Water Pump Fault Diagnosis Method and System Based on Neural Network

Abstract

The present invention provides a water pump fault diagnosis method and system based on a neural network, comprising the following steps: a data collection step: collecting water pump operation data; an identification model step: inputting the water pump operation data into the water pump fault identification model to obtain a first diagnosis result output by the water pump fault identification model; a fault judgment step: judging whether the water pump has a fault according to the first diagnosis result of the water pump fault identification model. The present invention can monitor the operating status of the water pump in real time and give a judgment prompt for the water pump that may have a fault, greatly improving the safety of the water pump during use and improving the problem of difficulty in troubleshooting the water pump.

Description

Technical Field

[0001] The present invention relates to the technical field of water pump operation status monitoring and diagnosis, and in particular to a water pump fault diagnosis method and system based on a neural network. Background Art

[0002] Water pumps are widely used in industrial systems. In typical heat exchange systems in energy industries, water pumps are always in an automated and continuous operation state. The safety and reliability of water pumps are of great significance to the stable operation of the system.

[0003] Current water pump operating status monitoring and fault diagnosis mainly rely on instrument measurement and subjective experience judgment, but water pumps have numerous operating parameters, complex mechanisms, strong time-varying properties, and many coupling factors, making it difficult to describe such complex equipment with precise mathematical models.

[0004] In addition, due to the high safety and reliability of water pumps, failure conditions rarely occur during their life cycle, which also leads to the water pump fault diagnosis system lacking sufficient fault data for analysis.

[0005] With respect to the above-mentioned related technologies, the inventors believe that it is difficult to describe a complex device such as a water pump using an accurate mathematical model, and the fault diagnosis system of the water pump lacks sufficient fault data for analysis, making it difficult to troubleshoot the water pump. Summary of the invention

[0006] In view of the defects in the prior art, the object of the present invention is to provide a water pump fault diagnosis method and system based on neural network.

[0007] A water pump fault diagnosis method based on a neural network provided by the present invention comprises the following steps:

[0008] Data collection steps: Collect water pump operation data;

[0009] Identification model step: inputting the water pump operation data into the water pump fault identification model to obtain a first diagnosis result output by the water pump fault identification model;

[0010] Fault judgment step: judge whether the water pump is faulty according to the first diagnosis result of the water pump fault identification model.

[0011] In a possible implementation, in the data collection step, the internal pressure data of the pipeline at different times are collected by pressure sensors deployed upstream and downstream of the water pump pipeline, and the head H(t) of the water pump is calculated based on the pressure data;

[0012] The flow data Q(t) is collected by the flow meter deployed on the water pump;

[0013] The power data P(t) is collected through the electricity meter deployed on the water pump;

[0014] The frequency data f(t) is collected through the frequency converter deployed on the water pump;

[0015] Among them, t represents the collection time.

[0016] In a possible implementation, the method further includes a state model step;

[0017] The state model step: inputting the water pump operation data into the water pump operation state model to obtain a prediction result of the water pump operation state model; determining a second diagnosis result according to the prediction result, wherein the second diagnosis result is used to indicate whether the water pump operation state is abnormal;

[0018] The fault judgment step: judging whether a fault occurs in the operation of the water pump according to the first diagnosis result and the second diagnosis result.

[0019] In a possible implementation, in the state model step, the water pump operation data is input into the water pump operation state model to obtain the water pump frequency prediction value and the water pump power prediction value within the set time period, respectively, and the variance determination coefficient and the average error percentage within the time period are calculated based on the water pump frequency prediction value and the water pump power prediction value, and used as the second diagnostic result; if the variance determination coefficient and the average error percentage are both lower than their respective thresholds, the water pump is deemed to be abnormal.

[0020] In a possible implementation, in the fault judgment step, if the water pump is determined to be abnormal according to both the first diagnostic result and the second diagnostic result, it is considered that the water pump is faulty; otherwise, it is considered that the water pump is in normal operation.

