Intelligent centrifugal pump and fault diagnosis method thereof

Abstract

The intelligent centrifugal pump comprises a comprehensive fault diagnosis system, the comprehensive fault diagnosis system comprises a multi-level sensor sensing network, and the multi-level sensor sensing network comprises a slowly-changing signal sensor group and a dynamic signal sensor group and is used for collecting multi-physical-quantity operation data of a centrifugal pump unit; the signal processing and hybrid intelligent diagnosis module is used for carrying out preprocessing, feature extraction and feature optimization on the collected signals and carrying out state recognition and fault classification by utilizing a plurality of special diagnosis models; and the multi-source information fusion and active control module is used for carrying out fusion decision on the diagnosis results of the models and linking an execution mechanism to realize closed-loop control. According to the intelligent centrifugal pump and the fault diagnosis method thereof, a full-dimension perception and hybrid intelligent diagnosis method is adopted, so that diagnosis is accurate and reliable, active intervention is achieved, the safety level is improved, predictive maintenance is achieved, and automation and intelligence are achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent water pump technology, and in particular to an intelligent centrifugal pump and its fault diagnosis method. Background Technology

[0002] Centrifugal pumps, as a critical fluid transport device, are widely used in important fields such as energy, power, aerospace, water conservancy, and petrochemicals, playing an irreplaceable role in numerous engineering applications. While ensuring their hydraulic performance and anti-cavitation capabilities, centrifugal pumps are currently developing towards higher speeds and higher power densities. This trend places higher demands on the safety and operational stability of the equipment. Simultaneously, centrifugal pumps have long faced multiple severe challenges, including corrosion, cavitation, vibration and shock, and variable loads, leading to common failures such as corrosion, wear, cracks, and even breakage in impellers and other flow components. These problems not only affect the reliability of the equipment itself but also directly threaten the stable operation of the entire system, becoming a focus of current technological attention.

[0003] In terms of information acquisition, data from a single sensor is incomplete and fails to fully reflect the complex state of the equipment, especially nonlinear faults. Regarding signal processing, issues arise such as the nonlinearity of fault signals, difficulty in early identification, and inconsistent characteristic frequency bandwidths. Fault signals (such as cavitation) exhibit nonlinear and non-stationary characteristics, with weak early features easily masked by noise. The sensitive frequency bandwidth of centrifugal pump faults varies significantly depending on pump type and operating conditions, making it difficult to establish universal and quantifiable judgment standards. In terms of optimized operation, for independently operating centrifugal pumps, actual operating conditions often deviate from the high-efficiency zone, leading to a significant decrease in efficiency, internal flow turbulence, and accelerated wear of components. When multiple pumps operate in parallel, it is difficult to dynamically match flow rate, pressure, and the optimal operating point of each pump, resulting in low overall operational efficiency and reliability. Regarding maintenance, traditional methods rely mainly on periodic inspections and manual experience, exhibiting significant lag and difficulty in achieving early warning and accurate diagnosis of faults. This not only easily leads to unplanned downtime but may also cause serious safety accidents.

[0004] Existing monitoring technologies are mostly limited to threshold alarms for single parameters (such as vibration or temperature), resulting in limited diagnostic capabilities and an inability to accurately distinguish between fault types such as cavitation, wear, and cracks. Although some research has attempted to introduce intelligent algorithms, most are still based on single-type signals (such as vibration signals alone), leading to insufficient diagnostic accuracy and reliability. Furthermore, these methods often fail to achieve deep integration with the pump's own control system (such as frequency conversion regulation and active vibration isolation), making it difficult to construct a complete "sensing-diagnosis-control" closed-loop intelligent operation and maintenance system, thus limiting their engineering application effectiveness. Summary of the Invention

[0005] In view of the aforementioned shortcomings of existing technologies, the technical problem to be solved by this invention is that existing centrifugal pump fault diagnosis and monitoring methods are singular, have limited diagnostic capabilities, lack multi-source information fusion, cannot achieve precise diagnosis, and suffer from information silos and redundancy, resulting in the loss of important fault characteristic signals, insufficient intelligence, reliance on manual experience, and a lack of predictive maintenance capabilities. This invention provides a smart centrifugal pump and its fault diagnosis method, which has full-dimensional perception, laying a data foundation for accurate diagnosis. The hybrid intelligent diagnostic method makes its diagnosis accurate and reliable, transforming passive alarms into proactive intervention, improving its safety level, and enabling predictive maintenance, achieving automation and intelligence.

[0006] To achieve the above objectives, the present invention provides a smart centrifugal pump, including a centrifugal pump body, a drive motor and a frequency converter, and also includes a comprehensive fault diagnosis system. The comprehensive fault diagnosis system includes...

[0007] A multi-level sensor network, which includes a group of slowly varying signal sensors and a group of dynamic signal sensors, is used to collect multi-physical quantity operation data of centrifugal pump units.

[0008] The signal processing and hybrid intelligent diagnostic module is used to preprocess, extract and optimize features of the acquired signals, and use multiple dedicated diagnostic models for state recognition and fault classification.

[0009] The multi-source information fusion and active control module is used to fuse the diagnostic results of various models to make decisions and link the actuators to achieve closed-loop control.

