Plunger pump fault online monitoring method and system based on multi-modal data

Abstract

The invention provides a plunger pump fault online monitoring method and system based on multi-modal data, and the method comprises the steps: constructing a training set, training a fault classification model, and obtaining a trained fault classification model; multi-modal data in the operation process of the plunger pump are obtained in real time and preprocessed; according to the preprocessed multi-modal data, constructing a first feature vector, inputting the trained fault classification model, and outputting a first prediction result; if the prediction result is in the abnormal state, analyzing the first feature vector when the prediction result is in the abnormal state, and constructing an incremental training set; repeatedly training the fault classification model by using the incremental training set to obtain a new model; and obtaining a first feature vector at the current moment, and inputting the first feature vector into the new model to obtain a second prediction result. According to the method, an uncertainty evaluation mechanism is introduced, quantitative analysis is carried out on uncertainty indexes of different modal data, high-quality incremental samples are further screened, and the model updating reliability is enhanced.

Description

[0001] The present application relates to the technical field of fault monitoring, and in particular to a method and system for online monitoring of plunger pump faults based on multimodal data. Background Art

[0002] In modern industrial environments, plunger pumps are critical equipment used in a wide range of fields, including the petroleum, chemical, aviation, and manufacturing industries. The proper operation of these devices is crucial for ensuring production efficiency and safety. However, due to long-term use or harsh operating conditions, plunger pumps can experience various failures, such as mechanical wear, seal aging, and pressure fluctuations. Therefore, developing an effective online monitoring method to monitor the health of plunger pumps in real time and predict potential failures has become a crucial step in ensuring safe equipment operation.

[0003] Traditional plunger pump fault diagnosis relies primarily on expert experience and periodic maintenance inspections. This approach is not only time-consuming and labor-intensive, but also often fails to detect signs of failure early, leading to increased repair costs and extended downtime. In recent years, with the development of machine learning technology, fault classification models based on fixed training sets have gradually been applied to plunger pump fault monitoring. These models can identify and classify specific types of faults by analyzing characteristic parameters in historical data (such as the wavelet energy ratio of vibration signals and the main frequency energy density of pressure signals). Although this approach significantly improves fault detection accuracy, it has a significant limitation: when new fault types emerge, the model struggles to accurately identify them due to a lack of corresponding training samples, thus affecting its long-term effectiveness and reliability.

[0004] Specifically, fault classification models trained with a fixed training set are typically only applicable to known fault modes. Once new, unforeseen fault types emerge during actual operation, the model may become ineffective. This is because existing models cannot dynamically adapt to new fault characteristics brought about by environmental changes and equipment degradation. Furthermore, traditional methods lack an understanding of the complex relationships between multimodal data (such as vibration, pressure, temperature, and acoustic emission signals), limiting their application in complex operating conditions. In this context, effectively integrating incremental learning mechanisms so that fault classification models can continuously learn and adapt to new fault types without losing existing knowledge has become a pressing technical challenge. Summary of the Invention

[0005] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] To this end, the first purpose of this application is to propose an online monitoring method for plunger pump faults based on multimodal data to improve the accuracy of effect evaluation.

[0007] The second objective of this application is to propose an online monitoring system for plunger pump faults based on multimodal data.

[0008] To achieve the above objectives, the first embodiment of the present application proposes an online monitoring method for plunger pump faults based on multimodal data, comprising: Step 1: Construct a training set and train the fault classification model to obtain a trained fault classification model; Step 2: Real-time acquisition of multimodal data during the operation of the plunger pump and preprocessing; Step 3: construct a first feature vector based on the preprocessed multimodal data, input the first feature vector into the trained fault classification model, and output a first prediction result; Step 4: If the prediction result is an abnormal state, analyze the first eigenvector when the prediction result is an abnormal state and construct an incremental training set; Step 5: Repeat the training of the fault classification model using the incremental training set to obtain a new model; Step 6: Obtain the first eigenvector at the current moment and input it into the new model to obtain the second prediction result.

