Hybrid deep learning model for fault diagnosis of drilling pump

Abstract

The invention discloses a hybrid deep learning model for drilling pump fault diagnosis, and relates to the field of drilling pump fault diagnosis, and the method comprises the steps: collecting the operation data of a drilling pump; preprocessing the data to obtain a training sample; constructing a hybrid deep learning model MS-AFN based on multi-source heterogeneous signal fusion and a multi-scale attention mechanism; and performing fault diagnosis on the constructed drilling pump data set by using an MS-AFN model to obtain an evaluation model of the drilling pump. The constructed hybrid deep learning model combines the advantages of a statistical data driving method and a deep learning method, and the model can effectively segment and shear multi-source heterogeneous signals, reduce redundant information, improve the calculation efficiency and reduce the over-fitting risk. And the model can adopt different attention mechanisms, so that the sensitivity of the model to fault details and global backgrounds is enhanced, and the diagnosis accuracy and reliability are improved.

Description

Technical Field

[0001] This patent relates to a hybrid deep learning model for drilling pump fault diagnosis, which belongs to the field of mechanical equipment fault diagnosis, and in particular to improving the accuracy and efficiency of fault diagnosis through multi-source heterogeneous signal fusion and multi-scale attention mechanism. Background Art

[0002] In the process of energy development, despite the rapid increase in demand and development of renewable energy sources such as wind energy, hydropower and solar energy, oil and natural gas are still indispensable energy sources in industrial production and daily life. As an important equipment for the development of oil and gas resources, the drilling pump is known as the "heart" of the drilling rig. However, since the drilling pump works in a harsh underground or submarine environment, it is subject to extremely high working pressure and strong corrosiveness, and is prone to failure. Traditional fault diagnosis methods mainly rely on single sensor signals or regular manual inspections, and have problems such as limited diagnostic range, high computational complexity and poor real-time performance. In addition, a single sensor signal is difficult to fully reflect the health status of the equipment and cannot adapt to complex and changeable working conditions, resulting in insufficient diagnostic accuracy.

[0003] With the development of sensing technology, equipment diagnosis using heterogeneous signals collected by multi-source sensors has become a research hotspot. However, there are significant differences between the signals collected by different sensors, such as the physical characteristics of the signals, sampling frequency, and time delay, which poses a huge challenge to signal fusion. Existing methods often fail to fully consider these differences, resulting in poor fusion effects and even misjudgments. In addition, traditional methods rely on manual feature extraction, which not only requires high expert knowledge, but also tends to overfit when processing complex multi-scale signals, affecting the generalization ability of diagnosis. Therefore, developing an intelligent diagnosis method that can effectively fuse multi-source heterogeneous signals and automatically extract multi-scale features has become the key to improving the accuracy and efficiency of drilling pump fault diagnosis. Summary of the invention

[0004] This patent proposes a hybrid deep learning model Multi-Scale Attention and Fusion Network (MS-AFN) based on multi-source heterogeneous signal fusion and multi-scale attention mechanism, which aims to improve the accuracy and efficiency of drilling pump fault diagnosis. This method combines multiple deep learning technologies and realizes efficient diagnosis of drilling pump faults in complex industrial scenarios through reasonable module design. The method includes the following steps:

[0005] S1. Collecting the operation data of the drilling pump;

[0006] S2, preprocess the data to obtain training samples;

[0007] S3. Construct a hybrid deep learning model MS-AFN based on multi-source heterogeneous signal fusion and multi-scale attention mechanism;

[0008] S4. Apply the MS-AFN model to perform fault diagnosis on the constructed drilling pump dataset and obtain the evaluation model of the drilling pump.

