Multi-stage pump middle section structure quality detection method and system

Abstract

The invention relates to a quality detection method and system for a multistage pump middle section structure, and relates to the field of quality detection, vibration signal monitoring is carried out on the multistage pump middle section structure in the operation process, and operation time parameters and water quality parameters of the middle section structure are collected for flow channel scaling prediction; collecting runner images in the middle section structure through an industrial endoscope, and respectively inputting the runner images into a plurality of scaling parameter classifiers to identify and obtain actual scaling parameters; guide vane deformation prediction is carried out, guide vane images in the middle section structure are collected through an industrial endoscope, and the guide vane images are input into a plurality of guide vane parameter classifiers to be recognized to obtain actual guide vane deformation parameters; according to the method, the actual scaling and guide vane deformation parameters are combined to obtain a quality detection result, and the technical problems that the flow channel scaling and guide vane deformation of the middle section structure of the multi-stage pump cannot be accurately and efficiently detected, the overall performance of the multi-stage pump cannot be controlled, and quality evaluation and maintenance are difficult are solved.

Description

Technical Field

[0001] The present invention relates to the field of quality inspection, and in particular to a method and system for inspecting the quality of a mid-section structure of a multi-stage pump. Background Art

[0002] As a key equipment for fluid transportation, multistage pumps are widely used in various industrial fields. The design and operating state of the mid-section structure of its core component have a crucial impact on the overall performance of the multistage pump. The mid-section structure mainly consists of two parts: the flow channel and the guide vane. Among them, the flow channel is the main channel for the fluid in the multistage pump. Taking water as an example, its structural design and material selection directly determine the smoothness and efficiency of the water flow. However, during long-term use, due to the deposition of impurities and chemical reactions in the water flow, scaling often occurs on the inner wall of the flow channel. These scalings not only reduce the efficiency of the water flow, but also change the flow state of the water flow, causing the multistage pump to vibrate during operation. Vibration not only affects the stability and life of the multistage pump, but may also cause damage to the surrounding equipment and building structures. On the other hand, the guide vane is an important component for realizing water flow energy conversion in the multistage pump. Its structural design and deformation state directly affect the hydraulic performance and efficiency of the multistage pump. Under the long-term scouring and stress of water flow, the guide vane often deforms due to changes in water flow. This deformation not only changes the hydraulic characteristics of the guide vane, but also may cause hydraulic imbalance and vibration problems in the multistage pump. Therefore, real-time monitoring and evaluation of the deformation state of the guide vanes is of great significance to ensure the stable operation of the multi-stage pump. Summary of the invention

[0003] The present invention aims to solve the technical problem that it is difficult to accurately and efficiently detect scaling of the flow channel and deformation of the guide vanes in the middle section structure of a multi-stage pump in the prior art, which leads to the inability to control the overall performance of the multi-stage pump, resulting in difficulties in quality assessment and maintenance. A method and system for detecting the quality of the middle section structure of a multi-stage pump is provided to solve the problem.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: In a first aspect, the present invention provides a method for detecting the quality of a mid-section structure of a multi-stage pump, the method comprising: monitoring vibration signals of the mid-section structure of a target multi-stage pump during operation to obtain vibration parameters, collecting operation time parameters and water quality parameters of the mid-section structure, predicting flow channel scaling, and obtaining predicted scaling parameters; collecting flow channel images in the mid-section structure through an industrial endoscope, and inputting the flow channel images into a plurality of scaling parameter classifiers according to the predicted scaling parameters to identify and obtain actual scaling parameters, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths; predicting guide vane deformation according to the actual scaling parameters to obtain predicted guide vane deformation parameters, collecting guide vane images in the mid-section structure through an industrial endoscope, and inputting the guide vane images into a plurality of guide vane parameter classifiers according to the predicted guide vane deformation parameters to identify and obtain actual guide vane deformation parameters; and obtaining a quality detection result of the mid-section structure by combining the actual scaling parameters and the actual guide vane deformation parameters.

[0005] In a second aspect, the present invention provides a multi-stage pump middle section structure quality detection system, the system comprising: a vibration monitoring module, used to monitor the vibration signal of the middle section structure of the target multi-stage pump during operation, obtain vibration parameters, and collect the operation time parameters and water quality parameters of the middle section structure, predict the flow channel scaling, and obtain the predicted scaling parameters; an image acquisition module, used to collect the flow channel image in the middle section structure through an industrial endoscope, and input the flow channel image into a plurality of scaling parameter classifiers according to the predicted scaling parameters, and identify and obtain the actual scaling parameters, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths; a classification recognition module, used to predict the guide vane deformation according to the actual scaling parameters, obtain the predicted guide vane deformation parameters, collect the guide vane image in the middle section structure through an industrial endoscope, and input the guide vane image into a plurality of guide vane parameter classifiers according to the predicted guide vane deformation parameters, and identify and obtain the actual guide vane deformation parameters; a result acquisition module, used to obtain the quality detection result of the middle section structure in combination with the actual scaling parameters and the actual guide vane deformation parameters.

[0006] The beneficial effects of the present invention are: through vibration signal monitoring, collection of water quality parameters and operating time parameters, combined with industrial endoscope image recognition and deep learning prediction technology, the flow channel scaling and guide vane deformation of the mid-section structure of the multi-stage pump can be accurately and efficiently detected, thereby achieving a comprehensive evaluation of the quality of the mid-section structure, and providing strong support for the maintenance and servicing of the multi-stage pump. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic flow chart of a method for detecting the quality of a mid-section structure of a multi-stage pump provided by the present invention.

[0008] Figure 2A schematic structural diagram of a multi-stage pump mid-section structure quality detection system provided by the present invention.

