Drilling rig fault diagnosis method and apparatus under complex working conditions, device, medium, and product

Abstract

The present application relates to the field of drilling rig fault diagnosis, and discloses a drilling rig fault diagnosis method and apparatus under complex working conditions, a device, a medium, and a product. The method comprises constructing a drilling rig fault data set comprising interference factors under complex working conditions; converting pressure time series data in the drilling rig fault data set into a time-frequency image by means of forward wavelet transform; using a convolutional autoencoder to denoise the time-frequency image; reconstructing the denoised time-frequency image into pressure time series data by means of inverse wavelet transform to obtain a reconstructed drilling rig fault data set; using the reconstructed drilling rig fault data set to train a composite neural network, the composite neural network comprising a bidirectional long short-term memory network and a KAN model; and using the trained composite neural network to perform real-time fault diagnosis on a drilling rig. The present application can maintain high-accuracy fault diagnosis for drilling rigs under complex noise caused by complex working conditions.

Description

Methods, devices, equipment, media, and products for diagnosing drilling machine faults under complex working conditions.

[0001] This application claims priority to Chinese Patent Application No. 202411222887.9, filed on September 2, 2024, entitled “Method, Apparatus, Equipment, Medium and Product for Fault Diagnosis of Drilling Machine under Complex Working Conditions”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of drilling machine fault diagnosis, and in particular to a method, device, equipment, medium and product for drilling machine fault diagnosis under complex working conditions. Background Technology

[0003] Fault diagnosis of drilling machines relies on sensors deployed on the equipment. The type of fault is determined by analyzing the signals collected by the sensors. The complex operating conditions of drilling machines introduce a large amount of complex non-steady-state noise, which distorts the signals collected by the sensors and obscures the true fault characteristics, thus increasing the difficulty of fault diagnosis.

[0004] Based on signals collected by sensors, the inventors know that methods for diagnosing drilling machine faults include multi-source data fusion, adaptive algorithms, expert systems, and real-time monitoring and early warning. Multi-source data fusion improves the accuracy of fault diagnosis by integrating data from multiple sensors (such as vibration, temperature, and current) and utilizing signal processing and data fusion techniques. However, multi-source data fusion requires a large amount of high-quality sensor data, which may be unstable or interfered with under complex operating conditions. Adaptive algorithms can automatically adjust model parameters according to changes in operating conditions to maintain diagnostic performance. However, the stability and robustness of adaptive algorithms under complex operating conditions still need improvement, especially when facing sudden faults or extreme conditions. Expert systems build a fault diagnosis rule base based on expert knowledge and combine it with data-driven methods to achieve rapid fault identification and location. Expert systems rely on experiential knowledge and struggle to handle new or unknown types of faults; the maintenance and updating of the rule base is also cumbersome. Real-time monitoring and early warning methods involve establishing a real-time monitoring system to continuously monitor key parameters, promptly detect anomalies, and issue early warnings to prevent fault escalation. However, real-time monitoring systems have high requirements for data transmission and processing. Under complex operating conditions, there may be data transmission delays or packet loss, which can affect the accuracy of fault diagnosis.

[0005] Therefore, none of the above technologies can effectively cope with the large amount of non-steady-state complex noise introduced by complex working conditions, and their impact on fault diagnosis remains significant. Summary of the Invention

[0006] The purpose of this application is to provide a method, device, equipment, medium, and product for diagnosing faults in drilling machines under complex working conditions, which can maintain high accuracy in fault diagnosis of drilling machines under complex noise caused by complex working conditions.

[0007] To achieve the above objectives, this application provides the following solution:

[0008] In a first aspect, this application provides a method for diagnosing drilling machine faults under complex working conditions, comprising: constructing a drilling machine fault dataset containing interference factors under complex working conditions; the drilling machine fault dataset including pressure time series data under normal conditions and pressure time series data under different fault types; converting the pressure time series data in the drilling machine fault dataset into a time-frequency image using a wavelet forward transform; denoising the time-frequency image using a convolutional autoencoder; reconstructing the denoised time-frequency image into pressure time series data using an inverse wavelet transform to obtain a reconstructed drilling machine fault dataset; and training the system using the reconstructed drilling machine fault dataset. A composite neural network is trained, comprising a bidirectional long short-term memory network and a KAN model. The bidirectional long short-term memory network is used to extract features from the pressure time series data to obtain the hidden state. The KAN model is used to identify drilling machine faults based on the hidden state and output the drilling machine fault diagnosis results. The real-time collected pressure time series data is sequentially processed through wavelet forward transform, convolutional autoencoder, and inverse wavelet transform to obtain real-time reconstructed pressure time series data. Based on the real-time reconstructed pressure time series data, the trained composite neural network is used to diagnose drilling machine faults and output real-time drilling machine fault diagnosis results.

