Condensate pump shafting vibration fault diagnosis method, device and medium based on multi-parameter analysis

Abstract

The present invention relates to the technical field of water pump fault diagnosis, and discloses a vibration fault diagnosis method, device and medium for the shafting of condensate pumps based on multi-parameter analysis. Through multi-source data fusion and environmental adaptive compensation, the present invention solves the problems of sensor signal distortion and noise interference of condensate pumps under high-temperature and high-humidity working conditions; adopts dynamic time-frequency domain feature extraction and an improved lightweight residual network to achieve efficient decoupling and accurate identification of multi-fault concurrent features; combines a fuzzy inference rule base with a shafting fluctuation propagation model, significantly improving the fault location accuracy and early fault detection rate, providing scientific and reliable technical support for the diagnosis and maintenance of the vibration faults of the condensate pump shafting, and effectively ensuring the safe and stable operation of factory equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of water pump fault diagnosis, and specifically to a method, device and medium for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis. Background Art

[0002] In factory equipment, the condensate pump is a core equipment in the public auxiliary project. Its shafting operates under high temperature, high humidity and variable load conditions for a long time, with frequent vibration faults and significantly higher diagnosis difficulty than conventional water pumps. In the prior art, the diagnosis methods for the vibration of the condensate pump shafting mainly rely on single-parameter analysis. This includes judging imbalance or misalignment by extracting the amplitudes of the 1x and 2x frequencies, and setting a fixed threshold for alarm using the linear combination of bearing temperature and amplitude. In addition, traditional methods generally use static neural network models or ISO 10816 vibration standards, relying only on single-sensor data (such as radial vibration amplitude), and lacking adaptability to the special working conditions of the condensate pump (such as start-stop transient and impeller cavitation).

[0003] The prior art has the following key problems:

[0004] (1) There is a lack of a compensation mechanism for sensor signal drift and noise interference in the high temperature and high humidity environment of the condensate pump, resulting in distorted feature extraction;

[0005] (2) When multiple faults occur simultaneously (such as coexistence of shaft bending and wear of the sealing ring), a single frequency domain analysis method cannot decouple the fault features;

[0006] (3) The dynamic correlation of phase difference, torque fluctuation and temperature parameters is not fully utilized, resulting in an early fault undetected rate exceeding 35%. For example, when the condensate pump operates at low load, the traditional method misjudges impeller cavitation as bearing looseness due to ignoring the non-linear coupling of torque signal and vibration phase.

[0007] Therefore, there is an urgent need for a diagnostic method that integrates multi-physical quantity dynamic analysis and environmental adaptive correction to achieve accurate fault identification under complex working conditions. For this reason, a method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis is proposed. Summary of the Invention

[0008] Aiming at the deficiencies of the prior art, the present invention provides a method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis to solve the problems in the background art.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis includes the following steps:

[0011] Step 1: Collect vibration signals through axial vibration sensors and radial vibration sensors, collect torque fluctuation signals through torque meters, collect cooling water temperature through temperature sensors, collect motor current parameters through Hall sensors, and obtain pump speed data in real time;

[0012] Step 2: Perform environment adaptive preprocessing on the vibration signals, including signal correction based on temperature drift compensation and anti-aliasing filtering, to generate denoised multi-channel time domain signals;

[0013] Step 3: Adopt a working condition adaptive time-frequency domain joint analysis method to perform dynamic wavelet packet decomposition and energy-phase correlation feature extraction on the denoised multi-channel time domain signals, and construct a multi-dimensional feature matrix including energy entropy, phase difference, and working condition parameters;

[0014] Step 4: Input the multi-dimensional feature matrix into an improved lightweight residual network for preliminary fault classification and output the fault probability distribution;

[0015] Step 5: Through the fuzzy inference engine, fuse the torque fluctuation signals collected by the torque meter, the torque fluctuation spectrum features obtained after spectrum analysis, and the fault probability distribution, and perform multi-level decision-making in combination with the historical fault case library to generate the final fault type, confidence level, and location information of the fault occurrence;

[0016] Step 6: Dynamically optimize the alarm threshold based on the cumulative operating duration, load rate, and environmental temperature of the equipment, and trigger a hierarchical early warning strategy.