[0009] Description of reference numerals: vibration monitoring module 11 , image acquisition module 12 , classification and recognition module 13 , result acquisition module 14 . DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0011] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0012] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or description". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0013] Embodiment 1: This scheme is based on the research of the middle section structure of the multi-stage pump. The middle section structure of the multi-stage pump is equipped with a flow channel inside, which is connected to the inlet and outlet of the guide vane impeller. It receives the high-speed water flow thrown out by the previous impeller, first decelerates it, then arranges the direction of the water flow, stabilizes the water flow, and then smoothly transmits it to the next impeller for continued pressurization. In layman's terms, it can be understood that the middle section is like a courier sorting center, and each level of impeller is like a courier. Each level of impeller accelerates the water flow. The middle section flow channel receives the high-speed water flow thrown out by the previous impeller, but the accelerated water will fly around under the action of the impeller. Therefore, the high-speed water flow needs to be "calmed down", that is, decelerated, so that the water flow speed is reduced, and then subsequent pressurization is carried out, that is, the acceleration of the next level of impeller. In this process, the direction of the water flow needs to be arranged to stabilize the water flow state, and the direction of the water flow is aligned with the inlet of the next level of impeller to avoid energy loss caused by inconsistent directions during water flow transmission and pressurization. Then the water flow is smoothly transmitted to the next level of impeller so that the next level of impeller can continue to pressurize the water flow. Based on the above working principles and actual impacts, the quality inspection of this plan is implemented to achieve a comprehensive assessment of the quality of the mid-section structure and provide strong support for the maintenance and servicing of multi-stage pumps.

[0014] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting the quality of a multi-stage pump mid-section structure, the method comprising: S10: monitoring vibration signals of the middle section structure of the target multistage pump during operation to obtain vibration parameters, collecting operation time parameters and water quality parameters of the middle section structure, predicting flow channel scaling, and obtaining predicted scaling parameters.

[0015] For example, vibration sensors are arranged to continuously monitor the vibration signals of a multi-stage pump during operation. These sensors can capture tiny vibration changes in the pump body and convert these vibration signals into analyzable vibration parameters, such as vibration frequency, amplitude, etc. These parameters can indirectly reflect the operating status of the internal flow path and component wear of the pump, and are important indicators for evaluating the operating status of the pump.

[0016] At the same time, water quality monitoring instruments are used to regularly collect various parameters of the water quality in the middle section of the multi-stage pump, such as water hardness, pH value, oxygen content, and impurity content. These water quality parameters are important bases for evaluating scaling tendency, and different water quality conditions will directly affect the speed and type of scaling.

[0017] In addition, the running time parameters of the mid-section structure are collected, that is, the cumulative running time of the multi-stage pump is recorded, which helps to understand the use history and potential aging problems of the pump, and provides time parameters for scale deposition.

[0018] After obtaining the vibration parameters, water quality parameters and running time parameters, the prediction model based on deep learning is used to predict the flow channel scaling. This model is trained and optimized through a large number of historical maintenance data of multi-stage pumps of the same model, and can accurately identify the complex relationship between various parameters and scaling size (such as scaling area and thickness). By inputting the currently monitored parameters into the model, the predicted scaling parameters can be obtained, so as to understand the scaling of the flow channel in advance and provide a scientific basis for subsequent maintenance decisions. The specific scaling parameters refer to the scaling size, that is, the scaling area and thickness. In addition, in other possible cases, the scaling type, scaling distribution, scaling density and scaling growth rate can also be predicted, which are not involved here. For example, if the prediction results show that the scaling area and thickness of a certain section of the flow channel have reached the level that affects the performance of the pump, then the cleaning work can be arranged in time to avoid the adverse effects of scaling on the pump performance and ensure the stable operation of the multi-stage pump.

[0019] S20: using an industrial endoscope, collecting a flow channel image in the middle section structure, and inputting the flow channel image into a plurality of scaling parameter classifiers according to the predicted scaling parameters to identify and obtain actual scaling parameters, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths.

[0020] Optionally, use industrial endoscope equipment to go deep into the middle structure of the multistage pump and collect clear images of the flow channel. These images intuitively show the scaling of the inner wall of the flow channel and provide information for subsequent analysis. Industrial endoscope equipment is a high-precision instrument used for industrial inspection. It uses optical or electronic imaging technology to clearly present the internal structure, defects, corrosion, scaling, etc. of the equipment, helping to accurately judge the internal condition of the equipment, discover potential problems in time, and provide strong support for equipment maintenance and overhaul.

[0021] After obtaining the flow channel image through the industrial endoscope, the scaling parameter classifier corresponding to the scaling parameter level close to the predicted scaling parameter obtained by the vibration signal, water quality and running time parameters is selected. These classifiers are trained in advance with a large amount of sample data and can accurately identify the characteristics of different scaling parameter levels. Among them, each scaling parameter classifier contains multiple scaling parameter classification paths, which are built based on deep learning algorithms and can perform integrated binary classification on the input flow channel image. In other words, they will determine whether the scaling in the image belongs to the scaling parameter level corresponding to the classifier. In this way, the scaling parameter classification results output by multiple classifiers are obtained.

[0022] In order to obtain the most accurate actual scaling parameters, the probability of each scaling parameter level appearing in multiple classification results is calculated. Specifically, the proportion of classification results judged as "yes" is counted to obtain the probability distribution of different scaling parameter levels. Finally, the scaling parameter level with the highest probability is selected as the actual scaling parameter.

[0023] In summary, this method not only improves the detection efficiency, but also greatly improves the accuracy. The integrated binary classification of multiple classifiers can more comprehensively capture the characteristics of flow channel scaling, avoiding the misjudgment that may exist in a single classifier. At the same time, since the classification model has been fully trained with a large amount of sample data, its convergence efficiency is high, and it can quickly give accurate detection results in practical applications, thereby more efficiently monitoring and maintaining the flow channel scaling of the middle section structure of the multi-stage pump, ensuring the stable operation of the multi-stage pump.

[0024] S30: predicting guide vane deformation according to the actual scaling parameters to obtain predicted guide vane deformation parameters, collecting guide vane images in the middle section structure through an industrial endoscope, and inputting the guide vane images into a plurality of guide vane parameter classifiers according to the predicted guide vane deformation parameters to identify and obtain actual guide vane deformation parameters.

[0025] S40: Combining the actual scaling parameters and the actual guide vane deformation parameters, obtaining a quality inspection result of the mid-section structure.

[0026] Specifically, the deformation of the guide vanes is predicted based on the actual scaling parameters. Scaling will change the flow channel structure, thereby affecting the water flow state, which in turn causes the deformation of the guide vanes. Therefore, based on the correlation between the structural parameters and the guide vane deformation, the predicted guide vane deformation parameters are obtained through model prediction. Next, an industrial endoscope is used to capture clear images of the guide vanes in the middle section structure. These images intuitively show the actual shape of the guide vanes.