[0009] Secondly, this application provides a fault diagnosis device for drilling machines under complex working conditions, including: a dataset construction module, a data conversion module, a noise reduction module, a reconstruction module, a training module, an acquisition module, and an application module.

[0010] The system comprises the following modules: a dataset construction module for building a drilling machine fault dataset containing interference factors under complex operating conditions; the dataset includes pressure time-series data under normal conditions and pressure time-series data under different fault types. A data conversion module converts the pressure time-series data in the drilling machine fault dataset into a time-frequency image using wavelet forward transform. A denoising module denoises the time-frequency image using a convolutional autoencoder. A reconstruction module reconstructs the denoised time-frequency image into pressure time-series data using inverse wavelet transform, obtaining a reconstructed drilling machine fault dataset. A training module trains a composite neural network using the reconstructed drilling machine fault dataset; the composite neural network includes a bidirectional long short-term memory network and a KAN model. The bidirectional long short-term memory network extracts features from the pressure time-series data to obtain hidden states; the KAN model identifies drilling machine faults based on the hidden states and outputs the drilling machine fault diagnosis results. An acquisition module sequentially processes the real-time acquired pressure time-series data through wavelet forward transform, convolutional autoencoder, and inverse wavelet transform to obtain real-time reconstructed pressure time-series data. The application module is used to perform fault diagnosis on the drilling machine based on real-time reconstructed pressure time series data and a trained composite neural network, and output real-time fault diagnosis results for the drilling machine.

[0011] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the drilling machine fault diagnosis method under complex working conditions described in the first aspect above.

[0012] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the drilling machine fault diagnosis method under complex working conditions described in the first aspect above.

[0013] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the drilling machine fault diagnosis method under complex working conditions described in the first aspect.

[0014] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0015] This application provides a method, device, equipment, medium, and product for fault diagnosis of drilling machines under complex working conditions. Mechanical faults typically exhibit characteristics in specific frequency bands. Wavelet transform, by converting pressure time-series data into a time-frequency image, allows observation of signal variations at different times and frequencies. Simultaneously, wavelet transform effectively removes noise from the signal while preserving fault characteristics, aiding in fault identification and location. Convolutional autoencoders can automatically extract high-level features from input data, removing noise and redundant information. Drilling machine faults often manifest as specific patterns in long-term data series. Bidirectional long short-term memory networks can capture these specific patterns, improving the accuracy of fault diagnosis. KAN models can capture and extract complex nonlinear features during drilling machine operation, thereby more accurately identifying fault modes. Through the combined use of wavelet transform, convolutional autoencoders, bidirectional long short-term memory networks, and KAN models, high-accuracy fault diagnosis of drilling machines is achieved even under complex noise conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 is an application environment diagram of a drilling machine fault diagnosis method under complex working conditions according to an embodiment of this application;

[0018] Figure 2 is a flowchart illustrating a fault diagnosis method for drilling machines under complex working conditions provided in an embodiment of this application;

[0019] Figure 3 is a simplified flowchart of a drilling machine fault diagnosis method under complex working conditions according to an embodiment of this application;

[0020] Figure 4 is a schematic diagram of the training process module provided in another embodiment of this application;

[0021] Figure 5 is a schematic diagram of the structure of a convolutional autoencoder provided in another embodiment of this application;

[0022] Figure 6 is a schematic diagram of a bidirectional long short-term memory network deep learning model provided in another embodiment of this application;

[0023] Figure 7 is a schematic diagram of the KAN deep learning model provided in another embodiment of this application;

[0024] Figure 8 is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] The drilling machine fault diagnosis method under complex working conditions provided in this application embodiment can be applied to the application environment shown in Figure 1. The terminal 102 communicates with the server 104 via a network. A data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated into the server 104, or placed on the cloud or other servers. The terminal 102 can send the video to be processed to the server 104. After receiving the pressure time series data, the server 104 constructs a drilling machine fault dataset containing interference factors under complex working conditions. The drilling machine fault dataset includes pressure time series data under normal conditions and pressure time series data under different fault types. A wavelet forward transform is used to convert the pressure time series data in the drilling machine fault dataset into a time-frequency image. A convolutional autoencoder is used to denoise the time-frequency image. An inverse wavelet transform is used to reconstruct the denoised time-frequency image into pressure time series data, obtaining the reconstructed drilling machine fault dataset. The reconstructed drilling machine fault dataset is used to train a composite neural network. This composite neural network includes a bidirectional long short-term memory network (LSTM) and a KAN model. The LSM is used to extract features from the pressure time-series data to obtain hidden states. The KAN model is used to identify drilling machine faults based on the hidden states and output drilling machine fault diagnosis results. The real-time collected pressure time-series data is sequentially processed through wavelet forward transform, convolutional autoencoder, and inverse wavelet transform to obtain real-time reconstructed pressure time-series data. Based on the real-time reconstructed pressure time-series data, the trained composite neural network is used to diagnose drilling machine faults and output real-time drilling machine fault diagnosis results. The server 104 can feed back the obtained real-time drilling machine fault diagnosis results to the terminal 102. Furthermore, in some embodiments, the drilling machine fault diagnosis method under complex working conditions can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly perform drilling machine fault diagnosis under complex working conditions based on the pressure time-series data, or the server 104 can obtain pressure time-series data from the data storage system and perform drilling machine fault diagnosis based on the pressure time-series data.