[0017] Further, the sensor self-check process is also included in the above Step 1:

[0018] Judge the health status of the sensor according to the amplitude of the coherence function of the vibration signal and the torque fluctuation signal at the power frequency. If <, mark the sensor as abnormal and enable the redundant channel data. The value range of A is .

[0019] Further, the specific implementation method of the temperature drift compensation in the above Step 2 is as follows:

[0020] According to the mapping relationship between the cooling water temperature and the thermal expansion coefficient of the shafting material, calculate the temperature compensation coefficient of the vibration signal. The expression is:

[0021]

[0022] where γ is the thermal expansion coefficient of the shafting material, T current is the cooling water temperature, is the temperature median value calculated by the sliding window.

[0023] Further, the operations of performing dynamic wavelet packet decomposition and energy-phase correlation feature extraction in step 3 include:

[0024] According to the real-time rotational speed change rate δ, δ = , dynamically select the wavelet basis function: when , the value range of B is 3% - 5%, and the Daubechies 6 wavelet basis is selected; otherwise, the Symlet 4 wavelet basis is selected; is the rotational speed difference, is the rotational speed detection cycle time; adaptively adjust the decomposition layer number according to the rotational speed frequency band ratio, and the calculation formula for its layer number L is:

[0025]

[0026] where, is the sampling frequency, is the maximum analysis frequency corresponding to the current rotational speed.

[0027] Further, the improvement of the lightweight residual network includes:

[0028] Embed a dual-path attention mechanism in the residual block to perform feature reweighting on the channel dimension and the spatial dimension respectively. The channel attention weight The calculation formula is:

[0029]

[0030] where, GAP is global average pooling, is the channel dimension weight matrix of the fully connected layer, W2 is the spatial dimension weight matrix of the fully connected layer, ReLU is the ReLU function, is the standard deviation of the real-time feature;

[0031] The pre-trained large ResNet-50 model is compressed into a lightweight network by the knowledge distillation method.

[0032] Further, the fuzzy inference engine in step 5 includes a special rule base for condensate pumps:

[0033] Rule 1: If the sudden increase in high-frequency energy entropy > H1 and the amplitude ratio of the 3rd harmonic of torque fluctuation > H2, then it is determined as an impeller cavitation fault, and the confidence level is H3. The value range of H1 is 35% - 40%, the value range of H2 is 20% - 25%, and the value range of H3 is 0.82 - 0.85;

[0034] Rule 2: When the axial / radial vibration energy ratio > N1 and the variance of phase difference fluctuation > N2, trigger the bearing wear detection sub-module. The value range of N1 is 2.25 - 2.35, and the value range of N2 is 9.5°² - 11°²;

[0035] Rule 3: If the cooling water temperature rise rate > S1 and the current harmonic distortion rate > S2, it is determined that the mechanical seal fails. The value range of S1 is 1.3 °C / min to 1.6 °C / min, and the value range of S2 is 7.5% to 8.5%.

[0036] Further, the generation of the positioning information of the fault occurrence location in step 5 is implemented by using a shafting wave propagation model: According to the peak time difference of the cross-correlation function of the vibration signals at both ends of the axis , calculate the length of the fault point from the end of the shafting connecting the condensate pump and the motor :

[0037]

[0038] where v is the stress wave velocity in the shaft material and δ is the correction amount of the sensor installation spacing.

[0039] Further, the method for dynamically optimizing the alarm threshold in step 6 includes:

[0040] Calculate the threshold baseline based on the Weibull distribution model , and its expression is:

[0041]

[0042] where α is the shape parameter of the equipment life distribution, β is the scale parameter of the equipment life distribution, η is the working condition correction factor, t is the time, and e is the natural constant;

[0043] Combine the standard deviation σ of the real-time features with the threshold baseline Generate the final alarm threshold :

[0044]

[0045] where k is the safety factor.

[0046] The present invention also provides a condensate pump shafting vibration fault diagnosis device based on multi-parameter analysis, including one or more processors for implementing a condensate pump shafting vibration fault diagnosis method as described above.

[0047] The present invention also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, it implements a condensate pump shafting vibration fault diagnosis method as described above.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] Through multi-source data fusion (vibration, torque, temperature, current) and environmental adaptive compensation (temperature drift correction, anti-aliasing filtering), the present invention solves the problems of sensor signal distortion and noise interference of condensate pumps under high-temperature and high-humidity conditions; by using dynamic time-frequency domain feature extraction (wavelet packet decomposition, energy-phase correlation analysis) and an improved lightweight residual network, it realizes the efficient decoupling and accurate identification of multi-fault concurrent features; combined with a fuzzy inference rule base and a shafting fluctuation propagation model, it significantly improves the fault location accuracy and early fault detection rate, providing scientific and reliable technical support for the diagnosis and maintenance of condensate pump shafting vibration faults, and effectively ensuring the safe and stable operation of power plant equipment.

