In-situ identification method and system for seabed sediments based on double-source differential vibration characteristics

Abstract

This invention relates to a method and system for in-situ identification of seabed sediments based on dual-source differential vibration characteristics. The method includes: simultaneously acquiring vibration signals at the power source location and the location of the working component of the engineering equipment; using the vibration signal acquired at the power source location as a reference signal and the vibration signal acquired at the working component location as a hybrid response signal; constructing a dual-source differential correction model to obtain a correction signal; extracting multidimensional sediment vibration characteristics from the correction signal to characterize the dynamic response of the sediments; and using a sediment category identification model based on the multidimensional sediment vibration characteristics to identify the seabed sediment category in situ. This invention achieves continuous in-situ identification of seabed sediments without altering existing seabed drilling, penetration, or piling operations, thereby providing reliable sediment information support for engineering activities such as seabed landslide identification and risk assessment, seabed resource exploration, and seabed infrastructure construction.

Description

Technical Field

[0001] This invention relates to the fields of marine engineering and geotechnical engineering, and in particular to a method and system for in-situ identification of seabed sediments based on dual-source differential vibration characteristics. Background Technology

[0002] Submarine landslides are widely distributed on continental shelves and slopes. Their occurrence is typically caused by the combined effects of external factors such as earthquakes, wave and ocean current disturbances, sea-level changes, and engineering activities, as well as the inherent conditions of the seabed sediments. After external disturbances apply additional loads or disturbances to the sediments, the physical and mechanical properties, structural state, and spatial distribution of the sediments become key internal factors determining the occurrence and evolution of landslides. Locally weak layers, loosely structured layers, or layers with abrupt property changes are more prone to strength attenuation and structural failure, thus inducing large-scale or gradual landslides. Landslides can damage submarine engineering facilities and pipelines, impact ecosystems, and potentially trigger secondary disasters. They also cause sediment fragmentation, remodeling, and redeposition, resulting in significant vertical and lateral changes in the properties of the landslide area, increasing the complexity of subsequent assessments. Therefore, understanding the types of sediments and their variation with depth and space is a crucial foundation for landslide identification and risk assessment. Besides landslides, sediment information is also needed in submarine engineering surveys, foundation construction, and resource exploration. Drilling, penetration, and piling operations enhance sediment heterogeneity, revealing hidden thin and weak interlayers that are easily missed by conventional discrete sampling or single-parameter testing. Existing methods mainly include static cone penetration testing (PCT), core drilling, and laboratory geotechnical testing. Core drilling is a discrete sampling method, which cannot continuously reflect the detailed characteristics that change rapidly with depth, and it lacks representativeness in areas with significant disturbance. PCT relies heavily on a few indicators such as cone tip resistance, which can easily lead to multiple interpretations when facing interlayers, abrupt changes in grain size, or weak interlayers, and the stability of the results is affected by differences in working conditions and equipment. Although laboratory geotechnical testing can obtain parameters, it is time-consuming, costly, and cannot provide real-time feedback on structural remodeling and strength decay during construction. At the same time, a large number of vibration signals during engineering are often filtered out as interference, failing to explore their value in sediment dynamic response.

[0003] Therefore, without changing the existing engineering operation process, if the vibration signals generated during the engineering operation can be effectively collected and analyzed, and combined with the response characteristics of sediments under engineering disturbance conditions, in-situ identification of seabed sediments can be achieved, which will help to make up for the shortcomings of traditional exploration methods in terms of continuity and fine identification. Summary of the Invention