[0027] In order to accurately identify the deformation parameters of the guide vanes, the guide vane parameter classifier corresponding to the closest guide vane deformation parameter level is selected according to the predicted guide vane deformation parameters. Each guide vane parameter classifier contains multiple classification paths based on deep learning algorithms, which can perform integrated binary classification on the input guide vane image to determine whether it belongs to the guide vane deformation parameter level corresponding to the classifier. The probability distribution of each guide vane deformation parameter level is calculated through the comprehensive judgment of multiple classifiers, and the level with the highest probability is selected as the actual guide vane deformation parameter. This step is similar to the acquisition of actual scaling parameters.

[0028] Finally, the actual scaling parameters and actual guide vane deformation parameters are combined to comprehensively evaluate the quality of the mid-section structure. For example, if the actual scaling parameters show that the flow channel is severely scaled, and the actual guide vane deformation parameters show that the guide vanes are significantly deformed, then it can be judged that there are quality problems with the mid-section structure and it needs to be cleaned and maintained in a timely manner. This process not only improves the detection efficiency, but also ensures the accuracy of the detection results, providing a strong guarantee for the stable operation of the multi-stage pump.

[0029] In a preferred embodiment, vibration signals of the middle section structure of the target multistage pump are monitored during operation to obtain vibration parameters, and operation time parameters and water quality parameters of the middle section structure are collected to predict flow path scaling and obtain predicted scaling parameters, including: vibration signals of the middle section structure of the target multistage pump are monitored during operation to obtain vibration parameters; during the operation of the target multistage pump, water quality parameters are regularly monitored to obtain a set of water quality parameters, and water quality parameters are calculated to obtain the water quality parameters; the accumulated operation time of the middle section structure is collected to obtain operation time parameters; and flow path scaling is predicted based on the vibration parameters, water quality parameters and operation time parameters to obtain predicted scaling parameters.

[0030] Preferably, the vibration signals of the multi-stage pump during operation are monitored in real time by means of a high-precision vibration sensor, so as to capture the tiny vibrations of the pump body caused by changes in the internal flow state or wear of components, and then obtain vibration parameters. These vibration parameters can reflect the operating stability of the pump and the health status of the internal structure.

[0031] At the same time, water quality parameters are collected regularly during the operation of the target multi-stage pump, including water hardness, pH value, oxygen content, and impurity content, to form a water quality parameter set. Through the comprehensive analysis of these water quality parameters, we can understand the impact of water quality on the scaling tendency of the pump's internal flow channel. For example, high-hardness water is more likely to form mineral deposits on the inner wall of the flow channel, thereby accelerating the scaling process.

[0032] In addition, record the cumulative operating time of the multistage pump as an important indicator to assess the pump's aging and operating history. The longer the operating time, the higher the risk of wear and scaling of the pump's internal components.

[0033] The prediction model takes vibration parameters, water quality parameters and running time parameters as input to predict flow channel scaling. The prediction model is trained with a large amount of historical data and can accurately identify the complex relationship between various parameters and scaling conditions, thereby outputting predicted scaling parameters. These predicted scaling parameters include not only the possibility and degree of scaling, but may also include information such as the type and distribution of scaling, providing a scientific basis for subsequent maintenance and cleaning work. However, the scaling parameters in this solution are specifically the scale size, that is, the scale area and thickness. If the prediction results show that a certain section of the flow channel may be severely scaled in the near future, a cleaning plan can be arranged in advance to avoid the adverse effects of scaling on pump performance.

[0034] In a preferred embodiment, flow channel scaling prediction is performed according to the vibration parameters, water quality parameters and running time parameters to obtain predicted scaling parameters, including: collecting a sample vibration parameter set, a sample water quality parameter set, and a sample running time parameter set according to historical maintenance data of the middle section structure of a multi-stage pump of the same model, and collecting scaling sizes of flow channel scaling, and annotating to obtain a sample scaling parameter set; constructing a flow channel scaling predictor based on deep learning; using the sample vibration parameter set, the sample water quality parameter set, the sample running time parameter set and the sample scaling parameter set as supervised training data and test data, and performing supervised training and testing on the flow channel scaling predictor, and completing the training and testing after the accuracy test is qualified; inputting the vibration parameters, water quality parameters and running time parameters into the flow channel scaling predictor, and predicting the output to obtain the predicted scaling parameters.

[0035] In detail, historical maintenance data of the middle section structure of multi-stage pumps of the same model are collected. These data include sample vibration parameter sets, sample water quality parameter sets, and sample running time parameter sets. At the same time, the actual scaling size of the flow channel scaling of these pumps during maintenance is collected, and the sample scaling parameter set is obtained based on the annotation. These sample data form the basis for predictive model training.

[0036] Subsequently, a flow channel scaling predictor was constructed using deep learning technology. This predictor is able to learn and identify the complex relationship between vibration parameters, water quality parameters, operating time parameters and scaling parameters. In order to train this predictor, the collected sample data is divided into supervised training data and test data, and the predictor is supervised trained and tested. During the training process, the predictor continuously adjusts its internal parameters to minimize the error between the predicted results and the actual scaling parameters. After multiple iterations of training, the training and testing process is completed when the accuracy of the predictor on the test data reaches a satisfactory level.

[0037] In actual use, the vibration parameters, water quality parameters and operating time parameters monitored in real time during the operation of the target multi-stage pump are input into the trained flow channel scaling predictor. The predictor quickly calculates the predicted scaling parameters based on these input parameters, which include the possible size information of the scaling. For example, if the prediction results show that a certain section of the flow channel may have a large scale in the future, a maintenance plan can be arranged in advance to clean or repair the pump to avoid the adverse effects of scaling on pump performance. This method not only improves the accuracy of the prediction, but also greatly improves the efficiency and reliability of multi-stage pump maintenance.

[0038] In a preferred embodiment, an image of the flow path in the middle section structure is collected through an industrial endoscope, and according to the predicted scaling parameters, the image of the flow path is respectively input into a plurality of scaling parameter classifiers to identify and obtain the actual scaling parameters, including: collecting an image of the flow path in the middle section structure through an industrial endoscope; according to the predicted scaling parameters, screening and obtaining N scaling parameter levels closest to the predicted scaling parameters, wherein N is a positive integer; according to the error amplitude between the N scaling parameter levels and the predicted scaling parameters, calculating and obtaining the number of N classification paths, and selecting the N scaling parameter levels; N scaling parameter classifiers corresponding to the scaling parameter levels, the flow channel images are respectively input into scaling parameter classification paths of N classification paths in the N scaling parameter classifiers, and N scaling parameter classification result sets are obtained as output, wherein each scaling parameter classifier includes multiple scaling parameter classification paths, and each scaling parameter classification result includes yes or no; according to the N scaling parameter classification result sets, the proportion of scaling parameter classification results of yes is calculated to obtain N scaling parameter level probabilities; the scaling parameter level with the largest scaling parameter level probability is output to obtain the actual scaling parameter.