[0050] Other features and advantages of the present invention will be described in the following specification, and part of them will be obvious from the specification or understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures pointed out in the specification, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flowchart of a method for diagnosing condensate pump shafting vibration faults based on multi-parameter analysis according to the present invention.

[0052] Figure 2 is a schematic structural diagram of a device for diagnosing condensate pump shafting vibration faults based on multi-parameter analysis according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by those of ordinary skill in the art in the technical field of the present invention without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0054] Please refer to Figure 1 , a method for diagnosing condensate pump shafting vibration faults based on multi-parameter analysis, includes the following steps:

[0055] Step 1: Collect vibration signals through axial vibration sensors and radial vibration sensors, collect torque fluctuation signals through a torque meter, collect cooling water temperature through a temperature sensor, collect motor current parameters through a Hall sensor, and obtain pump speed data in real time.

[0056] Self-check the sensors, and the process includes:

[0057] Judge the health status of the sensors through the amplitude of the coherence function of the vibration signal and the torque fluctuation signal at the power frequency. If , the sensor is marked as abnormal and the redundant channel data is enabled. The value range of A is , preferably 0.6.

[0058] Step 2: Perform environment adaptive preprocessing on the vibration signal, including signal correction based on temperature drift compensation and anti-aliasing filtering, to generate a denoised multi-channel time-domain signal. Anti-aliasing filtering is a well-known technology and will not be elaborated.

[0059] Among them, the specific implementation method of temperature drift compensation is:

[0060] According to the mapping relationship between the cooling water temperature and the thermal expansion coefficient of the shafting material, calculate the temperature compensation coefficient of the vibration signal , and the expression is:

[0061]

[0062] Among them, γ is the thermal expansion coefficient of the shafting material, T current Cooling water temperature is the temperature median value calculated through a sliding window.

[0063] Step 3: Adopt a working condition adaptive time-frequency domain joint analysis method to perform dynamic wavelet packet decomposition and energy-phase correlation feature extraction on the denoised multi-channel time-domain signal, and construct a multi-dimensional feature matrix including but not limited to energy entropy, phase difference and working condition parameters.

[0064] Among them, the operations of dynamic wavelet packet decomposition and energy-phase correlation feature extraction include:

[0065] According to the real-time rotational speed change rate δ, δ = , dynamically select the wavelet basis function: when , the value range of B is 3% - 5%, preferably 5%, and the Daubechies 6 wavelet basis is selected; otherwise, the Symlet 4 wavelet basis is selected; is the rotational speed difference, is the rotational speed detection cycle time. Adaptively adjust the decomposition layer number according to the rotational speed frequency band ratio, and the calculation formula of its layer number L is:

[0066]

[0067] Among them, is the sampling frequency, is the maximum analysis frequency corresponding to the current rotational speed.

[0068] Step 4: Input the multi-dimensional feature matrix into the improved lightweight residual network (Light-ResNet) for preliminary fault classification, and output the fault probability distribution. The working process of the lightweight residual network is a well-known technology and will not be elaborated.

[0069] Among them, the improvements to the lightweight residual network include:

[0070] Embed a dual-path attention mechanism in the residual block to perform feature re-weighting on the channel dimension and the spatial dimension respectively. The channel attention weight The calculation formula is:

[0071]

[0072] Among them, GAP is global average pooling, is the channel dimension weight matrix of the fully connected layer, W2 is the spatial dimension weight matrix of the fully connected layer, ReLU is the ReLU function, is the standard deviation of the real-time features.

[0073] The large ResNet-50 model trained is compressed into a lightweight network using the knowledge distillation method.

[0074] Step 5: Fusion the torque fluctuation signals collected by the torque meter through the fuzzy inference engine, obtain the torque fluctuation spectrum characteristics and the fault probability distribution after spectrum analysis, and perform multi-level decision-making in combination with the historical fault case library to generate the final fault type, confidence level and location information.

