Marine riser fault detection methods

Abstract

This invention provides a method for detecting faults in marine risers, relating to the field of marine engineering monitoring technology. The method includes the following steps: Step S1: Establishing a simulation model of the marine riser through finite element analysis and performing simulation calculations to obtain simulation data; Step S2: Building a deep learning model based on a self-attention-scalar long short-term memory network and training the deep learning model using the simulation data obtained in Step S1; Step S3: Constructing a generative adversarial domain reconstruction transfer network, building a generator and a discriminator, and performing adversarial training on the generator and discriminator using real and simulated data respectively; Step S4: Using the generator as a preprocessing part of the real data, and performing fault detection using the deep learning model trained in Step S2. This invention achieves the synergistic integration of multiple technologies and solves the shortcomings of existing methods for fault detection in marine risers.

Description

Technical Field

[0001] This invention relates to the field of marine engineering monitoring technology, specifically a method for detecting faults in marine risers. Background Technology

[0002] Marine risers, as transmission pipelines connecting subsea production systems and floating structures, are a crucial component of deep-sea oil and gas field development systems. Due to the time-varying nature of deep-sea environmental loads and the complexity of the interaction between risers and seabed soil, riser safety presents unique challenges and significant uncertainties. Furthermore, marine risers are not only affected by natural environmental factors such as seawater corrosion, but also suffer from weakened external protective layers due to long-term weathering. Additionally, transporting high-temperature, high-pressure liquefied petroleum gas within the riser accelerates internal wear and tear. Under the combined effect of these factors, a riser failure can jeopardize the safety of the entire deep-sea oil and gas field system and may even lead to a series of serious accidents.

[0003] Therefore, fault detection of marine risers is essential. However, existing technologies in this field have limitations in practical applications. Specifically, existing riser designs face challenges such as discrepancies between assumed and actual environmental loads, limitations of riser calculation and analysis models, uncertainties in riser damage, and potential deviations between riser installation and actual conditions. Therefore, preliminary design alone cannot accurately reflect the true loads and stresses acting on the riser. Common riser fault detection methods include sensor technology, fiber optic sensing technology, sonar detection technology, UAVs and remote sensing technology, data analysis, and pattern recognition algorithms. For example, the patent with publication number CN106697209A, which describes a deep-water tension leg platform device and its real-time riser monitoring method, employs sensor deployment. With the rise and development of deep learning technology, data-driven methods are making riser fault detection more intelligent. However, based on practical experience, these methods still have several shortcomings, such as poor model generalization ability, with existing deep learning models showing a significant performance drop when applied across different scenarios; high data collection and labeling costs, requiring data to be recollected for each different scenario; and the need to label the samples, which requires a lot of time and effort. Furthermore, in actual engineering, it is difficult to directly obtain risers with a certain type of fault. In such cases, the labeling is based on theoretical calculations, such as traditional peak analysis, which further challenges the accuracy of the results.

[0004] In summary, this invention provides a method for detecting faults in marine risers. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting faults in marine risers, in order to address the shortcomings of existing methods for detecting faults in marine risers as mentioned in the background art.

[0006] This invention is achieved using the following technical solution:

[0007] A method for detecting faults in marine risers includes the following steps:

[0008] Step S1: Establish a simulation model of the marine riser through finite element analysis and perform simulation calculations to obtain simulation data;

[0009] Step S2: Build a deep learning model based on a self-attention-scalar long short-term memory network, and train the deep learning model using the simulated data obtained in step S1;

[0010] Step S3: Construct a generative adversarial domain reconstruction transfer network, build a generator and a discriminator, and conduct adversarial training on the generator and discriminator using real data and simulated data respectively;

[0011] Step S4: Use the generator as a preprocessing part of the real data and use the deep learning model trained in step S2 for fault detection.

[0012] Further, step S1 includes the following sub-steps:

[0013] Step S1-1: Using finite element analysis, establish a simulation model of the marine riser based on its parameters;

[0014] Step S1-2: Define boundary conditions;

[0015] Step S1-3: Determine the monitoring points and set the fault type;

[0016] Step S1-4: Perform simulation calculations on the marine riser simulation model to obtain the vibration response signals of multiple specified sampling points of the marine riser under different working conditions;

[0017] Step S1-5: Preprocess the simulation data obtained in step S1-4 and divide the vibration response signal into a (batch, 1, seq) format.