[0021] In a possible implementation manner, establishing the water pump operation state model includes the following steps:

[0022] Sample acquisition step: acquiring the normal operation data of the water pump to obtain a normal operation sample set of the water pump; the normal operation sample set of the water pump includes basic input parameters and basic output parameters corresponding to the basic input parameters; the basic input parameters include the water pump head and water pump flow in the normal operation sample set of the water pump, and the basic output parameters include the water pump power and water pump frequency in the normal operation sample set of the water pump;

[0023] State model training steps: establish a BP neural network, wherein the basic input parameters serve as input data of the BP neural network, the basic output parameters serve as the labeled data of the corresponding basic input parameters, and the output data of the BP neural network is the predicted output parameters; based on the predicted output parameters and the basic output parameters corresponding to the predicted output parameters, adjust the network parameters of the BP neural network, and determine the trained BP neural network as the water pump operation state model.

[0024] In a possible implementation, the method further includes:

[0025] For at least one basic input parameter, a polynomial combination is used to perform feature expansion on the basic input parameter to obtain a parameter after feature expansion of the basic input parameter, and the basic input parameter and its feature expansion parameter are used as input data of the BP neural network; the correlation between the basic output parameter and the input data of the BP neural network is compared; and the input data of the BP neural network with a predetermined correlation between the basic output parameter and the input data of the BP neural network is retained.

[0026] In a possible implementation manner, establishing the water pump fault identification model includes the following steps:

[0027] Data set construction step: constructing a data set, wherein the data set includes normal samples and fault samples, wherein the normal samples are data when the water pump operates normally, and the fault samples are data when the water pump fails;

[0028] Recognition model training steps: establish a convolutional neural network, use the data set as the input data of the convolutional neural network, use the state label as the annotation data of the input data, and the state label includes a first state label that identifies the normal operation of the water pump and a second state label that identifies the water pump failure; the output data of the convolutional neural network is the predicted state, and the predicted state includes a normal state that identifies the normal operation of the water pump and a fault state that identifies the water pump failure; optimize the convolutional neural network based on the predicted state and the state label, and determine the trained convolutional neural network as a water pump fault recognition model.

[0029] In a possible implementation, in the data set construction step, water pump operation data is obtained; the water pump operation data is divided at set time intervals, and the water pump operation data is organized into the following form:

[0030]

[0031] Among them, t kl+l represents the kl+lth moment; Indicates t kl+l The lift of the moment; Indicates t kl+l Traffic flow at the moment; Indicates t kl+l The pump power at the moment; Indicates t kl+l The pump frequency at the moment; k represents the kth moment; l represents the length of the time interval, and n represents the sample of the pump operation data parameters;

[0032] A data set is constructed, and the input data of the water pump fault identification model is the time series matrix of the time period set in the data set.

[0033] In one possible implementation, the convolutional neural network structure includes: at least one convolutional layer, a flattening layer, a forgetting layer, and a fully connected layer.

[0034] A water pump fault diagnosis system based on a neural network provided by the present invention comprises data acquisition hardware, a host computer and a computer, wherein the data acquisition hardware is connected to the host computer, and the host computer is connected to the computer;

[0035] The data acquisition hardware collects the water pump operation data and sends the water pump operation data to the host computer;

[0036] The host computer receives and stores the water pump operation data, and sends the water pump operation data to the computer;

[0037] The computer processes the water pump operation data.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The water pump of the present invention has a wide range of applications in industrial systems. The safety and reliability of the water pump are of great significance to the stable operation of the system. The present invention can monitor the operating status of the water pump in real time and provide a judgment and prompt for the water pump that may fail, which greatly improves the safety of the water pump during use and improves the difficulty of troubleshooting the water pump.

[0040] 2. The present invention uses wireless sensors to collect lift data. Wireless sensors are easy to deploy, do not require wiring, have fast and stable connections, and use WiFi to access the control system;

[0041] 3. The present invention combines the prediction results of two neural network models, improves the accuracy of fault diagnosis and reduces the possibility of program misjudgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0043] Figure 1 A schematic diagram of water pump fault diagnosis of the present invention;

[0044] Figure 2 It is a schematic diagram of collecting data for the water pump of the present invention;

[0045] Figure 3 It is a basic flow chart of the present invention;

[0046] Figure 4It is a schematic diagram of the structure of the BP neural network of the present invention;

[0047] Figure 5 Schematic diagram of the structure of the convolutional neural network of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0049] The embodiment of the present invention discloses a water pump fault diagnosis method based on neural network, such as Figure 1 As shown, the method comprises the following steps: a data collection step: collecting water pump operation data. an identification model step: inputting the water pump operation data into the water pump fault identification model to obtain a first diagnosis result output by the water pump fault identification model. a fault judgment step: judging whether the water pump is faulty according to the first diagnosis result of the water pump fault identification model.