[0075] Among them, the fuzzy inference engine includes a special rule library for condensate pumps:

[0076] Rule 1: If the sudden increase in high-frequency energy entropy > H1 and the amplitude ratio of the 3rd harmonic of torque fluctuation > H2, then it is determined as the impeller cavitation fault, and the confidence level is H3. The value range of H1 is 35% - 40%, preferably 40%, the value range of H2 is 20% - 25%, preferably 25%, and the value range of H3 is 0.82 - 0.85, preferably 0.85;

[0077] Rule 2: When the axial / radial vibration energy ratio > N1 and the variance of phase difference fluctuation > N2, trigger the bearing wear detection sub-module. The value range of N1 is 2.25 - 2.35, preferably 2.3, and the value range of N2 is 9.5°² - 11°², preferably 10°²;

[0078] Rule 3: If the cooling water temperature rise rate > S1 is accompanied by the current harmonic distortion rate > S2, then it is determined as the mechanical seal failure. The value range of S1 is 1.3℃ / min - 1.6℃ / min, preferably 1.5℃ / min, and the value range of S2 is 7.5% - 8.5%, preferably 8%.

[0079] Among them, the location information of the fault occurrence position is realized by using the shafting fluctuation propagation model: Calculate the length of the fault point from the end of the shafting connecting the condensate pump and the motor according to the peak time difference of the cross-correlation function of the vibration signals at both ends of the axis Calculate the length of the fault point from the end of the shafting connecting the condensate pump and the motor :

[0080]

[0081] Among them, is the peak time difference of the cross-correlation function of the vibration signals at both axial ends, v is the stress wave velocity in the shaft material, and δ is the correction amount of the sensor installation spacing.

[0082] Step 6: Dynamically optimize the alarm threshold based on the cumulative operating duration, load rate, and ambient temperature of the device, and trigger a hierarchical early warning strategy. The hierarchical early warning strategy performs early warning according to different alarm thresholds. The hierarchical early warning strategy is a well-known technology and will not be elaborated here.

[0083] Among them, the method for dynamically optimizing the alarm threshold includes:

[0084] Calculating the threshold baseline based on the Weibull distribution model , and its expression is:

[0085]

[0086] Among them, α is the shape parameter of the device life distribution, β is the scale parameter of the device life distribution, η is the working condition correction factor, t is the time, and e is the natural constant.

[0087] Combining the standard deviation σ of the real-time features with the threshold baseline Generating the final alarm threshold :

[0088]

[0089] Among them, k is the safety factor, and its value range is 2.0 - 3.0.

[0090] Referring to Figure 2 , an apparatus for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis provided by an embodiment of the present invention includes one or more processors for implementing a method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis in the above embodiment.

[0091] An embodiment of an apparatus for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis of the present invention can be applied to any device with data processing capabilities. The any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, such as Figure 2As shown, it is a hardware structure diagram of any device with data processing capabilities where a condensate pump shafting vibration fault diagnosis device based on multi-parameter analysis of the present invention is located. In addition to Figure 2 the shown processor, memory, network interface, and non-volatile memory, in the embodiment, any device with data processing capabilities where the device is located usually also includes other hardware according to the actual functions of the any device with data processing capabilities, which will not be elaborated herein.

[0092] The implementation processes of the functions and roles of each unit in the above device are specifically detailed in the implementation processes of the corresponding steps in the above method, which will not be elaborated herein.

[0093] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combinations of these technical features do not conflict, they should be considered as within the scope described in this specification.

[0094] The embodiment of the present invention also provides a readable storage medium, on which a program is stored. When the program is executed by a processor, it implements a condensate pump shafting vibration fault diagnosis method based on multi-parameter analysis in the above embodiment.

[0095] The readable storage medium can be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The readable storage medium can also be an external storage device, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc., equipped on the device. Further, the readable storage medium can also include both the internal storage unit of any device with data processing capabilities and the external storage device. The readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or will be output.

[0096] Embodiment

[0097] Please refer to Figure 1 , a condensate pump shafting vibration fault diagnosis method based on multi-parameter analysis in the present invention includes the following content:

[0098] 1. System hardware configuration and data acquisition

[0099] The implementation object of this embodiment is a vertical condensate pump (model: 9LDTN-6P) supporting a 1000MW unit of a certain power plant. Its shafting structure includes four-stage impellers, an intermediate bearing housing, and a hydraulic coupling. The hardware system includes a multi-source sensor array and a data synchronization device. The multi-source sensor array includes a vibration monitoring unit, a torque monitoring unit, a temperature monitoring unit, and an electrical parameter unit.