[0009] In some implementations, the multimodal data includes: a vibration signal, a pressure signal, a temperature signal, and an acoustic emission signal.

[0010] In some implementations, the first eigenvector is composed of the following characteristic parameters: wavelet energy ratio of the vibration signal, main frequency energy density of the pressure signal, first-order difference mean of the temperature signal, and instantaneous frequency change rate of the acoustic emission signal.

[0011] In some implementations, if the prediction result is an abnormal state, analyzing the first feature vector when the prediction result is an abnormal state and constructing an incremental training set includes the following steps: Step 41: Call the first feature vector predicted to be in an abnormal state and define it as a candidate sample. Determine the consistency index between each candidate sample and the historical samples in the historical fault case library. Preset a first target threshold. Based on the relationship between the first target threshold and the average Mahalanobis distance, determine whether the current candidate sample is a new fault type sample or a repeated sample. If it is a new fault type sample, define it as the first pending sample. If it is a repeated sample, add it to the historical fault case library to update or remove the historical fault case. Step 42: Determine the similarity between each candidate sample and each historical sample in the historical fault case library, preset a second target threshold, and determine whether the current candidate sample is a first quality sample or a second quality sample based on the relationship between the second target threshold and the similarity; if the current candidate sample is a first quality sample, define the candidate sample as a second pending sample; if the current candidate sample is a second quality sample, define the candidate sample as a third pending sample; Step 43: If the candidate sample is defined as both the first pending sample and the second pending sample, an incremental training set is constructed based on the candidate sample.

[0012] In some implementations, constructing an incremental training set based on candidate samples includes the following steps: Step 431: call the pre-processed multimodal data of the candidate sample; Step 432: input the data of each modality into the trained fault classification model separately, and output a second prediction result; Step 433, defining at least three uncertainty indicators for the second prediction result; Step 434: construct a ratio between every two uncertainty indicators, preset a third target threshold, and when 2 / 3 of the ratios are greater than the third target threshold, define the second prediction result of the current modality as the first uncertainty, determine an uncertainty score based on the ratio, and determine an optimization threshold based on the uncertainty score; In step 435 , the candidate sample defined as both the first pending sample and the second pending sample is compared with the optimization threshold. If the candidate sample is smaller than the optimization threshold, the candidate sample is defined as an incremental sample; if the candidate sample is greater than or equal to the optimization threshold, the candidate sample is eliminated.

[0013] In some implementations, if the consistency index between each candidate sample and the historical samples in the historical fault case library is greater than the first target threshold, the current candidate sample is defined as a new fault type sample; if the consistency index between each candidate sample and the historical samples in the historical fault case library is less than or equal to the first target threshold, the current candidate sample is defined as a duplicate sample.

[0014] In some implementations, if the similarity between each candidate sample and each historical sample in the historical fault case library is greater than the second target threshold, the current candidate sample is defined as a first quality sample; if the similarity between each candidate sample and each historical sample in the historical fault case library is less than or equal to the second target threshold, the current candidate sample is defined as a second quality sample.

[0015] To achieve the above objectives, the second embodiment of the present application proposes an online monitoring system for plunger pump faults based on multimodal data, comprising: Fault classification model: Build a training set and train the fault classification model to obtain a trained fault classification model; Data acquisition and preprocessing module: connected to the fault classification model, used to obtain multimodal data during the operation of the plunger pump in real time and perform preprocessing; The first prediction module is connected to the data acquisition and preprocessing module, and is used to construct a first feature vector based on the preprocessed multimodal data, input the trained fault classification model, and output a first prediction result; Incremental training set construction module: connected to the first prediction module, used to analyze the first feature vector whose prediction result is in an abnormal state and construct an incremental training set; Incremental training module: connected to the incremental training set construction module, used to repeatedly train the fault classification model using the incremental training set to obtain a new model; Second prediction module: connected to the incremental training module, used to obtain the first feature vector at the current moment and input it into the new model to obtain the second prediction result. Fault classification model: constructs a training set and trains the fault classification model to obtain a trained fault classification model; Data acquisition and preprocessing module: connected to the fault classification model, used to obtain multimodal data during the operation of the plunger pump in real time and perform preprocessing; The first prediction module is connected to the data acquisition and preprocessing module, and is used to construct a first feature vector based on the preprocessed multimodal data, input the trained fault classification model, and output a first prediction result; Incremental training set construction module: connected to the first prediction module, used to analyze the first feature vector whose prediction result is in an abnormal state and construct an incremental training set; Incremental training module: connected to the incremental training set construction module, used to repeatedly train the fault classification model using the incremental training set to obtain a new model; The second prediction module is connected to the incremental training module, used to obtain the first feature vector at the current moment and input it into the new model to obtain the second prediction result.