[0039] Specifically, an industrial endoscope is used to collect clear images of the flow channel inside the middle structure of the multi-stage pump. These images intuitively show the scaling condition of the inner wall of the flow channel. Since scaling prediction may have certain errors, in order to more accurately identify the actual scaling parameters, multiple scaling parameter levels are pre-set and corresponding scaling parameter classifiers are constructed. Each classifier contains multiple scaling parameter classification paths, which are constructed based on deep learning algorithms and can perform binary classification judgments on the input flow channel images.

[0040] After obtaining the predicted scaling parameters, the N scaling parameter levels closest to the predicted scaling parameters (for example, N is set to 5) are selected as candidate actual scaling parameter levels. Next, the number of classification paths is calculated based on the error margin between each candidate scaling parameter level and the predicted scaling parameter. An example is now given to illustrate this calculation process. The error margin between the scaling parameter level and the predicted scaling parameter is the absolute difference divided by the predicted scaling parameter. If the error margin is 10%, 1-10%=90% is used, and then 90% is multiplied by the number of scaling parameter classification paths in the corresponding scaling parameter classifier (for example, 10), and the number of classification paths is 9. Subsequently, the flow channel images are input into the N scaling parameter classifiers respectively, and the randomly selected scaling parameter classification paths are classified into two categories according to the calculated number of classification paths. Each classification path will output a scaling parameter classification result, that is, to determine whether the scaling parameter in the flow channel image is the scaling parameter level corresponding to the classifier. In this way, a set of N scaling parameter classification results is obtained.

[0041] In order to determine the most likely actual scaling parameter level, the proportion of each scaling parameter level judged as "yes" in N classification results is calculated, that is, the scaling parameter level probability. For example, if a scaling parameter level is judged as "yes" 5 times in 9 classification paths, then its scaling parameter level probability is 5 / 9. Finally, the scaling parameter level with the highest scaling parameter level probability is output as the actual scaling parameter. This method selects the classifier corresponding to the close scaling parameter level, and configures the number of classification paths for binary classification according to the error amplitude calculation, and selects the most likely scaling parameter based on probability, which not only improves the detection and identification efficiency, but also reduces the waste of computing power.

[0042] For example, suppose that the scaling parameter of a certain section of flow channel is predicted to be moderate, but there may be a certain error in the prediction. Therefore, five scaling parameter classifiers corresponding to the five scaling parameter levels closest to the moderate level (such as slight, mild, moderate, severe, and serious) are selected. The number of classification paths required in each classifier is determined by calculating the error margin, and the flow channel image is input into these classification paths for binary classification. Finally, the probability of each scaling parameter level is calculated based on the classification results, and the scaling parameter level with the highest probability is selected as the actual scaling parameter. This can provide a more accurate understanding of the scaling situation of the flow channel and provide strong support for subsequent maintenance and cleaning work.

[0043] In a preferred embodiment, according to the error amplitudes between N scaling parameter levels and the predicted scaling parameters, the number of N classification paths is calculated, N scaling parameter classifiers corresponding to the N scaling parameter levels are selected, the flow channel images are respectively input into the scaling parameter classification paths of the N classification paths in the N scaling parameter classifiers, and N scaling parameter classification result sets are obtained by output, including: respectively calculating the error amplitudes between the N scaling parameter levels and the predicted scaling parameters, respectively subtracting the N error amplitudes from 1, multiplying the number of scaling parameter classification paths in each scaling parameter classifier and rounding, to obtain the number of N classification paths; selecting the N scaling parameter levels for the N scaling parameter classification paths; N scaling parameter classifiers corresponding to the plurality of scaling parameter levels are pre-trained, each scaling parameter classifier comprises a plurality of scaling parameter classification paths constructed and trained based on a convolutional neural network, the training data comprises a sample flow channel image and a sample scaling parameter classification result, each sample scaling parameter classification result comprises a binary classification result of whether the scaling parameter in the sample flow channel image is a scaling parameter level corresponding to the scaling parameter classifier; according to the number of the N classification paths, scaling parameter classification paths of the number of the N classification paths are randomly selected in the N scaling parameter classifiers, the flow channel image is input, and a set of N scaling parameter classification results are obtained as output.

[0044] Furthermore, for the predicted scaling parameters, the N scaling parameter levels closest to it are screened out, and the error margins between these N scaling parameter levels and the predicted scaling parameters are calculated respectively. Specifically, the result is multiplied by the number of scaling parameter classification paths preset in each scaling parameter classifier by 1 minus each error margin, and the number of N classification paths is rounded. These scaling parameter classifiers are pre-trained with a large amount of sample data. Each classifier is built based on a convolutional neural network and contains multiple scaling parameter classification paths. The training data includes sample flow channel images and corresponding sample scaling parameter classification results, wherein each sample scaling parameter classification result clearly marks whether the scaling parameter in the sample flow channel image is the scaling parameter level corresponding to the classifier, that is, a binary classification is performed.

[0045] After obtaining the number of N classification paths, for each fouling parameter level, the corresponding number of fouling parameter classification paths are randomly selected in its corresponding fouling parameter classifier. Then, the collected runner images are input into these selected classification paths, and each classification path outputs a fouling parameter classification result based on the image features, that is, to determine whether the fouling parameter in the runner image is the fouling parameter level corresponding to this classifier. Finally, the output results of all classification paths are summarized to form N sets of fouling parameter classification results. These result sets provide an important basis for determining the actual fouling parameter. For example, if a certain fouling parameter level is judged as "yes" in multiple classification paths, then this fouling parameter level may be the actual fouling parameter. Through this method, the fouling situation in the runner can be more accurately identified, providing strong support for the maintenance and cleaning work of the multistage pump.

[0046] In a preferred embodiment, according to the actual fouling parameter, a diffuser deformation prediction is carried out to obtain a predicted diffuser deformation parameter, including: collecting a sample fouling parameter set according to the historical maintenance data of the middle section structure of the same type of multistage pump, and collecting the deformation angles of the diffuser under different sample fouling parameters, which are marked as diffuser deformation parameters to obtain a sample diffuser deformation parameter set; constructing a diffuser deformation predictor based on deep learning; using the sample fouling parameter set and the sample diffuser deformation parameter set as supervised training data and test data to perform supervised training and testing on the diffuser deformation predictor, and completing the training and testing after passing the accuracy test; inputting the actual fouling parameter into the diffuser deformation predictor, and predicting and outputting to obtain the predicted diffuser deformation parameter.

