Centrifugal pump winding data model

Abstract

The invention belongs to the technical field of centrifugal pump fault diagnosis, and particularly relates to a centrifugal pump winding data model which specifically comprises the following steps: key sensing dimension and data acquisition: acquiring core data through four types of sensors, and extracting time sequence characteristics strongly related to impeller faults; according to the time dimension, the winding or sudden cracking fault of the impeller is represented as parameter sudden jump, winding jump is monitored through a tension sensor, and specifically, real-time data of flow, vibration and current of the centrifugal pump are sampled; according to the centrifugal pump winding data model, real-time data monitoring is conducted on the condition in the using process of a centrifugal pump through key sensing dimension and data collection, RNN model design and comprehensive fault index gamma and fault judgment, and corresponding problems of the centrifugal pump are directly diagnosed according to different data feedbacks in the using process. Therefore, the centrifugal pump can be overhauled or maintained conveniently and timely, the working quality of the centrifugal pump is guaranteed, and the service life of the centrifugal pump is prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of centrifugal pump fault diagnosis, in particular to a centrifugal pump winding data model. BACKGROUND

[0002] The centrifugal pump is a kind of blade pump, which uses the centrifugal force generated by the high-speed rotation of the impeller to transport liquid. Its core working principle involves filling the water before starting, and when the impeller rotates, it drives the liquid to rotate at high speed. Under the action of centrifugal force, the liquid is thrown to the outer edge of the impeller and pressed into the water outlet pipe, while a low pressure area is formed at the center of the impeller, and the external liquid is sucked into the pump by atmospheric pressure to circulate and transport.

[0003] The centrifugal pump still has the following technical problems in use:

[0004] At present, some centrifugal pumps are used in deep underwater locations. In order to ensure the normal use of the centrifugal pump, it needs to be pulled out of the water regularly for maintenance. However, this operation has the problem that the use of the centrifugal pump cannot be directly understood through the data monitoring of the centrifugal pump, so that the use of the centrifugal pump cannot be understood in real time.

[0005] Therefore, the present application provides a centrifugal pump winding data model to solve the above-mentioned problems. SUMMARY

[0006] The purpose of the present application is to provide a centrifugal pump winding data model to solve the problems mentioned in the background.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solution: a centrifugal pump winding data model, the specific content of which is as follows:

[0008] I. Key perception dimension and data acquisition: 4 types of sensors are used to collect core data, and time sequence features strongly related to impeller failure are extracted;

[0009] (1) Time dimension: the appearance of impeller winding or sudden cracking failure is characterized by sudden jump of parameters. The winding jump is monitored by a tension sensor, and the real-time data of the flow, vibration and current of the centrifugal pump are sampled;

[0010] (2) Pump efficiency and flow: the water outlet efficiency of the centrifugal pump and the model core input data are collected by the water outlet flow sensor, inlet and outlet pressure sensor and ammeter;

[0011] (3) Vibration: a three-axis acceleration sensor is installed inside the bearing box to collect vibration data;

[0012] (4) Current and stator temperature: the data of current and stator temperature are collected by three-phase ammeter and stator coil temperature sensor;

[0013] II. RNN model design: long short-term memory network is used to process time series data to solve the gradient disappearance problem of ordinary RNN, which is more suitable for capturing the dynamic change rule of faults;

[0014] (1) Alpha model (vibration shaft imbalance quantification): the input of the alpha model includes a vibration time series feature vector;

[0015] (2) Beta model (efficiency decay quantification): the input of the beta model includes a water outflow efficiency time series feature vector;

[0016] III. Comprehensive fault index γ and fault determination:

[0017] (1) γ value calculation and threshold: γ = 0.6α + 0.4β;

[0018] (2) Simple logic troubleshooting: the γ value early warning result is verified again through the rule base to exclude interference;

[0019] Non-impeller fault types include main shaft wear, oil chamber water ingress, water ingress and sand, and cavitation;

[0020] (3) Impeller fault subdivision: including impeller increase and impeller decrease.