[0004] To address the shortcomings of existing seabed sediment exploration and identification methods, such as difficulty in in-situ acquisition of sediment types during engineering operations, easy omission of hidden thin and weak interlayers, and ineffective separation and utilization of sediment vibration responses under engineering disturbance conditions, this invention aims to provide a method and system for in-situ identification of seabed sediments based on dual-source differential vibration characteristics. Without altering existing seabed drilling, penetration, or piling procedures, this invention simultaneously acquires and jointly processes vibration signals from different locations during engineering operations. It utilizes a dual-source differential correction model to obtain correction signals and performs feature analysis, achieving in-situ identification of seabed sediments. This provides reliable sediment information support for engineering activities such as seabed landslide identification and risk assessment, seabed resource exploration, and seabed infrastructure construction.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for in-situ identification of seabed sediments based on dual-source differential vibration characteristics, the method comprising the following: Vibration signals were simultaneously collected at the power source location and the working component location of the engineering equipment. The vibration signal collected at the power source location was used as the reference signal, and the vibration signal collected at the working component location was recorded as the hybrid response signal. Construct a dual-source differential correction model: Simultaneously construct a transfer mapping operator with the reference signal and the hybrid response signal as inputs. After processing the reference signal with the transfer mapping operator, obtain the estimated signal of the body vibration transfer component. Subtract the hybrid response signal from the estimated signal of the body vibration transfer component to obtain the correction signal. Extract multidimensional sediment vibration features from the correction signal to characterize the dynamic response of sediments; Based on the multidimensional sediment vibration characteristics, the sediment category identification model is used to identify the category of seabed sediments in situ.

[0006] Furthermore, the transfer mapping operator is implemented using an adaptive filter based on the minimum mean square error criterion, a normalized adaptive filter, correlation modeling, or statistical weighting. In the specific implementation process: the time window is divided according to the working depth interval, and the time window corresponding to the current working depth interval is used as a local update range; within the time window, the transfer mapping relationship is recursively updated using the real-time data of each discrete sampling point.

[0007] Furthermore, the multidimensional sediment vibration characteristics include: time-domain statistical characteristics characterizing vibration amplitude and stability, frequency-domain characteristics characterizing vibration energy distribution characteristics, and complexity stability characteristics characterizing the complexity or stability of vibration response.

[0008] Preferably, the time-domain statistical features include the root mean square value. The peak factor, skewness, and kurtosis are among the features; the frequency domain features include the power spectrum centroid frequency, the frequency band energy ratio, and the dominant frequency energy; the complexity stability features include sample entropy and permutation entropy.

[0009] Furthermore, samples and labels for training the sediment category identification model are constructed: during seabed engineering surveys, penetration or piling operations, the reference signal and mixed response signal are windowed according to the preset operation depth interval to form a time window sequence; dual-source differential correction is performed within each time window to obtain the corresponding correction signal sequence; multi-dimensional sediment vibration features are extracted from the correction signal sequence within each time window, and the multi-dimensional sediment vibration features are spliced ​​together to form a sample feature vector; Simultaneously obtain the sediment category label corresponding to the feature vector of each sample; The training set is composed of all sample feature vectors and their corresponding sediment category labels.

[0010] Secondly, the present invention provides an in-situ identification system for seabed sediments based on dual-source differential vibration characteristics, wherein the system performs the method described above, including: The data acquisition module is used to simultaneously acquire dual-source vibration signals from the power source location and the working component location of the engineering equipment; The signal preprocessing and windowing module is used to preprocess the acquired dual-source vibration signals and window the dual-source vibration signals along the working depth to determine the corresponding time window for each depth segment during continuous drilling; the preprocessing includes detrending and bandpass filtering; The dual-source differential correction module is used to output correction signals for the dual-source vibration signals using a dual-source differential correction model within each time window; The feature construction module is used to extract and construct multidimensional sediment vibration features based on the correction signal; The sediment category identification module is used to predict and output the identification result of the sediment category based on the multidimensional sediment vibration characteristics as input.

[0011] Furthermore, the data acquisition module includes a reference vibration acquisition unit and a hybrid response vibration acquisition unit. The reference vibration acquisition unit is located at the power source of the engineering equipment, and the hybrid response vibration acquisition unit is located at the working component. The hybrid response vibration acquisition unit is a short section structure. The body of the hybrid response vibration acquisition unit is a section of the drill rod. The upper and lower ends are rigidly connected in series with the drill rod and the lower working component, respectively, and are arranged coaxially with the drill rod. The hybrid response vibration acquisition unit has a sealed pressure chamber inside, which integrates a time recorder, a triaxial accelerometer, a data storage unit, and a battery unit. The time recorder is used to record the acquired hybrid response signal according to time, the triaxial accelerometer is used to acquire the triaxial acceleration vibration signal at the location of the working component, and the data storage unit is used to locally cache the data according to time under abnormal conditions (such as communication restrictions).