[0018] Step S1-6: Perform label processing on the simulated data obtained in step S1-4.

[0019] Further, step S2 includes the following sub-steps:

[0020] Step S2-1: Combine the self-attention mechanism with a neural network module containing a scalar long short-term memory network to obtain a deep learning model;

[0021] Step S2-2: Add a location detection head and a fault detection head to the deep learning model obtained in step S2-1, wherein the location detection head predicts the location of the fault and the fault detection head predicts the type of fault.

[0022] Step S2-3: Input the simulated data obtained in step S1 into the deep learning model to train the deep learning model.

[0023] Further, step S3 includes the following sub-steps:

[0024] Step S3-1: Define the set of real data as the target domain and the set of simulated data as the source domain. Use a generative adversarial network to align the target domain data with the source domain data.

[0025] Step S3-2: Based on the target domain data and the source domain data, a generator and a discriminator are built using a one-dimensional convolutional neural network.

[0026] Furthermore, in step S4, the generator parameters are frozen, real data is input into the generator to obtain data that aligns with the feature domain of the simulated data, and then the data is input into the deep learning model for fault monitoring.

[0027] Furthermore, in step S4, a confidence threshold is set for both the detection results of the position detection head and the detection results of the fault detection head. Detection results below the confidence threshold are not processed, while detection results above the confidence threshold are judged as the corresponding fault mode.

[0028] Furthermore, in step S1-1, the parameters of the marine riser include size, material, and connection method; in step S1-2, the boundary conditions include the installation and constraint method of the marine riser, as well as the action mode and magnitude of the external loads on the marine riser.

[0029] Furthermore, in steps S1-6, the label corresponding to the vibration response signal has two dimensions. The first dimension is the sampling position, represented by numbers 0 to n-1, where n is the number of sampling points. The second dimension is the fault type, including no fault, corrosion, crack, deformation, and mixed type faults.

[0030] Furthermore, in steps S2-3, during training, a dynamic weighting method is used to balance the two detection heads. The initial loss weight of the position detection head is set to 1, and the initial loss weight of the fault detection head is set to 0.5. The loss weight of the position detection head decreases linearly to 0.2 during training, and the loss weight of the fault detection head increases linearly to 1 during training.

[0031] Further, in step S3-2, a batch of data is randomly selected from the source domain and input into the discriminator for training, at which time the generator parameters are frozen; then, the discriminator parameters are frozen, the generator parameters are unfrozen, a batch of data is randomly selected from the target domain and input into the generator, and the data generated by the generator is input into the discriminator for training; in this step, the above training process needs to be repeated until the preset conditions are met.

[0032] The beneficial effects achieved by this invention are:

[0033] This invention provides a method for detecting faults in marine risers, encompassing the entire process of simulation modeling, network architecture design, and training strategies, forming a complete technical closed loop. Specifically, it achieves the synergistic integration of multiple technologies by employing finite element simulation to acquire data, building a deep learning model based on a self-attention-scalar long short-term memory network, and constructing a generative adversarial domain reconstruction and transfer network. Based on this, the invention addresses the inherent deficiencies of traditional fault detection methods in data acquisition, model recognition, and generalization capabilities, effectively improving the accuracy of marine riser fault detection. This provides advanced technical support for the safe operation and maintenance of marine risers, thereby ensuring the efficient and reliable operation of marine oil and gas production systems. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the marine riser fault detection method according to an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of the adversarial training process in the marine riser fault detection method described in this embodiment of the invention. Detailed Implementation

[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0037] Example 1

[0038] This embodiment provides a method for detecting faults in marine risers. Please refer to [link / reference]. Figure 1 It includes the following steps:

[0039] Step S1: Establish a simulation model of the marine riser through finite element analysis and perform simulation calculations to obtain simulation data.

[0040] Specifically, it includes the following sub-steps:

[0041] Step S1-1: Using finite element analysis, a simulation model of the marine riser is established based on its parameters, including dimensions, materials, and connection methods. In this embodiment, the specific parameters are the riser's wall thickness, diameter, length, and the material's elastic modulus, Poisson's ratio, and yield strength.