[0009] Principle of the invention: By adopting the mean difference signal fusion module (MDSFM) and multi-scale attention network (MSAN) proposed in the present invention, the problem of insufficient information of traditional single sensor signals under complex working conditions is effectively solved. MDSFM greatly reduces data redundancy and improves the representativeness of signals by accurately identifying and fusing private features in multi-source heterogeneous signals, thereby significantly reducing computational complexity while ensuring diagnostic accuracy. MSAN not only enhances the model's ability to capture local details and global information through multi-scale feature extraction, but also improves the processing efficiency of long time series data. Experimental results show that the present invention exhibits excellent fault identification capabilities under different working conditions, especially under high load conditions, and can achieve fast and accurate fault diagnosis, providing reliable technical support for real-time monitoring and maintenance of industrial sites.

[0010] Furthermore, in S1, the specific steps for collecting the operating data of the drilling pump include: correctly installing and connecting temperature, pressure, speed and other sensors on the drilling pump system; configuring the data transmission mode, sampling frequency and data storage format parameters of the sensor; ensuring the normal operation of the sensor, recording the operating data of the drilling pump, and storing the collected sensor data.

[0011] Furthermore, in said S2, the specific steps of preprocessing the data include: normalizing the collected operation data based on the Min-max standardization method to eliminate the influence of the data dimension; screening out sensor data that can accurately characterize the degradation process based on the change trend of the drilling pump operation data; accurately capturing the correlation between adjacent time points and converting multiple sensor data into the required sample size based on the sliding window processing technology;

[0012] Furthermore, in S3, the MDSFM module is constructed to achieve effective fusion of multi-source heterogeneous signals by calculating the mean difference between the main signal and the auxiliary signal. This module enhances the representativeness of the signal and reduces the data dimension by removing the public components and retaining the private components. MSAN is constructed, and MSAN includes two branches: CBAM branch and multi-head self-attention branch. The MDSFM module is combined with the MSAN module to form a new hybrid deep learning model MS-AFN;

[0013] Furthermore, in S4, ablation experiments and comparative experiments are performed to verify the superiority and generalization of the MS-AFN model proposed in this paper.

[0014] One or more technical solutions provided by the present invention have at least the following technical effects or advantages:

[0015] 1. The drilling pump fault diagnosis method based on data and model hybrid drive of the present invention can effectively combine the advantages of both statistical data driven method and deep learning method, significantly improving the efficiency and accuracy of drilling pump fault diagnosis.

[0016] 2. The drilling pump fault diagnosis method based on data and model hybrid drive of the present invention combines the main signal and auxiliary signal in the multi-source heterofrequency signal of the drilling pump through the constructed MDSFM module. The method improves the accuracy and efficiency of the model through signal filtering and fusion.

[0017] 3. The drilling pump fault diagnosis method based on data and model hybrid drive of the present invention can better capture global context information through the constructed MSAN network, thereby improving the model's understanding of global features. Through this novel structure, the multi-scale feature extraction method MSAN can improve the accuracy and reliability of drilling pump fault diagnosis.

[0018] 4. The drilling pump fault diagnosis method based on data and model hybrid drive of the present invention can provide greater flexibility for the maintenance team, optimize the maintenance plan, choose to repair or replace parts at the most appropriate time, and minimize downtime and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the present invention, and do not constitute a limitation on the embodiments of the present invention;

[0020] Figure 1 It is a flow chart of a hybrid deep learning model for drilling pump fault diagnosis in the present invention;

[0021] Figure 2 Schematic diagram of the structure of the mean difference signal fusion module (MDSFM) in the present invention;

[0022] Figure 3 Schematic diagram of the structure of the self-attention mechanism in the present invention;

[0023] Figure 4 Schematic diagram of the structure of the multi-scale attention network (MSAN) in the present invention;

[0024] Figure 5 It is a schematic diagram of the structure of the hybrid deep learning model (MS-AFN) for drilling pump fault diagnosis in the present invention;

[0025] Figure 6 is a comparison chart of ablation experiment results in the present invention;

[0026] Figure 7 are the model test results under different working conditions in the present invention; DETAILED DESCRIPTION