[0050] like Figure 2 and Figure 3 As shown, in the data collection step, the water pump operation data is obtained, and the water pump operation data includes: the water pump head H(t), the water pump flow Q(t), the water pump power P(t) and the water pump frequency f(t).

[0051] In some embodiments, pressure sensors deployed upstream and downstream of the water pump pipeline are used to collect internal pressure data of the pipeline at different times, and the head H(t) of the water pump is calculated based on the pressure data, where 1 meter of head = pressure 10 kPa.

[0052] In some embodiments, the flow data Q(t) is collected by a flow meter deployed on the water pump.

[0053] In some embodiments, the power data P(t) is collected by an electricity meter deployed on the water pump; and the frequency data f(t) is collected by a frequency converter deployed on the water pump.

[0054] Among them, t represents the collection time.

[0055] In the specific implementation process, when the water pump data is acquired, the measuring instruments (data acquisition hardware) include wireless pressure sensors, flow meters, electricity meters and frequency converters. Real-time collection of water pump operation data: The measuring instruments deployed on the pipelines and equipment are connected to the host computer, and the host computer collects the reading data of the measuring instruments that change over time. The specific measurement instrument settings and physical quantities collected are as follows: The wireless pressure sensors deployed before and after the water pump pipeline collect pressure data, access the host computer through WIFI wirelessly, and calculate the head H(t). It has the characteristics of no need for wiring, flexible deployment and stable connection. Wireless pressure sensors can also be connected by wire, such as using network cables or 485 buses, and wireless pressure sensors can also be connected by other wireless methods (NBIOT or LORA). The wired flow meter deployed on the water pump collects flow data Q(t) and transmits it to the host computer by wire. The flow meter can be installed upstream or downstream of the water pump, but the installation position should meet the requirements of the straight pipe end that is not less than 5 times the distance from the upstream to the water pump or the straight pipe section that is not less than 3 times the distance from the downstream to the water pump, and there should be no other pipes on the straight pipe section. Wireless or wired transmission can be used between the flow meter and the host computer, without special restrictions. The power data P(t) is collected by the electricity meter deployed on the water pump and transmitted to the host computer by wire. The electricity meter is installed on the power distribution switch cabinet of the water pump. The connection between the electricity meter and the host computer can be wireless or wired.

[0056] The frequency converter deployed on the water pump collects frequency data f(t) and transmits it to the host computer via wire. The flow meter collects flow data Q(t), the electricity meter collects power data P(t), and the frequency converter collects frequency data f(t), all of which are transmitted to the host computer via wire. The frequency converter is installed between the water pump electrode and the power distribution cabinet. Usually, the power of the power distribution cabinet is input to the frequency converter, and the output current of the frequency converter is input to the water pump motor. There are no specific requirements for the spatial position of the frequency converter. The transmission between the frequency converter and the host computer is usually wired, but wireless transmission can also be used. The host computer is responsible for collecting the training data required for modeling.

[0057] In the specific implementation process, recent data (usually data from the operating time to the previous hour) is input into the model to monitor the operation status of the pump in real time and diagnose faults. The data acquisition system established using the data acquisition steps performs real-time analysis on the recent data monitored in the most recent period. Recent pump operation data includes actual pump head, actual pump flow, actual pump power and actual pump frequency.

[0058] In the identification model step, the water pump operation data is input into the water pump fault identification model to obtain the first diagnosis result output by the water pump fault identification model. The recent data (actual water pump head, actual water pump flow, actual water pump power and actual water pump frequency) is input into the established convolutional neural network water pump fault identification model, and the model outputs the first diagnosis result output by the normal or abnormal water pump fault identification model. The output of the convolutional neural network is a normal or abnormal label.

[0059] The method also includes a state model step: inputting water pump operation data into the water pump operation state model to obtain a prediction result of the water pump operation state model; determining a second diagnostic result based on the prediction result, and the second diagnostic result is used to indicate whether the water pump operation state is abnormal.