[0016] The present application provides an online monitoring method for plunger pump faults based on multimodal data. When the prediction result shows that the system is in an abnormal state, the present invention further conducts an in-depth analysis of the first eigenvector under the abnormal state, and determines whether the current sample belongs to a new fault type or a repeated sample by determining its consistency index and similarity with the samples in the historical fault case library. If it is a new fault type, it will be included in the incremental training set for subsequent model updates and optimizations. In this way, the present invention can not only capture newly emerging fault modes in a timely manner, but also continuously improve the generalization ability and robustness of the fault classification model, ensuring its stable performance during long-term operation. In addition, the present invention also introduces an uncertainty assessment mechanism, which further screens out high-quality incremental samples by quantitatively analyzing the uncertainty indicators of different modal data, thereby enhancing the effectiveness and reliability of model updates. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A flow chart of a method for online monitoring of plunger pump failures based on multimodal data provided in an embodiment of the present application; Figure 2 A structural block diagram of a plunger pump fault online monitoring method system based on multimodal data provided in an embodiment of the present application; Figure 3 A block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0019] The following describes the method, device and equipment for evaluating the effectiveness of rain enhancement operations for intelligently predicting natural rainfall according to an embodiment of the present application with reference to the accompanying drawings.

[0020] Figure 1 A flow chart of a method for online monitoring of plunger pump faults based on multimodal data provided in an embodiment of the present application.

[0021] It should be noted that the executor of the plunger pump fault online monitoring method based on multimodal data in an embodiment of the present application is a plunger pump fault online monitoring system based on multimodal data in an embodiment of the present application. The plunger pump fault online monitoring system based on multimodal data can be configured in an electronic device so that the electronic device can perform the plunger pump fault online monitoring function based on multimodal data.

[0022] like Figure 1 As shown, the plunger pump fault online monitoring method based on multimodal data includes the following steps: Step 1: Build a training set and train the fault classification model to obtain a trained fault classification model.

[0023] The training set refers to the initial or fixed training set, consisting of multimodal sensor data from the plunger pump under various operating conditions. This multimodal data includes vibration, pressure, temperature, and acoustic emission signals. These signals reflect the dynamic characteristics of the equipment in terms of mechanical vibration, hydraulic fluctuations, thermal effects, and material fatigue. Each set of samples corresponds to a time window (e.g., 5 seconds), from which key characteristic parameters are extracted to form a training feature vector. Specifically, the extracted characteristic parameters include, but are not limited to, the wavelet energy ratio of the vibration signal, the main frequency energy density of the pressure signal, the first-order difference mean of the temperature signal, and the instantaneous frequency change rate of the acoustic emission signal.

[0024] Among them, the wavelet energy ratio is obtained by calculating the energy proportion of each scale after performing discrete wavelet transform (DWT) decomposition on the vibration signal; the main frequency energy density is obtained by extracting the energy density of the main frequency component and its neighborhood based on the fast Fourier transform (FFT) result of the pressure signal; the first-order difference mean reflects the temperature change trend by calculating the average absolute value of the temperature difference between adjacent moments; the instantaneous frequency change rate uses the Hilbert-Huang transform (HHT) or short-time Fourier transform (STFT) to obtain the time-frequency distribution of the acoustic emission signal, and further extracts the change slope of its instantaneous frequency.