[0012] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention proposes a dual-source differential correction model, which can dynamically cancel the propagation component of the body vibration under the background of strong mechanical noise from the drilling rig, thereby improving the extraction purity of the sediment dynamic response component. In actual operation, the signal acquired at the location of the working component is not pure sediment information, but rather a superposition of the propagation component of the body vibration and the component generated by the interaction between the working component and the sediment. To address this characteristic, this invention constructs a transfer mapping operator using a reference signal and a mixed response signal as input. The reference signal is modeled to approximate the behavior of the body vibration transmission component at the location of the working component. This component is then canceled out of the mixed response signal using differential correction, resulting in a corrected signal that better reflects the dynamic response characteristics of the sediment. Since the transmission relationship changes with depth, working conditions, and connectivity, an adaptive update method is used to adjust the mapping relationship in real time, ensuring that the cancellation process tracks changes in working conditions and remains relevant.

[0013] (2) In this invention, considering that the whole process is non-stationary, the time window is divided according to the preset operation depth interval. The time window is regarded as quasi-steady state, and the transfer mapping estimation and differential correction are completed in each time window, which can significantly improve the stability of the decoupling process.

[0014] (3) In this invention, the hybrid response vibration acquisition unit can be rigidly connected in series with the drill pipe as a structural component, and realize the synchronous acquisition of dual-source vibration signals. It adopts a short-section structure with threaded connection ends at both ends. After being directly connected in series with the drill pipe, it forms a rigid whole, avoiding the problems of loosening and displacement of externally attached and externally mounted sensors under strong impact and strong vibration conditions, which leads to inconsistent signals. The short-section structure is equipped with a sealed pressure chamber, which encapsulates the triaxial accelerometer, data storage unit, battery unit, time recorder, etc. in the sealed pressure chamber, which can continuously record vibration signals in harsh environments. The hybrid response vibration acquisition unit can export data after being recovered from the seabed. At the same time, the power source position acquisition end (reference vibration acquisition unit) and the working component position acquisition end (hybrid response vibration acquisition unit) are synchronously acquired using a unified time reference, so that the two signals can be accurately aligned in terms of time delay and phase, which is beneficial for subsequent high-precision analysis.

[0015] (4) This invention achieves non-destructive identification during drilling throughout the entire process. Without adding extra procedures and operating time, it enables continuous and real-time scanning of seabed sediment types and solves the problems of discontinuous and thin-layer-missing traditional discrete sampling, thereby improving exploration efficiency and stratification integrity. Traditional methods often rely on coring or discrete sampling. When the sampling interval is large, thin and weak interlayers are easily missed. Even with denser sampling, the construction period and cost will increase significantly. By using the vibration generated during drilling as a signal source, dual-source vibration signals are collected synchronously during the operation and preprocessed, dual-source differentially corrected, and identified. The identification results are generated continuously in time windows and form a continuous profile corresponding to the depth, making it easier to capture changes in thin and weak interlayers. At the same time, continuous stratification information can be obtained without additional downtime sampling, which can be used for real-time on-site decision support and transform the sampling work from comprehensive sampling to key stratum verification sampling, thereby improving efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall system layout and the internal structure of the hybrid response vibration acquisition unit in an embodiment of the present invention.

[0017] Figure 2 This is a flowchart illustrating one embodiment of the in-situ identification method for seabed sediments based on dual-source differential vibration characteristics according to the present invention.

[0018] In the diagram, 1-reference vibration acquisition unit, 2-data processing and analysis center, 3-hybrid response vibration acquisition unit, 4-time recorder, 5-triaxial accelerometer, 6-sealed pressure chamber, 7-data storage unit, and 8-battery unit. Detailed Implementation

[0019] The present invention will be further explained below with reference to the embodiments and accompanying drawings, but this is not intended to limit the scope of protection of this application.

