frequencies. For each decomposed mode component, its permutation entropy is calculated to assess the degree of randomness. The higher the permutation entropy, the more significant the noise attribute. Noise components with permutation entropy higher than a preset threshold are removed, and the remaining mode components are summed and reconstructed to obtain the denoised signal. Compared with traditional channel-specific filtering, VMD unified noise reduction avoids the phase differences introduced by different filters and preserves the true phase relationship between signals.

[0063] Since the sampling frequencies of each sensor are different, the noise-reduced signals need to be time-aligned. This embodiment uses the Dynamic Time Warping (DTW) algorithm: using the vibration signal as a reference sequence, the optimal warping path is found through dynamic programming, mapping the signals of the other channels onto the time axis of the reference sequence. DTW alignment can unify the time base while preserving the inherent hysteresis information of fault disturbances propagating between different physical quantities, providing a real data basis for subsequent calculations of propagation delays.

[0064] The time-aligned signals of each channel are organized into a multi-source synchronous state sequence according to the sampling time, and each column represents a snapshot of the state of all physical quantities at the same time.

[0065] Specifically, such as Figure 2 As shown, the extraction of associated features and the construction of the state feature chain specifically include:

[0066] Step S201: Divide the multi-source synchronization state sequence into multiple consecutive time-series state segments according to the dynamic time-series slice window and sliding step size;

[0067] Step S202: For each time-series state segment, extract the associated feature groups, and arrange the associated feature groups extracted from each time-series state segment in chronological order to form a state feature chain composed of N multi-source fusion feature units;

[0068] Specifically, a dynamic timing slice window Win and a sliding step size step are defined to divide the multi-source synchronization state sequence into K consecutive timing state segments, represented as: ;in, This indicates rounding down for each timing state segment. It contains multi-channel data sampled continuously by Win. For each time-series state segment Extract the following associated feature groups to form a multi-source fusion feature unit. ;

[0069] in:

[0070] The associated feature group includes a flow-flow joint feature for characterizing pressure changes under load changes. In the flow-flow joint feature, the position coordinates and instantaneous speed of the pump operating point on the head-flow (HQ) performance curve are determined by calculating the average outlet pressure and average flow rate within the time segment, which is used to characterize the dynamic changes in load.

[0071] The associated feature group also includes vibration energy coupling features for characterizing pressure pulsations, specifically by calculating the ratio coefficient between the frequency band energy of the pressure pulsation signal and the corresponding frequency band energy of the vibration signal, which characterizes the efficiency of hydraulic energy transfer to mechanical vibration.

[0072] The associated feature group also includes flow dynamics features of temperature used to characterize the thermodynamic response, specifically including the hysteresis correlation coefficient between the rate of temperature change and the rate of flow change, used to characterize the response delay of the thermodynamic process to changes in flow.

[0073] The associated feature group also includes noise coherence features of vibration used to characterize mechanical disturbances. These features use the peak values ​​of the coherence functions of vibration and noise signals in the characteristic frequency band to characterize the disturbance state of the mechanical structure.

[0074] Finally, in this embodiment, the associated feature groups extracted from each temporal state segment are arranged in chronological order to form a state feature chain consisting of N multi-source fusion feature units.

[0075] Please continue to refer to Figure 1 The centrifugal pump fault diagnosis method based on multi-source data fusion in this embodiment of the invention further includes:

[0076] Step S102: Based on the state feature chain, establish the nonlinear coupling mapping relationship between each feature, identify the causal transmission path of the disturbance between features, and form a multi-source coupled response network that reflects the direction of fault evolution;

[0077] For further details, please refer to Figure 3 The construction of the multi-source coupled response network specifically includes the following steps:

[0078] Step S301: Construct an initial topology network with each associated feature in the state feature chain as a network node and the physical causal relationship between the associated features as a directed edge; wherein, the direction of the directed edge represents the direction of transmission of fault disturbance between physical quantities, including at least the first transmission path from the pressure feature node to the vibration feature node, and the second transmission path from the flow feature node to the temperature feature node.