[0021] Preferably, the key perception dimension and the time dimension perception item of data acquisition have a frequency of 1-10 Hz, the feature extraction includes calculating the parameter instantaneous change rate (such as Δflow / Δt, Δvibration amplitude / Δt), and the time window is set to 10 seconds; when the change rate exceeds 3 times the standard deviation of the normal fluctuation range, it is marked as a jump event;

[0022] In the pump efficiency and flow perception item, the core indicators include real-time flow Q, pump efficiency η and efficiency decay rate, the real-time flow Q directly reflects the impeller water delivery capacity, and winding will cause Q to drop sharply; the pump efficiency H is the head, calculated by pressure difference, U and I are motor voltage and current; the efficiency decay rate is the decline amplitude of η within 30 seconds;

[0023] In the vibration perception item, the core indicators include 1 times frequency amplitude A1 and 1 times frequency proportion; the 1 times frequency amplitude A1 is synchronized with the vibration frequency of the rotating speed, and winding causes uneven impeller mass, A1 will significantly increase; the 1 times frequency proportion is the ratio of A1 to total vibration energy;

[0024] In the current and stator temperature perception item, the core indicators include current overload coefficient and stator temperature rise rate, the current overload coefficient is actual current / rated current; the stator temperature rise rate is the temperature rise value within 5 minutes.

[0025] Preferably, the network structure of the alpha model in the RNN model design comprises an input layer, an LSTM layer and an output layer, the input layer is 3 vibration features and 60 time steps; the LSTM layer is two layers, the first layer is 64 neurons, and the second layer is 32 neurons; the output layer includes one neuron, outputs the alpha value and the quantization axis imbalance degree.

[0026] Preferably, the network structure of the beta model in the RNN model design comprises an input layer, an LSTM layer and an output layer, the input layer is 3 efficacy features and 60 time steps; the LSTM layer is two layers, the first layer is 64 neurons, and the second layer is 32 neurons; the output layer includes one neuron, outputs the beta value and the quantization efficacy attenuation degree.

[0027] Preferably, in the calculation of the comprehensive failure index gamma and the gamma value of failure judgment and the threshold sensing item, the content of the threshold setting is: when gamma is greater than or equal to 0.5, it is a mild failure, and when gamma is greater than or equal to 0.8, it is a severe failure.

[0028] Preferably, in the simple logic troubleshooting sensing item of the comprehensive failure index gamma and the failure judgment, the troubleshooting rule of the main shaft wear includes no jump event; the alpha and beta values slowly rise over time; the vibration spectrum contains the bearing characteristic frequency;

[0029] The troubleshooting rule of oil chamber water entry is that the humidity sensor data is greater than 80% RH; the vibration signal is mixed with 20-50Hz low frequency noise;

[0030] The troubleshooting rule of water entry sand is that the vibration is 100-500Hz irregular broadband noise; the stator temperature is locally raised;

[0031] The troubleshooting rule of cavitation is that the inlet pressure is less than the medium vaporization pressure, the vibration contains 600-2500Hz high frequency noise, and the alpha value is less than 0.3.

[0032] Compared with the prior art, the beneficial effects of the present application are:

[0033] The centrifugal pump winding data model monitors the situation in the use process of the centrifugal pump in real time through key sensing dimensions, data acquisition, RNN model design and comprehensive failure index gamma and failure judgment, directly diagnoses the corresponding problems according to different data feedback in the use process, so as to facilitate timely maintenance or maintenance of the centrifugal pump, ensure the working quality of the centrifugal pump and prolong the service life thereof. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is a whole model schematic diagram of the present application. DETAILED DESCRIPTION

[0035] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0036] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore, it cannot be understood as a limitation on the present application.

[0037] Embodiment:

[0038] Please refer to Figure 1 The present application provides a technical solution:

[0039] A centrifugal pump winding data model, the specific content of the data model is as follows:

[0040] I. Key perception dimension and data acquisition: through 4 types of sensors to collect core data, extract time sequence features strongly related to impeller failure;

[0041] (1) Time dimension: the impeller appears winding or sudden burst failure, which is characterized by parameter sudden jump, the winding jump is monitored by the tension sensor, and the real-time data of the flow, vibration and current of the centrifugal pump are sampled;

[0042] (2) Pump efficiency and flow: the water outlet flow sensor, inlet and outlet pressure sensor and ammeter are used to collect the water outlet efficiency of the centrifugal pump and the model core input data;

[0043] (3) Vibration: the three-axis acceleration sensor is installed in the bearing box to collect vibration data;

[0044] (4) Current and stator temperature: the three-phase ammeter and stator coil temperature sensor are used to collect the data of current and stator temperature;

[0045] II. RNN model design: long short-term memory network is used to process time sequence data, solve the gradient disappearance problem of ordinary RNN, and is more suitable for capturing the dynamic change rule of failure;

[0046] (1) Alpha model (vibration shaft imbalance quantization): the input of the alpha model includes vibration time sequence feature vector;

[0047] (2) Beta model (quantification of water outflow efficiency decay): the input of the beta model includes the water outflow efficiency time series feature vector;

[0048] III. Comprehensive fault index γ and fault determination:

[0049] (1) γ value calculation and threshold: γ = 0.6α + 0.4β;

[0050] (2) Simple logic troubleshooting: secondary verification of γ value warning results through rule base to exclude interference;

[0051] Non-impeller fault types include main shaft wear, oil chamber water ingress, water ingress and sand, cavitation;

[0052] (3) Impeller fault subdivision: including impeller increase and impeller decrease.