[0010] Furthermore, the gradually changing signal sensor group includes a torque tachometer, an electromagnetic flowmeter, a temperature sensor, and a differential pressure sensor; the torque tachometer is used to measure the pump shaft speed and torque in real time, providing a basis for calculating hydraulic performance and identifying overload; the electromagnetic flowmeter is used to monitor the instantaneous flow rate in the pipeline; the temperature sensor is arranged in the motor bearing, pump bearing, or other parts to monitor temperature rise changes; the differential pressure sensor is installed at the pump inlet and outlet to calculate the pump head;

[0011] The dynamic signal sensor group includes pressure pulsation sensors arranged on the inner wall of the pump volute and pump cover, dynamic strain gauges embedded in the impeller blades, and multi-axis vibration sensors arranged at various mechanical key points of the unit; the signals of the dynamic strain gauges are transmitted through fiber optic grating sensors or remote sensing systems of rotating components.

[0012] Furthermore, the signal processing and hybrid intelligent diagnostics module is configured to perform the following steps:

[0013] The vibration signal is extracted using a stacked sparse autoencoder for deep feature extraction, then reduced in dimensionality by principal component analysis, and input into a support vector machine classifier to classify the equipment status into normal operation, abnormal vibration of mechanical system, or abnormal vibration of overcurrent system.

[0014] After filtering, denoising, and normalizing the pressure pulsation signal, a two-dimensional image is generated through Gram angle field transformation.

[0015] Two-dimensional images and dynamic strain gauge signals are input into a multi-channel convolutional neural network for feature extraction and fusion to identify cavitation, corrosion, wear, cracks, and fracture faults in the impeller.

[0016] Furthermore, the Gram angle field transformation specifically includes mapping the one-dimensional pressure pulsation signal to the polar coordinate system and generating a two-dimensional grayscale image by calculating the Gram angle sum field and the Gram angle difference field.

[0017] Furthermore, the multi-source information fusion and active control module includes

[0018] Over-temperature and overload protection system, used to automatically adjust speed or stop the machine via frequency converter when temperature or torque exceeds the limit;

[0019] An active vibration isolation compensation system is used to activate active vibration isolators to suppress vibration when the vibration amplitude exceeds the limit;

[0020] Intelligent hydraulic performance regulation system, used to automatically or suggest adjusting flow valves or rotation speed based on performance evaluation and fault diagnosis results;

[0021] The fault early warning and operation alarm system is used to generate graded alarm information and transmit it remotely.

[0022] Furthermore, the activation condition of the active vibration isolation compensation system is that when the mechanical system is diagnosed as having abnormal vibration, and the multi-source information fusion system determines that the vibration of the unit support exceeds the standard, the central controller triggers the active vibration isolator to work.

[0023] Furthermore, when an abnormal vibration of the flow system is diagnosed, a fine diagnostic process for the flow components based on pressure pulsation signals and dynamic strain gauge signals is triggered to further identify the root causes of cavitation, corrosion, wear, or blade cracks.

[0024] In another preferred embodiment of the present invention, a fault diagnosis method for the above-mentioned intelligent centrifugal pump is provided, comprising the following steps:

[0025] (1) Synchronously collect slowly changing signals and dynamic signals through a multi-level sensor network;

[0026] (2) Perform deep feature extraction and dimensionality reduction on the vibration signal, and use a classifier to make an initial judgment on the abnormal state;

[0027] (3) Perform GAF image transformation on the pressure pulsation signal, and combine it with the dynamic strain signal to achieve fine fault classification through a convolutional neural network;

[0028] (4) Integrate multi-source diagnostic information to trigger corresponding protection, vibration isolation, adjustment or early warning operations.

[0029] Furthermore, before the GAF image transformation, the length and quality of the pressure pulsation signal are checked, and the signal is reacquired if the requirements are not met.

[0030] Furthermore, the multi-channel convolutional neural network includes a multi-channel input layer, parallel convolution and pooling layers, a feature fusion layer, and a Softmax classifier, which can simultaneously process multiple types of input signals such as temperature, GAF images, vibration, dynamic strain, and hydraulic performance.

[0031] Technical effect

[0032] This invention provides a smart centrifugal pump and its fault diagnosis method, which has full-dimensional perception, covering all key parameters of hydraulics, mechanics, acoustics, and thermodynamics, laying a data foundation for accurate diagnosis; it has a hybrid intelligent diagnosis method to achieve accuracy and reliability. This centrifugal pump fault diagnosis method combines the advantages of stacked autoencoders in processing time-series signals and CNNs in processing image signals. It adopts the optimal algorithm for signals with different physical characteristics, realizing fine differentiation of faults such as cavitation, corrosion, wear, and cracks, and achieving high diagnostic accuracy.

[0033] This method links fault diagnosis with actuators (frequency converters, active vibration isolators), achieving a leap from "passive alarm" to "active intervention" and improving the inherent safety level of the system. Through performance trend analysis and early fault diagnosis, it guides condition-based maintenance, avoiding over-maintenance and sudden failures, and reducing total lifecycle costs. The entire process requires minimal manual feature engineering, achieving end-to-end automated fault diagnosis, reducing reliance on human experience, and demonstrating a high degree of intelligence.