[0079] In this embodiment, the network node includes:

[0080] Node N1: Load variation characteristics (operating point location);

[0081] Node N2: Pressure pulsation-vibration energy coupling characteristics;

[0082] Node N3: Temperature-flow hysteresis correlation coefficient;

[0083] Node N4: Vibration-noise coherent characteristics.

[0084] Define the following directed edges:

[0085] Edge e1→2: From N1 to N2, characterizing the effect of load change on pressure-vibration energy transfer;

[0086] Edge e2→4: From N2 to N4, it represents the transmission of abnormal pressure pulsation to mechanical vibration and noise;

[0087] Edge e2→3: from N2 to N3, characterizing the effect of energy loss on the thermodynamic response;

[0088] Step S302: Calculate the normalized propagation delay and dynamic response gain of each directed edge, and label them as attributes of the directed edge in the multi-source coupled response network.

[0089] In one specific implementation of the present invention, the calculation of normalized transmission delay specifically includes: performing cross-correlation analysis on two associated feature sequences connected by a directed edge, taking the time shift corresponding to the peak value of the cross-correlation function as the current transmission delay of the directed edge, and taking the ratio of the current transmission delay to the reference transmission delay under the device health reference state as the normalized transmission delay.

[0090] In one specific implementation of the present invention, the calculation of dynamic response gain specifically includes: establishing a transfer function for two associated feature sequences connected by a directed edge, calculating the average amplitude gain within a preset fault feature frequency band as the current response gain, and using the ratio of the current response gain to the reference response gain under the device health reference state as the dynamic response gain.

[0091] Specifically, in this embodiment of the invention, for each directed edge connecting two associated feature sequences, the normalized propagation delay is calculated; where: taking edge e2→4 as an example, the pressure pulsation-vibration energy coupling feature sequence is denoted as... The vibration-noise coherent feature sequence is ,calculate and cross-correlation function , represented as:

[0092]

[0093] in, The time lag variable represents the sequence. Relative to sequence The time offset, whose value range is preset based on the physical characteristics of centrifugal pump fault propagation. In this embodiment, The value is 2 seconds; This represents the mathematical expectation operation, which calculates the mean over time t, where T is the total duration of the feature sequence.

[0094] This invention takes the cross-correlation function The lag time corresponding to the maximum value As the transmission delay at the current moment , represented as: ; where argmax represents the operation of taking the value of the independent variable when the function reaches its maximum value.

[0095] Under the equipment health baseline condition, the baseline propagation delay is calculated using the same method as described above. The baseline data is taken from the continuous operation period during which the equipment was confirmed to be operating normally in the early stage of commissioning;

[0096] Furthermore, in this embodiment, the normalized propagation delay D of a directed edge is defined as the ratio of the current propagation delay to the reference propagation delay, expressed as: The normalized propagation delay D has the following meanings: when D≈1, it means that the propagation delay of the propagation path is within the normal range; when D<1, it means that the propagation speed of the disturbance on the path is faster than normal, which may be a sign of abnormally high energy transfer efficiency in the early stage of the fault; when D>1, it means that the propagation speed is slower, which may indicate that the propagation path is blocked or the damping is increased.

[0097] Please continue to refer to Figure 1 The centrifugal pump fault diagnosis method based on multi-source data fusion in this embodiment of the invention further includes:

[0098] Step S103: Based on the multi-source coupled response network, perform joint analysis on the feature offset amplitude, inter-feature transmission delay and response gain at consecutive time intervals, extract abnormal evolution components that deviate from the normal coupling law, and generate a fault risk index sequence corresponding to the current operating state.

[0099] Furthermore, the extraction step of the anomalous evolution component in this invention specifically includes:

[0100] For each directed edge in the multi-source coupled response network, a coupling offset vector is defined, the elements of which include the difference between the normalized propagation delay and 1, and the difference between the dynamic response gain and 1.