[0100] Vibration monitoring unit: Three-axis vibration sensors (PCB 352C33, range ±50g, sampling frequency 10kHz) are installed at the driving end and non-driving end of the pump shaft respectively, with an axial spacing of 1.2m and a radial installation angle interval of 120°.

[0101] Torque monitoring unit: A strain type torque meter (HBM T40B, accuracy ±0.1%FS) is installed on the output shaft of the hydraulic coupling to collect torque fluctuation signals in real time.

[0102] Temperature monitoring unit: PT100 platinum resistors (accuracy ±0.5°C) are arranged in the bearing housing and the cooling water circuit, with a sampling interval of 1 second.

[0103] Electrical parameter unit: The three-phase current of the motor is collected through a Hall sensor (LEM ITC 200-S), and the output frequency of the frequency converter is obtained synchronously.

[0104] Data synchronization device: A PXIe-8840 controller is used to carry a synchronous acquisition card (NI 9234) to achieve strict time scale alignment of multi-channel signals, with a synchronous error <1μs.

[0105] 2. Implementation details of environment adaptive preprocessing

[0106] Step 1: Temperature drift compensation

[0107] Establish a thermal expansion coefficient database of the shafting material (martensitic stainless steel 2Cr13): Measured through a laboratory temperature control box .

[0108] Reference temperature Dynamic update: Calculate the sliding median value based on the cooling water temperature data in the previous 12 hours at the current moment. For example, when the historical temperature sequence is [45.3°C, 46.1°C, 47.5°C, 45.8°C], take the median value of 46.1°C.

[0109] Signal correction example: If the current temperature , then the compensation coefficient , and scale the amplitude of the original vibration signal proportionally.

[0110] Step 2: Anti-aliasing filter design

[0111] According to the number of teeth of the pump shaft gear k = 28 and the maximum rotational speed , calculate the cutoff frequency :

[0112]

[0113] An 8th-order Butterworth filter is used with a stopband attenuation of >60dB to ensure that the signal is alias-free in the range of 0 to 11.2kHz.

[0114] Step 3: Sensor self-check and redundancy switching

[0115] Calculate the vibration signal (channel CH1) and torque signal (channel CH2) at the power frequency The coherence function at :

[0116]

[0117] in, is the autopower spectrum, is the cross power spectrum.

[0118] like , determine that CH1 is abnormal, and automatically switch to the backup vibration sensor CH3.

[0119] 3. Dynamic time-frequency domain feature extraction process

[0120] Step 1: Condition-adaptive wavelet packet decomposition

[0121] When the speed change rate Rated speed / second (rated speed 2975r / min corresponds to 5% of 148.75r / min / s), select Daubechies 6 wavelet basis.

[0122] Decomposition layer calculation: current speed corresponds to the maximum analysis frequency Sampling frequency ,but:

[0123]

[0124] 64 sub-bands are generated, covering the range 0-156.25 Hz.

[0125] Step 2: Energy-phase correlation analysis

[0126] Calculate the energy entropy of the k=24 sub-band (corresponding to 75-78.125Hz) :

[0127]

[0128] Extract the phase difference sequence of axial and radial vibration signals , calculate its variance .

[0129] 4. Fault Classification Implementation of Light-ResNet

[0130] Network Structure Parameters:

[0131] Input Layer: 64×64 Feature Matrix (corresponding to 64 sub-bands × 64 time frames);

[0132] Residual Block Configuration: 4 double-path attention residual blocks, with channel numbers [64, 128, 256, 512];

[0133] Attention Mechanism: Global average pooling followed by two fully connected layers (512→32→512), compression ratio 16:1, LeakyReLU negative slope 0.2;

[0134] Knowledge Distillation: The teacher network (ResNet-50) is pre-trained on the CWRU bearing dataset, and the student network retains the feature mapping relationship of the first 3 layers of the teacher network.

[0135] Training Process:

[0136] Dataset: Collect 3-year operation data of condensate pumps, including 12 types of fault modes (such as impeller cavitation, bearing spalling, shaft bending, etc.), with a total of 8500 groups of samples;

[0137] Optimizer: AdamW (learning rate 3e-4, weight decay 0.01);

[0138] Output Result: The fault probability distribution of the current sample is [impeller cavitation: 0.72, bearing wear: 0.18, normal: 0.10].