[0034] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the system structure of a smart centrifugal pump according to a preferred embodiment of the present invention;

[0036] Figure 2 This is a schematic diagram of the structure of a smart centrifugal pump according to a preferred embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the sensor arrangement of a smart centrifugal pump according to a preferred embodiment of the present invention;

[0038] Figure 4 This is a schematic diagram of the centrifugal pump impeller structure and the position of the pre-embedded dynamic strain gauges of a smart centrifugal pump according to a preferred embodiment of the present invention.

[0039] Figure 5 This is a flowchart of a system fault detection and unit vibration fault diagnosis method for a smart centrifugal pump based on vibration signals, which is a preferred embodiment of the present invention.

[0040] Figure 6 This is a flowchart of a fault diagnosis process for a unit overcurrent component of a smart centrifugal pump based on pressure pulsation signals and dynamic strain gauge signals, which is a preferred embodiment of the present invention.

[0041] Figure 7 This is a preferred embodiment of the processing flow of pressure pulsation signals of a smart centrifugal pump according to the present invention;

[0042] Figure 8 This is a block diagram of a multi-source operation and fault information fusion system for a smart centrifugal pump, according to a preferred embodiment of the present invention.

[0043] Among them, 1-unit base, 2-unit foundation, 3-unit motor, 4-active vibration isolator, 5-motor speed and torque meter coupling, 6-unit base, 7-speed and torque meter, 8-speed and torque meter support, 9-speed and torque meter centrifugal pump coupling, 10-centrifugal pump body support, 11-centrifugal pump suspension bearing assembly, 12-centrifugal pump body, 13-centrifugal pump outlet pressure measuring pipe section, 14-centrifugal pump outlet vertical pipe section, 15-centrifugal pump outlet horizontal pipe section, 16-centrifugal pump cover, 17-electromagnetic flowmeter, 18-centrifugal pump inlet pressure measuring pipe section, 19-centrifugal pump inlet pipe section, 20-flow regulating valve, 21-impeller front cover plate, 22-blade suction surface, 23-blade suction surface pre-embedded strain gauge, 24-impeller inlet ring, 25-impeller rear cover plate, 26-blade working surface, 27-blade working surface pre-embedded strain gauge, 28-impeller outlet surface. Detailed Implementation

[0044] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0045] In the following description, specific details, such as particular internal procedures and techniques, are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will appreciate that the invention may be practiced in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of the invention with unnecessary detail.

[0046] like Figure 1 As shown in the figure, this invention provides a smart centrifugal pump, the core of which lies in constructing a closed-loop operation and maintenance system that includes information sensing, information processing, optimized operation, and intelligent maintenance of the centrifugal pump system, realizing a system from physical perception to intelligent decision-making. The overall data processing and diagnostic process of the system follows the structure shown in the attached figures, and the specific implementation process is as follows:

[0047] 1. Data acquisition from multiple channels and types of sensors

[0048] The system first collects operational data of multiple physical quantities simultaneously through a sensor network deployed at key parts of the centrifugal pump unit:

[0049] (1) Slowly changing signal sensor: used to monitor the macroscopic state and performance trend of the system.

[0050] a. Torque-tachometer: Measures the pump shaft speed and torque in real time, providing a basis for calculating hydraulic performance and identifying overload.

[0051] b. Electromagnetic flow meter: monitors the instantaneous flow rate in pipelines.

[0052] c. Temperature sensor: placed in motor bearings, pump bearings, etc., to monitor temperature rise changes.

[0053] d. Differential pressure sensor: installed at the inlet and outlet of the pump to calculate the pump head.

[0054] (2) Dynamic signal sensor: used to capture dynamic high-frequency information related to faults.

[0055] a. Pressure pulsation sensor: Installed on the inner wall of the pump volute and pump cover, it sensitively captures dynamic pressure fluctuations caused by impeller rotation, and is particularly sensitive to hydraulic faults such as impeller blade cracks, fractures, and cavitation.

[0056] b. Dynamic strain gauges: Pre-embedded or attached to key stress-bearing parts such as impeller blades, they directly sense dynamic stress changes caused by cavitation, corrosion or cracks. The signals are received by fiber optic grating sensors or telemetry transmission systems arranged on the centrifugal pump body near the main shaft, and are supplemented by pressure pulsation sensors to sense fault information of impeller components.

[0057] c. Vibration sensors: Multi-axis vibration sensors are arranged in bearing housings, pump bases, and other locations to comprehensively monitor the mechanical vibration status.

[0058] 2. Signal Acquisition, Integration, and Preprocessing

[0059] The analog signals output from all sensors are synchronously acquired, converted from analog to digital, and aggregated into a unified digital signal stream by the data acquisition system. Subsequently, the raw signals undergo preprocessing to improve data quality.

[0060] (1) Smooth filtering and outlier processing are performed on slowly changing signals (such as temperature and efficiency) to eliminate measurement noise and pulse interference and retain the true performance trend.

[0061] (2) Filter, denoise (e.g., wavelet denoising) and normalize dynamic signals (such as vibration and pressure pulsation) to eliminate background noise, enhance the signal-to-noise ratio of fault features, and prepare for subsequent feature extraction.

