Oil and gas station machine and pump equipment fault diagnosis method and system based on multi-source data fusion

Abstract

The invention relates to an oil and gas station machine and pump equipment fault diagnosis method and system based on multi-source data fusion. The method comprises the steps of collecting high-frequency vibration data and low-frequency operation condition data of a machine and pump rolling bearing, such as inlet and outlet flow, inlet and outlet pressure and inlet and outlet temperature data; high-frequency vibration signals in normal and abnormal operation states are converted into two-dimensional image signals, and the two-dimensional image signals and operation state labels form a data set; comprising the following steps: performing feature extraction on a one-dimensional vibration signal, forming an operation feature set with a low-frequency operation signal, and forming a data set with an operation state label; a rolling bearing fault diagnosis model is constructed based on a two-dimensional convolutional neural network (2DCNN) and a support vector machine (SVM) respectively, two fault diagnosis results are comprehensively considered based on an improved D-S evidence theory to obtain a final fault diagnosis result, and through a decision-level multi-source data fusion method, time characteristics of data are considered, and the fault diagnosis accuracy is improved. And the spatial characteristics of the data are considered, so that the diagnosis result is more accurate.

Description

Technical Field

[0001] The present invention belongs to the field of mechanical equipment fault diagnosis, and specifically relates to a fault diagnosis method and system for oil and gas station pump equipment based on multi-source data fusion. Background Technique

[0002] With the rapid development of the oil and gas industry, the role of pump equipment in oil and gas stations is becoming increasingly important during the production process. However, due to factors such as complex equipment operating environments and variable working conditions, pump equipment is prone to various faults, resulting in reduced production efficiency, energy waste, and increased safety risks. Therefore, timely and effective fault diagnosis of pump equipment has become an important means to ensure the safety and stability of oil and gas production.

[0003] Currently, the fault diagnosis of pump equipment in oil and gas stations mainly relies on manual experience judgment and simple sensor monitoring data. However, this method has disadvantages such as strong subjectivity, single monitoring data, and inability to comprehensively reflect the equipment operating conditions, resulting in low fault diagnosis accuracy and difficulty in meeting the requirements of oil and gas production. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides a fault diagnosis method and system for oil and gas station pump equipment based on multi-source data fusion. This method constructs a multi-source data fusion model by fusing various sensor monitoring data, historical operation data, etc., to comprehensively and accurately analyze and evaluate the operating status of pump equipment. The original vibration signal of the pump equipment collected is converted into a one-dimensional time series signal and a two-dimensional image signal. Based on the one-dimensional time series vibration signal, features are extracted, and the extracted features and the low-frequency operation signal are together formed into an operation feature set, which together with the operation status label forms a data set. Based on the SVM algorithm, the basic probabilities of various fault types are calculated; based on the two-dimensional image vibration signal, a rolling bearing fault diagnosis model is constructed using a two-dimensional convolutional neural network (2DCNN) to obtain the basic probabilities of various fault types, and the improved D-S evidence theory is used to comprehensively consider the two fault diagnosis conclusions to obtain the final diagnosis result. Through the decision-level multi-source data fusion method, both the time characteristics and the spatial characteristics of the data are considered, making the diagnosis result more accurate.

[0005] The technical solution adopted by the present invention to achieve the above object is:

[0006] A fault diagnosis method for oil and gas station pump equipment based on multi-source data fusion, comprising the following steps:

[0007] Collect high-frequency vibration data and low-frequency operation condition data of various operating states of pump equipment;

[0008] Process the collected high-frequency vibration data and low-frequency operating condition data to obtain one-dimensional time-series data, two-dimensional image data, and operating status labels to form a fault dataset together;

[0009] Establish a fault diagnosis model. The fault diagnosis model includes a two-dimensional convolutional neural network model and a support vector machine model. Train the image data and time-series data in the fault dataset respectively, so that the models can each output fault diagnosis categories; use the improved DS evidence theory to fuse the two fault diagnosis results as the final diagnosis category of the fault diagnosis model;

[0010] Collect on-site data, perform data processing, input the optimized fault diagnosis model, and automatically output the status diagnosis category result.

[0011] The processing process of the high-frequency vibration data includes:

[0012] 1) Arrange the one-dimensional vibration data in chronological order;

[0013] 2) Select a number of data in sequence and normalize them to the gray value range of 0-255;

[0014] 3) Fill the normalized data into a two-dimensional matrix row by row to form a gray-scale image;

[0015] 4) Repeat steps 2) and 3) and combine the gray-scale image with the operating status label to form a dataset {image matrix of vibration data, status label}.