[0101] The weighted summation of each coupling offset vector yields the overall network anomaly evolution component at the current moment.

[0102] The network's overall abnormal evolution components, arranged in chronological order across multiple consecutive moments, constitute the fault risk indicator sequence; wherein, the morphological characteristics of the fault risk indicator sequence include growth slope, acceleration, and current amplitude.

[0103] Specifically, this invention applies to each directed edge in a multi-source coupled response network. Define its coupling offset vector V i , represented as: This vector quantifies the direction and magnitude of the offset of the i-th transmission path relative to the healthy baseline at the current moment, where the first component D of the vector... i -1 indicates the degree of offset in transmission delay, and the first component A of the vector i -1 indicates the degree of shift in response gain;

[0104] Furthermore, this invention performs a weighted summation of the coupling offset vectors of each directed edge in the network to obtain the overall abnormal evolution component of the network at the current moment, expressed as: Where M is the total number of directed edges in the multi-source coupled response network, w i The weight coefficient is the preset weight coefficient for the i-th directed edge. The weight coefficient is preset according to the importance of the propagation path in the fault evolution process.

[0105] For example, for cavitation faults, the weight of the edge pointing from pressure characteristics to vibration characteristics can be set to 0.5, and the weight of the edge pointing from flow rate characteristics to temperature characteristics can be set to 0.2. i Let be the coupling offset vector of the i-th directed edge. Represents vector V i The L2 norm.

[0106] Through the above calculations, the abnormal evolution component E comprehensively reflects the overall deviation of the entire multi-source coupled response network from the healthy baseline state at the current moment; among them, the larger the value of E, the more significant the deviation between the current operating state of the centrifugal pump and the normal coupling law.

[0107] Please continue to refer to Figure 1 The centrifugal pump fault diagnosis method based on multi-source data fusion in this embodiment of the invention further includes:

[0108] Step S104: Based on the fault risk index sequence, match the abnormal evolution component with the pre-established fault feature pattern library to determine fault attribute information;

[0109] Step S105: Output corresponding diagnostic information based on the fault attribute information, the diagnostic information including warning level and / or trend prediction results.

[0110] Furthermore, the fault feature pattern library stores typical topological change patterns of multi-source coupled response networks and typical growth curve shapes of fault risk index sequences under different evolution stages for different fault types.

[0111] Furthermore, the determination of fault attribute information includes:

[0112] The fault type is identified based on the location and orientation of the directed edges that have undergone significant shifts in the multi-source coupled response network;

[0113] Based on the growth slope, acceleration, and current amplitude of the fault risk indicator sequence, the typical growth curve shape in the fault characteristic pattern library is matched to determine the current evolution stage of the fault; wherein, the evolution stage includes at least the initial stage, the development stage, and the near-failure stage.

[0114] Specifically, a fault characteristic pattern library is pre-established, constructed using historical fault data and accelerated life test data. The content stored in the pattern library includes:

[0115] Network topology change patterns under different fault types at different evolution stages.

[0116] Take cavitation failure as an example:

[0117] In the initial stage: the normalized propagation delay D of edge e2→4 begins to decrease, from 1.0 to about 0.8, the dynamic response gain A increases slightly, from 1.0 to about 1.2, and the changes of other edges are not significant.

[0118] Development phase: D for edge e2→4 further decreases to 0.5-0.6, A increases to 1.5-1.8, edge e1→2 begins to shift, D decreases to around 0.9.

[0119] Nearing the failure period: Multiple edges show significant shifts simultaneously, the D of edge e2→4 is below 0.4 or shows unstable fluctuations, and the A of edge e2→3 increases significantly, indicating abnormal thermal response.

[0120] Typical growth curve shapes for different fault types at different evolution stages.