[0062] 3. Multi-dimensional feature extraction and optimization

[0063] The preprocessed data is then fed into the feature extraction module to extract fault information from different perspectives:

[0064] (1) Extracting from slowly varying signals:

[0065] a. Trend characteristics: such as performance degradation rate, temperature change slope, linear regression slope within the sliding window, curvature, etc., to capture performance degradation trends and reflect the long-term health status of the equipment.

[0066] b. Statistical characteristics: such as mean, variance, peak value, kurtosis, etc., which characterize the overall distribution of the signal.

[0067] c. Time-domain characteristics: such as mean, standard deviation, root mean square, etc.

[0068] d. Frequency domain characteristics: Low-frequency components are obtained through Fourier transform to analyze periodic, slowly varying faults.

[0069] (2) Extracting from dynamic signals:

[0070] a. Time-domain characteristics: including root mean square value, peak value, waveform indicators, kurtosis, etc., reflecting the amplitude energy and impulse characteristics of the signal.

[0071] b. Frequency domain characteristics: The spectrum is obtained through fast Fourier transform, and the amplitude, harmonic components, and spectral centroid of characteristic frequencies are extracted.

[0072] c. Time-frequency domain comprehensive characteristics: Short-time Fourier transform or wavelet transform is used to obtain the energy distribution of the signal in the joint time and frequency domains, which is suitable for non-stationary signal analysis.

[0073] d. Nonlinear characteristics: such as approximate entropy and Lyapunov exponent, used to reveal the complex characteristics in the dynamic behavior of a system.

[0074] (3) Subsequently, principal component analysis or maximum correlation minimum redundancy algorithm is used to select features and reduce data dimensionality of the constructed high-dimensional feature set, remove redundant information, retain the feature subset most sensitive to faults, and improve the efficiency and generalization ability of subsequent models.

[0075] 4. Fault diagnosis and operation monitoring model decision making

[0076] The optimized feature vectors after dimensionality reduction are input in parallel into multiple dedicated intelligent diagnostic and monitoring models:

[0077] (1) Motor fault diagnosis model: mainly based on current and vibration characteristics, to diagnose motor faults such as rotor bar breakage, bearing wear, and eccentricity.

[0078] (2) Fault diagnosis model for flow components: This is the core of the system. It integrates dynamic strain, pressure pulsation and vibration characteristics, and uses a trained hybrid intelligent algorithm (such as stacked autoencoder and convolutional neural network) to perform fine classification and identification of faults such as cavitation, corrosion, wear and cracks of the impeller.

[0079] (3) Unit hydraulic performance monitoring model: Real-time efficiency is calculated based on flow rate, head, torque and speed to evaluate the performance status and deterioration trend of the pump.

[0080] (4) Unit vibration performance monitoring model: Based on the vibration characteristics of each measuring point, the overall mechanical operating status of the unit is evaluated.

[0081] 5. Multi-source information fusion and intelligent control execution

[0082] The central controller fuses multi-source operational and fault information from the diagnostic and monitoring results of all the above models to form a comprehensive system status assessment conclusion. Based on this conclusion, the real-time intelligent adjustment and early warning system is triggered and executes corresponding operations:

[0083] (1) Over-temperature and overload protection system: When the temperature or torque exceeds the safety threshold, the frequency converter will automatically reduce the motor speed or perform a shutdown operation to prevent equipment damage.

[0084] (2) Active vibration isolation compensation system: When the vibration amplitude exceeds the preset limit, the active vibration isolator installed under the base is activated immediately to generate a counterforce to offset and suppress the vibration.

[0085] (3) Intelligent hydraulic performance adjustment system: If cavitation or performance degradation is diagnosed, the flow regulating valve can be automatically adjusted or the speed can be adjusted to restore the pump to high-efficiency operation.

[0086] (4) Fault warning and operation alarm system: Based on the severity of the fault, different levels of warning or alarm information (such as "early wear warning", "moderate cavitation warning" and "serious crack alarm") are generated and sent to the operators and shore-based monitoring center through human-machine interface or remote transmission unit to guide maintenance decisions.

[0087] like Figure 2 As shown, this invention uses a traditional single-stage single-suction centrifugal pump as an example to demonstrate the structure of the centrifugal pump, as well as the arrangement of the torque and speed meter, inlet and outlet pressure measuring sections, and flow regulating valve, which will not be described in detail here.

[0088] like Figure 3 As shown, this invention constructs a complete multi-parameter sensor network based on the traditional centrifugal pump structure, realizing comprehensive perception of the unit's hydraulic, mechanical, and thermodynamic states. Specifically, 3a1, 3a2, 3a3 are vibration sensors on the motor; 3b is a temperature sensor on the motor; 3a1, 3a2, 3a3 are vibration sensors on the unit base; 7a is a vibration sensor on the torque-speed meter; 10a is a vibration sensor on the centrifugal pump body support; 11a is a vibration sensor near the centrifugal pump suspension bearing; 11b is a temperature sensor near the centrifugal pump suspension bearing; 11e is a fiber optic grating sensor or telemetry transmission system; and 12a is a vibration sensor on the centrifugal pump body. 12c1, 12c2, 12c3, 12c4, 12c5 - Pressure pulsation sensors on the inner wall of the centrifugal pump body; 13a1, 13a2 - Vibration sensors on the outlet pressure measuring pipe section of the centrifugal pump; 13d1, 13d2 - Differential pressure sensors on the outlet pressure measuring pipe section of the centrifugal pump; 16a1, 16a2 - Vibration sensors on the pump cover of the centrifugal pump; 16c1, 16c2 - Pressure pulsation sensors on the inner wall of the pump cover of the centrifugal pump; 18d1, 18d2 - Differential pressure sensors on the inlet pressure measuring pipe section of the centrifugal pump.