[0016] The processing process of the low-frequency data includes:

[0017] 1) Arrange the one-dimensional vibration data in chronological order;

[0018] 2) Select a number of data in sequence and calculate the time-domain indexes of each data;

[0019] 3) Combine the time-domain indexes with the inlet and outlet temperature, pressure, and flow data obtained in the same historical time period for data fusion, and form an operating status feature set {time-domain indexes of vibration data, inlet and outlet temperature, pressure, and flow data} after normalization;

[0020] 4) Repeat steps (2) and (3) and combine the operating status feature set with the operating status label to form a dataset {operating status feature, status label}.

[0021] Training the image data using the two-dimensional convolutional neural network model includes the following steps:

[0022] 1) Construct a fault dataset, combine the obtained two-dimensional grayscale image data with the corresponding status label to form 1 sample, perform one-hot encoding on the fault label, and divide it into a training set and a test set according to a certain proportion;

[0023] 2) Construct a two-dimensional convolutional neural network fault diagnosis model, including an input layer, a two-dimensional convolutional layer, a Dropout layer, a two-dimensional pooling layer, a fully connected layer, and an output layer;

[0024] 3) Further optimize the model by using adaptive adjustment of the learning rate and taking the training error as the stopping condition;

[0025] 4) Set the hyperparameters of the model, including the number of convolutional kernels, the size of convolutional kernels, the sampling method of the pooling layer, the region size, the dropout rate, and the batch size;

[0026] 5) Use the training set data to train the fault diagnosis model so that it can identify the characteristics of different fault types and output the state category probability value m1;

[0027] 6) Input the test training set data into the trained fault diagnosis model to test the fault diagnosis accuracy w1 of the model.

[0028] The training of the time series data using the support vector machine model includes the following steps:

[0029] 1) Construct a fault data set. Combine the obtained one-dimensional time series data with the corresponding state labels to form 1 sample. Perform one-hot encoding on the fault labels and divide them into a training set and a test set according to a certain proportion;

[0030] 2) Construct a support vector machine (SVM) fault diagnosis model. Select the Gaussian radial basis kernel as the kernel function and set the kernel function parameter g and the penalty coefficient C;

[0031] 3) Use the training set data to train the fault diagnosis model so that it can identify the characteristics of different fault types and output the state category probability value m2;

[0032] 4) Input the test set data into the trained fault diagnosis model to test the diagnosis accuracy w2 of the model.

[0033] The final diagnosis result is obtained by using the improved D-S evidence theory with the two fault diagnosis conclusions through the following steps:

[0034] 1) Convert the preliminary diagnosis results obtained from the above two fault diagnoses into evidence bodies;

[0035] 2) Define the basic probability assignment (BPA) assignment principle for each evidence body;

[0036] 3) Fuse the evidence bodies to obtain the final diagnosis result. The fusion calculation formula is as follows:

[0037]

[0038] Among them, A is a subset of all possible operating states, including state labels {normal, inner ring fault, outer ring fault, and rolling element fault}. w1' and w2' represent the weights of two algorithms in the D-S evidence theory, which are obtained by normalizing the diagnostic accuracies w1 and w2 of the two algorithms in the sample test set. m1 and m2 represent the probabilities of two different fault diagnosis algorithms diagnosing the state type A. K is a normalization coefficient, and the calculation formula is as follows:

[0039]

[0040] Among them, both B and C represent the current operating state of the device diagnosis.

[0041] An oil and gas station pump equipment fault diagnosis system based on multi-source data fusion, comprising:

[0042] Industrial data acquisition equipment, which are sensors and instruments set at the industrial equipment end; used to collect high-frequency vibration data and low-frequency operating condition data of various operating states of pump equipment.

[0043] A fault diagnosis host computer, which receives the data of the industrial data acquisition equipment, trains and optimizes the built-in fault diagnosis program to obtain an optimized fault diagnosis model. The fault diagnosis program executes the method steps described in any one of claims 1-6; and uses the optimized and trained fault diagnosis model to identify faults for the real-time collected data on site and outputs the fault diagnosis category result.

[0044] The beneficial effects of the present invention are:

[0045] 1. By comprehensively utilizing multi-source data, including high-frequency vibration data and low-frequency operating condition data, the present invention can comprehensively reflect the operating state and potential faults of the equipment.

[0046] 2. By introducing a decision-level multi-source data fusion method, the present invention considers both the time characteristics and the spatial characteristics of the data, can more accurately identify fault patterns, and reduce the possibility of misdiagnosis and missed diagnosis. At the same time, combined with the improved D-S evidence theory, the two fault diagnosis results are comprehensively considered, further improving the reliability of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flowchart of the fault diagnosis method of the present invention.

