A machine learning-based machine pump fault diagnosis method and system
By labeling historical vibration data and training the model using Fourier transform, the problems of parameter complexity and high professional requirements in pump fault diagnosis are solved, achieving higher diagnostic accuracy and intuitiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- QINGDAO AENES INFORMATION TECH CO LTD
- Filing Date
- 2023-05-11
- Publication Date
- 2026-05-15
AI Technical Summary
Existing pump and motor fault diagnosis methods have too many parameters, complex reasoning processes, high requirements for the expertise of commissioning personnel, and difficulty in improving accuracy when data is insufficient.
A machine learning-based fault diagnosis method is adopted. By labeling historical vibration data and performing fast Fourier transform, unbalanced, misaligned, and under-lubricated fault models are trained. Frequency domain information and ISO international standard amplitudes are used for diagnosis, and the model is optimized by combining backpropagation algorithm.
It reduces the professional requirements for personnel handling the problem, improves the accuracy and intuitiveness of diagnostic results, and better reflects the equipment status and fault type.
Smart Images
Figure CN116720112B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of fault diagnosis technology, specifically, it relates to a machine pump fault diagnosis method and system based on machine learning. Background Technology
[0002] Vibration analysis is a commonly used technique for diagnosing pump and motor faults. It involves measuring the vibration of the pump to identify the cause of any malfunction. Vibration analysis typically uses vibration sensors to measure the pump's vibration and specialized vibration analysis instruments or software to analyze the measured data. Generally, vibration analysis can help identify problems such as bearing failures, impeller failures, and pump casing failures. Furthermore, it can help diagnose issues such as uneven pump speed and uneven flow rate.
[0003] Currently, the industry still relies primarily on manual analysis for pump and compressor fault analysis. Vibration data is collected using vibration monitoring equipment such as spot inspection instruments, and then analyzed manually. This method is labor-intensive and only suitable for analyzing individual devices, and can no longer meet the needs of automated diagnosis for large-scale equipment.
[0004] In the field of intelligent diagnostics, the industry already has diagnostic methods based on machine learning models such as deep neural networks and random forest classification. Deep neural network algorithms typically use a large number of parameters and layers, including numerous hidden layers. Due to the large number of parameters, the inference process is extremely complex, requiring a high level of expertise from the debugging personnel; the inference process can only be operated by professionals. Furthermore, another problem faced by traditional intelligent diagnostic technologies is that when the amount of data is insufficient, it is difficult to optimize the model, thus hindering the improvement of accuracy. When the inference results are incorrect, debugging and correction are even more difficult.
[0005] The information disclosed in this background section is only intended to enhance the understanding of the background technology of this application, and therefore may include prior art that is not known to those skilled in the art. Summary of the Invention
[0006] This invention addresses the technical problems of existing intelligent diagnostic methods for pump and motor faults, such as excessive parameters, extremely complex reasoning processes, and high professional requirements for commissioning personnel. It proposes a new method for diagnosing pump and motor faults that can solve these problems.
[0007] To achieve the above-mentioned objectives, the present invention employs the following technical solution:
[0008] A machine learning-based method for diagnosing pump faults includes:
[0009] The historical vibration data tagging process involves tagging the historical vibration data. The tagging content includes at least the fault level, fault type, and fault percentage. The fault percentage is the proportion of the fault type to all types of faults in the pump. The fault types include at least imbalance faults, misalignment faults, and insufficient lubrication faults.
[0010] Perform Fast Fourier Transform on historical vibration data to obtain frequency domain information of historical vibration data, including at least the harmonic amplitude and high frequency energy amplitude of the data in the frequency domain;
[0011] The frequency domain information of historical vibration data is used to train models for each fault type using the backpropagation algorithm.
[0012] Acquire vibration data, perform fast Fourier transform on the vibration data, and obtain the frequency domain information of the vibration data;
[0013] The frequency domain information of the vibration data is input into the model of each fault type, and the fault level and fault percentage of the vibration data are calculated based on the output results of the model.
[0014] In some embodiments, the frequency domain information also includes the envelope energy value of the data in the frequency domain.
[0015] In some embodiments, training the model for each fault type also includes using the ISO international standard amplitude as an input value to train the model.
[0016] In some embodiments, the harmonic amplitude includes 1 to 8 harmonic amplitudes.
[0017] In some embodiments, the model includes an imbalance fault model f(x), a misalignment fault model h(x), and an underlubrication fault model, wherein:
[0018]
[0019]
[0020]
[0021] The steps for training models for each fault type include training the parameters 'a' for each model. ji b ji and c ji Where j = 1, 2, 3, i = 1, 2, ... 10, x1 to x8 represent the 1st to 8th harmonic amplitude values of the vibration data, x9 represents the high-frequency energy amplitude of the vibration data, and x 10 The envelope energy value of the vibration data, x 11 This indicates the ISO international standard amplitude.