[0025] To achieve accurate fault classification, each sample must be assigned a clear fault type label. Based on the typical failure modes common in plunger pumps, the following fault categories are defined and assigned integer labels: normal state (0), plunger wear (1), valve plate leakage (2), suction valve stuck (3), discharge valve failure (4), and bearing failure (5). Labels are obtained through manual calibration based on historical maintenance records and alarm log information.

[0026] This embodiment uses a deep neural network as the underlying architecture to build a fault classification model. Considering the need for multimodal data fusion, a multi-channel convolutional neural network (CNN) combined with fully connected layers is used to process data streams from different modalities. The model's input layer receives the training feature vector, which then passes through a feature extraction module consisting of multiple convolutional layers, batch normalization layers, and ReLU activation functions. A global average pooling layer then compresses the spatial dimensions. Feature vectors from different modalities are then concatenated in a fusion layer, and finally, two fully connected hidden layers output the probability distribution of each fault category. The output layer uses a Softmax activation function to complete the final classification task.

[0027] During model training, the cross-entropy loss function is used to calculate the error. Training stops when the error is sufficiently small or the maximum number of iterations is reached. The maximum number of iterations is 50.

[0028] After training, the model's performance is evaluated on an independent test set to ensure its generalization ability to unknown samples. If the test set accuracy reaches above 90%, the model is considered to have good initial classification capabilities and a trained fault classification model is obtained.

[0029] Step 2: Acquire multimodal data during the operation of the plunger pump in real time and perform preprocessing.

[0030] First, multimodal data from the operation of the plunger pump is acquired in real time. The data is then denoised using low-pass filtering (to remove high-frequency interference) and wavelet threshold denoising (to retain effective features while suppressing noise). Normalisation is then performed to unify the data from each channel to the [0,1] interval to eliminate the impact of dimensional differences. For data points with missing or outliers, linear interpolation or sliding window mean substitution strategies are used to repair them.

[0031] Step 3: construct a first feature vector based on the preprocessed multimodal data, input the first feature vector into the trained fault classification model, and output a first prediction result.

[0032] The preprocessed signal is split into frames according to a fixed time window (such as 5 seconds), and the corresponding feature parameters are extracted to form the first feature vector that can be used for model input, thereby completing the conversion from the original signal to structured feature data, providing high-quality input for subsequent fault prediction and incremental learning. The trained fault classification model is input and the first prediction result, i.e., the fault type, is output.

[0033] Step 4: If the prediction result is an abnormal state, analyze the first eigenvector when the prediction result is an abnormal state and construct an incremental training set.

[0034] Step 41, call the first feature vector in the abnormal state with the prediction result and define it as a candidate sample, determine the consistency index between each candidate sample and the historical samples in the historical fault case library, preset the first target threshold, and determine whether the current candidate sample is a new fault type sample or a repeated sample based on the relationship between the first target threshold and the average Mahalanobis distance; if it is a new fault type sample, it is defined as the first pending sample; if it is a repeated sample, it is added to the historical fault case library to update or remove the historical fault case.

[0035] If the consistency index between each candidate sample and the historical samples in the historical fault case library is greater than the first target threshold, the current candidate sample is defined as a new fault type sample; if the consistency index between each candidate sample and the historical samples in the historical fault case library is less than or equal to the first target threshold, the current candidate sample is defined as a duplicate sample.

