Drainage facility fault diagnosis method and system in water networking environment

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a drainage facility fault diagnosis method and system in a water networking environment, and the method comprises the steps: installing a sensor, and obtaining drainage facility information and shaft seal monitoring information; respectively carrying out feature engineering on the obtained drainage facility information and shaft seal monitoring information data to obtain frequency domain data and time frequency data; constructing first time sequence data and second time sequence data according to the obtained frequency domain data and time frequency data in combination with original time sequence data; constructing a fault detection model, inputting the constructed first time sequence data and second time sequence data into the fault detection model, and outputting a fault detection result; according to the invention, the shaft seal is monitored more comprehensively while the sewage lift pump is monitored, so that some faults which are large in hidden danger and high in monitoring difficulty when cavitation is generated can be found quickly and efficiently, and thus timely maintenance is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and more specifically, to a method and system for diagnosing drainage facility faults in a water networking environment. Background Art

[0002] The sewage lifting pump station is mainly used to raise the water level of sewage to ensure that the sewage can flow smoothly in the drainage pipe system by gravity. When the sewage lifting pump is in operation, since the pressure at the impeller inlet is lower than the saturated vapor pressure of the liquid, part of the liquid in the sewage will vaporize to form bubbles. These bubbles will burst quickly when they flow to the area with higher pressure. At the moment the bubble bursts, the surrounding liquid rushes to the center of the bubble at a very high speed, generating a strong impact force.

[0003] The strong vibration and impact caused by cavitation are transmitted to the shaft seal, causing the sealing element of the shaft seal to bear additional uneven force. Under normal circumstances, the shaft seal can effectively prevent sewage leakage in the pump and external air from entering the pump body. However, under the influence of cavitation, the sealing element is frequently deformed by force, a small gap may appear between the fitting surfaces of the dynamic and static rings of the mechanical seal, and the packing of the packing seal may become loose and wear more severe. With the emergence of these problems, the sealing performance of the shaft seal gradually decreases, initially manifested as slight sewage dripping. If not handled in time, the leakage will become more and more serious, eventually leading to a large amount of sewage leakage, which not only pollutes the working environment, but also may cause the lubricating fluid in the pump to be diluted or washed away by sewage, causing bearings and other components to be damaged due to lack of lubrication, further affecting the normal operation of the pump.

[0004] Since the shaft seal itself is located in a relatively hidden part of the pump body, it has poor visibility and is difficult to directly observe its sealing status. For mechanical seals, small gap changes and wear of the sealing surface are difficult to detect by conventional means, and will only be discovered when leakage is obvious. The situation of packing seals is similar. The looseness and wear of the packing will not cause obvious changes in appearance in the early stage, and it is difficult to monitor the leakage amount, because in actual operation, a small amount of sewage dripping may be considered a normal phenomenon, which is not easy to attract enough attention, thus missing the opportunity for early repair, and then causing more serious failures. Summary of the invention

[0005] The present invention provides a drainage facility fault diagnosis method and system in a water network environment, which solves the technical problem in the related art that the initial performance of the fault is not obvious and the fault is difficult to find.

[0006] The present invention provides a drainage facility fault diagnosis method in a water networking environment, comprising:

[0007] Step 100, installing sensors to obtain drainage facility information and shaft seal monitoring information for n consecutive time steps before the current time step, wherein the drainage facility information includes: electrical system information, mechanical system information, hydraulic system information, liquid level control system information, and water quality related information;

[0008] Step 200, performing feature engineering on the drainage facility information and shaft seal monitoring information data acquired at n time steps to obtain frequency domain data and time-frequency data;

[0009] Step 300, constructing first time series data and second time series data according to the obtained frequency domain data and time-frequency data combined with the original time series data;

[0010] Step 400, constructing a fault detection model, and inputting the constructed first time series data and second time series data into the fault detection model, and outputting a fault detection result;

[0011] The specific execution process of step 400 is as follows:

[0012] Step 401, outputting a first hidden state and a second hidden state according to the constructed first time series data and the second time series data as input;

[0013] Step 402, performing hidden state fusion on the first hidden state and the second hidden state, and outputting a fused hidden state;

[0014] Step 403, mapping the fused hidden state to the output through the first fully connected layer to obtain a fault detection result.