[0027] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those within the scope of this description. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0029] Embodiment 1

[0030] refer to Figure 1-Figure 7 , this embodiment provides a hybrid deep learning model method for drilling pump fault diagnosis, the method comprising:

[0031] S1. Collecting the operation data of the drilling pump;

[0032] S2, preprocess the data to obtain training samples;

[0033] Among them, the specific steps of preprocessing the drilling pump data to obtain training samples include: clustering the collected operation data according to the working conditions, and normalizing the data according to the working conditions to eliminate the influence of the data dimension; screening out sensor data that can accurately characterize the degradation process according to the changing trend of the drilling pump operation data; and using the sliding window technology to process the screened sensor data to construct training samples.

[0034] S3. The construction of the hybrid deep learning model MS-AFN first constructs the mean difference signal fusion module (MDSFM) to achieve effective fusion of multi-source heterogeneous signals by calculating the mean difference between the main signal and the auxiliary signal. This module enhances the representativeness of the signal and reduces the data dimension by removing the common components and retaining the private components. Then the multi-scale attention network (MSAN) is constructed. MSAN includes two branches: CBAM branch and multi-head attention branch. The CBAM branch is used to capture local features, while the multi-head attention branch is used to capture global features. Through this multi-scale feature extraction method, MSAN can effectively extract rich information from multi-source heterogeneous signals. CBAM is a hybrid attention mechanism module for enhancing the performance of convolutional neural networks. It consists of two sub-modules: channel attention module (CAM) and spatial attention module (SAM). The main goal of CBAM is to improve the network's ability to represent the input by dynamically adjusting the weights of channel and spatial features, thereby enhancing the network's perception and generalization capabilities. The multi-head attention mechanism focuses on different subspaces through multiple independent attention heads, thereby capturing relevant information from different parts of the sequence. The outputs of multiple heads are concatenated and linearly transformed to generate the final attention output.

[0035] S4. Compare the MS-AFN network model with other advanced models to verify the superiority and generalization of the MS-AFN model proposed in this paper.

[0036] In order to verify the effectiveness of each submodule in the multi-scale network module, an ablation experiment was constructed. In the ablation experiment, The Proposed Method was compared with Without MHSA, Without CBAM, Without BiLSTM, Without MDSFM and Exchange Attention Branch. Finally, the experiment showed that the MDSFM module proposed in this paper effectively improved the efficiency and accuracy of the model.

[0037] In order to evaluate the effectiveness of the proposed method, comparative experiments were conducted under each working condition, comparing MS-AFN with WCCN-BiLSTM, MDSFM-WCCN-BiLSTM, CNN, MDSFM-CNN, ZFNet, MDSFM-ZFNet, AlexNet and MDSFM-AlexNet. The experiments were conducted five times in total, and the results were recorded and averaged. The experimental results show that the invented model has significant advantages in both efficiency and accuracy.

[0038] The above results clearly demonstrate the superiority of MS-AFN and the effectiveness of MDSFM.

[0039] Embodiment 2

[0040] Based on the above embodiment 1, in this embodiment, the calculation formula of Min-max normalization is:

[0041]

[0042] in, and X i,j Represent the data after and before normalization, i and j represent the i-th data point and j-th sensor, respectively. and represent the minimum and maximum values ​​of the j-th sensor respectively.

[0043] The calculation formulas for the three performance indicators of accuracy, precision, recall, and F1 score are as follows:

[0044]

[0045] Where TP, FP, FN, and TN represent the counts of true positive, false positive, false negative, and true negative results, respectively.

[0046] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0047] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A hybrid deep learning model for drilling pump fault diagnosis, characterized in that: The method comprises: S1. Collecting the operating data of oil drilling pumps; S2, preprocess the data to obtain training samples; S3. Construct a hybrid deep learning model (MS-AFN) based on multi-source heterogeneous signal fusion and multi-scale attention mechanism; S4. Apply the MS-AFN model to perform fault diagnosis on the constructed drilling pump dataset and evaluate the model.