[0053] In the time dimension perception item of the key perception dimension and data collection, the frequency of data collection is 1-10 Hz, the feature extraction includes calculating the parameter instantaneous change rate (such as Δflow / Δt, Δvibration amplitude / Δt), and the time window is set to 10 seconds; when the change rate exceeds 3 times the standard deviation of the normal fluctuation range, it is marked as a jump event;

[0054] In the pump efficiency and flow perception item, the core indicators include real-time flow Q, pump efficiency η and efficiency decay rate, the real-time flow Q directly reflects the impeller water delivery capacity, and winding will cause Q to drop sharply; the pump efficiency H is the head, calculated by pressure difference, U and I are motor voltage and current; the efficiency decay rate is the decline amplitude of η within 30 seconds;

[0055] In the vibration perception item, the core indicators include 1 times frequency amplitude A1 and 1 times frequency proportion; the 1 times frequency amplitude A1 is synchronized with the vibration frequency of the rotating speed, and winding causes uneven impeller mass, A1 will significantly increase; the 1 times frequency proportion is the ratio of A1 to total vibration energy;

[0056] In the current and stator temperature perception item, the core indicators include current overload coefficient and stator temperature rise rate, the current overload coefficient is the actual current / rated current; the stator temperature rise rate is the temperature rise value within 5 minutes.

[0057] In the RNN model design, the network structure of the alpha model includes an input layer, an LSTM layer and an output layer, the input layer is 3 vibration features, 60 time steps; the LSTM layer is two layers, the first layer is 64 neurons, and the second layer is 32 neurons; the output layer includes one neuron, outputs the α value and quantifies the imbalance degree of the axis.

[0058] The network structure of the β model in the RNN model design includes an input layer, an LSTM layer and an output layer, the input layer is 3 efficacy characteristics and 60 time steps; the LSTM layer is two layers, the first layer is 64 neurons and the second layer is 32 neurons; the output layer includes one neuron, outputs the β value and the quantified efficacy attenuation degree.

[0059] In the calculation of the comprehensive failure index γ and the γ value of failure determination and the threshold perception item, the content of threshold setting is: when γ≥0.5, it is a mild failure, and when γ≥0.8, it is a severe failure.

[0060] In the simple logical troubleshooting perception item of the comprehensive failure index γ and failure determination, the troubleshooting rules of main shaft wear include no jump event; α, β value slowly rises with time; vibration spectrum contains bearing characteristic frequency;

[0061] The troubleshooting rules of water entering the oil chamber are: humidity sensor data>80% RH; vibration signal mixed with 20-50Hz low frequency noise;

[0062] The troubleshooting rules of water and sand are: vibration is 100-500Hz irregular broadband noise; local temperature rise of stator;

[0063] The troubleshooting rules of cavitation are: inlet pressure<medium vaporization pressure, vibration contains 600-2500Hz high frequency noise, and α value<0.3.

[0064] In the winding data model of centrifugal pump, the evaluation of winding degree (such as 40%, 80%) needs to be quantitatively learned by RNN model on time series data of key perception dimensions, combined with the degree grading of vibration shaft imbalance α and water discharge efficacy attenuation β, and finally corresponding to the specific winding ratio through the weighted fusion value γ.

[0065] The following is the specific implementation method:

[0066] I. Definition and grading standard of winding degree

[0067] The proportion of the effective working area of the impeller covered by the winding is taken as the physical definition of the winding degree (such as 40% of the blade area is wound), and it is related to the quantifiable sensor data features, and the grading standard is as follows:

[0068]

[0069] II. Quantitative evaluation process of RNN model on winding degree

[0070] 1. Time series feature enhancement of input data

[0071] For the four perception dimensions of "time dimension jump", "vibration 1x frequency", "water flow", and "current and temperature", extract time series features highly correlated with winding degree as inputs for RNN:

[0072] Vibration features: time series change rate of 1x frequency amplitude (ΔA1 / Δt), standard deviation of amplitude fluctuation within 5 consecutive periods (reflecting the stability of imbalance);

[0073] Water output efficiency features: flow decay slope (ΔQ / Δt), cumulative decay amount of pump efficiency (∑Δη);

[0074] Current and temperature features: current overload duration (time length exceeding rated value), stator temperature rise rate (ΔT / Δt).