[0015] In a preferred embodiment, the electrical system information includes: motor voltage, motor current, insulation resistance value, control element action time;

[0016] Mechanical system information includes: horizontal, vertical and axial vibration acceleration, vibration spectrum, temperature, noise decibel value;

[0017] Hydraulic system information includes: water inlet flow, water outlet flow, water inlet pressure value, water outlet pressure value;

[0018] Liquid level control system information includes: liquid level measurement value, liquid level change rate;

[0019] Water quality related information includes: pH value, sediment content, chemical oxygen demand (COD);

[0020] The shaft seal monitoring information includes: shaft seal temperature, seal chamber pressure, and shaft seal vibration frequency.

[0021] In a preferred embodiment, the frequency of data collection is fixed, and the default value is to collect data once every 1 minute.

[0022] In a preferred embodiment, the first time series data and the second time series data are both obtained by CNN and LSTM to obtain the first hidden state and the second hidden state.

[0023] In a preferred embodiment, the output fault detection result is a vector, the first The component represents the The probability value of a fault. If the probability value is higher than 0.5, it means the fault has occurred. If it does not exist, it means it is normal.

[0024] In a preferred embodiment, the second hidden state is connected to the second fully connected layer for separate pre-training, and the output of the second fully connected layer is the fault state of the shaft seal; the fault state of the shaft seal includes: normal, mild, moderate, and severe.

[0025] In a preferred embodiment, the fault detection model needs to update parameters through training, and the training is performed through a supervised training method, and the multi-classification cross entropy loss function is used as the loss function of the model. The designed loss function is as follows:

[0026] ;

[0027] : The number of samples in the training set for each iteration;

[0028] : The number of fault categories;

[0029] :sample In category The real label on

[0030] :Model for samples Belongs to category The predicted probability of

[0031] : The logarithm of the predicted probability, used to calculate the difference between the predicted and true labels;

[0032] : The average cross entropy loss value of the entire batch of samples.

[0033] In a preferred embodiment, feature engineering includes frequency domain feature extraction and time-frequency feature extraction.

[0034] In a preferred embodiment, frequency domain feature extraction: Fast Fourier transform is applied to each time series data to convert it into frequency domain features:

[0035] ;

[0036] : original time series data;

[0037] : Fourier transform;

[0038] : frequency domain data;

[0039] Time-frequency feature extraction: Apply wavelet transform to each time series data to extract time-frequency features:

[0040] ;

[0041] : original time series data;

[0042] : Wavelet transform;

[0043] : Time-frequency data.

[0044] The beneficial effect of the present invention is that by performing more comprehensive monitoring of the shaft seal while monitoring the sewage lift pump, the present invention can quickly and efficiently discover some faults caused by cavitation that have greater hidden dangers but are difficult to monitor, thereby performing timely maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The invention relates to a drainage facility fault diagnosis method in a water networking environment. DETAILED DESCRIPTION

[0046] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0047] At least one embodiment of the present invention discloses a method for diagnosing drainage facility faults in a water network environment, such as Figure 1 As shown, including:

[0048] Step 100, installing sensors to obtain drainage facility information and shaft seal monitoring information for n consecutive time steps before the current time step, wherein the drainage facility information includes: electrical system information, mechanical system information, hydraulic system information, liquid level control system information, and water quality related information;

[0049] Electrical system information includes: motor voltage, motor current, insulation resistance value, control element action time;

[0050] Mechanical system information includes: horizontal, vertical and axial vibration acceleration, vibration spectrum, temperature, noise decibel value;

[0051] Hydraulic system information includes: water inlet flow, water outlet flow, water inlet pressure value, water outlet pressure value;

[0052] Liquid level control system information includes: liquid level measurement value, liquid level change rate;

[0053] Water quality related information includes: pH value, sediment content, chemical oxygen demand (COD);

[0054] The shaft seal monitoring information includes: shaft seal temperature, seal chamber pressure, and shaft seal vibration frequency;

[0055] In some embodiments of the present invention, a high-precision vibration sensor is installed on the water pump to measure the vibration acceleration of the water pump in three axes (X, Y, Z), in units of m / s²; the temperature sensor uses a platinum resistance temperature sensor to measure the temperature of the water pump motor winding, with an accuracy of up to ±0.1°C and a range of 0-150°C; the pressure sensor uses a piezoresistive pressure sensor, which is installed at the inlet and outlet of the water pump to measure the inlet and outlet pressures in units of Pa with an accuracy of ±0.01%FS. An ultrasonic liquid level sensor is installed in the water collection tank to measure the liquid level height, with a resolution of up to 1mm. The measurement range depends on the depth of the water collection tank, and the output signal is a 4-20mA current signal corresponding to the liquid level range. An electromagnetic flowmeter is installed on the pipeline to measure the sewage flow, with an accuracy of ±0.5%, and the flow unit is m³ / h. The flow velocity and flow data of the sewage in the pipeline can be obtained in real time.