2. A hybrid deep learning model for drilling pump fault diagnosis according to claim 1, characterized in that: In the step S1, collecting the operation data of the drilling pump in step S1 includes: S11. Correctly install and connect temperature, pressure, speed and other sensors on the drilling pump system; S12, configuring the data storage mode, sampling frequency and data storage format parameters of the sensor; S13, ensuring the normal operation of the sensor, recording the operation data of the drilling pump, and storing the collected sensor data.

3. A hybrid deep learning model for drilling pump fault diagnosis according to claim 1, characterized in that: In S2, the specific steps of preprocessing the data include: S21. Normalize the collected operating data according to the working conditions to eliminate the influence of the data dimension; use the Min-max normalization method to make the sensor value fluctuate within the range of [-1,1]; in, and X i,j Represent the data after and before normalization, i and j represent the i-th data point and j-th sensor, respectively. and Represent the minimum and maximum values ​​of the jth sensor respectively; S22, selecting sensor data that can accurately describe the degradation process according to the change trend of the drilling pump operation data; S23. Use sliding window processing technology to process the filtered sensor data and construct training samples.

4. A hybrid deep learning model for drilling pump fault diagnosis according to claim 1, characterized in that: The specific steps of building a hybrid deep learning model based on multi-source heterogeneous signal fusion and multi-scale attention mechanism include: S31. Connect the main signal and the auxiliary signal in the first dimension to obtain the fused multi-source heterogeneous signal as the input signal of the model, whose dimension is 2, and its formula is as follows: Among them, S S Represents the distribution difference between the main signal and the auxiliary signal, S P With S A Represent the main signal and auxiliary signal respectively, n P and n A represents the sample length of the signal, · represents the absolute value; S32. The features are arranged in descending order by the mean. The mean difference in descending order is mainly used to conveniently identify and select the part with the largest difference to optimize signal processing. The following formula is used: T S =Rank(S S ) Where T represents the similarity of the feature, and Rank() represents descending order according to the sequence number; S33, selecting the features with high similarity in the two auxiliary signals as common features for subsequent feature interaction, and retaining the remaining features as private features. The common features and private features of the main signal and the auxiliary signal can be expressed as: Among them, FR SV Represents private features. After obtaining the private features in the auxiliary signal, it is connected with the main signal to obtain the input of the multi-scale network: Input=Ξ{S P ,Ξ[(PP A V ,FR A P ),dim=2]},dim=1 Among them, PP A V Represents the private part of the vibration signal, PP A P Represents the private part of the pressure signal, and dim represents the dimension of data merging.

5. A hybrid deep learning model for drilling pump fault diagnosis according to claim 1, characterized in that: The MS-AFN model is used to diagnose the faults of the constructed drilling pump dataset and evaluate the model, including: S41. Construct the CBAM branch in the multi-scale network. The calculation formula involved in the overall CBAM process is as follows: Where F is the input feature map, σ is the Sigmoid activation function, MLP represents the multi-layer perceptron, c (F) represents the channel attention feature map, M s (F c ) represents the spatial attention feature map, Represents element multiplication operation, AvgPool represents average pooling, and MaxPool represents maximum pooling; S42. Construct a multi-head self-attention branch in a multi-scale network. The calculation formula is as follows: Where d represents the dimension, Q refers to the query matrix, K corresponds to the key matrix, V is related to the value matrix, and X f represents the input feature matrix, W Q Related to the weight matrix corresponding to the query matrix, W K Related to the weight matrix corresponding to the key matrix, W V Related to the weight matrix corresponding to the value matrix; S43. The performance indicators used for evaluating a hybrid deep learning model for drilling pump fault diagnosis are as follows: Where TP, FP, FN, and TN represent the counts of true positive, false positive, false negative, and true negative results, respectively.

Images