[0048] Figure 2 is a two-dimensional grayscale image generated based on vibration data. DETAILED DESCRIPTION OF THE INVENTION

[0049] The following further describes the present invention in detail with reference to the drawings and embodiments.

[0050] AsFigure 1 As shown, it is the flowchart of the method of the present invention.

[0051] The fault diagnosis method for pump and motor equipment in oil and gas stations based on multi-source data fusion mainly includes three steps: data acquisition step, data processing step, and fault diagnosis step. In the data acquisition step, high-frequency vibration data of the pump and motor rolling bearings and low-frequency operating condition data such as inlet and outlet flow rates, inlet and outlet pressures, and inlet and outlet temperatures are collected. The data processing module converts the high-frequency vibration signal into a two-dimensional image signal, extracts features from the one-dimensional vibration signal, and combines the low-frequency operating signal for normalization to form an operating feature set. The fault diagnosis module constructs a rolling bearing fault diagnosis model based on 2DCNN, constructs a rolling bearing fault diagnosis model based on SVM, and finally constructs a comprehensive fault diagnosis model based on the improved D-S evidence theory.

[0052] The specific steps of the present invention are as follows:

[0053] Step 1: Data acquisition step, collect high-frequency vibration data of the pump and motor rolling bearings and low-frequency operating condition data such as inlet and outlet flow rates, inlet and outlet pressures, and inlet and outlet temperatures, etc.;

[0054] Step 2: Data processing step, including high-frequency vibration data processing and low-frequency time-series data processing;

[0055] Step 2.1: The high-frequency vibration data processing includes the following steps:

[0056] Step 2.1.1: Arrange the one-dimensional vibration data in chronological order;

[0057] Step 2.1.2: Select 1024 data in sequence and normalize the arranged vibration data to the gray value range of 0-255;

[0058] Step 2.1.3: Fill the normalized data into a two-dimensional matrix row by row to form a gray-scale image, as Figure 2 shown;

[0059] Step 2.1.4: Repeat Step 2.1.2 and Step 2.1.3 and combine the gray-scale image with the operation status label to form a data set. For example, the normal operation label is 0, the inner ring fault label is 1, the outer ring fault label is 2, and the rolling element fault label is 3.

[0060] Step 2.2: The low-frequency time-series data processing includes the following steps:

[0061] Step 2.2.1: Arrange the one-dimensional vibration data in chronological order;

[0062] Step 2.2.2: Select 1024 data in sequence, and calculate their mean differences, absolute average values, variance standard deviations, root mean square amplitudes, root mean square values, peak values, maximum values, minimum values, waveform indicators, peak indicators, pulse indicators, margin indicators, skewness, and kurtosis and other time-domain indicators. The calculation methods of each indicator are as follows in the table:

[0063]

[0064] Step 2.2.3: Combine the extracted time-domain indicators with the inlet and outlet temperature, pressure, and flow data obtained during the same historical time period for data fusion, and form a set of operating state characteristics after normalization;

[0065] Step 2.2.4: Repeat Step 2.2.2 and Step 2.2.3 and combine the set of operating state characteristics with the operating state labels to form a data set. For example, the normal operating label is 0, the inner ring fault label is 1, the outer ring fault label is 2, and the rolling element fault label is 3.

[0066] Step 3: Fault diagnosis steps, including constructing a rolling bearing fault diagnosis model based on 2DCNN, constructing a rolling bearing fault diagnosis model based on SVM, and finally a comprehensive fault diagnosis model based on the improved D-S evidence theory;

[0067] Step 3.1: Constructing a rolling bearing fault diagnosis model based on 2DCNN includes the following steps:

[0068] Step 3.1.1: Construct a fault data set. Combine the obtained two-dimensional grayscale image data with the corresponding state labels to form 1 sample. Perform one-hot encoding on the fault labels and divide them into a training set and a test set according to a ratio of 4:1;

[0069] Step 3.1.2: Construct a two-dimensional convolutional neural network (2DCNN) fault diagnosis model, which consists of an input layer, a two-dimensional convolutional layer, a Dropout layer, a two-dimensional pooling layer, a fully connected layer, and an output layer.

[0070] Step 3.1.3: Further optimize the model by using a method of adaptively adjusting the learning rate and using the training error as the stopping condition. The specific method is as follows: Set the initial learning rate to 0.009, the minimum learning rate to 0.001, adjust the learning rate every five iterations, and set the decay factor of the learning rate to 0.9.