[0121] The growth curve shape of the fault risk indicator sequence:

[0122] Early stage: The abnormal evolutionary component E shows a slow linear growth, with a growth slope S < 0.1 / hour;

[0123] Development stage: E shows accelerated growth, with a growth slope of 0.1 ≤ S < 0.5 / hour and an acceleration Acc > 0;

[0124] Nearing expiration: E increases exponentially and rapidly, with a growth slope S ≥ 0.5 / hour.

[0125] Match the current state of the multi-source coupled response network with typical topology change patterns in the pattern library.

[0126] The recognition rules are as follows:

[0127] If edge e2→4 shows a significant shift, and the shift direction is D decreasing and A increasing, then it is initially determined to be a cavitation-related fault.

[0128] If both edge e1→2 and edge e2→4 show significant offsets and the vibration-noise coherence characteristics are abnormal, it is preliminarily determined to be a bearing wear type fault.

[0129] If edge e2→3 shows a significant shift and the temperature-flow hysteresis correlation characteristics are abnormal, it is initially determined to be a sealing leakage fault.

[0130] In this embodiment, monitoring revealed that the normalized propagation delay D of edge e2→4 decreased from 1.0 to 0.72, the dynamic response gain A increased from 1.0 to 1.35, and the changes of other edges were not significant, matching the topological change pattern of the initial cavitation stage in the matching pattern library.

[0131] The evolution stage is determined based on the growth pattern of the fault risk indicator sequence.

[0132] In this embodiment, the anomalous evolution component at the current moment =0.42, the growth slope of the last 10 moments is S=0.06 / hour, and the acceleration Acc≈0.

[0133] Match the above parameters with typical growth curve shapes in the pattern library:

[0134] S = 0.06 < 0.1, which is consistent with the slope characteristics of the initial stage;

[0135] Acc≈0 indicates that the accelerated growth phase has not yet been entered.

[0136] =0.42, which is within the typical amplitude range of the initial stage (0.2-0.8).

[0137] Overall assessment: The current fault type is early cavitation, and the evolution stage is the initial stage.

[0138] Finally, in the fault information of step S105, the fault type is early cavitation, the evolution stage is the initial stage, the warning level is Level 1 warning (attention level), and the trend prediction is that according to the current growth slope (0.06 / hour), it is expected to enter the development stage in about 120 hours. It is recommended to arrange a shutdown inspection within 72 hours, clean the suction pipeline filter screen, and check whether there is throttling or blockage at the pump inlet.

[0139] Diagnostic information can be output in the following forms: text prompts and audible and visual alarms on the local display panel, pop-up warnings and trend curves on the host computer monitoring interface, and SMS or APP notifications pushed to the mobile terminals of maintenance personnel via 4G / 5G networks.

[0140] To further illustrate the universality of this method, the diagnostic process is briefly described below using bearing wear failure as an example.

[0141] During the operation of a centrifugal pump, the following changes were monitored by a multi-source coupled response network:

[0142] The normalized propagation delay D of edge e1→2 decreases from 1.0 to 0.85;

[0143] As edge e2→4, D decreases from 1.0 to 0.65, and A increases from 1.0 to 1.8.

[0144] Vibration-noise coherence characteristic C6 at bearing fault characteristic frequency The value at that location increased from 0.2 to 0.6.

[0145] The current value of the abnormal evolution component E of the fault risk indicator sequence is 0.78, the growth slope S = 0.25 / hour, and the acceleration Acc = 0.02 / h².

[0146] Matching fault feature pattern library: Multiple edges offset simultaneously, involving load changes and vibration noise transmission, which conforms to the topology change pattern of bearing wear. S=0.25 is in the 0.1-0.5 range, and Acc>0, which conforms to the curve shape of the development period.

[0147] The output diagnostic information is as follows: the fault type is bearing wear, the evolution stage is the development stage, the warning level is the level 2 warning, and the trend prediction is that, based on the current acceleration, it is expected to enter the near failure stage in about 50 hours. It is recommended to arrange for bearing replacement within 24 hours.