[0028] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0029] In an exemplary embodiment, as shown in FIG2, a fault diagnosis method for drilling machines under complex working conditions is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using the server 104 in FIG1 as an example, and includes the following steps 201 to 207. Wherein:

[0030] Step 201: Construct a drilling machine fault dataset containing interference factors under complex working conditions; the drilling machine fault dataset includes pressure time series data under normal conditions and pressure time series data under different fault types.

[0031] Step 202: Use wavelet forward transform to convert the pressure time series data in the drilling machine fault dataset into a time-frequency image;

[0032] Step 203: Denoise the time-frequency image using a convolutional autoencoder.

[0033] Step 204: Reconstruct the denoised time-frequency image into pressure time-series data by inverse wavelet transform to obtain the reconstructed drilling machine fault dataset.

[0034] Step 205: Train a composite neural network using the reconstructed drilling machine fault dataset; the composite neural network includes a bidirectional long short-term memory network and a KAN model, the bidirectional long short-term memory network is used to extract features from the pressure time series data to obtain the hidden state; the KAN model is used to identify drilling machine faults based on the hidden state and output the drilling machine fault diagnosis result.

[0035] Step 206: The real-time collected pressure time series data are sequentially processed through wavelet forward transform, convolutional autoencoder and inverse wavelet transform to obtain real-time reconstructed pressure time series data.

[0036] Step 207: Based on the real-time reconstructed pressure time series data, use the trained composite neural network to perform fault diagnosis on the drilling machine and output the real-time fault diagnosis results of the drilling machine.

[0037] By implementing steps 201 to 207 above, mechanical faults typically exhibit characteristics within specific frequency bands. Wavelet transform, by converting pressure time-series data into a time-frequency image, allows observation of signal variations across different times and frequencies. Simultaneously, wavelet transform effectively removes noise from the signal, preserving fault characteristics and aiding in fault identification and localization. Convolutional autoencoders can automatically extract high-level features from input data, removing noise and redundant information. Drilling machine faults often manifest as specific patterns in long-term data series. Bidirectional long short-term memory networks can capture these specific patterns, improving the accuracy of fault diagnosis. The KAN model can capture and extract complex nonlinear characteristics during drilling machine operation, thereby more accurately identifying fault modes. Through the combined use of wavelet transform, convolutional autoencoders, bidirectional long short-term memory networks, and the KAN model, high-accuracy fault diagnosis of drilling machines is achieved even under complex noise conditions.

[0038] Figure 3 is a simplified flowchart of the drilling machine fault diagnosis method under complex working conditions according to this application. Figure 4 is a block diagram of the training process from steps 201 to 205. Figure 4 shows that the training process from steps 201 to 205 is mainly divided into three parts. The first part is to construct a drilling machine fault dataset containing interference factors. In this stage, pressure time series data of the drilling machine under different faults are collected. The interference faced by the drilling machine under complex working conditions is simulated by a composite noise composed of Gaussian noise and equal noise, and finally a drilling machine fault dataset containing interference factors is obtained. The second part is noise filtering. This part consists of wavelet transform and convolutional autoencoder. First, the time series is converted into a time-frequency image by wavelet forward transform, thereby decomposing the signal into different frequency bands and achieving preliminary noise filtering. Then, a convolutional autoencoder is used for further denoising. The convolutional autoencoder retains the main features and removes noise components by learning the low-dimensional representation of the signal. Furthermore, it can utilize local connectivity and weight sharing mechanisms to better capture local features in time-frequency images, improving denoising performance. Finally, the time-frequency image is converted back to a time series using inverse wavelet transform to recover the original denoised signal. The third part is a composite neural network. This part consists of a bidirectional long short-term memory (BILSTM) network and a KAN model. The bidirectional structure of the BILSTM network enables it to capture the feature information of sequential dependencies, enhancing the expressive power of fault features. Finally, the KAN model completes the drilling machine fault identification. Unlike the traditional multilayer perceptron (MLP), the KAN model uses learnable activation functions on its weights. This means that each weight parameter is replaced by a univariate function, which is parameterized by spline functions, allowing the KAN model to adjust the model more flexibly and better fit the fault data features.

[0039] In another exemplary embodiment of this application, the fault types include: loss of A-type sealing ring, loss of B-type sealing ring, damage to the return accumulator, abnormal damper hole, abnormal flow in the damper circuit, and low charge in the high-voltage accumulator.