[0022] In some embodiments, the pump fault diagnosis method further includes the step of establishing a fault severity value-fault level correspondence table;
[0023] The method for calculating the fault level of the vibration data includes:
[0024] Calculate the fault severity value = k1*f(x) + k2*h(x) + k3*g(x);
[0025] Look up the fault severity value-fault level correspondence table to determine the fault level corresponding to the fault severity value;
[0026] Where k1, k2, and k3 are the weighting coefficients of f(x), h(x), and g(x), respectively.
[0027] In some embodiments, the method for calculating the fault percentage of the vibration data includes:
[0028] Fault percentage = k1*f(x):k2*h(x):k3*g(x).
[0029] In some embodiments, the method also includes a step of manually verifying the output of the model;
[0030] When the model outputs incorrect results, analyze the reasons for the misdiagnosis and correct the output results. When the model outputs correct results, add the data used for validation to the case library and continue to use it to train the model.
[0031] This invention also proposes a pump fault diagnosis system, comprising:
[0032] A wireless vibration sensor is installed on the pump under test to collect vibration data of the pump.
[0033] A wireless gateway, which is wirelessly connected to the wireless vibration sensor;
[0034] A cloud server is connected to the wireless gateway via a wireless network, and the wireless gateway uploads the vibration data sent by the wireless vibration sensor to the cloud server.
[0035] The cloud server is configured to perform fault diagnosis according to any of the pump fault diagnosis methods described above.
[0036] In some embodiments, the wireless vibration sensor communicates with the wireless gateway via Zigbee / Lora wireless communication, and the wireless gateway communicates with the cloud server via 4G / 5G wireless communication.
[0037] Compared with the prior art, the advantages and positive effects of the present invention are:
[0038] The pump fault diagnosis method of this invention marks verified historical vibration data, including at least the fault level, fault type, and fault percentage, and trains a diagnostic model based on this data. The input parameters are simple, and the output results are more intuitive and clear, reducing the professional requirements for operators. By acquiring information such as the octave amplitude and high-frequency energy amplitude of historical vibration data in the frequency domain, the amplitude of different octaves can reflect information such as imbalance faults, misalignment faults, loose equipment foundations, and loose shafts, while the high-frequency energy amplitude can reflect the lubrication and wear conditions of the mechanical equipment. Therefore, the model diagnosis results of this scheme are more accurate.
[0039] Other features and advantages of the present invention will become clearer after reading the detailed embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0041] Figure 1 This is a flowchart of the model training steps in one embodiment of the pump fault diagnosis method proposed in this invention;
[0042] Figure 2 This is a flowchart of the fault diagnosis steps in one embodiment of the pump fault diagnosis method proposed in this invention;
[0043] Figure 3 This is a schematic diagram of a fault diagnosis model in one embodiment of the pump fault diagnosis method proposed in this invention;
[0044] Figure 4 This is a spectrum diagram of misalignment fault in one embodiment of the pump fault diagnosis method proposed in this invention;
[0045] Figure 5 This is a schematic diagram of an embodiment of the pump fault diagnosis system proposed in this invention. Detailed Implementation
[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0047] It should be noted that in the description of this invention, terms such as "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," indicating directional or positional relationships, are based on the directional or positional relationships shown in the accompanying drawings. These are merely for ease of description and do not indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation; therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0048] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0049] Example 1
[0050] This embodiment proposes a machine learning-based method for diagnosing pump and motor faults, such as... Figure 1 , Figure 2 As shown, it includes:
[0051] The historical vibration data tagging process involves tagging the historical vibration data. The tagging content includes at least the fault level, fault type, and fault percentage. The fault percentage is the proportion of the fault type to all types of faults in the pump. The fault types include at least imbalance faults, misalignment faults, and insufficient lubrication faults.
[0052] Perform Fast Fourier Transform on historical vibration data to obtain frequency domain information of historical vibration data, including at least the harmonic amplitude and high frequency energy amplitude of the data in the frequency domain;
[0053] The frequency domain information of historical vibration data is used to train models for each fault type using the backpropagation algorithm.
[0054] Acquire vibration data, perform fast Fourier transform on the vibration data, and obtain the frequency domain information of the vibration data;
[0055] The frequency domain information of the vibration data is input into the model of each fault type, and the fault level and fault percentage of the vibration data are calculated based on the output results of the model.
[0056] The pump fault diagnosis method in this embodiment marks verified historical vibration data, including at least the fault level, fault type, and fault percentage. Based on this, a diagnostic model is trained. The input parameters are simple, and the output results are more intuitive and clear, reducing the professional requirements for operators. By acquiring information such as the octave amplitude and high-frequency energy amplitude of historical vibration data in the frequency domain, different octave amplitudes can reflect information such as imbalance faults, misalignment faults, loose equipment foundations, and loose shafts. The high-frequency energy amplitude can reflect the lubrication and wear conditions of the mechanical equipment. Therefore, the model diagnosis results of this scheme are more accurate.