[0036] When the first prediction result output by the fault classification model is an abnormal state, the corresponding first feature vector is extracted and defined as a candidate sample. Subsequently, the set of existing sample feature vectors in the historical fault case library is called to calculate the average Mahalanobis distance between the candidate sample and all historical samples as an indicator to measure its consistency with the existing fault mode, that is, the consistency index. Specifically, the Mahalanobis distance effectively reflects the degree to which the candidate sample deviates from the distribution of historical samples in the feature space by considering the covariance structure between each feature; if the distance is large, it means that the sample is significantly different from the existing fault type and may belong to a new fault type; otherwise, it may be a repeated sample of an existing type. In order to set the judgment basis, this embodiment presets a first target threshold, which is determined based on the statistical characteristics of the historical sample consistency index and is set to the mean of the average Mahalanobis distance of the historical samples plus a standard deviation to ensure a high recognition sensitivity for new samples. If the consistency index of the current candidate sample is greater than the first target threshold, it is determined to be a new fault type sample and defined as the first pending sample; if it is less than or equal to the threshold, it is determined to be a duplicate sample. At this time, it is added to the historical fault case library for update operation, and the decision on whether to remove the old sample is made based on the timestamp or similarity retention strategy, thereby realizing dynamic maintenance and optimization of the case library.

[0037] Step 42: Determine the similarity between each candidate sample and each historical sample in the historical fault case library, preset a second target threshold, and determine whether the current candidate sample is a first quality sample or a second quality sample based on the relationship between the second target threshold and the similarity; if the current candidate sample is a first quality sample, define the candidate sample as a second pending sample; if the current candidate sample is a second quality sample, define the candidate sample as a third pending sample; If the similarity between each candidate sample and each historical sample in the historical fault case library is greater than the second target threshold, the current candidate sample is defined as a first quality sample; if the similarity between each candidate sample and each historical sample in the historical fault case library is less than or equal to the second target threshold, the current candidate sample is defined as a second quality sample.

[0038] In fact, steps 42 and 41 simultaneously analyze the first eigenvector for the first prediction result indicating an abnormal state. When a candidate sample is identified as abnormal, the system extracts its corresponding first eigenvector and compares it one by one with existing samples in the historical fault case library, calculating similarity. By traversing all samples in the historical fault case library, the system obtains similarity values ​​between the candidate sample and each historical sample, and further calculates the mean or the average of the top K highest values ​​as the final comprehensive similarity score for quality assessment.

[0039] Compare the similarity distribution between historical samples in the historical fault case library and other samples, calculate their mean and standard deviation, and set the second target threshold to the mean plus one standard deviation. This method filters out samples that closely match existing fault types, ensuring that only samples with high confidence are considered high-quality samples.

[0040] If the comprehensive similarity between the candidate sample and the samples in the historical fault case library is greater than the second target threshold, the sample is considered to be highly matched with the existing fault mode and has high reliability. It is defined as a first-quality sample and further marked as a second-pending sample. If the comprehensive similarity of the candidate sample is less than or equal to the second target threshold, it is considered that its matching degree with the historical sample is low, and there may be noise interference or incomplete information problems. It is defined as the second quality sample and marked as the third pending sample.

[0041] Among them, the first-quality samples represent high-quality data with high matching accuracy and can be used for subsequent incremental training; while the second-quality samples are considered to be low-quality samples. Although they are still retained in the system for subsequent analysis, they will not directly participate in the model update process and need to be further combined with the uncertainty assessment mechanism to decide whether to eliminate or retain them.

[0042] Step 43: If the candidate sample is defined as both the first pending sample and the second pending sample, an incremental training set is constructed based on the candidate sample.

[0043] Step 431: call the pre-processed multimodal data of the candidate sample.

[0044] Step 432: input the data of each modality into the trained fault classification model separately, and output a second prediction result.

[0045] Step 433: define at least three uncertainty indicators for the second prediction result.

[0046] Uncertainty indicators, including entropy, KL divergence, and mutual information, can reflect the model's confidence in the current sample from different perspectives.

[0047] In step 434, a ratio is constructed between every two uncertainty indicators, and a third target threshold is preset. When 2 / 3 of the ratios are greater than the third target threshold, the second prediction result of the current modality is defined as the first uncertainty, and the uncertainty score is determined based on the ratio, and the optimization threshold is determined based on the uncertainty score.