[0040] In another exemplary embodiment of this application, in order to accurately train the composite neural network so that it can identify whether the drilling machine has malfunctioned under complex working conditions, and the type of malfunction when it does, the drilling machine malfunction dataset is designed to consist of normal data and various malfunction data. If there are 10 types of malfunctions, then the label for normal data is set to 0, and the label for malfunction data is set to 1 to 10. Therefore, the specific process of constructing the drilling machine malfunction dataset containing interference factors under complex working conditions in step 201 above is replaced by the following steps 301 to 303:

[0041] Step 301: Collect raw pressure time series data of the drilling machine under normal conditions and under different fault types.

[0042] Step 302: Combine Gaussian noise and equalized noise to form composite noise to simulate the interference faced by the drilling machine under complex working conditions; where Gaussian noise represents random environmental interference and unpredictable fluctuations inside the equipment, and equalized noise represents systematic and uniformly distributed external interference.

[0043] Step 303: Add the composite noise to the original pressure time series data to generate a drilling machine fault dataset containing interference factors under complex working conditions.

[0044] Step 302 above uses a composite noise system composed of Gaussian noise and equalized noise to simulate the interference faced by the drilling machine under complex working conditions. Gaussian noise simulates random environmental interference and unpredictable fluctuations within the equipment, while equalized noise represents systematic and uniformly distributed external interference, such as equipment vibration and mechanical resonance. The two types of noise act together on the drilling machine's pressure time-series data X(t) in a superimposed form, thus more realistically reflecting the complex interference factors in the actual working environment. By superimposing the composite noise onto the original pressure time-series data, a simulated dataset containing various interference factors can be generated. This dataset not only helps us better understand the impact of noise on the drilling machine's pressure signal but also provides a foundation for subsequent fault identification under interference. (The last sentence appears to be incomplete and possibly refers to a different topic.) G (t) represents Gaussian noise, I E (t) represents the equalized noise, and the composite noise is represented by I. N (t)=I G (t)+I E (t). The pressure time series data after applying noise interference is represented as: X noise (t)=X(t)+I N (t).

[0045] Gaussian noise I G This simulates random environmental disturbances and unpredictable fluctuations within equipment. Its specific expression is derived from the normal distribution and Gaussian noise I. GThe probability density function is expressed as:

[0046] Where μ represents the average noise level, and η represents the noise fluctuation range. 2 This indicates the degree of noise dispersion.

[0047] Gaussian noise I at time t G (t) can be represented as a random variable following the above normal distribution. For a Gaussian noise with mean μ and standard deviation η, it can be represented as: I G (t)=μ+η·Z(t).

[0048] Where Z(t) is a standard normally distributed random variable.

[0049] To equalize noise I E This simulates systematic and uniformly distributed external disturbances, such as equipment vibration and mechanical resonance. Equalized noise is a type of noise with uniform distribution characteristics, and its probability density function is constant within a specific interval. The specific expression for equalized noise can be represented using a uniform distribution. For equalized noise over an interval [a, b], its probability density function is expressed as:

[0050] Equalization noise I over time t E (t) can be represented as a random variable following a uniform distribution U(a,b), and its expression is: I E (t)>U(a,b).

[0051] Generate a standard uniformly distributed random number U(t) in the interval [0,1]. Transform the uniformly distributed random number to the target interval [a,b] to obtain the equalization noise I. E (t), the expression for the related process is: I E (t)=a+(ba)·U(t).

[0052] Where a and b are the upper and lower limits of the target interval, respectively.

[0053] In another exemplary embodiment of this application, wavelet transform can simultaneously provide time and frequency information. By converting time-series data into a time-frequency image, the changes in the signal at different times and frequencies can be observed. This is very useful for fault diagnosis because mechanical faults often exhibit characteristics in specific frequency bands. Wavelet transform has multi-resolution analysis capabilities, allowing signals to be decomposed into different scales for analysis. This means that both global features and local details of the signal can be observed simultaneously, aiding in fault identification and localization. Wavelet transform has a significant effect on denoising. By selecting appropriate wavelet basis functions and thresholds, noise in the signal can be effectively removed while preserving fault characteristics. Wavelet transform can extract useful features from the original time-series signal, including energy distribution, singularity, etc. These features can be used for subsequent fault classification and identification. The wavelet forward transform process used in step 202 above is as follows: for the stress time series X under disturbance... noise Perform wavelet transform on the time series data X, selecting the wavelet mother function ψ(t) and scale a. noise The wavelet transform of (t) is defined as:

[0054] Where 'a' is the scaling parameter, controlling the scaling of the wavelet function; and 'b' is the position parameter, controlling the translation of the wavelet function. The above formula essentially calculates the similarity between the input signal and wavelet basis functions at different scales and translations. For each pair (a, b), W is calculated. T (a, b) yields wavelet coefficients for each corresponding scale and location. These coefficients represent the original pressure time series X. noise (t) Energy information at this scale and location.