[0042] Step S1-2: Define boundary conditions, which include the installation and constraint methods of the marine riser, as well as the mode and magnitude of the external loads acting on the marine riser. The installation methods of the marine riser include rigid connections, flexible connections, and other special connection methods; in this example, a flexible connection is selected. External loads include ocean waves and currents. In this embodiment, the JONSWAP spectrum is used to simulate irregular ocean waves. The effective wave height is set to 5 meters, the average period to 8 seconds, and the wave direction to form a 45° angle with the platform axis. The ocean current is simplified as a stratified uniform flow, with a surface current velocity of 1.2 m / s, decreasing linearly with depth, reaching 0.5 m / s at the seabed. The current direction is the same as the wave direction.

[0043] Steps S1-3: Determine monitoring points and set fault types, where commonly considered fault types include corrosion, cracking, and deformation. In this embodiment, two points are selected each at the top, middle, and bottom of the marine riser as sampling points, and three different values ​​are set for corrosion, cracking, and deformation to simulate different degrees of damage. Corrosion is achieved by modifying the geometry of the deep learning model. Specifically, a small region needs to be created at the corresponding location in the marine riser model, and the wall thickness of this region needs to be reduced to simulate material loss caused by corrosion. Cracks are simulated using the fracture mechanics module in ABAQUS. Specifically, a crack surface needs to be defined in the model, and then the crack initiation location, propagation direction, and propagation length need to be specified. Deformation is simulated by changing the geometry of the model. Specifically, a concave geometry needs to be created at the corresponding location.

[0044] Steps S1-4: Simulate the marine riser model using dedicated software to obtain vibration response signals from multiple specified sampling points under different operating conditions. In this embodiment, the dedicated software used is ABAQUS.

[0045] Step S1-5: Preprocess the simulated data obtained in Step S1-4, dividing the vibration response signal into a (batch, 1, seq) format. The first dimension, batch, represents the batch size, indicating the number of data stacks during a single forward propagation in training. The second dimension, 1, indicates that only one sampling point's data is input each time. The third dimension, seq, represents the sampling of the vibration response signal. In this embodiment, the number of data stacks is 64, and the number of vibration response signal samples is 2000, so it is written as (64, 1, 2000), and the data is packaged randomly.

[0046] Steps S1-6: The simulated data obtained in steps S1-4 is labeled. The labels corresponding to the vibration response signals have two dimensions. The first dimension is the sampling location, represented by numbers 0 to n-1, where n is the number of sampling points. The second dimension is the fault type, including no fault, corrosion, crack, deformation, and mixed fault types. In this embodiment, the first dimension is represented by numbers 0 to 5. In the second dimension, 0 represents no fault, 1, 2, and 3 represent low, medium, and high levels of corrosion, respectively; 4, 5, and 6 represent low, medium, and high levels of crack, respectively; 7, 8, and 9 represent low, medium, and high levels of deformation, respectively; 10 represents a mixed corrosion and crack fault; 11 represents a combined corrosion and deformation fault; 12 represents a combined crack and deformation fault; and 13 represents the presence of all three types of faults.

[0047] Step S2: Build a deep learning model based on a self-attention-scalar long short-term memory network, and train the deep learning model using the simulated data obtained in step S1.

[0048] Specifically, it includes the following sub-steps:

[0049] Step S2-1: Combine the self-attention mechanism with a neural network module containing a scalar long short-term memory network to obtain a deep learning model. Wherein:

[0050] Self-attention mechanisms can capture long-range dependencies, and are even better at capturing relationships between elements that are far apart in a sequence; furthermore, different input sequences generate different attention weights, which enhances the capabilities of deep learning models. The core formula of the self-attention mechanism is:

[0051]

[0052] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k The dimension of the key vector.