[0057] In the field of industrial vibration analysis, low harmonics have strong diagnostic value. For example, an elevated 1st harmonic indicates a high probability of equipment imbalance, an elevated 2nd harmonic may indicate misalignment, elevated 3rd and 4th harmonics may indicate loose foundations, and elevated 5th and 6th harmonics may indicate loose shafts. This system selects the 1st to 8th harmonic amplitudes for analysis, as these have higher analytical value. High-frequency energy amplitudes can reflect the lubrication and wear conditions of mechanical equipment, and are therefore also an important calculation parameter. Thus, in some embodiments, the harmonic amplitudes include those from the 1st to 8th harmonics. Figure 4 The image shows a typical spectrum diagram of a misalignment fault, which shows the amplitude of the misalignment fault at the 1st, 2nd, and 3rd harmonics.
[0058] The parameters mentioned above are difficult to analyze in isolation. For example, a loose equipment foundation could cause the 1st octave to be too high, so a high 1st octave does not necessarily indicate an imbalance fault. Therefore, this solution uses an artificial intelligence model to perform batch analysis of the above parameters and draw conclusions, thus improving accuracy.
[0059] Envelope energy values are of relatively clear guiding significance for some faults, such as rolling bearing faults, gear faults, and poor lubrication. Therefore, in some embodiments, frequency domain information also includes the envelope energy value of the data in the frequency domain.
[0060] In some embodiments, training the models for each fault type also includes inputting the ISO international standard amplitude as an input value into the model for training. The ISO international standard amplitude is used to provide a reference value for the octave amplitude.
[0061] In some embodiments, the model includes an imbalance fault model f(x), a misalignment fault model h(x), and an underlubrication fault model g(x), wherein:
[0062]
[0063]
[0064]
[0065] The steps for training models for each fault type include training the parameters 'a' for each model. ji b ji and c ji Where j = 1, 2, 3, i = 1, 2, ... 10, x1 to x8 represent the 1st to 8th harmonic amplitude values of the vibration data, x9 represents the high-frequency energy amplitude of the vibration data, and x 10 The envelope energy value of the vibration data, x 11 This indicates the ISO international standard amplitude.
[0066] like Figure 3 As shown, the inputs to the three model functions f(x), h(x), and g(x) are x1, x2, x3...x11, and the outputs are single scalars. The input to o(x) is the output of f(x), h(x), and g(x), and the output is the overall severity score of the fault and the weight of each fault cause.
[0067] This solution trains imbalance, misalignment, and under-lubrication types separately as the visible layer in the entire system. It not only analyzes the equipment's operating status and possible causes of failure, but also outputs derivation processes that can be understood by human logic, facilitating equipment maintenance engineers' comprehension and troubleshooting.
[0068] In some embodiments, the pump fault diagnosis method further includes the step of establishing a fault severity value-fault level correspondence table;
[0069] Methods for calculating the fault level of vibration data include:
[0070] Calculate the fault severity value = k1*f(x) + k2*h(x) + k3*g(x);
[0071] Look up the fault severity value-fault level mapping table to determine the fault level corresponding to the fault severity value;
[0072] Where k1, k2, and k3 are the weighting coefficients of f(x), h(x), and g(x), respectively.
[0073] In some embodiments, the method for calculating the fault percentage of vibration data includes:
[0074] Fault percentage = k1*f(x):k2*h(x):k3*g(x).
[0075] In some embodiments, the method also includes a step of manually verifying the output of the model;
[0076] When the model outputs incorrect results, analyze the reasons for the misdiagnosis and correct the output. When the model outputs correct results, add the validation data to the case library and continue training the model. This allows for further model optimization.
[0077] In the training steps for each fault type, the unbalanced model, misalignment model, and underlubrication model are calculated using the backpropagation algorithm. The backpropagation algorithm is an algorithm that derives the model from an existing dataset.
[0078] This embodiment uses historical vibration data that has been manually verified to train and solve the model.
[0079] Taking the training of an imbalanced fault model f(x) as an example, the dataset uses eleven input parameters (x1 to x11) as a set of input data, and y1, y2, and y3 as a set of expected data for f(x), h(x), and g(x). A total of m sets of data are used. The bias J function of f(x) is calculated using the residual sum of squares algorithm. Let x(i) represent the i-th set of input data, and y(i) represent the i-th set of output data corresponding to x(i) in the f(x) function. Then:
[0080]
[0081] It is known that the smaller the result of the bias function J, the higher the model accuracy. The parameters of f(x) are solved using the gradient descent method, and the calculation steps are as follows:
[0082]
[0083]
[0084] …
[0085]
[0086]
[0087] …
[0088]
[0089]
[0090] …
[0091]
[0092]
[0093] …
[0094] Where α is the preset learning rate, for example, it can be set to 0.01.