[0055] Then, all the calculated wavelet coefficients are organized into a two-dimensional matrix M. T , of which element M i,j Corresponding to a specific scale a i and time location b j Wavelet coefficients. Two-dimensional matrix M T The expression is:

[0056] Matrix M T Capture signal X noise The characteristics of (t) at different scales and time locations allow for effective analysis and visualization of the signal's time-frequency properties in matrix form. Furthermore, this organization provides a foundation for further signal processing, such as signal reconstruction and feature extraction. Since wavelet coefficients can have a very wide range, they need to be normalized to better represent them in the image, as expressed by:

[0057] Among them, M T[i,j] are elements of the original wavelet coefficient matrix, min(M T ) and max(M T ) represent the coefficient matrix M respectively T The minimum and maximum values ​​in the matrix are then determined. The normalized coefficient matrix M is then used to find these values. Tnorm The mapping of [i,j] to the time-frequency image is expressed as: I[i,j]=255·M Tnorm [i,j].

[0058] Where I[i,j] represents the grayscale value of the corresponding pixel in the time-frequency image, ranging from 0 (complete black) to 255 (complete white). This ensures that the dynamic range of the wavelet coefficients is fully expressed in the grayscale levels of the time-frequency image. Each element in I[i,j] corresponds to a pixel in the time-frequency image I, therefore the matrix I[i,j] can be equivalent to the time-frequency image I.

[0059] In another exemplary embodiment of this application, a convolutional autoencoder can automatically extract high-level features from the input data and compress high-dimensional data into a low-dimensional latent space through the encoder part. This helps to extract useful features from the disturbed original signal and remove noise and redundant information. The convolutional autoencoder can learn the main structure of the data by reconstructing the loss, thereby reconstructing the denoised signal in the decoder part. This denoising capability is particularly useful for drilling machinery data under complex working conditions, because sensor signals are easily affected by environmental interference. Through the reconstruction process of the autoencoder, the reconstructed signal can be compared with the original signal, and the differences can be analyzed to identify the fault type and location. This process can help engineers understand the cause of the fault more intuitively. The process of the convolutional autoencoder used in step 203 above is as follows: the time-frequency image I is input into the convolutional autoencoder, and the time-frequency image I[i,j] is denoised through a series of convolution, pooling, fully connected and deconvolution operations.

[0060] Figure 5 shows a specific structure of a convolutional autoencoder, comprising three convolutional layers, one pooling layer, one fully connected layer (FC), and three deconvolutional layers. The expression for the first convolutional layer, Conv1, processing the time-frequency image I is shown below:

[0061] Among them, I fconv1 This represents the feature map of the time-frequency image I after processing by the first convolutional layer, where σ1 is the activation function of the first convolutional layer. λ1 is the weight of the k-th filter in the first convolutional layer, b1 is the bias term of the first convolutional layer, and λ1 is the regularization parameter of the first convolutional layer. It is the L2 norm of w1.

[0062] The feature map I obtained afterwards fconv1 Input to the second convolutional layer Conv2, the second convolutional layer Conv2 processes feature map I fconv1 The expression to be processed is:

[0063] Among them, I fconv2 Representative Feature Map I fconv1 The feature map after processing by the second convolutional layer, where σ² is the activation function of the second convolutional layer. λ1 is the weight of the k-th filter in the second convolutional layer, b2 is the bias term of the second convolutional layer, and λ2 is the regularization parameter of the second convolutional layer. It is the L2 norm of w2.

[0064] The feature map I obtained afterwards fconv2 Input to the third convolutional layer Conv3, the third convolutional layer Conv3 processes feature map I fconv2 The expression to be processed is:

[0065] Among them, I fconv3 Representative Feature Map I fconv2 The feature map after processing by the third convolutional layer, where σ3 is the activation function of the third convolutional layer. λ is the weight of the k-th filter in the third convolutional layer, b3 is the bias term of the third convolutional layer, and λ3 is the regularization parameter of the third convolutional layer. It is the L2 norm of w3.

[0066] Feature Map I fconv3 The Flatten operation yields a one-dimensional array I. flat The data is then fed into the fully connected layer. The Flatten operation linearizes the spatial and channel information from the convolutional layer, allowing the data to undergo a weighted summation operation between each neuron in the fully connected layer, expressed as: I flat =flatten(I fconv3 );

[0067] Where h represents a one-dimensional array after weighted summation in the fully connected layer, σ fc It is the activation function of the fully connected layer, w fc It is the weight matrix of the fully connected layer, b fc It is the bias term of the fully connected layer, λ fc These are the regularization parameters for fully connected layers. It is w fc The L2 norm of .