[0047] Exemplarily, the historical maintenance data of the middle section structure of the same type of multistage pump are collected, and these data contain a sample fouling parameter set, that is, the fouling situations of different pumps under different operating conditions. At the same time, the deformation angles of the diffuser of these pumps under the corresponding fouling parameters are collected, and these angles are marked as diffuser deformation parameters, thus forming a sample diffuser deformation parameter set. These sample data provide a basis for constructing the prediction model.

[0048] Subsequently, a diffuser deformation predictor is constructed using deep learning technology, and this predictor can learn and identify the complex relationship between the fouling parameter and the diffuser deformation parameter. To train this predictor, the collected sample fouling parameter set and the sample diffuser deformation parameter set are used as supervised training data and test data. During the training process, the predictor continuously adjusts its internal parameters to minimize the error between the prediction result and the true diffuser deformation parameter. After multiple iterative trainings, when the accuracy of the predictor on the test data reaches a satisfactory level, the training and testing process is completed.

[0049] After that, the actual detected scaling parameters are input into the trained guide vane deformation predictor. The predictor quickly calculates the predicted guide vane deformation parameters based on these input parameters, which usually include information such as the deformation angle and deformation direction of the guide vanes. For example, if the prediction results show that the guide vanes of a pump will deform at a large angle under specific scaling parameters, a maintenance plan can be arranged in advance to make necessary adjustments or repairs to the pump to avoid the adverse effects of guide vane deformation on pump performance. This method not only improves the accuracy of the prediction, but also greatly improves the efficiency and reliability of multi-stage pump maintenance.

[0050] In a preferred embodiment, an industrial endoscope is used to collect a guide vane image in the middle section structure, and according to the predicted guide vane deformation parameter, the guide vane image is respectively input into a plurality of guide vane parameter classifiers to identify and obtain the actual guide vane deformation parameter, including: collecting a guide vane image in the middle section structure through an industrial endoscope; selecting M guide vane deformation parameter levels closest to the predicted guide vane deformation parameter, wherein M is a positive integer; selecting M guide vane parameter classifiers corresponding to the M guide vane deformation parameter levels, wherein a plurality of guide vane parameter classifiers corresponding to a plurality of guide vane deformation parameter levels are pre-trained, and each guide vane parameter classifier includes a plurality of guide vane parameter classifiers constructed and trained based on a convolutional neural network. The guide vane parameter classification result is a binary classification result of whether the guide vane deformation parameter in the sample guide vane image is a guide vane deformation parameter level corresponding to the guide vane parameter classifier; the guide vane image is respectively input into a plurality of guide vane parameter classification paths in the M guide vane parameter classifiers to obtain M guide vane parameter classification result sets, wherein each guide vane parameter classification result includes yes or after; the proportion of the guide vane parameter classification results of yes in the M guide vane parameter classification result sets is respectively calculated to obtain M guide vane deformation parameter level probabilities; the guide vane deformation parameter level with the largest guide vane deformation parameter level probability is output as the actual guide vane deformation parameter.

[0051] Optionally, based on the same principle as the above-mentioned scaling parameter classifier, an industrial endoscope is used to penetrate into the middle structure of the multi-stage pump to collect clear images of the guide vanes, which intuitively show the actual shape of the guide vanes.

[0052] Similarly, since the predicted guide vane deformation parameters may have certain errors, in order to more accurately identify the actual guide vane deformation parameters, the M guide vane deformation parameter levels closest to the predicted guide vane deformation parameters (M is a positive integer and can be the same as N in the scaling parameter classification) are selected as candidates. Then, according to the error amplitude between each candidate guide vane deformation parameter level and the predicted guide vane deformation parameter, a classification path number is calculated. This calculation process is similar to the scaling parameter classification: for example, the error amplitude between the guide vane deformation parameter level and the predicted guide vane deformation parameter is the absolute difference divided by the predicted guide vane deformation parameter. If the error amplitude is a certain proportion (such as 10%), 1 is subtracted from the proportion (such as 1-10%=90%), and the result is multiplied by the number of guide vane parameter classification paths in the corresponding guide vane parameter classifier (the number is pre-set, for example, 10), thereby obtaining the number of classification paths.

[0053] Subsequently, M guide vane parameter classifiers corresponding to the M guide vane deformation parameter levels are selected. These classifiers are obtained by pre-training with a large amount of sample data. Each classifier is constructed based on a convolutional neural network and contains multiple guide vane parameter classification paths. The training data includes sample guide vane images and corresponding sample guide vane parameter classification results, wherein each sample guide vane parameter classification result clearly marks whether the guide vane deformation parameter in the sample guide vane image is the guide vane deformation parameter level corresponding to the classifier, that is, a binary classification is performed. Afterwards, the collected guide vane images are respectively input into the multiple guide vane parameter classification paths in the M guide vane parameter classifiers, and each classification path will output a guide vane parameter classification result based on the image features, that is, it is determined whether the guide vane deformation parameter in the guide vane image is the guide vane deformation parameter level corresponding to the classifier. In this way, a set of M guide vane parameter classification results is obtained.

[0054] In order to determine the most likely actual guide vane deformation parameter level, the proportion of guide vane parameter classification results that are "yes" in the M guide vane parameter classification result sets, that is, the guide vane deformation parameter level probability, is calculated respectively. For example, if a guide vane deformation parameter level has the highest proportion of being judged as "yes" in multiple classification paths, then the corresponding guide vane deformation parameter level probability is the largest. Finally, the guide vane deformation parameter level with the highest guide vane deformation parameter level probability is output as the actual guide vane deformation parameter. This method selects the classifier corresponding to the close guide vane deformation parameter level, and configures the number of classification paths for binary classification according to the error amplitude calculation, and selects the most likely guide vane deformation parameter based on probability, which not only improves the detection and recognition efficiency, but also reduces the waste of computing power.