[0095] The calculation of each parameter needs to be repeated to guide the output of the deviation J function to stabilize.
[0096] Similarly, the parameters in the misalignment fault model h(x) and the underlubrication fault model g(x) are calculated using the above steps.
[0097] In the vibration data diagnosis step, the summary model o(f(x),h(x),g(x)) takes f(x),h(x),g(x) as inputs and outputs as outputs. After weight calculation, the final conclusion is obtained.
[0098] Example 2
[0099] This embodiment proposes a pump fault diagnosis system, such as Figure 5 As shown, it includes:
[0100] A wireless vibration sensor is installed on the pump under test to collect vibration data of the pump.
[0101] A wireless gateway, which is wirelessly connected to the wireless vibration sensor;
[0102] A cloud server is connected to the wireless gateway via a wireless network, and the wireless gateway uploads the vibration data sent by the wireless vibration sensor to the cloud server.
[0103] The cloud server is configured to perform fault diagnosis according to the pump fault diagnosis method described in Embodiment 1.
[0104] In some embodiments, the wireless vibration sensor communicates with the wireless gateway via Zigbee / Lora wireless communication, and the wireless gateway communicates with the cloud server via 4G / 5G wireless communication.
[0105] The method for diagnosing pump and motor malfunctions can be found in Example 1, and will not be repeated here.
[0106] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions claimed by the present invention.
Claims
1. A machine learning-based method for diagnosing pump faults, characterized in that, include: The historical vibration data tagging process involves tagging the historical vibration data. The tagging content includes at least the fault level, fault type, and fault percentage. The fault percentage is the proportion of the fault type to all types of faults in the pump. The fault types include at least imbalance faults, misalignment faults, and insufficient lubrication faults. Perform Fast Fourier Transform on historical vibration data to obtain frequency domain information of historical vibration data, including at least the harmonic amplitude and high-frequency energy amplitude of the data in the frequency domain; The frequency domain information of historical vibration data is used to train models for each fault type using the backpropagation algorithm. Acquire vibration data, perform fast Fourier transform on the vibration data, and obtain the frequency domain information of the vibration data; The frequency domain information of the vibration data is input into the model of each fault type, and the fault level and fault percentage of the vibration data are calculated based on the output results of the model. The model includes an imbalance fault model f(x), a misalignment fault model h(x), and an underlubrication fault model g(x), wherein: ； ； ； The steps for training models for each fault type include training the parameters of each model. , as well as , Where j=1,2,3, i=1,2,…10, ~ These represent the amplitude values of the vibration data at the 1st to 8th harmonics, respectively. This represents the high-frequency energy amplitude of the vibration data. The envelope energy value represents the vibration data. Indicates the ISO international standard amplitude; The pump fault diagnosis method also includes the step of establishing a fault severity value-fault level correspondence table; The method for calculating the fault level of the vibration data includes: The fault severity value is calculated as: k1 * f(x) + k2 * h(x) + k3 * g(x); Look up the fault severity value-fault level correspondence table to determine the fault level corresponding to the fault severity value; Where k1, k2, and k3 are the weighting coefficients of f(x), h(x), and g(x), respectively.
2. The pump fault diagnosis method according to claim 1, characterized in that, The frequency domain information also includes the envelope energy value of the data in the frequency domain.
3. The pump fault diagnosis method according to claim 1, characterized in that, The method for calculating the fault percentage of the vibration data includes: Fault percentage = k1 * f(x) : k2 * h(x) : k3 * g(x).
4. The pump fault diagnosis method according to any one of claims 1-3, characterized in that, It also includes the step of manually validating the model's output; When the model outputs incorrect results, analyze the reasons for the misdiagnosis and correct the output results. When the model outputs correct results, add the data used for validation to the case library and continue to use it to train the model.
5. A pump fault diagnosis system, characterized in that, include: A wireless vibration sensor is installed on the pump under test to collect vibration data of the pump. A wireless gateway, which is wirelessly connected to the wireless vibration sensor; A cloud server is connected to the wireless gateway via a wireless network, and the wireless gateway uploads the vibration data sent by the wireless vibration sensor to the cloud server. The cloud server is configured to perform fault diagnosis according to the pump fault diagnosis method described in any one of claims 1-4.
6. The pump fault diagnosis system according to claim 5, characterized in that, The wireless vibration sensor communicates with the wireless gateway via Zigbee / Lora wireless communication, and the wireless gateway communicates with the cloud server via 4G / 5G wireless communication.