[0053] Unlike traditional Long Short-Term Memory (LSTM) networks, scalar LSTM networks introduce exponential gating, normalization, and stabilizers. Exponential gating allows the input and forget gates to have exponential activation functions, enabling the model to dynamically adjust information storage and forgetting based on input, effectively solving the problem of traditional LSTM networks' inability to flexibly revise storage decisions. Specifically:

[0054] The forward propagation process of a traditional Long Short-Term Memory (LSTM) network is as follows:

[0055] c t =f t ⊙c t-1 +it ⊙z t ;

[0056]

[0057] Among them, c t Represents cell state, h t Represents the hidden state, z t Represents cell input, i t Represents the input gate, f t Represents the Gate of Oblivion, o t Represents the output gate; the "~" symbol indicates an intermediate state in the calculation of the corresponding parameter; W z W i W f W o The input weight vectors correspond to the input and cell input, input gate, forget gate, and output gate, respectively, with weights R. z R i R f R o and correspond to the recurrent weights between the hidden state and the cell input, input gate, forget gate, and output gate, respectively; b z b i b f b o And are the corresponding bias terms; ψ()=tanh(),σ(x)=1 / (1+exp(-x)).

[0058] In scalar long short-term memory networks, exponential gating, normalized states, and stable states are added to traditional long short-term memory networks. The forward propagation process is as follows:

[0059] c t =f t ⊙c t-1 +i t ⊙z t ;

[0060] n t =f t ⊙n t-1 +i t ;

[0061]

[0062] By adding a stable state, we can achieve the following:

[0063] m t =max(log(f) t )+m t-1 ,log(i t )),

[0064] i′ t=exp(log(i t )-m t ),

[0065] f′ t =exp(log(f t )+m t-1 -m t ),

[0066] Where n t It is the normalized state, m t It is a stable gating state to prevent value overflow caused by the exponential activation function. The meanings of the other letters are the same as in the formula of the forward propagation process of traditional long short-term memory networks.

[0067] Based on the core mechanism of Scalar Long Short-Term Memory (SLSM) networks, layer normalization (LN), causal convolution, swish activation function, Gaussian error linear unit (GELU activation function), block diagonal layers, and linear projection layers are added and connected through residual structures to construct a neural network module with Scalar SLSM. In this embodiment, when combining the self-attention mechanism with the neural network module with Scalar SLSM, training is performed on an RTX 4070 graphics card, using one self-attention module and two neural network modules with Scalar SLSM.

[0068] Step S2-2: Add a location detection head and a fault detection head to the deep learning model obtained in Step S2-1. Both the location detection head and the fault detection head contain activation function layers, random dropout layers, and linear layers, and are identical except for the output dimension. The location detection head is used to predict the location of the fault. In this embodiment, the predicted value dimension is (64, 6), where 64 represents that 64 samples are stacked in one training session, and 6 represents that there are 6 sampling point locations. Each specific value represents the confidence level of the location prediction. The fault detection head is used to predict the type of fault. In this embodiment, the predicted value dimension is (64, 14), where 64 represents that 64 samples are stacked in one training session, and 14 represents that there are 14 sampling point locations. Each specific value represents the confidence level of the fault prediction. Step S2-3: Input the simulated data obtained in Step S1 into the deep learning model to train the deep learning model. During training, a dynamic weighting method is used to balance the two detectors. The initial loss weight for the position detector is set to 1, and the initial loss weight for the fault detector is set to 0.5. The loss weight for the position detector decreases linearly to 0.2 during training, while the loss weight for the fault detector increases linearly to 1. In this embodiment, training stops if the accuracy and loss do not decrease for 10 consecutive epochs after 200 or 100 epochs. The training loss function includes position loss and classification loss, and the formula is as follows:

[0069] Ltotal =αL pos +βL cls ;

[0070] Among them, L pos For location classification loss, L cls For fault classification loss, the cross-entropy loss function is used; α and β are dynamic weights.

[0071] The specific formula is as follows:

[0072] L pos =L CE (f pos (x), y pos );

[0073] L cls =L CE (f cls (x), y fal );

[0074] Among them, L CE f represents the cross-entropy loss function. pos (·) represents the function of the network from input to the position detection head output, y pos For location labels, y fal For fault type labels;

[0075]

[0076] Where t is the current training round, and T is the total number of training rounds.

[0077] Step S3: Construct a generative adversarial domain reconstruction transfer network, build a generator and a discriminator, and conduct adversarial training on the generator and discriminator using real data and simulated data respectively.