[0089] The specific arrangement and function of each sensor are as follows:

[0090] (1) Hydraulic performance monitoring sensor

[0091] Real-time monitoring of hydraulic performance parameters is fundamental to evaluating the efficiency and condition of centrifugal pumps. This system achieves this through the following sensors:

[0092] a. Electromagnetic flow meter (17): Installed on the pump outlet pipeline, used to accurately measure the instantaneous flow rate of the conveyed medium.

[0093] b. Differential pressure sensor: including inlet differential pressure sensor (18d1, 18d2) and outlet differential pressure sensor (13d1, 13d2), which are installed in the inlet pressure measuring pipe section (18) and outlet pressure measuring pipe section (13) of the centrifugal pump, respectively, to obtain the inlet and outlet pressure of the pump and then calculate the head.

[0094] c. Speed ​​and torque meter (7): Connected between the unit motor (3) and the centrifugal pump body (12) via couplings (5, 9) respectively, for directly measuring the speed, torque and input power of the pump shaft.

[0095] (2) Vibration monitoring sensor

[0096] To capture the mechanical dynamics of the unit during operation, vibration sensors are placed on all critical mechanical structures:

[0097] a. Motor vibration monitoring: Three single-axis (or three-axis) vibration sensors (3a1, 3a2, 3a3) are arranged on the generator motor (3) to monitor the vibration status of the motor body.

[0098] b. Vibration monitoring of transmission chain and pump body: Vibration sensors are sequentially arranged on the speed and torque meter (7a), centrifugal pump body support (10a), centrifugal pump suspension bearing component (11a), centrifugal pump body (12a), centrifugal pump outlet pressure measuring pipe section (13a1, 13a2) and centrifugal pump cover (16a1, 16a2), forming a complete vibration transmission path monitoring network.

[0099] (3) Impeller dynamic and special monitoring sensors

[0100] This is the core of detailed fault diagnosis; the sensor directly or indirectly monitors the working status of the impeller (21-28):

[0101] a. High-frequency pressure pulsation sensors: To directly capture the dynamic pressure generated by the interaction between impeller rotation and fluid, five pressure pulsation sensors (12c1 to 12c5) are arranged on the inner wall of the pump volute, and two sensors (16c1, 16c2) are arranged on the inner wall of the pump cover. These signals are extremely sensitive to faults such as cavitation and blade passage frequency.

[0102] b. Dynamic strain monitoring: To enable early diagnosis of blade stress state and cracks, dynamic strain gauges are pre-embedded on the blade suction surface (23) and the blade working surface (27). The signals are reliably transmitted from the rotating impeller to the stationary data acquisition system via fiber optic grating sensors or a rotating component telemetry system (11e).

[0103] (4) Temperature monitoring sensor

[0104] Used for over-temperature warning and thermal fault diagnosis:

[0105] Temperature sensors are placed at key heat source locations such as the motor bearing (3b) and the centrifugal pump suspension bearing (11b) to monitor temperature rise in real time.

[0106] In summary, the sensor system layout of this invention can be clearly summarized in the following table:

[0107]

[0108] All sensor data is ultimately connected to the central controller via shielded cables or optical fibers, providing a comprehensive, synchronous, and high-quality data foundation for subsequent hybrid intelligent diagnostic algorithms.

[0109] like Figure 4 As shown, in this embodiment of the invention, considering manufacturing factors, stress concentration may occur at the connection between the impeller blade and the front and rear cover plates due to casting sand inclusions, porosity, or welding. Near the blade outlet, load concentration occurs due to the large pressure difference between the blade pressure surface and suction surface. Therefore, dynamic strain gauges are pre-embedded on the blade pressure surface near the blade inlet, near the front cover plate, near the rear cover plate, and near the blade outlet. Figure 4 The layout location was shown.

[0110] like Figure 5 As shown in the figure, this invention provides a fault diagnosis method for a smart centrifugal pump, which is a multi-level intelligent fault diagnosis and active control system based on vibration signals. The specific implementation process is as follows:

[0111] (1) First step: Vibration signal feature extraction and compression

[0112] The system collects raw vibration signals in real time through a network of vibration sensors pre-deployed at key parts of the unit. These signals are directly sent to the core analysis module without the need for complex sample construction and preprocessing.

[0113] a. Deep feature extraction: The original vibration signal is nonlinearly transformed using a stacked sparse autoencoder (SSAE) to automatically learn and extract high-order hidden features that can characterize the health status of the equipment.

[0114] b. Feature compression: Subsequently, principal component analysis (PCA) is used to reduce the dimensionality of the above-mentioned hidden features, resulting in highly condensed and non-redundant compressed features, laying the foundation for accurate and rapid fault diagnosis in the future.