[0020] This invention addresses three main problems in high-intensity mechanical operations on the seabed: First, the response signal of sediments is weak and easily overwhelmed by the strong vibrations and noise generated by the equipment itself; second, when vibrations travel long distances along structures such as drill pipes, they can cause time delays, frequency band selectivity, and even certain nonlinear distortions, making the reference signal unusable for direct subtraction; third, if the timing of dual-source acquisitions is misaligned, even a small phase delay error can cause subsequent cancellation to fail.

[0021] This invention achieves adaptive vibration cancellation and sediment feature separation based on a dual-source physical architecture. The dual-source physical architecture, with its specific spatial topology of "power source location reference acquisition + working component location response acquisition," differs from traditional single-point vibration monitoring or discrete sampling structures. The power source location is used to stably acquire reference signals, while the working component location acquires a mixed response signal containing the body vibration transmission component and the sediment response. In the construction of the dual-source differential correction model, an adaptive filter based on the minimum mean square error criterion is preferentially introduced as a transfer mapping operator, achieving dynamic stripping of nonlinear mechanical noise. Using the dual-source differential correction model, a pure sediment dynamic response signal is extracted in real time from strong mechanical background noise, enabling high signal-to-noise ratio sediment category identification during continuous penetration drilling operations. Simultaneously, drawing on the concept of multi-source fusion, the single vibration index is upgraded to a "multi-dimensional sediment vibration feature" that includes power spectrum centroid frequency, sample entropy, and arrangement entropy.

[0022] In terms of hardware, the position acquisition end of the working component is designed as a detachable short section structure that can be rigidly connected in series with the drill pipe. A sealed pressure chamber is set inside the high-strength alloy steel shell. The sealed pressure chamber integrates a triaxial accelerometer, a data storage unit, a battery unit, and a time recorder, and reserves a data export interface to ensure stable acquisition under high pressure, strong impact, and strong noise environments, and to export data after retrieval. At the same time, the time recorder provides a unified time reference to ensure strict alignment of the two signals at the acquisition level. Considering the overall non-stationary nature of the operation process, this invention performs windowing processing. Within the windowed time window, the transfer mapping estimation and cancellation are completed, and the parameters are updated according to the working conditions of the working depth, so that subsequent feature construction and recognition can stably output a continuous profile that changes with depth.

[0023] During subsea engineering operations, the engineering equipment driven by the power system continuously interacts with the seabed sediments, inevitably generating a mixed response signal that includes both the vibration of the engineering equipment itself and the dynamic response components of the sediments. This invention addresses this by deploying a reference vibration acquisition unit and a mixed response vibration acquisition unit at the power source location and the location of the working component, respectively, to simultaneously acquire dual-source vibration signals. Using the vibration signal acquired at the power source location as a reference signal, a signal transmission mapping relationship is constructed from the power source location to the working component location. This compensates for transmission effects such as amplitude attenuation and phase lag in the reference signal, making it closer to the vibration transmission components reaching the working component location. Differential cancellation is then performed to preserve the sediment response as much as possible, statistically suppressing interference components highly correlated with the vibration of the engineering equipment itself, and making the corrected signal more prominent in terms of the dynamic response characteristics of the sediments.

[0024] Because the vibration signal generated by the power source undergoes nonlinear amplitude attenuation and phase lag when transmitted to the working component location via a long drill pipe, the bulk noise component in the reference signal acquired at the power source location and the mixed response signal acquired at the working component location cannot be directly aligned and subtracted. Therefore, an adaptive filter is used to construct a transfer mapping operator from the power source location to the working component location to perform real-time filtering on the reference signal, simulating the distortion state after transmission through the mechanical structure, thus obtaining an estimated signal of the bulk vibration transmission component. Then, the estimated signal of the bulk vibration transmission component is subtracted from the mixed response signal acquired at the working component location, thereby canceling the mechanical bulk noise in both the time and frequency domains and extracting a purer sediment dynamic response characteristic, which is the correction signal. This yields the dual-source differential correction model. The dual-source differential correction model is a closed-loop signal processing process that first simulates channel distortion and then performs precise cancellation.

[0025] Based on the correction signal, multidimensional sediment vibration features are extracted and processed for identification, thereby enabling continuous in-situ identification of different sediment categories.