[0078] Specifically, it includes the following sub-steps:

[0079] Step S3-1: Define the set of all real data as the target domain and the set of all simulated data as the source domain. Use a generative adversarial network to align the target domain data with the source domain data.

[0080] Step S3-2: Based on the target domain data and source domain data, a generator and discriminator are built using a one-dimensional convolutional neural network. The adversarial training process is as follows: Figure 2 As shown, in this embodiment, ResNet18 is selected as the generator and discriminator. The loss function of the discriminator is the binary cross-entropy loss function, and the loss function of the generator is the negative of the cross-entropy function. The parameters of the generator and discriminator are not shared.

[0081] During training, it is agreed that label 0 represents the real data in the source domain and label 1 represents the generated data or target domain data. First, the generator parameters are frozen, and a batch of data is randomly selected from the source domain and input into the discriminator for training. The data format of this batch of data is the same as that of the simulated data, which is (64, 1, 2000). Then, the output dimension of the discriminator is obtained as (64, 2), which represents the probability prediction of each sample belonging to the two classes (real data in the source domain / generated data). Then, the binary cross-entropy loss function is used to calculate the error between the discriminator output and the real domain label (label 0 corresponding to the source domain data). The training goal at this stage is to minimize the discrimination error, that is, to improve the discriminator's ability to accurately distinguish between real and fake data.

[0082] Then, the parameters of the discriminator are frozen and the parameters of the generator are unfrozen. A batch of data is randomly selected from the target domain, with the format (64, 1, 2000). This batch of target domain data is input into the generator, and the generator outputs data in the same format (64, 1, 2000). The generator's role here is to attempt to learn the distribution characteristics of the target domain data, thereby generating samples that are closer to the real data. The data generated by the generator is then input into the discriminator, and the discriminator's output dimension is (64, 2), which represents the discriminator's probability prediction that each sample belongs to one of two classes (real data from the source domain / generated data from the target domain). The binary cross-entropy loss function is then used to calculate the error between the discriminator's output and the domain label. At this point, the goal of training the generator is to maximize the discrimination error, that is, to make it more difficult for the discriminator to distinguish between the data generated by the generator and the real target domain data, thereby improving the generator's ability to deceive the discriminator.

[0083] During training, the above steps need to be repeated until the loss no longer decreases for 10 consecutive rounds after 200 rounds or 100 rounds, at which point training can be terminated.

[0084] Step S4: Freeze the generator parameters, save the parameters of the trained generator, and use the generator as a preprocessing part of the real data. When using the deep learning model built in Step S2 for fault detection, first input the real data into the generator to obtain data aligned with the simulated data feature domain, and then input it into the deep learning model for fault monitoring. A confidence threshold is set for the position head detection results and the fault detection head results; results below the threshold are not processed, and results above the confidence threshold are determined to be the corresponding fault mode. This scheme completes domain transfer and eliminates the need to retrain the deep learning model built in Step 2.

[0085] The aforementioned marine riser fault detection method achieves the synergistic integration of multiple technologies by employing finite element simulation to acquire data, building a deep learning model based on a self-attention-scalar long short-term memory network, and constructing a generative adversarial domain reconstruction transfer network. This results in several beneficial effects, specifically:

[0086] Firstly, using simulation to acquire a large amount of data, and using the high-fidelity data generated by finite element simulation as the training basis, helps to solve the problem of the difficulty in obtaining and the scarcity of actual marine riser fault data.

[0087] Secondly, a deep learning model based on self-attention-scalar long short-term memory network was built, which has the advantage of high recognition accuracy. Unlike CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network), the self-attention mechanism allows the model to better construct global relationships and establish global connections between sampling points. The scalar long short-term memory network has both fast training and fast reasoning capabilities, can dynamically adjust stored information, and capture information from multiple angles. That is, the model can dynamically adjust its behavior according to different inputs, which enhances the model's generalization ability. At the same time, the scalar long short-term memory network also introduces a steady state to avoid the gradient explosion problem during training.

[0088] Third, to address the problem of poor recognition results when directly inputting real data due to domain differences between simulated and real data, a generative adversarial network domain transfer method is adopted. This method can perform fault detection even when real data is unlabeled, and does not require retraining of the recognition model, which meets the needs of practical use.