[0056] The frequency of data collection is fixed, and the default value is to collect data once every 1 minute.

[0057] Step 200, performing feature engineering on the drainage facility information and shaft seal monitoring information data acquired at n time steps to obtain frequency domain data and time-frequency data;

[0058] In one embodiment of the present invention, feature engineering includes frequency domain feature extraction and time-frequency feature extraction, specifically including:

[0059] Frequency domain feature extraction: Apply fast Fourier transform to each time series data to convert it into frequency domain features:

[0060] ;

[0061] : original time series data;

[0062] : Fourier transform;

[0063] : frequency domain data;

[0064] Time-frequency feature extraction: Apply wavelet transform to each time series data to extract time-frequency features:

[0065] ;

[0066] : original time series data;

[0067] : Wavelet transform;

[0068] : time-frequency data;

[0069] Step 300, constructing first time series data and second time series data according to the obtained frequency domain data and time-frequency data combined with the original time series data;

[0070] In one embodiment of the present invention, the constructed time series data is in the form of:

[0071] ;

[0072] : original time series data;

[0073] : frequency domain data;

[0074] : time-frequency data;

[0075] : constructed time series data;

[0076] Step 400, constructing a fault detection model, and inputting the constructed first time series data and second time series data into the fault detection model, and outputting a fault detection result;

[0077] Step 401, outputting a first hidden state and a second hidden state according to the constructed first time series data and the second time series data as input;

[0078] In one embodiment of the present invention, CNN and LSTM are used to perform step 401, and the specific formula is as follows:

[0079] ;

[0080] : constructed time series data;

[0081] CNN: convolution function;

[0082] : Convolution output features;

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] : Convolution output features;

[0090] , , : Input gate, forget gate, output gate;

[0091] , , , : The first, second, third, fourth weight matrices;

[0092] , , , : first, second, third, fourth input matrix;

[0093] , , , : The first, second, third, and fourth bias terms;

[0094] : The hidden state at the tth time step;

[0095] : cell status;

[0096] : Hyperbolic tangent function;

[0097] : S-type function.

[0098] In one embodiment of the present invention, the first time series data and the second time series data are both obtained by CNN and LSTM to obtain the first hidden state and the second hidden state.

[0099] Step 402, performing hidden state fusion on the first hidden state and the second hidden state, and outputting a fused hidden state;

[0100] In one embodiment of the present invention, the calculation formula of hidden state fusion is as follows:

[0101] ;

[0102] : No. The first hidden state of the time step;

[0103] : No. The second hidden state of the time step;

[0104] : splicing function;

[0105] : Fusion hidden state;

[0106] Step 403, mapping the fused hidden state to the output through the first fully connected layer to obtain a fault detection result;

[0107] In one embodiment of the present invention, the calculation formula of the first fully connected layer is as follows:

[0108] ;

[0109] : The first output weight coefficient;

[0110] : Fusion hidden state;

[0111] : The first output bias term;

[0112] : S-type function;

[0113] : The output is the fault detection result.

[0114] In one embodiment of the present invention, the output fault detection result is a vector, the first The component represents the The probability value of a fault. If the probability value is higher than 0.5, it means the fault has occurred. If it does not exist, it means it is normal.

[0115] In one embodiment of the present invention, the faults that occur include: bearing failure: wear, damage, lack of lubrication; impeller failure: cracks, wear, cavitation; shaft seal failure: seal failure, leakage; cavitation failure: suction port pressure is too low; motor failure: overload, winding damage; electrical control system failure: control element failure; flow abnormality: mismatch between inlet and outlet flow; liquid level control problem: abnormal liquid level measurement; abnormal water quality: abnormal pH value, COD value, and sand content; system overload failure: excessive pressure, too high current; operation and setting problems: start-up delay, abnormal liquid level control; lubrication failure: lack of lubrication, contamination.