[0075] Group the above features into time series sequences according to time steps (e.g., every 1 second), with a length of 60 (i.e., 1 minute of historical data) as the input vector for RNN.

[0076] 2. RNN-based winding degree grading learning

[0077] Use a bidirectional long short-term memory network (Bi-LSTM, an improved version of RNN) to quantify the degree of α (vibration axis imbalance) and β (water output efficiency decay) respectively:

[0078] α model (vibration dimension):

[0079] Input: time series feature sequence of vibration 1x frequency;

[0080] Hidden layer: 2 layers of LSTM, capturing the nonlinear growth rule of 1x frequency amplitude when winding degree increases through the gating mechanism;

[0081] Output: α value (0-1), corresponding to the winding degree of the vibration dimension (e.g., α=0.4 corresponds to 40% winding vibration features, α=0.8 corresponds to 80% winding vibration features).

[0082] β model (water output efficiency dimension):

[0083] Input: time series feature sequence of flow, efficiency, and current;

[0084] Hidden layer: 2 layers of LSTM, learning the time series correspondence between winding degree and efficiency decay (e.g., 80% winding rate of flow drop);

[0085] Output: β value (0-1), corresponding to the winding degree of the water output efficiency dimension (e.g., β=0.4 corresponds to 40% winding efficiency features, β=0.8 corresponds to 80% winding efficiency features).

[0086] 3. Final evaluation of winding degree (γ value mapping)

[0087] By weighting the outputs of alpha and beta, a comprehensive measure value gamma = 0.6alpha + 0.4beta is obtained, and then the gamma value is mapped to the specific winding degree:

[0088]

[0089] III. Model training and precision guarantee

[0090] Label data generation: Through experimental simulation of different winding degrees (such as artificial winding of impeller 40% and 80% area), corresponding sensor data is collected and labeled as 40%, 80%, etc.

[0091] Loss function: Mean square error (MSE) loss is used to minimize the difference between the gamma value of the model output and the actual winding degree label.

[0092] Precision verification: On the test set, the winding degree evaluation error should be ≤5% (such as actual 40% winding, the model evaluation result is in the range of 35%-45%).

[0093] IV. Example description (winding degree evaluation process)

[0094] Data input: During the operation of a centrifugal pump, the 1x frequency vibration amplitude increases from 2mm / s to 5.6mm / s (increases by 180%), the flow rate decreases from 500m 3 / h to 200m 3 / h (decreases by 60%), and the current increases from 80A to 115A (increases by 43.75%), and the time series data is input into the model for 1 minute.

[0095] RNN output: The alpha model identifies the vibration feature corresponding to 80% winding (alpha = 0.82), and the beta model identifies the efficiency feature corresponding to 80% winding (beta = 0.78).

[0096] Gamma value calculation: gamma = 0.6x0.82 + 0.4x0.78 = 0.804, which is mapped to 80% winding.

[0097] Conclusion: The impeller winding degree reaches 80%, which needs to be urgently stopped for cleaning.

[0098] Through this method, the RNN model can accurately quantify the winding degree of the centrifugal pump, providing a hierarchical decision basis for "mild winding (40%) needs planned shutdown" and "severe winding (80%) needs emergency treatment" for operation and maintenance personnel, avoiding equipment overload damage.

[0099] It should be noted that the electrical components mentioned in the present application have been combed according to the actual situation during manufacturing, and will not cause wire harness winding or affect the work.

[0100] The standard parts used in the present application can be purchased from the market, and the special-shaped parts can be ordered according to the description and drawings, and the specific connection mode of each part adopts the conventional means such as bolts, rivets and welding in the prior art, the mechanical parts and equipment adopt the conventional type in the prior art, and the circuit connection adopts the conventional connection mode in the prior art, which will not be described in detail here, the control mode of the present application is controlled by the controller, and the control circuit of the controller can be realized by simple programming of the person skilled in the art, and it should be noted that the electrical components mentioned in the present application have been combed according to the actual situation when manufactured, which will not cause the wire harness to be entangled or affect the work, and the contents not described in detail in the specification belong to the prior art known to the person skilled in the art.

[0101] In the description of the present application, it should be understood that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0102] The basic principles and main features of the present application and the advantages of the present application are shown and described above, and it is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be realized in other specific forms without departing from the spirit or essential characteristics of the present application; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application, and any reference signs in the claims should not be regarded as limiting the claims involved.