[0071] Step 3.1.4: Set the hyperparameters of the model according to experience. The number of convolutional kernels is 12, the size of the convolutional kernel is 3×3, the pooling layer uses mean sampling, the region size is selected as 3×3 and the region can overlap, the dropout rate is 0.5, and the batch size is 32;

[0072] Step 3.1.5: Use the training set data to train the fault diagnosis model so that it can identify the characteristics of different fault types and output the probability value m1 of the status diagnosis category;

[0073] Step 3.1.6: Input the test training set data into the trained fault diagnosis model to test the fault diagnosis accuracy rate w1 of the model.

[0074] Step 3.2: Constructing a rolling bearing fault diagnosis model based on SVM includes the following steps:

[0075] Step 3.2.1: Construct a fault data set. Combine the obtained one-dimensional time series data with the corresponding status labels to form 1 sample. Perform one-hot encoding on the fault labels and divide them into a training set and a test set according to a ratio of 4:1;

[0076] Step 3.2.2: Construct a support vector machine (SVM) fault diagnosis model. Select the Gaussian radial basis kernel as the kernel function, set the kernel function parameter g to 0.25, and set the penalty coefficient C to 1.0.

[0077] Step 3.2.3: Use the training set data to train the fault diagnosis model so that it can identify the characteristics of different fault types and output the probability value m2 of the status diagnosis category;

[0078] Step 3.2.4: Input the test set data into the trained fault diagnosis model to test the diagnosis accuracy rate w2 of the model.

[0079] Step 3.3: The comprehensive fault diagnosis model based on the improved D-S evidence theory includes the following steps:

[0080] Step 3.3.1: Convert the preliminary diagnosis results obtained from the above two fault diagnoses into evidence bodies;

[0081] Step 3.3.2: Define the basic probability assignment (BPA) assignment principle for each evidence body;

[0082] Step 3.3.3: Fuse the evidence bodies to obtain the final diagnosis result. The fusion calculation formula is as follows:

[0083]

[0084] Where A is a subset in the recognition framework, that is, an operating state of the device, such as normal, inner ring fault, outer ring fault, and rolling element fault. The recognition framework is the set of all possible operating states. w1′ and w2′ represent the weights of the two algorithms in the D-S evidence theory, which are obtained by normalizing the diagnosis accuracy rates w1 and w2 of the two algorithms in the sample test set. m1 and m2 represent the probabilities of two different fault diagnosis algorithms diagnosing the state type A of faults, and K is a normalization coefficient. The calculation formula is as follows:

[0085]

[0086] Where both B and C represent an operating state of the current diagnosis of the device.

[0087] In practical applications, the on-site bearing data is collected and input into two trained fault diagnosis models to automatically output the results of the fault type labels.

[0088] This example also provides a fault diagnosis system for oil and gas station pumps and other equipment based on multi-source data fusion, including:

[0089] Industrial data acquisition equipment, which are sensors and instruments set at the industrial equipment end; used to collect high-frequency vibration data and low-frequency operating condition data of various operating states of pumps and other equipment;

[0090] The upper computer for fault diagnosis receives the data of the industrial data acquisition equipment, trains and optimizes the built-in fault diagnosis program to obtain an optimized fault diagnosis model. The fault diagnosis program executes the method steps as described above; and uses the optimized and trained fault diagnosis model to identify faults in the real-time collected data on site and outputs the results of the fault diagnosis categories.

[0091] The above preferred embodiments are only typical examples of the disclosed examples. All equivalent replacements or equal changes and modifications made according to the idea of this example are within the scope protected by this patent.

Claims

1. A fault diagnosis method for oil and gas station pump equipment based on multi-source data fusion, characterized in that It includes the following steps: Collect high-frequency vibration data and low-frequency operating condition data of various operating states of pumps and motors; Process the collected high-frequency vibration data and low-frequency operating condition data to obtain one-dimensional time series data, two-dimensional image data, and operating state labels, which together form a fault dataset; Establish a fault diagnosis model. The fault diagnosis model includes a two-dimensional convolutional neural network model and a support vector machine model. Train the image data and time series data in the fault dataset respectively, so that each model can output the fault diagnosis category; Use the improved DS evidence theory to fuse the two fault diagnosis results as the final diagnosis category of the fault diagnosis model; Collect on-site data, process the data, input it into the optimized fault diagnosis model, and automatically output the state diagnosis category result.