[0055] For example, assume that the predicted deformation parameter of a certain pump guide vane is medium-level bending, but there may be a certain error in the prediction. Then, three guide vane parameter classifiers corresponding to the three guide vane deformation parameter levels closest to medium-level bending (such as slight bending, medium bending, and severe bending) are selected. The number of classification paths to be used within each classifier is determined by calculating the error margin, and the guide vane image is input into these classification paths for binary classification. Finally, the probability of each guide vane deformation parameter level is calculated based on the classification results, and the guide vane deformation parameter level with the highest probability is selected as the actual guide vane deformation parameter. In this way, the deformation situation of the guide vane can be understood more accurately, providing strong support for subsequent maintenance and adjustment work.

[0056] In a preferred embodiment, by combining the actual scaling parameter and the actual guide vane deformation parameter, a quality inspection result of the middle section structure is obtained, including: integrating the actual scaling parameter and the actual guide vane deformation parameter as the quality inspection result of the middle section structure of the target multistage pump.

[0057] Specifically, integrating the actual scaling parameter and the actual guide vane deformation parameter to form the quality inspection result of the middle section structure of the target multistage pump. This result not only reflects the scaling situation of the flow channel but also reveals the deformation degree of the guide vane, providing comprehensive and accurate data support for the maintenance and upkeep of the multistage pump. For example, if the actual scaling parameter shows severe scaling in the flow channel and the actual guide vane deformation parameter indicates significant deformation of the guide vane, it can be determined that there is a problem with the quality of the middle section structure of the multistage pump, and timely cleaning and maintenance are required to avoid adverse effects on the performance and lifespan of the pump.

[0058] A method for inspecting the quality of the middle section structure of a multistage pump provided by an embodiment of the present invention has at least the following technical effects: 1. By fusing multi-dimensional data such as vibration parameters, water quality parameters, and operation time parameters, a deep learning-based flow channel scaling predictor is constructed, realizing accurate prediction of the flow channel scaling situation. The fusion of such multi-dimensional parameters not only improves the accuracy of the prediction but also makes the prediction results more comprehensive and reliable. At the same time, based on the predicted scaling parameter, further prediction of the guide vane deformation is carried out, forming a complete prediction chain from scaling to guide vane deformation, providing a more in-depth and detailed analysis for the quality inspection of the middle section structure of the multistage pump.

[0059] 2. When identifying the actual scaling parameter and the actual guide vane deformation parameter, an intelligent classification and recognition technology is adopted. By screening multiple parameter levels closest to the predicted parameter and calculating the corresponding number of classification paths, the image is input into multiple classifiers for binary classification judgment. Finally, the most likely parameter level is determined through probability analysis. This method combining intelligent classification and recognition with probability analysis effectively improves the accuracy and efficiency of the recognition, reducing the possibility of misjudgment and missed judgment.

[0060] 3. Combining industrial endoscope technology with deep learning enables intuitive observation and accurate analysis of the internal conditions of the middle section of the multi-stage pump. The clear images of the flow channel and guide vanes collected by the endoscope provide rich training data and test data for the deep learning model. At the same time, the processing results of the deep learning model provide strong support for guiding further observation and analysis by the endoscope. This organic combination makes the quality inspection of the middle section of the multi-stage pump more intuitive, accurate and efficient.

[0061] Embodiment 2: like Figure 2 As shown, based on the same inventive concept as the method for detecting the quality of the middle section structure of a multistage pump provided in the first embodiment, the embodiment of the present invention further provides a system for detecting the quality of the middle section structure of a multistage pump, the system comprising: The vibration monitoring module 11 is used to monitor the vibration signal of the middle section structure of the target multi-stage pump during operation, obtain vibration parameters, collect the operation time parameters and water quality parameters of the middle section structure, predict the flow channel scaling, and obtain the predicted scaling parameters.

[0062] The image acquisition module 12 is used to collect the flow channel image in the middle section structure through an industrial endoscope, and input the flow channel image into multiple scaling parameter classifiers according to the predicted scaling parameters to identify and obtain actual scaling parameters, wherein each scaling parameter classifier includes multiple scaling parameter classification paths.

[0063] The classification and recognition module 13 is used to predict the guide vane deformation according to the actual scaling parameters and obtain the predicted guide vane deformation parameters. The guide vane image in the middle section structure is collected through an industrial endoscope. According to the predicted guide vane deformation parameters, the guide vane image is input into a plurality of guide vane parameter classifiers respectively to identify and obtain the actual guide vane deformation parameters.

[0064] The result acquisition module 14 is used to obtain the quality detection result of the mid-section structure by combining the actual scaling parameter and the actual guide vane deformation parameter.

[0065] Furthermore, the vibration monitoring module 11 is further configured to perform the following steps: The vibration signal of the middle section structure of the target multi-stage pump is monitored during operation to obtain vibration parameters; during the operation of the target multi-stage pump, water quality parameters are regularly monitored to obtain a water quality parameter set, and the water quality parameters are calculated to obtain the water quality parameters; the cumulative operation time of the middle section structure is collected to obtain the operation time parameters; and flow channel scaling is predicted based on the vibration parameters, water quality parameters and operation time parameters to obtain predicted scaling parameters.

[0066] Furthermore, the vibration monitoring module 11 is further configured to perform the following steps: According to the historical maintenance data of the middle section structure of the same model of multi-stage pump, a sample vibration parameter set, a sample water quality parameter set, and a sample running time parameter set are collected, and the scaling size of the flow channel scaling is collected and annotated to obtain the sample scaling parameter set; based on deep learning, a flow channel scaling predictor is constructed; the sample vibration parameter set, the sample water quality parameter set, the sample running time parameter set, and the sample scaling parameter set are used as supervised training data and test data, and the flow channel scaling predictor is supervised trained and tested, and the training and testing are completed after the accuracy test is qualified; the vibration parameters, water quality parameters, and running time parameters are input into the flow channel scaling predictor, and the predicted scaling parameters are obtained by prediction output.

[0067] Furthermore, the image acquisition module 12 is further configured to perform the following steps: The flow channel image in the middle section structure is collected through an industrial endoscope; according to the predicted scaling parameter, N scaling parameter levels closest to the predicted scaling parameter are screened, wherein N is a positive integer; according to the error amplitude between the N scaling parameter levels and the predicted scaling parameter, the number of N classification paths is calculated, and N scaling parameter classifiers corresponding to the N scaling parameter levels are selected, and the flow channel image is respectively input into the scaling parameter classification paths of the N classification paths in the N scaling parameter classifiers, and N scaling parameter classification result sets are obtained as output, wherein each scaling parameter classifier includes multiple scaling parameter classification paths, and each scaling parameter classification result includes yes or no; according to the N scaling parameter classification result sets, the proportion of scaling parameter classification results that are yes is calculated to obtain N scaling parameter level probabilities; the scaling parameter level with the largest scaling parameter level probability is output to obtain the actual scaling parameter.