[0060] Specifically, the water pump operation data is input into the water pump operation state model, and the water pump frequency prediction value and water pump power prediction value within the set time period (the set time period and the recent data refer to the same time period, usually the data from the program running time to the previous hour) are obtained respectively, and the variance determination coefficient and the average error percentage within the time period are calculated according to the water pump frequency prediction value and the water pump power prediction value, and used as the second diagnosis result. If the variance determination coefficient and the average error percentage are both lower than their respective thresholds, the water pump is considered to be abnormal, otherwise the water pump is considered to be normal.

[0061] In some embodiments, recent data (actual pump head and actual pump flow) are input into the established BP neural network pump operation status model to obtain the pump frequency prediction value and pump power prediction value within the time period, the pump frequency prediction value is compared with the actual pump frequency, the pump power prediction value is compared with the actual pump power, and the variance determination coefficient and the average error percentage within the time period are calculated.

[0062] One possible implementation to determine the coefficient of determination of variance:

[0063]

[0064] Among them, R represents the coefficient of determination of variance; SSR is the full name of Sum of Squares Total, which represents the variance of variable y relative to the center. The change of SST is called Sum of Squares Regression, which means the variable Relative to the center 's abnormal movement; It represents the predicted value of the i-th sample; i refers to the number of samples in the evaluation time period. The sample is usually one per minute, so there are generally 60 samples in one hour. The time range of i is usually 1 hour, that is, i=60. Represents the average value of the sample actual value; yi represents the actual value of the i-th sample, where and i It can correspond to both the pump power and the pump frequency. When is the predicted value of water pump power, and i are the average value of the actual power of the water pump and the actual power value of the water pump respectively. When it is the predicted value of pump frequency, and i They are the average value of the actual frequency of the water pump and the actual frequency value of the water pump. The general evaluation method introduced here is that the water pump power needs to be evaluated using the formula of R2, and the water pump frequency also needs to be evaluated in the same way.

[0065] One possible implementation to determine the average error percentage is:

[0066]

[0067] Among them, MAPE represents the mean absolute percentage error; actual(t) represents the actual value of the tth sample; forecast(t) represents the predicted value of the tth sample; n represents the number of samples, among which actual(t) and forecast(t) can represent both the pump power and the pump frequency. The average error percentage is not the same as the average absolute error percentage. The average error percentage includes the average absolute percentage error. The difference lies in whether the absolute value calculation is included. When actual(t) represents the actual power value of the pump, forecast(t) represents the predicted value of the pump power. When actual(t) represents the actual frequency value of the pump, forecast(t) represents the predicted value of the pump frequency. What is introduced here is the general calculation method. The pump power needs to be calculated using the MAPE formula, and the pump frequency also needs to be calculated in the same way.

[0068] If the variance determination coefficient and the average error percentage are both lower than the threshold, the pump is considered to be operating abnormally. The deep neural network is responsible for comparing the predicted frequency & power (pump frequency prediction value and pump power prediction value) with the actual measured frequency & power (pump actual frequency prediction value and pump actual power prediction value). If the determination coefficient is lower than 0.85 and the average error percentage is higher than 10%, it is diagnosed as a fault. & represents sum.

[0069] The present invention does not limit the order of the process of determining the first diagnostic result and the process of determining the second diagnostic result. The first diagnostic result may be determined first, and then the second diagnostic result; the second diagnostic result may be determined first, and then the first diagnostic result; or the first diagnostic result and the second diagnostic result may be determined simultaneously.

[0070] In the fault judgment step, it is determined whether the water pump is faulty according to the first diagnosis result and the second diagnosis result. Based on the judgment results of the two models, if the water pump is abnormal according to the first diagnosis result and the second diagnosis result, it is considered that the water pump is faulty; otherwise, it is considered that the water pump is in normal operation. Based on the results of the two models, if both show abnormality, it is determined that the water pump is currently faulty, otherwise it is not.

[0071] Perform BP neural network and convolutional neural network modeling.

[0072] like Figure 3 and Figure 4 As shown, an embodiment of establishing a water pump operation status model provided by the present invention. The establishment of the water pump operation status model includes the following steps: (1) Sample acquisition step: Acquire the normal operation data of the water pump to obtain a water pump normal operation sample set. Among them, the water pump normal operation sample set includes basic input parameters and basic output parameters corresponding to the basic input parameters. The basic input parameters include the water pump head (denoted as H) and the water pump flow (denoted as Q) in the water pump normal operation sample set, and the basic output parameters include the water pump power (denoted as P) and the water pump frequency (denoted as f) in the water pump normal operation sample set. Obtain n samples of the water pump normal operation data parameters at different times, where n is a positive integer.