2. The fault diagnosis method for oil and gas station pump equipment based on multi-source data fusion according to claim 1, wherein The high-frequency vibration data processing process includes: 1) Arrange the one-dimensional vibration data in chronological order; 2) Select several data in sequence and normalize them to the gray value range of 0-255; 3) Fill the normalized data into a two-dimensional matrix row by row to form a grayscale image; 4) Repeat steps 2) and 3) and combine the grayscale image with the operating state label to form a dataset {image matrix of vibration data, state label}.

3. The fault diagnosis method for oil and gas station pump equipment based on multi-source data fusion according to claim 1, wherein The low-frequency data processing process includes: 1) Arrange the one-dimensional vibration data in chronological order; 2) Select several data in sequence and calculate the time domain indexes of each data; 3) Combine the time domain indexes with the inlet and outlet temperature, pressure, and flow data obtained in the same historical time period for data fusion, and form an operating state feature set {time domain indexes of vibration data, inlet and outlet temperature, pressure, and flow data} after normalization; 4) Repeat steps (2) and (3) and combine the operating state feature set with the operating state label to form a dataset {operating state feature, state label}.

4. The fault diagnosis method and system for oil and gas station pump equipment based on multi-source data fusion according to claim 1, characterized in that Training the image data using the two-dimensional convolutional neural network model includes the following steps: 1) Construct a fault dataset, combine the obtained two-dimensional grayscale image data with the corresponding state label to form 1 sample, perform one-hot encoding on the fault label, and divide it into a training set and a test set according to a certain proportion; 2) Construct a two-dimensional convolutional neural network fault diagnosis model, including an input layer, a two-dimensional convolutional layer, a Dropout layer, a two-dimensional pooling layer, a fully connected layer, and an output layer; 3) Further optimize the model by adaptively adjusting the learning rate and using the training error as the stopping condition; 4) Set the hyperparameters of the model, including the number of convolutional kernels, the size of the convolutional kernels, the sampling method of the pooling layer, the region size, the dropout rate, and the batch size; 5) Use the training set data to train the fault diagnosis model so that it can identify the characteristics of different fault types and output the state category probability value m1; 5. The fault diagnosis method and system for oil and gas station pump equipment based on multi-source data fusion according to claim 1, characterized in that, 6) Input the test training set data into the trained fault diagnosis model to test the fault diagnosis accuracy w1 of the model. Training the time series data using the support vector machine model includes the following steps: 1) Construct a fault dataset, combine the obtained one-dimensional time series data with the corresponding state label to form 1 sample, perform one-hot encoding on the fault label, and divide it into a training set and a test set according to a certain proportion; 2) Construct a support vector machine (SVM) fault diagnosis model, select the Gaussian radial basis kernel as the kernel function, and set the kernel function parameter g and the penalty coefficient C; 3) Use the training set data to train the fault diagnosis model to enable it to identify the probability value m2 of the output state category of different fault type features; 4) Input the test set data into the trained fault diagnosis model to test the diagnostic accuracy w2 of the model.

6. The fault diagnosis method and system for oil and gas station pump equipment based on multi-source data fusion according to claim 1, characterized in that The final diagnostic result is obtained by using the improved D-S evidence theory with two fault diagnosis conclusions through the following steps: 1) Convert the preliminary diagnostic results obtained from the above two fault diagnoses into evidence bodies; 2) Define the basic probability assignment (BPA) assignment principle for each evidence body; 3) Fuse the evidence bodies to obtain the final diagnostic result. The fusion calculation formula is as follows: where A is the subset of all possible operating states including the state labels {normal, inner ring fault, outer ring fault, and rolling element fault}, w1' and w2' represent the weights of the two algorithms in the D-S evidence theory, which are obtained by normalizing the diagnostic accuracies w1 and w2 of the two algorithms in the sample test set, m1 and m2 represent the probabilities of the two different fault diagnosis algorithms diagnosing the state type A, and K is the normalization coefficient. The calculation formula is as follows: where both B and C represent the current operating state of the equipment diagnosis.

7. A fault diagnosis system for oil and gas station pump equipment based on multi-source data fusion, characterized in that, Including: Industrial data acquisition devices, which are sensors and instruments installed at the industrial equipment end; Used to collect high-frequency vibration data and low-frequency operating condition data of various operating states of pumps; The fault diagnosis host computer, which receives the data of the industrial data acquisition device, trains and optimizes the built-in fault diagnosis program to obtain an optimized fault diagnosis model. The fault diagnosis program executes the method steps described in any one of claims 1-6; And uses the optimized and trained fault diagnosis model to identify faults in the real-time collected data on-site and output the fault diagnosis category result.

Images