[0048] First, three uncertainty metrics are calculated for the second prediction results under each modality: entropy (E), KL divergence (KL), and mutual information (MI). Three sets of ratios are then constructed: E / KL, MI / E, and KL / MI. Subsequently, the prediction results for the same modality under historical operating conditions are retrieved, and the historical distribution of the three sets of ratios is statistically analyzed. The mean (μ1, μ2, μ3) and standard deviation (σ1, σ2, σ3) of these ratios are calculated. Based on this, a third target threshold is set as the mean of the corresponding ratio plus one standard deviation (T1, T2, T3). For the current candidate sample, if any two of its ratios are simultaneously greater than the corresponding third target threshold, the prediction for that modality is considered to have high uncertainty and is marked as a first uncertainty state; otherwise, it is considered a low uncertainty state. To further quantify the degree of uncertainty, the absolute value of the difference between the current three sets of ratios and their respective means is calculated to form an uncertainty score. Next, the mean uncertainty score for all modalities is calculated and weighted based on the degree of deviation of each modality's uncertainty score from its mean, resulting in a comprehensive optimization score. Weights can be pre-set based on modality importance or dynamically adjusted based on historical performance. Finally, the optimization threshold is based on the third target threshold and the comprehensive optimization score, which is used for subsequent screening of incremental samples.

[0049] In step 435 , the candidate sample defined as both the first pending sample and the second pending sample is compared with the optimization threshold. If the candidate sample is smaller than the optimization threshold, the candidate sample is defined as an incremental sample; if the candidate sample is greater than or equal to the optimization threshold, the candidate sample is eliminated.

[0050] Candidate samples, defined as both the first and second pending samples, are evaluated. These candidate samples have been preliminarily screened for data points with potential incremental value through consistency index and similarity analysis. Next, their uncertainty scores are compared with the optimization threshold. If the uncertainty score is less than the optimization threshold, the candidate sample is judged to have low uncertainty, high confidence, good representativeness, and stability, and can be included in the incremental training set. If the uncertainty score is greater than or equal to the optimization threshold, the sample is judged to have high prediction uncertainty or feature instability risk, which may affect the effectiveness of the model update and is therefore eliminated.

[0051] Step 5: Repeat the training of the fault classification model using the incremental training set to obtain a new model.

[0052] The construction of the incremental training set primarily relies on an initial sample screening mechanism. Using consistency indices, similarity analysis, and uncertainty assessment, the system identifies candidate samples that are both representative of new fault types and meet high-quality standards. These samples typically originate from abnormal conditions encountered during equipment operation and can provide new knowledge boundaries for the model. To ensure the quality of the incremental training set, the following extraction strategies are employed: Uncertainty score sorting: samples with lower uncertainty scores are prioritized to ensure higher prediction confidence; Time window filtering: retain recently collected data (such as the last three months) to reflect the latest changes in working conditions; Feature matching verification: Check whether the key features of the sample meet the model input requirements to avoid introducing noise or invalid data.

[0053] Directly adding incremental training sets to the initial training set may cause the dataset size to expand rapidly, affecting training efficiency and computing resource consumption. Therefore, certain data management strategies should be adopted to control the total data volume before merging: Set a maximum capacity limit: for example, limit the entire training set to contain no more than 10,000 samples; Eliminate aging data: remove samples of fault types that have not appeared for a long time (e.g., no update for more than one year); Redundant data compression: Use clustering algorithms (such as K-means) to remove highly similar samples and retain representative samples.

[0054] The recommended removal ratios are as follows: At least 20% of historical samples are retained for each fault type to maintain the model's memory capacity; The proportion of incremental samples should not exceed 20% of the total data set to prevent the introduction of excessive noise; Regularly (e.g., every six months or a year) conduct a comprehensive review of the dataset and adjust the retention ratio.

[0055] During the model training phase, a loss function is defined that combines the old knowledge retention term and the new sample fitting term to balance the relationship between the two: ; and They represent the loss functions on the original training set and the newly added samples, both using cross entropy loss; λ∈[0.3,0.7] represents the loss weight coefficient of the old samples, represents the regularization term, which is used to prevent drastic fluctuations in model parameters, θ0 represents the model parameters at the last completion of training, and γ is the regularization strength coefficient.