[0068] Next, a reshape operation is performed on the one-dimensional array h. This reshape adjusts the output of the fully connected layer back to a dimension and shape acceptable to the deconvolutional layer. The expression is: h reshape =reshape(h).

[0069] Then h reshape The input is fed into the third deconvolutional layer DeConv3, and the third deconvolutional layer DeConv3 processes h. reshape The expression to be processed is:

[0070] Among them, h d3 It is the output of the third deconvolutional layer, DeConv3, h reshape It is a multidimensional array, σ d3 w d3 b d3 and λ d3 These are the activation function, weight matrix, bias term, and regularization parameter of the third deconvolutional layer, respectively. It is w d3 The L2 norm.

[0071] Then h d3 The input is fed into the second deconvolutional layer DeConv2, and the second deconvolutional layer DeConv2 processes h. d3 The expression to be processed is:

[0072] Among them, h d2 It is the output of the third deconvolution layer, σ d2 w d2 b d2 and λ d2 These are the activation function, weight matrix, bias term, and regularization parameter of the second deconvolutional layer, respectively. It is w d2 The L2 norm of .

[0073] Finally, h d2 The first deconvolutional layer DeConv1 is input to obtain the output time-frequency image. The first deconvolutional layer DeConv1 is used for h d2 The expression to be processed is:

[0074] Where, σ d1 w d1 b d1 and λ d1 These are the activation function, weight matrix, bias term, and regularization parameter of the first deconvolutional layer, respectively. It is w d1 The L2 norm of .

[0075] In another exemplary embodiment of this application, the specific process of using inverse wavelet transform in step 204 is as follows: the time-frequency image obtained by the convolutional autoencoder. Inverse wavelet transform is used to restore the time-frequency image into time-series data to meet the requirements of subsequent models.

[0076] First, the time-frequency image data is remapped back to wavelet coefficient form using the following formula:

[0077] in, It is a time-frequency image output from an autoencoder. The normalized coefficient values ​​extracted from M T It is the original wavelet coefficient matrix. This is the result of converting these values ​​into the scale of the original wavelet coefficients.

[0078] Next, inverse wavelet transform is used to recover the wavelet coefficients. Converting to time-series data involves using the same wavelet mother function as the forward wavelet transform, but the operation is inverse, as shown in the following expression:

[0079] in, This is the reconstructed drilling machine fault timing data. These are the recovered wavelet coefficients. The wavelet basis functions after corresponding scaling and translation adjustments.

[0080] In another exemplary embodiment of this application, the composite neural network in step 205 above includes a bidirectional long short-term memory network and a KAN model.

[0081] As shown in Figure 6, the BILSTM network can simultaneously consider both forward and backward information of the sequence data, capturing the dynamic characteristics of time series data more comprehensively through bidirectional processing. Under complex operating conditions, drilling machine fault signals may be affected by various environmental noises and interferences. The BILSTM network can effectively utilize information from previous and subsequent time points, improving the robustness and accuracy of feature extraction. Through memory and forgetting mechanisms, the BILSTM network effectively filters irrelevant or interfering information, highlighting important fault features. In complex operating conditions, there are many interfering signals. The BILSTM network can reduce the impact of noise on fault diagnosis and enhance the extraction of useful signals through its memory units and forgetting gates. The BILSTM network excels at processing long-term series data, extracting deeper patterns and features from the time series. Drilling machine faults often manifest as specific patterns in long-term series data; the BILSTM network can capture these patterns, improving the accuracy of fault diagnosis.

[0082] As shown in Figure 7, the KAN model can capture and extract complex nonlinear characteristics during the operation of drilling machines, thereby more accurately identifying fault modes. Its multi-layered structure can approximate arbitrarily complex nonlinear relationships, helping to handle various nonlinear disturbances in complex operating conditions. Compared to traditional linear models, the KAN model is more suitable for handling nonlinearity and variability in real industrial environments. Through its deep structure and complex feature space mapping, the KAN model effectively filters out noise and interference in sensor signals, improving the robustness of fault diagnosis. Even under the influence of environmental interference and mechanical vibration, the KAN model maintains high diagnostic accuracy. The KAN model can process multi-source heterogeneous data (such as vibration signals, temperature signals, and current signals), improving the comprehensiveness and accuracy of fault diagnosis by fusing data from multiple sensors. Multi-source data fusion enhances the model's ability to detect different types of faults and adapts to diverse fault modes under complex operating conditions. The KAN model has strong adaptive capabilities, continuously adjusting model parameters based on real-time data to adapt to different operating conditions, ensuring the continuity and stability of fault diagnosis. Adaptability enables the KAN model to respond and adjust in a timely manner when faced with dynamic environments and changes in operating conditions, thereby improving the real-time performance and reliability of fault diagnosis.