[0073] In some embodiments, the water pump operation status model adopts a BP neural network, wherein BP stands for Back Propagation in English and its Chinese translation is back propagation algorithm.

[0074] In some embodiments, when obtaining the input parameters and output parameters of the water pump operation status model, the data acquisition system established by the data acquisition step acquires the normal operation data of the water pump within a set time period, and organizes the data into the following form:

[0075]

[0076] Among them, t i represents the i-th moment; Indicates t i The lift of the moment; Indicates t i Traffic flow at the moment; Indicates t i The pump power at the moment; Indicates t i The pump frequency at the moment. Set the duration to Normally, the value is taken and calculated once every hour. Each time the value is taken, the data from the program running time to the previous hour will be obtained, with a total of 60 samples.

[0077] Through the above method, one-dimensional samples n of the normal operating parameters of the water pump at different times are obtained. The data collection step is similar to the data collection step of the above fault diagnosis, which will not be repeated here.

[0078] In some embodiments, in order to improve the fitting accuracy of the model, the basic input parameters are automatically expanded with independent variables, specifically including: for at least one basic input parameter, a polynomial combination can be used to perform feature expansion on the basic input parameter to obtain the parameter after feature expansion of the basic input parameter, and the basic input parameter and its feature expanded parameter are used as input data of the BP neural network.

[0079] Specifically, the basic input parameters are feature expanded using a logarithmic and polynomial combination method, and the correlation between the input data of the BP neural network (basic input parameters and parameters after feature expansion) and the basic output parameters is compared; the input data of the BP neural network with a predetermined correlation between the basic output parameters and the input data of the BP neural network is retained (retained). Only variable features with a predetermined correlation greater than 0.5 or less than -0.5 among the basic input parameters and their expanded parameters will be retained, and the rest will be deleted. Logarithmic conversion is to take the natural logarithm of the original data (basic input parameters).

[0080] Compare the correlation between the input data (basic input parameters and expanded input parameters) and the output parameters of the BP neural network, and retain the expanded variables with higher predetermined correlation. The Pearson correlation coefficient is used to evaluate the correlation coefficient of each input and output between the input data and the output parameters of the BP neural network, and the input features with a correlation coefficient greater than 0.5 or less than -0.5 are retained. The predetermined higher correlation means that the correlation is greater than 0.5 or less than -0.5. The output parameters are: P: water pump power; f: water pump frequency.

[0081] (2) State model training steps: Establish a BP neural network, wherein the basic input parameters serve as the input data of the BP neural network, the basic output parameters serve as the labeling data of the corresponding basic input parameters, and the output data of the BP neural network serves as the predicted output parameters; based on the predicted output parameters and the basic output parameters corresponding to the predicted output parameters, adjust the network parameters of the BP neural network, and determine the trained BP neural network as the water pump operation state model.

[0082] Specifically, the BP neural network includes the following parameters: the number of input layer nodes is 6, the number of output layer nodes is 1, the number of hidden layers is 10, and the activation function is linear activation. The parameters here are used to describe the structure of the BP neural network model. In the model training, based on the normal operation sample set of the water pump obtained in the sample acquisition step, the back propagation algorithm is used to calculate the network parameters that minimize the prediction error of the output parameters, and the water pump operation status model is established accordingly.

[0083] like Figure 3 and Figure 5 As shown, the embodiment of the present invention provides a method for establishing a water pump fault identification model. Specifically, the method comprises the following steps:

[0084] (1) Dataset construction step: construct a data set, the data set includes normal samples and fault samples, the normal samples are data when the water pump operates normally, and the fault samples are data when the water pump fails.

[0085] Specifically, the water pump operation data is obtained; the water pump operation data is divided at set time intervals, and based on the one-dimensional sample form of the normal operation parameters of the water pump at different times, the data is further divided at set time intervals l, and the data is organized into a time series matrix in the following form:

[0086]

[0087] Among them, t kl+l represents the kl+lth moment; Indicates t kl+l The lift of the moment; Indicates t kl+l Traffic flow at the moment; Indicates t kl+l The pump power at the moment; Indicates t kl+l The pump frequency at the moment; k represents the kth moment; l represents the length of the time interval, which is usually 20 minutes, i.e. 20 samples, but will also be adjusted as the model is trained; n represents the number of samples of the pump operation data parameters.