[0148] Please refer to Figure 4 In another embodiment of the present invention, a centrifugal pump fault diagnosis system based on multi-source data fusion is provided, the fault diagnosis system comprising:

[0149] The state feature chain construction module 401 is used to extract associated features representing load changes, energy transfer, thermal response and mechanical disturbance based on multi-source synchronous state sequences, construct multi-source fusion feature units corresponding to each operating stage, and form a state feature chain that can continuously represent the evolution of the working condition.

[0150] The coupled response network construction module 402 is used to establish a nonlinear coupling mapping relationship between each associated feature based on the state feature chain, identify the causal transmission path of disturbance between features, and form a multi-source coupled response network that reflects the direction of fault evolution.

[0151] The abnormal evolution analysis module 403 is used to perform joint analysis on the feature offset amplitude, inter-feature transmission delay and response gain at continuous time based on the multi-source coupled response network, extract abnormal evolution components that deviate from the normal coupling law, and generate a fault risk index sequence corresponding to the current operating state.

[0152] The fault mode matching module 404 is used to match the abnormal evolution component with a pre-established fault feature pattern library based on the fault risk index sequence to determine fault attribute information; the fault attribute information includes fault type and fault evolution stage.

[0153] The diagnostic output module 405 is used to output corresponding diagnostic information based on the fault attribute information, the diagnostic information including warning level and / or trend prediction results.

[0154] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0155] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] In the above embodiments of this application, the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with relevant laws, regulations and standards, take necessary protective measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0157] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0158] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0160] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A centrifugal pump fault diagnosis method based on multi-source data fusion, characterized in that, include: Based on the multi-source synchronous state sequence, the associated features representing load change, energy transfer, thermal response and mechanical disturbance are extracted, and multi-source fusion feature units corresponding to each operating stage are constructed to form a state feature chain that can continuously represent the evolution of the operating condition. Based on the state feature chain, a nonlinear coupling mapping relationship between each feature is established, the causal transmission path of disturbance between features is identified, and a multi-source coupled response network reflecting the direction of fault evolution is formed. Based on the multi-source coupled response network, the characteristic offset amplitude, inter-feature transmission delay and response gain at consecutive time moments are jointly analyzed to extract abnormal evolution components that deviate from the normal coupling law and generate a fault risk index sequence corresponding to the current operating state. Based on the fault risk index sequence, the abnormal evolution component is matched with a pre-established fault feature pattern library to determine fault attribute information. Based on the fault attribute information, corresponding diagnostic information is output, including warning level and / or trend prediction results.

2. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 1, characterized in that, By collecting multi-source heterogeneous data from centrifugal pumps and preprocessing it, a time-aligned multi-source synchronous state sequence is formed. in: Multi-source heterogeneous data includes at least two of the following: vibration data, noise data, current data, temperature data, pressure data, and flow rate data; the preprocessing includes: Unified noise reduction processing based on variational mode decomposition is performed on the signals of each channel; Using the vibration signal as a reference sequence, a dynamic time warping algorithm is used to align the noise-reduced signals of each channel in time, forming a multi-source synchronous state sequence that corresponds to the sampling time and retains the fault propagation delay information.

3. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The extraction of associated features and the construction of state feature chains specifically include: Based on the dynamic time-series slice window and sliding step size, the multi-source synchronization state sequence is divided into multiple consecutive time-series state segments; For each temporal state segment, an associated feature group is extracted. The associated feature groups extracted from each temporal state segment are then arranged in chronological order to form a state feature chain consisting of N multi-source fused feature units. The associated feature group includes: Flow combined characteristics are used to characterize the pressure of load changes, including the position coordinates of the pump operating point on the HQ performance curve and the instantaneous velocity of movement; Vibration energy coupling characteristics are used to characterize pressure pulsations that transmit energy, including the ratio coefficient of the frequency band energy of the pressure pulsation signal to the corresponding frequency band energy of the vibration signal. Flow dynamics, used to characterize the temperature of the thermal response, including the hysteresis correlation coefficient between the rate of temperature change and the rate of flow change; Noise coherence characteristics are used to characterize the vibration of mechanical disturbances, including the peak values ​​of the coherence function of the vibration and noise signals in the characteristic frequency band.

4. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The construction of the multi-source coupled response network includes: An initial topology network is constructed using each associated feature in the state feature chain as a network node and the physical causal relationship between the associated features as a directed edge. The direction of the directed edge represents the direction of transmission of fault disturbance between physical quantities, including at least the first transmission path from the pressure feature node to the vibration feature node and the second transmission path from the flow feature node to the temperature feature node. Calculate the normalized propagation delay and dynamic response gain of each directed edge, and label them as attributes of the directed edges in the multi-source coupled response network.

5. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The calculation of the normalized transmission delay specifically includes: Cross-correlation analysis is performed on the two associated feature sequences connected by the directed edge. The time shift corresponding to the peak value of the cross-correlation function is taken as the current transmission delay of the directed edge. The ratio of the current transmission delay to the reference transmission delay under the device health baseline state is taken as the normalized transmission delay.

6. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The calculation of the dynamic response gain specifically includes: For two associated feature sequences connected by a directed edge, a transfer function is established, and the average amplitude gain within a preset fault feature frequency band is calculated as the current response gain. The ratio of the current response gain to the reference response gain under the device health reference state is used as the dynamic response gain.

7. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The extraction of the anomalous evolution components specifically includes: For each directed edge in the multi-source coupled response network, a coupling offset vector is defined, the elements of which include the difference between the normalized propagation delay and 1, and the difference between the dynamic response gain and 1. The weighted summation of each coupling offset vector yields the overall network anomaly evolution component at the current moment. The network's overall abnormal evolution components, arranged in chronological order across multiple consecutive moments, constitute the fault risk indicator sequence; wherein, the morphological characteristics of the fault risk indicator sequence include growth slope, acceleration, and current amplitude.

8. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The fault feature pattern library stores typical topological change patterns of multi-source coupled response networks and typical growth curve shapes of fault risk index sequences under different evolution stages for different fault types.

9. The centrifugal pump fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, The fault attribute information to be determined includes: The fault type is identified based on the location and orientation of the directed edges that have undergone significant shifts in the multi-source coupled response network; Based on the growth slope, acceleration, and current amplitude of the fault risk indicator sequence, the typical growth curve shape in the fault characteristic pattern library is matched to determine the current evolution stage of the fault; wherein, the evolution stage includes at least the initial stage, the development stage, and the near-failure stage.

10. A centrifugal pump fault diagnosis system based on multi-source data fusion, used to implement the centrifugal pump fault diagnosis method as described in any one of claims 1-9, characterized in that, include: The state feature chain construction module is used to extract related features representing load changes, energy transfer, thermal response, and mechanical disturbances based on multi-source synchronous state sequences, construct multi-source fusion feature units corresponding to each operating stage, and form a state feature chain that can continuously represent the evolution of the operating conditions. The coupled response network construction module is used to establish a nonlinear coupling mapping relationship between each associated feature based on the state feature chain, identify the causal transmission path of disturbances between features, and form a multi-source coupled response network that reflects the direction of fault evolution. The abnormal evolution analysis module is used to perform joint analysis on the feature offset amplitude, inter-feature transmission delay and response gain at continuous time based on the multi-source coupled response network, extract abnormal evolution components that deviate from the normal coupling law, and generate a fault risk index sequence corresponding to the current operating state. The fault mode matching module is used to match the abnormal evolution component with a pre-established fault feature pattern library based on the fault risk index sequence to determine fault attribute information; the fault attribute information includes fault type and fault evolution stage; The diagnostic output module is used to output corresponding diagnostic information based on the fault attribute information, including warning level and / or trend prediction results.

Images