[0068] Furthermore, the image acquisition module 12 is further configured to perform the following steps: Calculate the error margins between N scaling parameter levels and the predicted scaling parameter respectively, and use 1 minus the N error margins, multiply by the number of scaling parameter classification paths in each scaling parameter classifier and round up to obtain the number of N classification paths; select N scaling parameter classifiers corresponding to the N scaling parameter levels, wherein multiple scaling parameter classifiers corresponding to multiple scaling parameter levels are pre-trained, each scaling parameter classifier includes multiple scaling parameter classification paths constructed and trained based on a convolutional neural network, the training data includes a sample flow channel image and a sample scaling parameter classification result, and each sample scaling parameter classification result includes a binary classification result of whether the scaling parameter in the sample flow channel image is a scaling parameter level corresponding to the scaling parameter classifier; according to the number of N classification paths, randomly select scaling parameter classification paths of the number of N classification paths in the N scaling parameter classifiers, input the flow channel image, and output to obtain a set of N scaling parameter classification results.

[0069] Furthermore, the classification identification module 13 is further configured to perform the following steps: According to the historical maintenance data of the middle section structure of the same model of multi-stage pump, a set of sample scaling parameters is collected, and the deformation angles of the guide vanes under different sample scaling parameters are collected and marked as guide vane deformation parameters to obtain a set of sample guide vane deformation parameters; based on deep learning, a guide vane deformation predictor is constructed; the sample scaling parameter set and the sample guide vane deformation parameter set are used as supervised training data and test data to perform supervised training and testing on the guide vane deformation predictor, and the training and testing are completed after the accuracy test is qualified; the actual scaling parameters are input into the guide vane deformation predictor, and the predicted guide vane deformation parameters are obtained by prediction output.

[0070] Furthermore, the classification identification module 13 is further configured to perform the following steps: The guide vane image in the middle section structure is collected through an industrial endoscope; M guide vane deformation parameter levels closest to the predicted guide vane deformation parameter are selected, wherein M is a positive integer; M guide vane parameter classifiers corresponding to the M guide vane deformation parameter levels are selected, wherein a plurality of guide vane parameter classifiers corresponding to a plurality of guide vane deformation parameter levels are pre-trained, each guide vane parameter classifier includes a plurality of guide vane parameter classification paths constructed and trained based on a convolutional neural network, the training data includes a sample guide vane image and a sample guide vane parameter classification result, and each sample guide vane parameter classification result includes a sample guide vane parameter classification path. Whether the guide vane deformation parameter in the guide vane image is a binary classification result of the guide vane deformation parameter level corresponding to the guide vane parameter classifier; the guide vane image is respectively input into the multiple guide vane parameter classification paths in the M guide vane parameter classifiers to obtain M guide vane parameter classification result sets, wherein each guide vane parameter classification result includes yes or after; respectively calculate the proportion of the guide vane parameter classification results of yes in the M guide vane parameter classification result sets to obtain M guide vane deformation parameter level probabilities; output the guide vane deformation parameter level with the largest guide vane deformation parameter level probability as the actual guide vane deformation parameter.

[0071] Furthermore, the result acquisition module 14 is further configured to perform the following steps: The actual scaling parameters and the actual guide vane deformation parameters are integrated as the quality inspection result of the middle section structure of the target multi-stage pump.

[0072] Through the above-mentioned detailed description of a method for detecting the quality of the middle section structure of a multi-stage pump, those skilled in the art can clearly know a system for detecting the quality of the middle section structure of a multi-stage pump in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0073] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting the quality of a multistage pump mid-section structure, characterized in that: The method comprises: Monitor the vibration signal of the middle section structure of the target multi-stage pump during operation to obtain vibration parameters, collect the operation time parameters and water quality parameters of the middle section structure, predict the flow channel scaling, and obtain the predicted scaling parameters; The flow channel image in the middle section structure is collected by an industrial endoscope, and the flow channel image is respectively input into a plurality of scaling parameter classifiers according to the predicted scaling parameters to identify and obtain actual scaling parameters, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths; According to the actual scaling parameters, guide vane deformation prediction is performed to obtain predicted guide vane deformation parameters, and guide vane images in the middle section structure are collected through an industrial endoscope. According to the predicted guide vane deformation parameters, the guide vane images are respectively input into a plurality of guide vane parameter classifiers to identify and obtain actual guide vane deformation parameters; The quality inspection result of the mid-section structure is obtained by combining the actual scaling parameters and the actual guide vane deformation parameters.

2. The method for detecting the quality of the middle section structure of a multistage pump according to claim 1, characterized in that: The vibration signal of the middle section structure of the target multi-stage pump is monitored during operation to obtain vibration parameters. The operation time parameters and water quality parameters of the middle section structure are collected to predict flow channel scaling and obtain predicted scaling parameters, including: Monitor the vibration signal of the middle section structure of the target multi-stage pump during operation to obtain vibration parameters; During the operation of the target multi-stage pump, water quality parameters are regularly monitored, a water quality parameter set is obtained, and water quality parameters are calculated; Collect the accumulated running time of the middle section structure and obtain the running time parameters; According to the vibration parameters, water quality parameters and operation time parameters, flow channel scaling prediction is performed to obtain predicted scaling parameters.

3. The method for detecting the quality of the middle section structure of a multistage pump according to claim 2, characterized in that: According to the vibration parameters, water quality parameters and operation time parameters, flow channel scaling prediction is performed to obtain predicted scaling parameters, including: According to the historical maintenance data of the middle section structure of the same model multi-stage pump, collect the sample vibration parameter set, sample water quality parameter set, sample running time parameter set, and collect the scaling size of the flow channel scaling, and mark the sample scaling parameter set; Based on deep learning, a flow channel scaling predictor is constructed; Using the sample vibration parameter set, the sample water quality parameter set, the sample running time parameter set and the sample scaling parameter set as supervised training data and test data, the flow channel scaling predictor is supervised trained and tested, and the training and testing are completed after the accuracy test is qualified; The vibration parameter, water quality parameter and operation time parameter are input into the flow channel scaling predictor, and the predicted scaling parameter is obtained by prediction output.