[0056] By designing the loss function described above, we can effectively learn new knowledge from newly added samples while maintaining the original model knowledge. During the actual training process, we use gradient descent or other optimization algorithms to minimize this loss function and update the model parameters.

[0057] Finally, the system comprehensively evaluates the new model using multiple evaluation metrics to determine whether it outperforms the old model. Common evaluation metrics include accuracy, precision, recall, F1 score, and AUC. If the new model outperforms the old model on at least three of these metrics, it is considered superior and can replace the old model for subsequent fault classification tasks.

[0058] Step 6: Obtain the first eigenvector at the current moment and input it into the new model to obtain the second prediction result.

[0059] like Figure 2 As shown, this embodiment further discloses a plunger pump fault online monitoring system based on multimodal data, which is used to execute the above-mentioned plunger pump fault online monitoring method based on multimodal data, including: Fault classification model: Build a training set and train the fault classification model to obtain a trained fault classification model; Data acquisition and preprocessing module: connected to the fault classification model, used to obtain multimodal data during the operation of the plunger pump in real time and perform preprocessing; The first prediction module is connected to the data acquisition and preprocessing module, and is used to construct a first feature vector based on the preprocessed multimodal data, input the trained fault classification model, and output a first prediction result; Incremental training set construction module: connected to the first prediction module, used to analyze the first feature vector whose prediction result is in an abnormal state and construct an incremental training set; Incremental training module: connected to the incremental training set construction module, used to repeatedly train the fault classification model using the incremental training set to obtain a new model; The second prediction module is connected to the incremental training module, used to obtain the first feature vector at the current moment and input it into the new model to obtain the second prediction result.

[0060] In order to implement the above embodiment, the present application also proposes an electronic device. Figure 3 , Figure 3 Schematic diagram of the structure of the electronic device provided in the embodiment of the present application. Figure 3 As shown, the electronic device 500 includes: a processor 501, and a memory 502 communicatively connected to the processor 501; the memory 502 stores computer-executable instructions; the processor 501 executes the computer-executable instructions stored in the memory to implement the method provided in the aforementioned embodiment.

[0061] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0062] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0063] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0064] It is important to note that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold beyond these legitimate uses. Furthermore, such collection / sharing should be conducted only after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes the relevant user information before using the feature. Furthermore, any necessary steps must be taken to safeguard and secure access to such personal information and ensure that others with access to personal information comply with its privacy policy and procedures.

[0065] This application contemplates providing implementations that allow users to selectively block the use or access of personal information data. Specifically, this disclosure contemplates providing hardware and / or software to prevent or block access to such personal information data. Risks can be minimized by limiting data collection and deleting data once it is no longer needed. Furthermore, where applicable, such personal information can be de-identified to protect user privacy.

[0066] In the descriptions of the foregoing embodiments, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and features of different embodiments or examples, unless they are mutually inconsistent.

[0067] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0068] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0069] The logic and / or steps represented in a flowchart or otherwise described herein, for example, can be considered a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" is any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (not exhaustive) of computer-readable media include: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0070] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0071] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0072] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0073] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for online monitoring of plunger pump failure based on multimodal data, characterized in that: The following steps are involved: Step 1: Construct a training set and train the fault classification model to obtain a trained fault classification model; Step 2: Real-time acquisition of multimodal data during the operation of the plunger pump and preprocessing; Step 3: construct a first feature vector based on the preprocessed multimodal data, input the first feature vector into the trained fault classification model, and output a first prediction result; Step 4: If the prediction result is an abnormal state, analyze the first eigenvector when the prediction result is an abnormal state and construct an incremental training set; Step 5: Repeat the training of the fault classification model using the incremental training set to obtain a new model; Step 6: Obtain the first eigenvector at the current moment and input it into the new model to obtain the second prediction result.

2. The method according to claim 1, characterized in that The multimodal data includes: vibration signals, pressure signals, temperature signals and acoustic emission signals.