[0088] In some embodiments, 1 to 4 columns are randomly selected from the 4 columns of data H, Q, P, and f (here they are selected vertically, 60 samples are divided per hour by time, and there are 4 types according to dimension (feature / physical quantity), and the operation here is a unified operation on one or several dimensions of multiple samples), and a random noise not exceeding [-30%, +30%] is added to each position of the randomly selected data to form a new data sample, and a faulty data label is added, and a normal data label is added to the original sample to form a data set in which faulty samples account for a large proportion.

[0089] The data set is constructed, and the input data of the water pump fault identification model is the time series matrix of the time period set in the data set. The data when the water pump fails includes the data formed by adding random noise that does not exceed the predetermined range to the normal operation data of the water pump, and also includes the actual operation data collected when the water pump fails.

[0090] (2) Recognition model training steps: establish a convolutional neural network, use the data set as the input data of the convolutional neural network, and use the state label as the annotation data of the input data. The state label includes a first state label indicating the normal operation of the water pump and a second state label indicating a water pump failure. The output data of the convolutional neural network is the predicted state. The predicted state includes a normal state indicating the normal operation of the water pump and a fault state indicating a water pump failure. The convolutional neural network is optimized based on the predicted state and the state label, and the trained convolutional neural network is determined as a water pump fault recognition model.

[0091] Specifically, the second state label includes a fault label of H with random noise, a fault label of Q with random noise, a fault label of P with random noise, and a fault label of f with random noise. When random noise is added to H, random noise is added to H and Q, random noise is added to H and P, random noise is added to H and f, random noise is added to H, Q and P, random noise is added to H, Q and f, random noise is added to H, P and f, and random noise is added to H, Q, P and f, any one of which corresponds to the fault label of H with random noise in the second state label (as long as random noise is added to H, it is defined as the fault label of H with random noise). When random noise is added to Q, random noise is added to Q and P, random noise is added to Q and f, and random noise is added to Q, P and f, any one of which corresponds to the fault label of Q with random noise in the second state label. When random noise is added to P, random noise is added to P and f, both correspond to the fault label of P with random noise in the second state label. When random noise is added only to f, it corresponds to the fault label of f with random noise in the second state label.

[0092] In some embodiments, the convolutional neural network structure includes: at least one convolutional layer, a flattening layer, a forgetting layer and a fully connected layer. The convolutional layer includes a first convolutional layer (16 convolution kernels) and a second convolutional layer (32 convolution kernels). The fully connected layer includes a first fully connected layer and a second fully connected layer, and the second fully connected layer is a Softmax output layer. Softmax is translated into Chinese as a reversible probability distribution function. The first convolutional layer, the second convolutional layer, the flattening layer, the forgetting layer, the first fully connected layer and the Softmax output layer are connected in sequence. The fully connected layer is the concept of a layer inside the neural network model. Each fully connected layer also has its own separate input and output. The input of the softmax layer is used here as the input of the final model. Therefore, in the large classification of the model layer, softmax is the output layer.

[0093] Among them, the convolution layer plays the role of filtering, which can extract the features of images or data; the flattening layer converts multi-dimensional data into one dimension, and acts as a bridge between the convolution layer and the forgetting layer; the forgetting layer discards specific connections to prevent overfitting; the fully connected layer acts as a classifier; and the softmax layer probabilizes the classification results.

[0094] In some embodiments, the convolutional neural network structure is determined to include: a first convolutional layer, a second convolutional layer, a flattening layer, a forgetting layer (forgetting rate is 0.4), a first fully connected layer, a second fully connected layer and a softmax output layer, and no pooling layer. The purpose of pooling is to reduce the dimension and reduce the complexity of the model. The model used here is not too complex, so a pooling layer is not required. Each of the above layers needs to use an activation function, and the activation function uses the RELU activation function:

[0095] y=max(x,0).

[0096] Among them, x represents the model input; y represents the model output. RELU is the full name of Rectified Linear Unit in English, and its Chinese translation is linear rectification function, also known as corrected linear unit. The main function of the activation function is to make the neural network model present nonlinear characteristics. The function form is y=max(0,x). Each layer of the convolutional neural network structure uses this function form.