[0083] The BILSTM network and KAN model will be introduced below.

[0084] BILSTM network: Reconstructs drilling machine fault time series data after wavelet transform, convolutional autoencoder, and inverse wavelet transform. The information is input into the composite neural network BILSTM-KAN for fault identification. The BILSTM network model consists of bidirectional LSTMs. The forget gate in the LSTM determines how much past information should be forgotten at the current time step. The formula related to the forget gate is expressed as:

[0085] Where σ is the sigmoid activation function, used to scale the input values ​​to the range [0,1]; h t-1 W represents the hidden state of the previous time step; f b is the weight matrix for the forget gate, applied to the hidden state of the previous time step and the input of the current time step; f For the bias term of the forget gate; f t This is the output of the forget gate.

[0086] The input gate in an LSTM determines how much new information should be added to the memory cell at the current time step. The formula for the input gate is as follows:

[0087] Among them, W i and b i Let i be the weight matrix and bias term of the input gate, respectively. tThis is the output of the input gate. Simultaneously, it contains information about new candidate memory cells. Using the tanh function to generate the value ensures that the value is in the range [-1, 1]. The relevant formula is expressed as:

[0088] Among them, W c and b c These are represented as the weight matrix and bias term of the candidate memory units, respectively. Based on the information of the candidate memory units... The formula for updating memory units is as follows:

[0089] Among them, c t-1 It represents the state of the memory cell at the previous moment.

[0090] The output gate in an LSTM determines the output information at the current time step, and the relevant formula is expressed as follows:

[0091] Among them, W o and b o These are represented as the weight matrix and bias term of the output gate, respectively.

[0092] The final hidden state information is typically scaled using the tanh function to represent the state of the memory cell, and the relevant formula is expressed as: h t =o t ·tanh(c t ).

[0093] Through the above process, drilling machine fault timing data is obtained. positive hidden state With reverse hidden state The final hidden state output by BILTSM These hidden states contain the time series of drilling machine failures. The characteristics are then input into the KAN network for further processing and fault identification.

[0094] KAN Model: In KAN, the weight parameters are replaced by univariate functions, typically represented by B-spline functions. The univariate function is represented as φ. j,i , where j and i represent the indices of the output and input, respectively. For the input x of each layer... i Calculate and output y j The calculation formula is:

[0095] The deep structure of KAN is implemented by stacking multiple layers of univariate functions. The output of each layer serves as the input of the next layer. Assuming there are L layers, the computation of each layer is as follows:

[0096] in, It is the output of the (l+1)th layer. It is the output of the l-th layer, n l It is the number of nodes in the l-th layer. It is a univariate function of the l-th layer. The final output of KAN is given by the calculation results of the last layer, and the relevant calculation formula is expressed as:

[0097] Where Φ represents the spline function matrix, y output This indicates the result of fault identification.

[0098] Based on the same inventive concept, this application also provides a drilling machine fault diagnosis device for implementing the drilling machine fault diagnosis method under complex working conditions described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the drilling machine fault diagnosis device under complex working conditions provided below can be found in the limitations of the drilling machine fault diagnosis method under complex working conditions described above, and will not be repeated here.

[0099] In one exemplary embodiment, a fault diagnosis device for drilling machines under complex working conditions is provided, comprising: a dataset construction module, a data conversion module, a noise reduction module, a reconstruction module, a training module, an acquisition module, and an application module.

[0100] The dataset construction module is used to construct a drilling machine fault dataset containing interference factors under complex working conditions; the drilling machine fault dataset includes pressure time series data under normal conditions and pressure time series data under different fault types.

[0101] The data conversion module is used to convert the pressure time series data in the drilling machine fault dataset into a time-frequency image using wavelet forward transform.

[0102] A denoising module is used to denoise the time-frequency image using a convolutional autoencoder.

[0103] The reconstruction module is used to reconstruct the denoised time-frequency image into pressure time-series data through inverse wavelet transform, thereby obtaining the reconstructed drilling machine fault dataset.

[0104] The training module is used to train a composite neural network using the reconstructed drilling machine fault dataset. The composite neural network includes a bidirectional long short-term memory network and a KAN model. The bidirectional long short-term memory network is used to extract features from the pressure time series data to obtain the hidden state. The KAN model is used to identify drilling machine faults based on the hidden state and output the drilling machine fault diagnosis results.

[0105] The acquisition module is used to sequentially process the real-time acquired pressure time series data through wavelet forward transform, convolutional autoencoder, and inverse wavelet transform to obtain real-time reconstructed pressure time series data.

[0106] The application module is used to perform fault diagnosis on the drilling machine based on real-time reconstructed pressure time series data and a trained composite neural network, and output real-time fault diagnosis results for the drilling machine.

[0107] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram is shown in Figure 8. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores real-time drilling machine fault diagnosis results. The I / O interfaces of the computer device are used for information exchange between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a drilling machine fault diagnosis method under complex working conditions.