[0103] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A centrifugal pump winding data model, characterized in that, The specific details of this data model are as follows: I. Key Perception Dimensions and Data Acquisition: Core data is collected through four types of sensors to extract time-series features strongly correlated with impeller failures; (1) Time dimension: The impeller entanglement or sudden breakage failure manifests as a sudden change in parameters. The entanglement change is monitored by a tension sensor, specifically by sampling the real-time data of the centrifugal pump's flow rate, vibration, and current. (2) Pump efficiency and flow rate: The pump efficiency and core input data of the model are collected by the outlet flow sensor, inlet and outlet pressure sensor and ammeter. (3) Vibration: A triaxial accelerometer is installed inside the bearing housing to collect vibration data; (4) Current and stator temperature: Data on current and stator temperature are collected using a three-phase ammeter and a stator coil temperature sensor; II. RNN Model Design: Long Short-Term Memory (LSTM) network is used to process time-series data, which solves the gradient vanishing problem of ordinary RNNs and is more suitable for capturing the dynamic change patterns of faults. (1) α model (quantization of vibration axis imbalance): The input of the α model includes vibration time sequence feature vector; (2) β model (quantification of water discharge efficiency decay): The input of the β model includes the time-series feature vector of water discharge efficiency; III. Comprehensive Fault Index γ and Fault Judgment: (1) Calculation of γ value and threshold: γ = 0.6α + 0.4β; (2) Simple logic check: The γ value warning result is verified twice through the rule base to eliminate interference; Non-impeller failure types include main shaft wear, water ingress into the oil chamber, cement and sand ingress, and cavitation. (3) Impeller-related faults are further subdivided into impeller increase and impeller decrease.

2. The centrifugal pump winding data model according to claim 1, characterized in that: In the key perception dimension and the time dimension of data acquisition, the data acquisition frequency is 1-10Hz, and the feature extraction includes calculating the instantaneous change rate of parameters (such as Δflow rate / Δt, Δvibration amplitude / Δt), with the time window set to 10 seconds; when the change rate exceeds 3 times the standard deviation of the normal fluctuation range, it is marked as a jump event; Among the pump efficiency and flow sensing parameters, the core indicators include real-time flow rate Q, pump efficiency η, and efficiency degradation rate. The real-time flow rate Q directly reflects the impeller's water delivery capacity; entanglement will cause Q to drop sharply. The pump efficiency... H represents the head, calculated from the pressure difference; U and I represent the motor voltage and current, respectively; the efficiency decay rate is the rate of decrease of η over 30 consecutive seconds. Among the vibration sensing items, the core indicators include the first harmonic amplitude A1 and the first harmonic percentage; the first harmonic amplitude A1 is the vibration frequency synchronized with the rotational speed. When the impeller mass is uneven due to winding, A1 will increase significantly; the first harmonic percentage is the ratio of A1 to the total vibration energy. Among the current and stator temperature sensing parameters, the core indicators include the current overload coefficient and the stator temperature rise rate. The current overload coefficient is the actual current / rated current; the stator temperature rise rate is the temperature rise value within 5 minutes.

3. The centrifugal pump winding data model according to claim 1, characterized in that: The RNN model design includes an input layer, an LSTM layer, and an output layer. The input layer has 3 vibrational features and 60 time steps. The LSTM layer has two layers: the first layer has 64 neurons and the second layer has 32 neurons. The output layer includes 1 neuron, outputting the α value and the quantization axis imbalance degree.

4. The centrifugal pump winding data model according to claim 1, characterized in that: The RNN model design includes an input layer, an LSTM layer, and an output layer. The input layer has three power features and 60 time steps. The LSTM layer consists of two layers: the first layer has 64 neurons and the second layer has 32 neurons. The output layer includes one neuron, outputting the β value and quantifying the power decay.

5. A centrifugal pump winding data model according to claim 1, characterized in that: In the calculation of the comprehensive fault index γ and the fault determination γ value and the threshold perception item, the threshold setting is as follows: γ≥0.5 is a minor fault, and γ≥0.8 is a severe fault.

6. The centrifugal pump winding data model according to claim 1, characterized in that: In the comprehensive fault index γ and the simple logic troubleshooting and perception items for fault determination, the troubleshooting rules for spindle wear include no jump events; α and β values ​​increase slowly over time; and the vibration spectrum includes bearing characteristic frequencies. The troubleshooting rules for water ingress into the oil chamber are: humidity sensor data > 80% RH; vibration signal mixed with 20-50Hz low-frequency noise; The inspection rules for cement and sand ingress are: irregular broadband noise with vibration of 100-500Hz; and localized increase in stator temperature. The rules for investigating cavitation are: inlet pressure < medium vaporization pressure, vibration containing high-frequency noise of 600-2500Hz, and α value < 0.3.

Images