[0139] 5. Example of Fuzzy Inference Multilevel Decision Making

[0140] Rule Triggering Example:

[0141] High-frequency Energy Entropy Sudden Increase Detection: Current E_{24}=2.37, baseline E_{24_base}=1.68, sudden increase amplitude (2.37 - 1.68) / 1.68≈41% > 40%.

[0142] Torque Third Harmonic Analysis: Perform FFT on the torque signal, calculate the amplitude ratio of the third harmonic (149.4Hz) = 28.5% > 25%.

[0143] Judgment Result: Trigger Rule 1, output "impeller cavitation fault" (confidence 0.85×0.72 = 0.612).

[0144] Fault Location Calculation:

[0145] Time difference of shock waves received by sensors at both axial ends , the sound velocity of the shaft material v = 5100 m / s, and the correction amount δ = 0.15 m.

[0146]

[0147] The location of the fault point is 1.84 m from the drive end, corresponding to the position of the second-stage impeller.

[0148] 6. Dynamic Threshold Optimization and Early Warning

[0149] Example of threshold calculation:

[0150] The equipment has been running for a cumulative time of t = 25000 hours, the Weibull parameter α = 35000 hours, β = 2.3, and the operating condition factor η = 0.92.

[0151]

[0152] The current standard deviation of the feature σ = 0.18, and the safety factor k = 2.5:

[0153]

[0154] The measured feature value is 1.23> , triggering a secondary early warning (yellow alert), indicating that maintenance should be arranged within 72 hours.

[0155] 7. Verification of Implementation Effect

[0156] Deploy the system of this embodiment on 6 identical condensate pumps in a certain power plant, and compare with the traditional method:

[0157] Missed detection rate: decreased from 35.2% to 6.8% (through continuous monitoring of 12 early cavitation faults for 3 months);

[0158] False alarm rate: decreased from 28.7% to 9.4% (eliminating 11 false alarms caused by temperature drift);

[0159] Location accuracy: the average error was improved from ±0.8 m to ±0.2 m (verified by a laser alignment instrument).

[0160] This embodiment realizes the accurate diagnosis and location of the vibration faults of the condensate pump shaft system through multi-source data fusion (vibration, torque, temperature, current), environmental adaptive compensation (temperature drift correction, anti-aliasing filtering), dynamic time-frequency domain feature extraction (wavelet packet decomposition, energy-phase correlation analysis), and collaborative decision-making of an improved lightweight residual network and fuzzy inference, significantly reducing the missed detection rate (from 35.2% to 6.8%) and the false alarm rate (from 28.7% to 9.4%), and improving the fault location accuracy to ±0.2 meters through the shaft system wave propagation model, providing scientific and reliable technical support for the maintenance of power plant equipment.

[0161] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis, characterized in that, It includes the following steps: Step 1: Collect vibration signals through an axial vibration sensor and a radial vibration sensor, collect torque fluctuation signals through a torque meter, collect the cooling water temperature through a temperature sensor, collect motor current parameters through a Hall sensor, and obtain the pump speed data in real time; Step 2: Perform environment-adaptive preprocessing on the vibration signals, including signal correction based on temperature drift compensation and anti-aliasing filtering, to generate a denoised multi-channel time-domain signal; Step 3: Adopt a working condition-adaptive time-frequency domain joint analysis method to perform dynamic wavelet packet decomposition and energy-phase correlation feature extraction on the denoised multi-channel time-domain signal, and construct a multi-dimensional feature matrix including energy entropy, phase difference, and working condition parameters; Step 4: Input the multi-dimensional feature matrix into an improved lightweight residual network for preliminary fault classification, and output the fault probability distribution; Step 5: Fusion the torque fluctuation signal collected by the torque meter through a fuzzy inference engine, the torque fluctuation spectrum features obtained after spectrum analysis and the fault probability distribution, and perform multi-level decision-making in combination with the historical fault case library to generate the final fault type, confidence level, and the positioning information of the fault occurrence location. The fuzzy inference engine includes a dedicated rule library for condensate pumps: Rule 1: If the sudden increase in high-frequency energy entropy > H1 and the amplitude ratio of the 3rd harmonic of torque fluctuation > H2, it is determined as an impeller cavitation fault, and the confidence level is H3. The value range of H1 is 35% - 40%, the value range of H2 is 20% - 25%, and the value range of H3 is 0.82 - 0.85; Rule 2: When the axial / radial vibration energy ratio > N1 and the variance of phase difference fluctuation > N2, trigger the bearing wear detection sub-module. The value range of N1 is 2.25 - 2.35, and the value range of N2 is 9.5°² - 11°²; Rule 3: If the cooling water temperature rise rate > S1 accompanied by the current harmonic distortion rate > S2, it is determined as a mechanical seal failure. The value range of S1 is 1.3℃ / min - 1.6℃ / min, and the value range of S2 is 7.5% - 8.5%; Step 6: Dynamically optimize the alarm threshold based on the cumulative operation duration, load rate, and ambient temperature of the equipment, and trigger a hierarchical early warning strategy.