[0097] In the specific implementation process, the input parameters of the water pump fault identification model are the time series matrix of H, Q, P, and f in a certain time period, and the output parameters are the normal or faulty state labels. The input parameters of the water pump fault identification model are the time series matrix of a certain time period in the data set. The time series matrix is ​​a matrix (a subset of the time series matrix) composed of randomly selecting a moment in the data set and taking out continuous samples from that moment to that moment + N. Usually N is 20 minutes. This moment is generated during the model training process and is a random moment. Based on the data set constructed in the data set construction step, the parameters of the composed matrix are optimized. The solution method uses the gradient descent method. The solution goal is to minimize the loss function of the model's predicted value and actual value. The loss function uses the Euclidean distance, that is, the Euclidean distance is minimized, and finally a convolutional neural network water pump fault identification model is established.

[0098] The data is organized into one-dimensional training data (one sample per time) and two-dimensional data (one sample is composed of 1 minute of continuous data). The one-dimensional data is fed into the deep neural network (BP neural network), and the two-dimensional data is fed into the convolutional neural network for training respectively.

[0099] The embodiment of the present invention also provides a water pump fault diagnosis system based on neural network, including data acquisition hardware, a host computer and a computer, wherein the data acquisition hardware is connected to the host computer, and the host computer is connected to the computer. The data acquisition hardware collects water pump operation data and sends the water pump operation data to the host computer; the host computer receives and stores the water pump operation data and sends the water pump operation data to the computer; the computer processes the water pump operation data. The data acquisition hardware includes a wireless pressure sensor, a flow meter, an electricity meter and a frequency converter. The wireless pressure sensor is deployed before and after the water pump pipeline, and the wireless pressure sensor is connected to the host computer. The flow meter is deployed upstream or downstream of the water pump. The flow meter is connected to the host computer. The electricity meter is installed on the power distribution switch cabinet of the water pump. The electricity meter is connected to the host computer. The frequency converter is installed between the water pump electrode and the power distribution cabinet, and the frequency converter is connected to the host computer. A BP neural network model and a convolutional neural network model are established in the computer. The data collected by the host computer from the wireless pressure sensor, the flow meter, the electricity meter and the frequency converter are input into the BP neural network model and the convolutional neural network model in the computer for processing.

[0100] The real-time monitoring and fault diagnosis system for the operation status of a water pump of the present invention is composed of data acquisition hardware and an artificial intelligence monitoring system. The present invention relates to computer-aided networking technology and the field of artificial intelligence, and provides real-time monitoring and fault diagnosis for the operation status of a water pump based on a neural network. In order to verify the accuracy and reliability of the present invention, corresponding test data are selected and brought into the established neural network model for performance analysis. Classification evaluation indicators such as accuracy, precision, recall rate, ROC curve and AUC are introduced, and the performance of the present invention is rigorously evaluated and analyzed. After analysis of field data, the present invention is reliable and effective. The above-mentioned classification evaluation indicators will be used when doing model training, which is a general method for model training.

[0101] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0102] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A water pump fault diagnosis method based on neural network, characterized in that: The steps include: Data collection steps: Collect water pump operation data; Identification model step: inputting the water pump operation data into the water pump fault identification model to obtain a first diagnosis result output by the water pump fault identification model; Fault judgment step: judging whether the water pump is faulty according to the first diagnosis result of the water pump fault identification model; The method also includes a state model step; The state model step: inputting the water pump operation data into the water pump operation state model to obtain a prediction result of the water pump operation state model; determining a second diagnosis result according to the prediction result, wherein the second diagnosis result is used to indicate whether the water pump operation state is abnormal; The fault judgment step: judging whether a fault occurs in the operation of the water pump according to the first diagnosis result and the second diagnosis result; In the state model step, the water pump operation data is input into the water pump operation state model to obtain the water pump frequency prediction value and the water pump power prediction value within the set time period, respectively. The variance determination coefficient and the average error percentage within the time period are calculated based on the water pump frequency prediction value and the water pump power prediction value, and used as the second diagnostic result; if the variance determination coefficient and the average error percentage are both lower than their respective thresholds, the water pump is deemed to be abnormal.