[0026] In this invention, a reference vibration acquisition unit located at the power source of the engineering equipment is used to acquire triaxial acceleration vibration signals that reflect the operating state of the engineering equipment itself; a hybrid response vibration acquisition unit located at the working component is used to acquire hybrid response signals that include the vibration of the engineering equipment itself and the dynamic response of the sediment.

[0027] Assume the triaxial acceleration vibration signal acquired at the location of the power source is the reference signal:

[0028] in, - Continuous time variable, in seconds , representing the time progression of vibration signal acquisition and drilling identification; - The reference signal represents the overall triaxial acceleration vibration signal collected at the power source location; - Power source location data collected Acceleration vibration signal in the axial direction; - Power source location data collected Acceleration vibration signal in the axial direction; - Power source location data collected Acceleration vibration signal in the axial direction.

[0029] Assume the triaxial acceleration vibration signal acquired at the location of the working component is a mixed response signal:

[0030] in, -exist The mixed response signal acquired at any time; -The location of the working components was collected in Acceleration vibration signal in the direction; -The location of the working components was collected in Acceleration vibration signal in the direction; -The location of the working components was collected in Acceleration vibration signal in the direction.

[0031] The hybrid response signal can be physically represented as a component of the body vibration transmission. With sediment dynamic response components Superposition relationship:

[0032] in, This refers to the body vibration transmission component that is highly correlated with the operating status of the power system of the engineering equipment and is transmitted to the sediment interface through the working components; This represents the sediment dynamic response component generated by the interaction between the working component and the sediment.

[0033] because and Mixed in Since it cannot be directly separated from the source, this invention uses a reference signal. As input, through passing the mapping operator The body vibration transmission component in the mixed response signal is estimated to obtain the estimated signal of the body vibration transmission component. :

[0034] This is a transfer mapping operator used to describe the mapping relationship between the reference signal and the estimated signal of the body vibration transmission component, representing the signal transmission mapping relationship from the power source location to the working component location. In this embodiment, it is used to characterize the transmission effects such as amplitude attenuation, phase lag, and frequency selectivity generated when mechanical waves propagate downwards along the drill pipe. Therefore, In a statistical approximation sense, it can be regarded as a component of the body vibration transmission. The estimated value.

[0035] In obtaining the estimated signal of the body vibration transmission components Then, it is extracted from the mixed response signal. The two signals are canceled out to obtain the corrected signal. :

[0036] in, This represents the correction signal in the continuous time domain; in the sense of statistical approximation, This can be considered as a component of the sediment dynamic response after stripping away the bulk vibration noise. The estimated value.

[0037] The dual-source differential correction model includes two levels: transfer mapping estimation and differential correction: 1) reference signal Transitive mapping operator Estimate the body vibration transmission components to obtain the estimated signal of the body vibration transmission components. ;2) Then mix the response signals Estimated signal of the vibration transmission component of the body The difference is calculated to obtain the correction signal. Among them, the transfer mapping estimation is used to obtain the estimated signal of the body vibration transfer component, and the differential correction outputs the correction signal, which is the output of the dual-source differential correction model.

[0038] In a preferred embodiment, the transitive mapping operator An adaptive filter based on the minimum mean square error criterion is employed. Considering that the engineering operation progresses continuously along the depth direction, time windows can be divided according to the operation depth interval, and the time window corresponding to the current operation depth interval is taken as a local update range. Within this time window, the transfer mapping relationship is recursively updated using real-time data from each discrete sampling point, thereby obtaining the correction signal corresponding to this time window. As the operation depth advances, the transfer mapping relationship can be dynamically adjusted according to changes in operation depth and working conditions.

[0039] Its discrete form can be expressed as:

[0040]

[0041]

[0042] in, - The index of the sampling points of the discrete time series within the current time window; -No. Discrete reference signals input at each sampling time; -No. Discrete hybrid response signal input at each sampling time; -in the The adaptive filter weight vector at the next iteration is used to characterize the transitive mapping operator. The parameter status; - Weight vector The transpose of the matrix; - The output signal of the adaptive filter, i.e., the first The estimated signal of the vibration transmission component of the working component at each sampling time; - indicates the number corresponding to the current time window. The prediction error at each sampling time; -After updating the weights based on the current prediction error, the result is the adaptive filter weight vector for the next time step; - The step size factor of the adaptive update directly determines the magnitude of the adaptive filter weight adjustment. It can be set according to the sampling frequency, signal amplitude range and convergence stability. This invention does not limit this.