[0116] In one embodiment of the present invention, the second hidden state is connected to the second fully connected layer for separate pre-training, and the calculation formula of the second fully connected layer is as follows:

[0117] ;

[0118] : The second output weight coefficient;

[0119] : No. The second hidden state of the time step;

[0120] : The second output bias term;

[0121] : S-type function;

[0122] : The output is the fault status of the shaft seal.

[0123] In one embodiment of the present invention, the fault status of the shaft seal includes: normal, slight, moderate, and severe.

[0124] The fault detection model needs to update parameters through training. In one embodiment of the present invention, the training is performed through a supervised training method, and the multi-classification cross entropy loss function is used as the loss function of the model. The designed loss function is as follows:

[0125] ;

[0126] : The number of samples in the training set at each iteration.

[0127] : The number of fault categories;

[0128] :sample In category The real label on

[0129] :Model for samples Belongs to category The predicted probability of

[0130] : The logarithm of the predicted probability, used to calculate the difference between the predicted and true labels;

[0131] : The average cross entropy loss value of the entire batch of samples.

[0132] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A drainage facility fault diagnosis method in a water networking environment, characterized in that: The following steps are involved: Step 100, installing sensors to obtain drainage facility information and shaft seal monitoring information for n consecutive time steps before the current time step, wherein the drainage facility information includes: electrical system information, mechanical system information, hydraulic system information, liquid level control system information, and water quality related information; Step 200, performing feature engineering on the drainage facility information and shaft seal monitoring information data acquired at n time steps to obtain frequency domain data and time-frequency data; Step 300, constructing first time series data and second time series data according to the obtained frequency domain data and time-frequency data combined with the original time series data; Step 400, constructing a fault detection model, and inputting the constructed first time series data and second time series data into the fault detection model, and outputting a fault detection result; The specific execution process of step 400 is as follows: Step 401, outputting a first hidden state and a second hidden state according to the constructed first time series data and the second time series data as input; Step 402, performing hidden state fusion on the first hidden state and the second hidden state, and outputting a fused hidden state; Step 403, mapping the fused hidden state to the output through the first fully connected layer to obtain a fault detection result.

2. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: Electrical system information includes: motor voltage, motor current, insulation resistance value, control element action time; Mechanical system information includes: horizontal, vertical and axial vibration acceleration, vibration spectrum, temperature, noise decibel value; Hydraulic system information includes: water inlet flow, water outlet flow, water inlet pressure value, water outlet pressure value; Liquid level control system information includes: liquid level measurement value, liquid level change rate; Water quality related information includes: pH value, sediment content, chemical oxygen demand (COD); The shaft seal monitoring information includes: shaft seal temperature, seal chamber pressure, and shaft seal vibration frequency.

3. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: The frequency of data collection is fixed, and the default value is to collect data once every 1 minute.

4. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: The first time series data and the second time series data are both derived from the first hidden state and the second hidden state through CNN and LSTM.

5. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: The output fault detection result is a vector. The component represents the The probability value of a fault. If the probability value is higher than 0.5, it means the fault has occurred. If it does not exist, it means it is normal.

6. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: The second hidden state is connected to the second fully connected layer for separate pre-training, and the output of the second fully connected layer is the fault state of the shaft seal; The fault conditions of the shaft seal include: normal, slight, moderate, and severe.

7. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: The fault detection model needs to update parameters through training. It is trained through supervised training and uses the multi-classification cross entropy loss function as the loss function of the model. The designed loss function is as follows: ； : The number of samples in the training set for each iteration; : The number of fault categories; :sample In category The real label on :Model for samples Belongs to category The predicted probability of : The logarithm of the predicted probability, used to calculate the difference between the predicted and true labels; : The average cross entropy loss value of the entire batch of samples.

8. The method for diagnosing drainage facility faults in a water network environment according to claim 1, characterized in that: Feature engineering includes frequency domain feature extraction and time-frequency feature extraction.

9. A drainage facility fault diagnosis method in a water network environment according to claim 8, characterized in that: Frequency domain feature extraction: Apply fast Fourier transform to each time series data to convert it into frequency domain features: ； : original time series data; : Fourier transform; : frequency domain data; Time-frequency feature extraction: Apply wavelet transform to each time series data to extract time-frequency features: ； : original time series data; : Wavelet transform; : Time-frequency data.

10. A drainage facility fault diagnosis system in a water networking environment, characterized in that: It is used to execute a drainage facility fault diagnosis method in a water networking environment as described in any one of claims 1-9.

Images