[0089] It should be noted that the parts not described in detail or in elaboration in the above solutions are all prior art and do not constitute improvements made by this invention to existing technology, nor are they within the protection scope of this invention's technical solutions. Therefore, they will not be elaborated upon further in this document. Of course, the above content is merely a preferred embodiment of this invention and should not be considered as limiting the scope of the embodiments of this invention. This invention is also not limited to the above examples; equivalent changes and improvements made by those skilled in the art within the substantial scope of this invention should all fall within the patent coverage of this invention.

Claims

1. A method for detecting faults in marine risers, characterized in that, Includes the following steps: Step S1: Establish a simulation model of the marine riser through finite element analysis and perform simulation calculations to obtain simulation data; Step S2: Build a deep learning model based on a self-attention-scalar long short-term memory network, and train the deep learning model using the simulated data obtained in step S1; Step S3: Construct a generative adversarial domain reconstruction transfer network, build a generator and a discriminator, and conduct adversarial training on the generator and discriminator using real data and simulated data respectively; Step S4: Use the generator as a preprocessing part of the real data and use the deep learning model trained in step S2 for fault detection. Step S1 includes the following sub-steps: Step S1-1: Using finite element analysis, establish a simulation model of the marine riser based on its parameters; Step S1-2: Define boundary conditions; Step S1-3: Determine the monitoring points and set the fault type; Step S1-4: Perform simulation calculations on the marine riser simulation model to obtain the vibration response signals of multiple specified sampling points of the marine riser under different working conditions; Step S1-5: Preprocess the simulation data obtained in step S1-4 and divide the vibration response signal into a (batch, 1, seq) format. Step S1-6: Perform label processing on the simulated data obtained in step S1-4; In step S1-1, the parameters of the marine riser include size, material and connection method; in step S1-2, the boundary conditions include the installation and constraint method of the marine riser, as well as the action mode and magnitude of the external load on the marine riser. In steps S1-6, the label corresponding to the vibration response signal has two dimensions. The first dimension is the sampling position, represented by numbers 0 to n-1, where n is the number of sampling points. The second dimension is the fault type, including no fault, corrosion, crack, deformation, and mixed type faults. Step S3 includes the following sub-steps: Step S3-1: Define the set of real data as the target domain and the set of simulated data as the source domain. Use a generative adversarial network to align the target domain data with the source domain data. Step S3-2: Based on the target domain data and the source domain data, a generator and discriminator are built using a one-dimensional convolutional neural network; In step S3-2, a batch of data is randomly selected from the source domain and input into the discriminator for training. At this time, the generator parameters are frozen. Then, freeze the discriminator parameters, unfreeze the generator parameters, randomly select a batch of data from the target domain and input it into the generator, and then input the data generated by the generator into the discriminator for training. This step requires repeating the training process until the preset conditions are met.

2. The method for detecting faults in marine risers according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S2-1: Combine the self-attention mechanism with a neural network module containing a scalar long short-term memory network to obtain a deep learning model; Step S2-2: Add a location detection head and a fault detection head to the deep learning model obtained in step S2-1, wherein the location detection head predicts the location of the fault and the fault detection head predicts the type of fault. Step S2-3: Input the simulated data obtained in step S1 into the deep learning model to train the deep learning model.

3. The method for detecting faults in marine risers according to claim 1, characterized in that: In step S4, the generator parameters are frozen, real data is input into the generator to obtain data that aligns with the feature domain of the simulated data, and then the data is input into the deep learning model for fault detection.

4. The method for detecting faults in marine risers according to claim 3, characterized in that: In step S4, a confidence threshold is set for the detection results of both the position detection head and the fault detection head. Detection results below the confidence threshold are not processed, while detection results above the confidence threshold are identified as the corresponding fault mode.

5. The method for detecting faults in marine risers according to claim 2, characterized in that: In steps S2-3, during training, a dynamic weighting method is used to balance the two detection heads. The initial loss weight of the position detection head is set to 1, and the initial loss weight of the fault detection head is set to 0.

5. The loss weight of the position detection head decreases linearly to 0.2 during training, and the loss weight of the fault detection head increases linearly to 1 during training.

Images