[0115] (2) Second step: Primary fault diagnosis and classification

[0116] The compressed feature vector is input into a trained Support Vector Machine (SVM) classifier. This classifier classifies the device's current operating state into one of the following three categories:

[0117] a. Normal operation;

[0118] b. Abnormal vibration of the mechanical system;

[0119] c. Abnormal vibration in the current-carrying system.

[0120] (3) Third step: Decision-making and hierarchical fault tracing and control

[0121] Based on the classification results of the SVM, the central controller triggers different advanced diagnostic and control processes, forming a decision-making closed loop:

[0122] a. If the diagnosis is "abnormal vibration of the overcurrent system":

[0123] i. Trigger secondary diagnostics: The system automatically initiates a detailed fault diagnosis process for unit overcurrent components based on pressure pulsation signals and dynamic strain gauge signals.

[0124] ii. Root cause analysis: This process will integrate the high-frequency hydraulic characteristics of pressure pulsation and the direct stress information of dynamic strain to deeply analyze the root cause of vibration and accurately identify specific faults such as cavitation, corrosion, wear, or blade cracks.

[0125] b. If the diagnosis is "abnormal vibration of the mechanical system":

[0126] i. Triggering Systematic Troubleshooting: The system analyzes vibration signals from the unit support, motor body, speed and torque meter, centrifugal pump suspension bearing components, centrifugal pump body, centrifugal pump outlet pressure measuring pipe section, and centrifugal pump cover in a preset logical sequence to accurately pinpoint the source of the fault.

[0127] ii. Information Fusion and Active Control: The investigation results are transmitted to the multi-source operation and fault information fusion system. If the final determination is that the vibration of the unit support exceeds the standard, the system immediately sends a command to the active vibration isolation compensation system to activate the active vibration isolators installed under the base, generating a counterforce to offset and suppress the vibration in real time, thus realizing a closed loop from diagnosis to control.

[0128] like Figure 6 As shown, the present invention provides a refined fault diagnosis for flow-through components of a generator unit (such as a centrifugal pump impeller) by fusing pressure pulsation and dynamic strain gauge signals, and employing advanced signal-to-image conversion and deep learning technologies. The specific implementation process is as follows:

[0129] (1) Synchronous acquisition and preprocessing of dual-channel signals

[0130] The system simultaneously acquires two types of key dynamic signals:

[0131] a. Pressure pulsation raw signal: captured by a high-frequency pressure pulsation sensor arranged on the inner wall of the pump volute and pump cover. This signal is extremely sensitive to hydraulic excitations caused by impeller rotation (such as cavitation and blade passing frequency).

[0132] b. Dynamic strain gauge signal: generated by dynamic strain gauges embedded or attached to the impeller blades and received by fiber optic grating sensors or telemetry transmission systems. This signal directly reflects the dynamic stress changes of the blades caused by fluid action and is direct evidence for identifying cracks and fatigue.

[0133] The acquired raw signals first enter the signal preprocessing module, where they are processed sequentially:

[0134] a. Filtering: Eliminating irrelevant high-frequency noise and power frequency interference from the signal.

[0135] b. Noise reduction: Advanced algorithms such as wavelet threshold denoising are adopted to further improve the signal-to-noise ratio and highlight fault characteristics.

[0136] c. Normalization: Eliminates the influence of amplitude dimensions, adjusts the signal to a uniform scale, and lays the foundation for accurate subsequent analysis.

[0137] (2) Time-domain signal imaging based on GAF

[0138] The preprocessed pressure pulsation signal is sent to the Gram angle field transformation module. This is the core innovative step, and its process is as follows:

[0139] a. A one-dimensional time-domain pressure pulsation sequence is mapped into a two-dimensional GAF image that preserves its time-domain correlation information through mathematical transformation (Gram angle sum field or Gram angle difference field).

[0140] b. The image is a high-resolution grayscale image, and the geometric structure between its pixels reflects the time dependence and dynamic characteristics of the original signal, thus transforming the fault diagnosis problem into a more suitable image recognition problem.

[0141] (3) Multi-channel convolutional neural network image recognition and diagnosis

[0142] The generated GAF ​​image and the preprocessed dynamic strain gauge data (which can be used as another channel or auxiliary feature) are jointly input into a pre-trained multi-channel convolutional neural network for intelligent diagnosis.

[0143] a. Multi-channel input layer: Responsible for receiving GAF images and dynamic strain features to construct multi-source information input.

[0144] b. Feature extraction convolutional layer: The network automatically extracts spatial features (such as edges, textures, and complex patterns) from the GAF image in layers from shallow to deep. These features are highly correlated with specific faults of the flow components (such as cavitation vortices and stress concentration caused by cracks).

[0145] c. Feature Fusion and Classification Layer: The multi-path features from pressure pulsation and dynamic strain gauges are deeply fused, and finally the Softmax classifier outputs accurate diagnostic results, such as "moderate cavitation with early wear" or "micro-cracks appear on the suction surface of the blade".