[0108] Those skilled in the art will understand that the structure shown in FIG8 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0109] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0110] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0111] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0112] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0113] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0114] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0115] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for diagnosing faults in drilling machines under complex working conditions, characterized in that, include: A drilling machine fault dataset containing interference factors under complex working conditions is constructed; the drilling machine fault dataset includes pressure time series data under normal conditions and pressure time series data under different fault types; The pressure time series data in the drilling machine fault dataset is converted into a time-frequency image using wavelet forward transform; The time-frequency image is denoised using a convolutional autoencoder; The denoised time-frequency images are reconstructed into pressure time-series data by inverse wavelet transform, thus obtaining the reconstructed drilling machine fault dataset. A composite neural network is trained using the reconstructed drilling machine fault dataset. The composite neural network includes a bidirectional long short-term memory network and a KAN model. The bidirectional long short-term memory network is used to extract features from the pressure time series data to obtain the hidden state. The KAN model is used to identify drilling machine faults based on the hidden state and output the drilling machine fault diagnosis results. The real-time collected pressure time series data is sequentially processed through wavelet forward transform, convolutional autoencoder and inverse wavelet transform to obtain real-time reconstructed pressure time series data; Based on real-time reconstructed pressure time series data, a trained composite neural network is used to diagnose faults in the drilling machine and output real-time fault diagnosis results.

2. The fault diagnosis method for drilling machines under complex working conditions according to claim 1, characterized in that, Construct a drilling machine fault dataset that includes interference factors under complex working conditions, specifically including: Collect raw pressure time series data of the drilling machine under normal conditions and different fault types; Gaussian noise and equal noise are combined to form composite noise to simulate the interference faced by the drilling machine under complex working conditions. Among them, Gaussian noise represents random environmental interference and unpredictable fluctuations inside the equipment, while equal noise represents systematic and uniformly distributed external interference. The composite noise is superimposed onto the original pressure time series data to generate a drilling machine fault dataset containing interference factors under complex working conditions.

3. The fault diagnosis method for drilling machines under complex working conditions according to claim 2, characterized in that, The expression for the Gaussian noise is: I G (t)=μ+η·Z(t); In the formula, I G Z(t) represents Gaussian noise at time t, μ is the mean of the normal distribution, η is the standard deviation of the normal distribution, and Z(t) is a standard normally distributed random variable.

4. The fault diagnosis method for drilling machines under complex working conditions according to claim 2, characterized in that, The expression for the equalization noise is: I E (t)=a+(b-a)·U(t); In the formula, I E (t) represents the equalization noise at time t, a and b are the upper and lower limits of the target interval, respectively, and U(t) is a standard uniformly distributed random number in the interval [0,1].

5. The fault diagnosis method for drilling machines under complex working conditions according to claim 1, characterized in that, Different fault types include: loss of A-type seal ring, loss of B-type seal ring, damage to the return accumulator, abnormal damper hole, abnormal flow in the damper circuit, and low charge in the high-voltage accumulator.

6. The fault diagnosis method for drilling machines under complex working conditions according to claim 1, characterized in that, The hidden states include: positive hidden states and negative hidden states.

7. A fault diagnosis device for drilling machines under complex working conditions, characterized in that, The drilling machine fault diagnosis device under complex working conditions includes: The dataset construction module is used to construct a drilling machine fault dataset containing interference factors under complex working conditions; the drilling machine fault dataset includes pressure time series data under normal conditions and pressure time series data under different fault types; The data conversion module is used to convert the pressure time series data in the drilling machine fault dataset into a time-frequency image using wavelet forward transform; A denoising module is used to denoise the time-frequency image using a convolutional autoencoder; The reconstruction module is used to reconstruct the denoised time-frequency image into pressure time-series data through inverse wavelet transform, thereby obtaining the reconstructed drilling machine fault dataset. The training module is used to train a composite neural network using the reconstructed drilling machine fault dataset. The composite neural network includes a bidirectional long short-term memory network and a KAN model. The bidirectional long short-term memory network is used to extract features from the pressure time series data to obtain the hidden state. The KAN model is used to identify drilling machine faults based on the hidden state and output the drilling machine fault diagnosis results. The acquisition module is used to process the real-time acquired pressure time series data through wavelet forward transform, convolutional autoencoder and inverse wavelet transform in sequence to obtain real-time reconstructed pressure time series data; The application module is used to perform fault diagnosis on the drilling machine based on real-time reconstructed pressure time series data and a trained composite neural network, and output real-time fault diagnosis results for the drilling machine.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the drilling machine fault diagnosis method under complex working conditions as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the drilling machine fault diagnosis method under complex working conditions as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the drilling machine fault diagnosis method under complex working conditions as described in any one of claims 1-6.

Images