[0043] In the discrete implementation, the adaptive filter is used to adjust the current time window based on the reference signal. The transmission mapping relationship is updated point by point, and the estimated signal of the body vibration transmission component is output. The adaptive filter, updated point by point, can output the estimated signal of the body vibration transmission component in the current time window; based on this, the correction signal for the current time window is obtained by subtracting the mixed response signal from the estimated signal of the body vibration transmission component at the corresponding time.

[0044] In engineering implementation, a normalized adaptive filter can also be used to improve stability under different operating conditions.

[0045] It should be noted that the above implementation is only an example, and the transitive mapping operator can also be implemented through correlation modeling, statistical weighting or other equivalent methods. This invention does not limit this.

[0046] During engineering operations, the overall operational behavior exhibits a non-stationary disturbance process. However, under the condition that the engineering operation parameters remain relatively stable, for a finite time window within the same sediment layer, the vibration response generated by the interaction between the working component and the sediment can be statistically approximated as a quasi-steady-state vibration process. Therefore, several time windows are divided along the working depth direction, and the vibration within the divided time windows is approximated as a quasi-steady-state vibration. This window is denoted as the quasi-steady-state time window.

[0047] In this invention, a quasi-steady-state time window is set to correspond to the working depth interval during the drilling process, that is, each time window corresponds to a current working depth interval; within the window, dual-source differential correction, feature construction and identification output are completed to characterize the sediment response state within the working depth interval. As the drilling progresses, the identification results are sequentially assigned to the corresponding working depth interval, thereby forming a sediment profile that changes continuously with the working depth.

[0048] Multidimensional sediment vibration features for characterizing sediment dynamic response are extracted from the correction signal, including: —Time-domain statistical characteristics characterizing vibration amplitude and stability; —Frequency domain characteristics that characterize the distribution of vibration energy; —Complexity stability features that characterize the complexity or stability of vibration response.

[0049] Based on the multidimensional sediment vibration characteristics, a classification model is input and trained to establish a sediment category recognition model, thereby enabling the differentiation of different sediment categories.

[0050] The classification model can be implemented in, but is not limited to, the following ways: Traditional supervised learning classifiers: Support Vector Machine, Random Forest, Gradient Boosting Tree, Neighbors, etc.; Neural network classifiers: multilayer perceptrons, convolutional neural networks, recurrent neural networks, or attention networks, etc. Ensemble learning: multi-model voting or stacked fusion, etc.

[0051] The overall processing flow of this invention (e.g.) Figure 2 (As shown) can be summarized as: Dual-source vibration synchronous acquisition → signal preprocessing and windowing → dual-source differential correction → extraction of multi-dimensional sediment vibration characteristics → sediment category identification output.

[0052] This invention also provides an in-situ identification system for seabed sediments based on dual-source differential vibration characteristics, comprising: The data acquisition module is used to simultaneously acquire dual-source vibration signals from the power source location and the working component location of the engineering equipment; The signal preprocessing and windowing module is used to preprocess the acquired dual-source vibration signals and window them along the working depth to determine the corresponding time window for each working depth segment during continuous drilling. The preprocessing includes detrending and bandpass filtering to suppress low-frequency drift and high-frequency random interference unrelated to sediment response. Detrending is used to suppress low-frequency drift and structural gradual change terms. Bandpass filtering is used to suppress high-frequency random interference unrelated to sediment response. The filtering passband can be selected based on the sampling frequency and equipment structural characteristics. The time window is divided along the working depth, and each time window is approximately considered as a quasi-steady-state time window. The dual-source differential correction module is used to output correction signals for dual-source vibration signals within each time window using the dual-source differential correction model. The feature construction module is used to extract and construct multidimensional sediment vibration features based on the correction signal; The sediment category identification module is used to predict and output the identification result of the sediment category based on the multidimensional sediment vibration characteristics as input.