2. The condensate pump shafting vibration fault diagnosis method based on multi-parameter analysis according to claim 1, wherein The sensor self-check process is also included in Step 1: By the coherence function of the vibration signal and the torque fluctuation signal Judge the health state of the sensor according to the amplitude at the power frequency. If , mark the sensor as abnormal and enable the redundant channel data. The value range of A is .

3. A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis according to claim 1, characterized in that, The specific implementation method of the temperature drift compensation in Step 2 is: Calculate the temperature compensation coefficient of the vibration signal according to the mapping relationship between the cooling water temperature and the thermal expansion coefficient of the shafting material , and the expression is: Among them, γ is the thermal expansion coefficient of the shafting material, T current is the cooling water temperature, is the median temperature calculated by the sliding window.

4. A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis according to claim 1, characterized in that The operations of performing dynamic wavelet packet decomposition and energy-phase correlation feature extraction in Step 3 include: According to the real-time rotational speed change rate δ, δ = , dynamically select the wavelet basis function: When , the value range of B is 3% - 5%, and the Daubechies 6 wavelet basis is selected; otherwise, the Symlet 4 wavelet basis is selected; is the rotational speed difference, is the rotational speed detection cycle time; adaptively adjust the decomposition level according to the rotational speed frequency band ratio, and the calculation formula for its level L is: Among them, is the sampling frequency, is the maximum analysis frequency corresponding to the current rotational speed.

5. A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis according to claim 1, characterized in that The improvement of the lightweight residual network includes: Embed the dual-path attention mechanism in the residual block to perform feature re-weighting on the channel dimension and the spatial dimension respectively. The channel attention weight The calculation formula is as follows: Among them, GAP is global average pooling, is the weight matrix of the channel dimension of the fully connected layer, W2 is the weight matrix of the spatial dimension of the fully connected layer, ReLU is the ReLU function, is the standard deviation of the real-time feature; The pre-trained large ResNet-50 model is compressed into a lightweight network by the knowledge distillation method.

6. The condensate pump shafting vibration fault diagnosis method based on multi-parameter analysis according to claim 1, characterized in that The generation of the positioning information of the fault occurrence location in the step 5 is realized by using the shafting fluctuation propagation model: according to the peak time difference of the cross-correlation function of the vibration signals at both ends of the axis , calculate the length of the fault point from the end of the shafting connecting the condensate pump and the motor : Wherein, v is the stress wave velocity in the shaft material, and δ is the correction amount of the sensor installation spacing.

7. A method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis according to claim 1, characterized in that The method for dynamically optimizing the alarm threshold in Step 6 includes: Calculating the threshold baseline based on the Weibull distribution model , and its expression is: Wherein, α is the shape parameter of the equipment life distribution, β is the scale parameter of the equipment life distribution, η is the working condition correction factor, t is the time, and e is the natural constant; Combine the standard deviation σ of real-time features with the threshold baseline Generate the final alarm threshold : Wherein, k is the safety factor.

8. A condensate pump shafting vibration fault diagnosis device based on multi-parameter analysis, characterized in that, It includes one or more processors for implementing a method for diagnosing the vibration fault of the condensate pump shafting based on multi-parameter analysis according to any one of claims 1 - 7.

9. A readable storage medium, characterized in that, A program is stored thereon. When the program is executed by a processor, it implements a condensate pump shafting vibration fault diagnosis method according to any one of claims 1-7 based on multi-parameter analysis.

Images