2. The water pump fault diagnosis method based on neural network according to claim 1 is characterized in that: In the data collection step, the internal pressure data of the pipeline at different times are collected by pressure sensors deployed upstream and downstream of the water pump pipeline, and the head H(t) of the water pump is calculated based on the pressure data; The flow data Q(t) is collected by the flow meter deployed on the water pump; The power data P(t) is collected through the electricity meter deployed on the water pump; The frequency data f(t) is collected through the frequency converter deployed on the water pump; Among them, t represents the collection time.

3. The water pump fault diagnosis method based on neural network according to claim 1 is characterized in that: In the fault judgment step, if the water pump is determined to be abnormal according to both the first diagnosis result and the second diagnosis result, it is considered that the water pump is faulty; otherwise, it is considered that the water pump is in normal operation.

4. The water pump fault diagnosis method based on neural network according to claim 1 is characterized in that: The establishment of the water pump operation state model comprises the following steps: Sample acquisition step: acquiring the normal operation data of the water pump to obtain a normal operation sample set of the water pump; the normal operation sample set of the water pump includes basic input parameters and basic output parameters corresponding to the basic input parameters; the basic input parameters include the water pump head and water pump flow in the normal operation sample set of the water pump, and the basic output parameters include the water pump power and water pump frequency in the normal operation sample set of the water pump; State model training steps: establish a BP neural network, wherein the basic input parameters serve as input data of the BP neural network, the basic output parameters serve as the labeled data of the corresponding basic input parameters, and the output data of the BP neural network is the predicted output parameters; based on the predicted output parameters and the basic output parameters corresponding to the predicted output parameters, adjust the network parameters of the BP neural network, and determine the trained BP neural network as the water pump operation state model.

5. The water pump fault diagnosis method based on neural network according to claim 4 is characterized in that: The method further comprises: For at least one basic input parameter, a polynomial combination is used to perform feature expansion on the basic input parameter to obtain a parameter after feature expansion of the basic input parameter, and the basic input parameter and its feature expansion parameter are used as input data of the BP neural network; the correlation between the basic output parameter and the input data of the BP neural network is compared; and the input data of the BP neural network with a predetermined correlation between the basic output parameter and the input data of the BP neural network is retained.

6. The water pump fault diagnosis method based on neural network according to claim 1 is characterized in that: The establishment of the water pump fault identification model includes the following steps: Data set construction step: constructing a data set, wherein the data set includes normal samples and fault samples, wherein the normal samples are data when the water pump operates normally, and the fault samples are data when the water pump fails; Recognition model training steps: establish a convolutional neural network, use the data set as the input data of the convolutional neural network, use the state label as the annotation data of the input data, and the state label includes a first state label that identifies the normal operation of the water pump and a second state label that identifies the water pump failure; the output data of the convolutional neural network is the predicted state, and the predicted state includes a normal state that identifies the normal operation of the water pump and a fault state that identifies the water pump failure; optimize the convolutional neural network based on the predicted state and the state label, and determine the trained convolutional neural network as a water pump fault recognition model.

7. The water pump fault diagnosis method based on neural network according to claim 6 is characterized in that: In the data set construction step, the water pump operation data is obtained; the water pump operation data is divided at set time intervals and organized into the following form: Among them, t kl+l represents the kl+lth moment; Indicates t kl+l The lift of the moment; Indicates t kl+l Traffic flow at the moment; Indicates t kl+l The pump power at the moment; Indicates t kl+l The pump frequency at the moment; k represents the kth moment; l represents the length of the time interval, and n represents the sample of the pump operation data parameters; A data set is constructed, and the input data of the water pump fault identification model is the time series matrix of the time period set in the data set.

8. The water pump fault diagnosis method based on neural network according to claim 6 is characterized in that: The convolutional neural network structure includes: at least one convolutional layer, a flattening layer, a forgetting layer and a fully connected layer.

9. A water pump fault diagnosis system based on neural network, characterized in that: The neural network-based water pump fault diagnosis method according to any one of claims 1 to 8 comprises data acquisition hardware, a host computer and a computer, wherein the data acquisition hardware is connected to the host computer, and the host computer is connected to the computer; The data acquisition hardware collects the water pump operation data and sends the water pump operation data to the host computer; The host computer receives and stores the water pump operation data, and sends the water pump operation data to the computer; The computer processes the water pump operation data.

Images