[0053] In some embodiments, the system further includes a sample library for storing training samples (samples in the training set), identification samples, and their corresponding labels.

[0054] Example 1 This invention acquires dual-source vibration signals simultaneously by arranging reference vibration acquisition units and hybrid response vibration acquisition units at the power source location and the working component location of the engineering equipment, respectively. The vibration signal acquired at the power source location is used as the reference signal. By constructing a transfer mapping relationship, interference components that are highly correlated with the vibration of the engineering equipment body are suppressed in a statistical sense, making the correction signal more prominent in the dynamic response characteristics of the sediment itself.

[0055] Considering that the vibration response throughout the entire engineering operation is a non-stationary process, this invention does not directly process the full-time dual-source vibration signal uniformly. Instead, it divides the continuous drilling process into segments along the drilling direction according to preset working depth intervals. Specifically, the continuous drilling process can be divided into multiple adjacent working depth segments, the first... The corresponding working depth range for each working depth segment is: ,in Indicates the first The starting depth of each working depth section This indicates the preset operating depth interval. Based on the operating depth-time correspondence recorded synchronously during the engineering operation, the corresponding time range of each operating depth segment in the continuous drilling process is determined. Signal segments within the corresponding time period are extracted from the pre-processed dual-source vibration signals as the analysis signal for that operating depth segment. Since the operating depth variation within a single operating depth segment is limited, and the equipment operating parameters and the interaction state between the operating components and sediment are relatively stable, dual-source differential correction and multi-dimensional sediment vibration feature construction can be completed separately within each operating depth segment, further obtaining the identification results of the sediment categories corresponding to each operating depth segment. As the drilling process continues, the identification results of each operating depth segment are sequentially mapped to the corresponding operating depth segment according to the operating depth sequence, thus forming a sediment profile that continuously changes with the operating depth.

[0056] In this embodiment, after completing the synchronous acquisition of dual-source vibration and performing signal preprocessing and windowing, dual-source differential correction is performed within each time window to obtain the correction signal; then, based on the correction signal corresponding to each time window, training samples and identification samples for sediment category identification model are constructed, thereby realizing automatic identification of sediment category and continuous output with working depth.

[0057] 1) Sample Construction and Label Acquisition During subsea engineering surveys, penetrations, or pile driving operations, the pre-processed dual-source vibration signals are windowed according to preset operating depth intervals to form a time window sequence. Dual-source differential correction is then performed within each time window to obtain the corresponding corrected signal sequence. Multidimensional sediment vibration features are extracted from the corrected signal sequence within each time window, and these features are concatenated to form a sample feature vector. .

[0058] in, Indicates the sequence number or index of the time window; This indicates that after dividing according to the preset work depth interval, the first... The sequence of correction signals within a time window; Indicates from the first The sample feature vector is composed of multidimensional sediment vibration features extracted from each time window.

[0059] To obtain the labels needed for supervised learning The true category can be provided in the following ways: Results of core sampling at the same depth and laboratory geotechnical tests; Static cone penetration test, in-situ test, or engineering geological stratification record; The combined interpretation results of construction records and known stratigraphic data.

[0060] Based on this, a training set is constructed. ,in Labels for sediment categories.

[0061] 2) Feature Construction and Input Format In a preferred embodiment, the sample feature vector A combination of the following three types of features, used to simultaneously characterize differences in vibration amplitude, energy distribution, and stability, is: Time-domain statistical characteristics: root mean square value Peak factor, skewness, kurtosis, etc.; Frequency domain energy characteristics: power spectrum centroid frequency, frequency band energy proportion, dominant frequency energy, etc.; Complexity stability characteristics: sample entropy, permutation entropy, or equivalent stability statistics, etc.

[0062] In another embodiment, the sample feature vector It may also include non-explicitly constructed manual features, which are: the three directions of each time window ( , , The correction signal sequences (in three directions) are concatenated in time series form to form a matrix input:

[0063] in, Indicates the first The input matrix is ​​formed by directly splicing the correction signals in the three directions of each time window in time series form. The dimension is The real space matrix, where "3" represents , , Three directions, This represents the total number of discrete sampling points contained within a single time window, which is the number of sampling points in the time window and is used for end-to-end model learning.