[0146] (4) Integration of diagnostic results and decision-making

[0147] The final diagnostic results will be sent to the system's multi-source fault information and operation information fusion center, where they will be comprehensively analyzed along with vibration diagnostic results, performance evaluation data, etc., providing the most critical basis for subsequent decisions such as whether to reduce speed, implement early warnings, or carry out planned maintenance.

[0148] Figure 7 This invention describes the processing flow for pressure pulsation signals. To achieve fault diagnosis based on deep learning, the one-dimensional pressure pulsation signal needs to be converted into a two-dimensional image that can be recognized by a convolutional neural network. The specific implementation of this signal processing flow is as follows:

[0149] (1) Acquisition and preprocessing of pressure pulsation signals

[0150] The system continuously acquires raw pressure pulsation time-domain signals through high-frequency pressure pulsation sensors installed on the inner walls of the pump casing and pump cover. These raw signals contain noise and have inconsistent dimensions, requiring immediate input to a preprocessing module for optimization. Filtering, noise reduction, and normalization are used to eliminate amplitude fluctuations caused by changes in operating conditions, establishing a unified standard for subsequent precise transformations.

[0151] (2) Signal quality and length verification

[0152] After preprocessing, the system automatically determines whether the length of the signal segment meets the requirements (for example, it must contain data from at least several impeller rotation cycles to ensure that periodic fault characteristics can be captured).

[0153] a. If “No”, meaning the signal length is insufficient or there is missing data, the system will feed back to the control unit to reacquire a pressure pulsation time-domain signal that meets the minimum length requirement.

[0154] b. If "yes", meaning that both the signal quality and length meet the standards, then it is allowed to enter the core GAF conversion module.

[0155] (3) GAF image generation

[0156] This step encodes a one-dimensional time-series signal into a two-dimensional image through mathematical transformation. Specifically, it involves three stages: polar coordinate transformation (mapping the preprocessed time-domain signal to a polar coordinate system), calculating the Gram matrix (capturing the correlation between time points from different perspectives), and generating a two-dimensional GAF image (outputting the matrix as a two-dimensional grayscale GAF image). This process fully encodes the temporal dependence and dynamic characteristics of the original pressure pulsation signal through the texture and structural patterns of the image.

[0157] (4) Output to the diagnostic module

[0158] Finally, the generated GAF ​​image is output to the fault diagnosis module as the direct input of the pre-trained multi-channel convolutional neural network for subsequent image feature extraction and fault classification, thereby realizing intelligent identification of faults in flow components such as cavitation, wear, and cracks.

[0159] Figure 8 The multi-source operation and fault information fusion system of this invention is based on a specially designed multi-channel convolutional neural network. This network can process different types of data in parallel and perform deep fusion at a high-level feature level, thereby achieving high-precision comprehensive fault diagnosis. The specific implementation process is as follows:

[0160] (1) Multi-channel input layer: Standardized access to heterogeneous data

[0161] The system standardizes and integrates multi-source data from different sensors, with varying physical meanings and formats, into different input channels of the network:

[0162] a. Channel 1: Receives a one-dimensional sequence of preprocessed temperature signals.

[0163] b. Channel 2: Receives a two-dimensional GAF image generated by the pressure pulsation signal through GAF transformation.

[0164] c. Channel 3: Receives a one-dimensional sequence of preprocessed vibration signals (which can be converted into a spectrogram or directly used for one-dimensional convolution).

[0165] d. Channel 4: Receives a one-dimensional sequence of dynamic strain gauge signals.

[0166] e. Channel 5: Receives a one-dimensional feature vector composed of hydraulic performance and operating parameters (such as efficiency, head, flow rate, etc.).

[0167] (2) Parallel feature extraction: targeted deep learning

[0168] Each input channel is configured with an independent convolutional neural network submodule to extract the deepest features of this type of data:

[0169] a. Convolution and Activation: Each submodule contains multiple convolutional layers and the ReLU activation function. The convolutional layers are responsible for locally sensing and extracting features from the raw data (such as extracting texture from GAF images and extracting impact patterns from vibration signals), while the ReLU function introduces non-linearity to enhance the model's expressive power.

[0170] b. Pooling layer dimensionality reduction: Max pooling layers are interspersed between convolutional layers to compress the size of feature maps, preserve the most salient features while reducing computational cost, and enhance the spatial invariance of features. For specific channels (as shown in the figure), min pooling layers may also be used to capture different feature responses.

[0171] (3) Feature fusion layer: Coordination and splicing of multi-source information

[0172] After deep feature extraction by each parallel submodule, the high-level feature maps output from all channels (which, for one-dimensional signals, are typically flattened into feature vectors) are fed to the feature fusion layer. At this layer, feature vectors from temperature, pressure fluctuations, vibration, strain, and performance parameters are combined into a global, comprehensive joint feature vector using a concatenation method. This step achieves deep fusion of information from different physical sources at the feature level.

[0173] (4) Comprehensive classification and diagnostic output

[0174] The fused joint feature vector contains a complete description of the device state and is then fed into the network's decision-making part.

[0175] a. Fully connected layer: One or more fully connected layers are responsible for learning the complex nonlinear mapping between these fused features and the final fault category.

[0176] b. Softmax classifier: The last layer of the network uses the Softmax function to convert the output of the fully connected layer into a probability distribution corresponding to each fault type.