3. The method according to claim 2, characterized in that The first eigenvector is composed of the following characteristic parameters: wavelet energy ratio of the vibration signal, main frequency energy density of the pressure signal, first-order difference mean of the temperature signal, and instantaneous frequency change rate of the acoustic emission signal.

4. The method according to claim 1, wherein If the prediction result is an abnormal state, analyzing the first eigenvector in the abnormal state and constructing an incremental training set includes the following steps: Step 41: Call the first feature vector predicted to be in an abnormal state and define it as a candidate sample. Determine the consistency index between each candidate sample and the historical samples in the historical fault case library. Preset a first target threshold. Based on the relationship between the first target threshold and the average Mahalanobis distance, determine whether the current candidate sample is a new fault type sample or a repeated sample. If it is a new fault type sample, define it as the first pending sample. If it is a repeated sample, add it to the historical fault case library to update or remove the historical fault case. Step 42: Determine the similarity between each candidate sample and each historical sample in the historical fault case library, preset a second target threshold, and determine whether the current candidate sample is a first quality sample or a second quality sample based on the relationship between the second target threshold and the similarity; if the current candidate sample is a first quality sample, define the candidate sample as a second pending sample; if the current candidate sample is a second quality sample, define the candidate sample as a third pending sample; Step 43: If the candidate sample is defined as both the first pending sample and the second pending sample, an incremental training set is constructed based on the candidate sample.

5. The method according to claim 4, characterized in that Constructing an incremental training set based on candidate samples includes the following steps: Step 431: call the pre-processed multimodal data of the candidate sample; Step 432: input the data of each modality into the trained fault classification model separately, and output a second prediction result; Step 433, defining at least three uncertainty indicators for the second prediction result; Step 434: construct a ratio between every two uncertainty indicators, preset a third target threshold, and when 2 / 3 of the ratios are greater than the third target threshold, define the second prediction result of the current modality as the first uncertainty, determine an uncertainty score based on the ratio, and determine an optimization threshold based on the uncertainty score; In step 435 , the candidate sample defined as both the first pending sample and the second pending sample is compared with the optimization threshold. If the candidate sample is smaller than the optimization threshold, the candidate sample is defined as an incremental sample; if the candidate sample is greater than or equal to the optimization threshold, the candidate sample is eliminated.

6. The method according to claim 1, characterized in that If the consistency index between each candidate sample and the historical samples in the historical fault case library is greater than the first target threshold, the current candidate sample is defined as a new fault type sample; if the consistency index between each candidate sample and the historical samples in the historical fault case library is less than or equal to the first target threshold, the current candidate sample is defined as a duplicate sample.

7. The method according to claim 5, characterized in that If the similarity between each candidate sample and each historical sample in the historical fault case library is greater than the second target threshold, the current candidate sample is defined as a first quality sample; if the similarity between each candidate sample and each historical sample in the historical fault case library is less than or equal to the second target threshold, the current candidate sample is defined as a second quality sample.

8. A plunger pump fault online monitoring system based on multimodal data, used to execute the plunger pump fault online monitoring method based on multimodal data according to any one of claims 1 to 7, characterized in that: include: Fault classification model: Build a training set and train the fault classification model to obtain a trained fault classification model; Data acquisition and preprocessing module: connected to the fault classification model, used to obtain multimodal data during the operation of the plunger pump in real time and perform preprocessing; The first prediction module is connected to the data acquisition and preprocessing module, and is used to construct a first feature vector based on the preprocessed multimodal data, input the trained fault classification model, and output a first prediction result; Incremental training set construction module: connected to the first prediction module, used to analyze the first feature vector whose prediction result is in an abnormal state and construct an incremental training set; Incremental training module: connected to the incremental training set construction module, used to repeatedly train the fault classification model using the incremental training set to obtain a new model; The second prediction module is connected to the incremental training module, used to obtain the first feature vector at the current moment and input it into the new model to obtain the second prediction result.

Images