4. The method for detecting the quality of the middle section structure of a multistage pump according to claim 1, characterized in that: The flow channel image in the middle section structure is collected by an industrial endoscope, and the flow channel image is input into a plurality of scaling parameter classifiers according to the predicted scaling parameters to identify and obtain the actual scaling parameters, including: Using an industrial endoscope, collecting a flow channel image in the middle section structure; According to the predicted scaling parameter, N scaling parameter levels closest to the predicted scaling parameter are screened and obtained, wherein N is a positive integer; According to the error amplitudes between N scaling parameter levels and the predicted scaling parameters, the number of N classification paths is calculated, N scaling parameter classifiers corresponding to the N scaling parameter levels are selected, the flow channel images are respectively input into the scaling parameter classification paths of the N classification paths in the N scaling parameter classifiers, and N scaling parameter classification result sets are obtained as output, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths, and each scaling parameter classification result includes yes or no; According to the N scaling parameter classification result sets, the proportion of scaling parameter classification results that are yes is calculated to obtain N scaling parameter level probabilities; The scaling parameter level with the maximum scaling parameter level probability is output to obtain the actual scaling parameter.

5. The method for detecting the quality of the middle section structure of a multistage pump according to claim 4, characterized in that: According to the error amplitude between N scaling parameter levels and the predicted scaling parameter, the number of N classification paths is calculated, N scaling parameter classifiers corresponding to the N scaling parameter levels are selected, the flow channel images are respectively input into the scaling parameter classification paths of the N classification paths in the N scaling parameter classifiers, and N scaling parameter classification result sets are obtained by output, including: Calculate the error margins between N scaling parameter levels and the predicted scaling parameter respectively, and use 1 minus the N error margins, multiply by the number of scaling parameter classification paths in each scaling parameter classifier and round up to obtain the number of N classification paths; Selecting N scaling parameter classifiers corresponding to the N scaling parameter levels, wherein a plurality of scaling parameter classifiers corresponding to a plurality of scaling parameter levels are pre-trained, each scaling parameter classifier includes a plurality of scaling parameter classification paths constructed and trained based on a convolutional neural network, the training data includes a sample flow channel image and a sample scaling parameter classification result, and each sample scaling parameter classification result includes a binary classification result of whether a scaling parameter in the sample flow channel image is a scaling parameter level corresponding to the scaling parameter classifier; According to the number of N classification paths, N scaling parameter classification paths of the number of classification paths are randomly selected in the N scaling parameter classifiers, the flow channel image is input, and N scaling parameter classification result sets are obtained as output.

6. The method for detecting the quality of the middle section structure of a multistage pump according to claim 1, characterized in that: According to the actual scaling parameters, guide vane deformation prediction is performed to obtain predicted guide vane deformation parameters, including: According to the historical maintenance data of the middle section structure of the same model multi-stage pump, a set of sample scaling parameters is collected, and the deformation angles of the guide vanes under different sample scaling parameters are collected and marked as guide vane deformation parameters to obtain a set of sample guide vane deformation parameters; Based on deep learning, a guide vane deformation predictor is constructed; Using the sample scaling parameter set and the sample guide vane deformation parameter set as supervised training data and test data, supervised training and testing are performed on the guide vane deformation predictor, and the training and testing are completed after the accuracy test is qualified; The actual fouling parameters are input into the guide vane deformation predictor, and the prediction output is used to obtain the predicted guide vane deformation parameters.

7. The method for detecting the quality of the middle section structure of a multistage pump according to claim 1, characterized in that: The guide vane image in the middle section structure is collected through an industrial endoscope, and according to the predicted guide vane deformation parameter, the guide vane image is respectively input into a plurality of guide vane parameter classifiers to identify and obtain the actual guide vane deformation parameter, including: Using an industrial endoscope, collecting an image of the guide vanes in the middle section structure; Selecting M guide vane deformation parameter levels closest to the predicted guide vane deformation parameter, where M is a positive integer; Selecting M guide vane parameter classifiers corresponding to the M guide vane deformation parameter levels, wherein a plurality of guide vane parameter classifiers corresponding to a plurality of guide vane deformation parameter levels are pre-trained, each guide vane parameter classifier includes a plurality of guide vane parameter classification paths constructed and trained based on a convolutional neural network, the training data includes a sample guide vane image and a sample guide vane parameter classification result, and each sample guide vane parameter classification result includes a binary classification result of whether a guide vane deformation parameter in the sample guide vane image is a guide vane deformation parameter level corresponding to the guide vane parameter classifier; Input the guide vane images into a plurality of guide vane parameter classification paths in the M guide vane parameter classifiers respectively to obtain M guide vane parameter classification result sets, wherein each guide vane parameter classification result includes yes or no; Calculate the proportion of the guide vane parameter classification results that are "yes" in the M guide vane parameter classification result sets respectively, and obtain the M guide vane deformation parameter level probabilities; The guide vane deformation parameter level with the highest probability is output as the actual guide vane deformation parameter.

8. The method for detecting the quality of the middle section structure of a multistage pump according to claim 1, characterized in that: In combination with the actual scaling parameters and the actual guide vane deformation parameters, the quality inspection results of the mid-section structure are obtained, including: The actual scaling parameters and the actual guide vane deformation parameters are integrated as the quality inspection result of the middle section structure of the target multi-stage pump.

9. A multi-stage pump mid-section structure quality detection system, characterized in that: A method for detecting the quality of a multistage pump mid-section structure according to any one of claims 1 to 8, the system comprising: The vibration monitoring module is used to monitor the vibration signal of the middle section structure of the target multi-stage pump during operation, obtain vibration parameters, collect the operation time parameters and water quality parameters of the middle section structure, predict the flow channel scaling, and obtain the predicted scaling parameters; An image acquisition module is used to acquire the flow channel image in the middle section structure through an industrial endoscope, and input the flow channel image into a plurality of scaling parameter classifiers respectively according to the predicted scaling parameters to identify and obtain the actual scaling parameters, wherein each scaling parameter classifier includes a plurality of scaling parameter classification paths; a classification and recognition module, configured to predict guide vane deformation according to the actual scaling parameters, obtain predicted guide vane deformation parameters, collect guide vane images in the middle section structure through an industrial endoscope, input the guide vane images into a plurality of guide vane parameter classifiers according to the predicted guide vane deformation parameters, and identify and obtain actual guide vane deformation parameters; The result acquisition module is used to obtain the quality detection result of the mid-section structure by combining the actual scaling parameter and the actual guide vane deformation parameter.

Images