[0177] c. Fault diagnosis classification output: The system ultimately outputs the fault category with the highest probability as the diagnostic result, such as "bearing wear with mild cavitation" or "impeller crack", which realizes a fine and reliable diagnosis based on multi-source information fusion.

[0178] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A smart centrifugal pump, comprising a centrifugal pump body, a drive motor, and a frequency converter, characterized in that, It also includes a comprehensive fault diagnosis system, which includes... A multi-level sensor sensing network, comprising a group of slowly varying signal sensors and a group of dynamic signal sensors, is used to collect multi-physical quantity operating data of the centrifugal pump unit. The signal processing and hybrid intelligent diagnostic module is used to preprocess, extract and optimize features of the acquired signals, and use multiple dedicated diagnostic models for state recognition and fault classification. The multi-source information fusion and active control module is used to fuse the diagnostic results of various models to make decisions and link the actuators to achieve closed-loop control.

2. The intelligent centrifugal pump as described in claim 1, characterized in that, The gradually changing signal sensor group includes a torque tachometer, an electromagnetic flowmeter, a temperature sensor, and a differential pressure sensor. The torque tachometer is used to measure the pump shaft speed and torque in real time, providing a basis for calculating hydraulic performance and identifying overload. The electromagnetic flowmeter is used to monitor the instantaneous flow rate in the pipeline. The temperature sensor is arranged in the motor bearing, pump bearing, or other parts to monitor temperature rise changes. The differential pressure sensor is installed at the pump inlet and outlet to calculate the pump head. The dynamic signal sensor group includes pressure pulsation sensors arranged on the inner wall of the pump volute and pump cover, dynamic strain gauges embedded in the impeller blades, and multi-axis vibration sensors arranged at various key mechanical points of the unit; the signals of the dynamic strain gauges are transmitted through fiber optic grating sensors or rotating component telemetry systems.

3. The intelligent centrifugal pump as described in claim 1, characterized in that, The signal processing and hybrid intelligent diagnostic module is configured to perform the following steps: The vibration signal is extracted using a stacked sparse autoencoder for deep feature extraction, then reduced in dimensionality by principal component analysis, and input into a support vector machine classifier to classify the equipment status into normal operation, abnormal vibration of mechanical system, or abnormal vibration of overcurrent system. After filtering, denoising, and normalizing the pressure pulsation signal, a two-dimensional image is generated through Gram angle field transformation. The two-dimensional image and the dynamic strain gauge signal are input into a multi-channel convolutional neural network for feature extraction and fusion to identify cavitation, corrosion, wear, cracks and fracture faults in the impeller.

4. The intelligent centrifugal pump as described in claim 3, characterized in that, The Gram angle field transformation specifically includes mapping a one-dimensional pressure pulsation signal to a polar coordinate system and generating a two-dimensional grayscale image by calculating the Gram angle sum field and the Gram angle difference field.

5. A smart centrifugal pump as described in claim 1, characterized in that, The multi-source information fusion and active control module includes Over-temperature and overload protection system, used to automatically adjust speed or stop the machine via frequency converter when temperature or torque exceeds the limit; An active vibration isolation compensation system is used to activate active vibration isolators to suppress vibration when the vibration amplitude exceeds the limit; Intelligent hydraulic performance regulation system, used to automatically or suggest adjusting flow valves or rotation speed based on performance evaluation and fault diagnosis results; The fault early warning and operation alarm system is used to generate graded alarm information and transmit it remotely.

6. A smart centrifugal pump as described in claim 5, characterized in that, The activation condition of the active vibration isolation compensation system is that when the mechanical system is diagnosed as having abnormal vibration, and the multi-source information fusion system determines that the vibration of the unit support exceeds the standard, the central controller triggers the active vibration isolator to work.

7. A smart centrifugal pump as described in claim 3, characterized in that, When an abnormal vibration of the flow system is diagnosed, a fine diagnostic process for the flow components based on pressure pulsation signals and dynamic strain gauge signals is triggered to further identify the root causes of cavitation, corrosion, wear, or blade cracks.

8. A fault diagnosis method for a smart centrifugal pump as described in any one of claims 1-7, characterized in that, Includes the following steps: (1) The slowly changing signal and the dynamic signal are collected synchronously through the multi-level sensor network; (2) Perform deep feature extraction and dimensionality reduction on the vibration signal, and use a classifier to make an initial judgment on the abnormal state; (3) Perform GAF image transformation on the pressure pulsation signal, and combine it with the dynamic strain signal to achieve fine fault classification through a convolutional neural network; (4) Integrate multi-source diagnostic information to trigger corresponding protection, vibration isolation, adjustment or early warning operations.

9. The fault diagnosis method for a smart centrifugal pump as described in claim 8, characterized in that, Before the GAF image transformation, the length and quality of the pressure pulsation signal are checked, and the signal is reacquired if the requirements are not met.

10. The fault diagnosis method for a smart centrifugal pump as described in claim 8, characterized in that, The multi-channel convolutional neural network includes a multi-channel input layer, parallel convolution and pooling layers, a feature fusion layer, and a Softmax classifier, and can simultaneously process multiple types of input signals such as temperature, GAF images, vibration, dynamic strain, and hydraulic performance.

Images