[0064] In this invention, the subscript k of the sample feature vector is used to indicate that the feature vector originates from the k-th time window and corresponds to the k-th operation depth segment.

[0065] The two types of input can coexist. The first type is explicitly constructed features, which are combinations of features extracted manually or from prior physical knowledge. These include time-domain statistical features, frequency-domain energy features, and complexity stability features. The second type is non-explicitly constructed handcrafted features, which are directly input by arranging and concatenating the correction signal sequences in the three directions within each time window into a matrix, containing complete continuous sampling.

[0066] 3) Training of sediment category recognition model In this embodiment, a sediment category identification model is constructed. Based on the training set Training sediment category recognition model This establishes a nonlinear mapping relationship between the sample feature vector and the sediment category:

[0067] in, Sediment category identification model For the first The sediment category output after inference from the feature vectors of each sample.

[0068] The training process can employ cross-validation or hold-out methods to determine the hyperparameters of the sediment category identification model; and class weights or resampling strategies can be introduced to address the class imbalance problem. This embodiment does not limit the specific training strategy and parameter selection, and can be implemented based on existing technologies.

[0069] 4) Identification and continuous output as the work depth progresses In the actual engineering identification phase, sample feature vectors are generated for each working depth segment corresponding to the time window. Perform reasoning to obtain the predicted category and its relationship with the working depth Correlation, forming a sequence of sediment profiles that continuously vary with the depth of the operation. .

[0070] In some implementations, to improve the stability of continuous profiles, depth smoothing can be introduced without changing the single-window identification results to suppress occasional abnormal window jumps, but this is not intended to limit the sediment category identification model.

[0071] 5) Sediment category identification model update Because sediment composition and construction conditions vary across different sea areas, this embodiment can incrementally update or perform transfer learning on the sediment category identification model after obtaining a small number of labeled samples from a new area. This allows the sediment category identification model to adapt to the new strata and construction conditions, thereby improving the stability of cross-regional identification. This invention does not limit the update method.

[0072] Example 2 This embodiment takes the process of penetration piling in submarine engineering survey or foundation construction as the application scenario. Without changing the original operation process, it realizes continuous in-situ identification of sediment types through dual-source synchronous acquisition and dual-source differential correction.

[0073] I. Simultaneous acquisition of dual-source vibration like Figure 1As shown, the system used in this embodiment includes a reference vibration acquisition unit 1, a data processing and analysis center 2, and a hybrid response vibration acquisition unit 3. The reference vibration acquisition unit 1 is located at the power source and is used to acquire reference signals reflecting the operating status of the engineering equipment body. It is connected to the data processing and analysis center 2 via a wired connection. The hybrid response vibration acquisition unit 3 is located at the working component and is used to acquire hybrid response signals containing the vibration transmission component of the body and the dynamic response component of the sediment. It is electrically connected to the data processing and analysis center. The hybrid response vibration acquisition unit 3 has a short-section structure. Its body acts as a section of the drill pipe, and its upper and lower ends are rigidly connected in series with the drill pipe and the lower working component (in this embodiment, the working component is the drill bit) respectively, maintaining a coaxial arrangement with the drill pipe. In this embodiment, the body of the hybrid response vibration acquisition unit is made of high-strength alloy steel short section, with standard drilling tapered pipe thread interfaces at both ends. It is installed on the drill pipe near the working component via a threaded series connection. The output shaft of the power source is connected to the upper end of the drill pipe, and the lower end of the drill pipe is connected to the working component through the short-section structure.

[0074] The hybrid response vibration acquisition unit 3 has a sealed pressure chamber 6 inside, which integrates a time recorder 4, a triaxial accelerometer 5, a data storage unit 7, and a battery unit 8. The time recorder 4 records the acquired hybrid response signal over time; the triaxial accelerometer 5 acquires the triaxial acceleration vibration signal at the location of the working component (i.e., obtains the hybrid response signal); the data storage unit 7 locally caches the data over time under abnormal conditions (such as communication limitations) for later export; and the battery unit 8 powers the hybrid response vibration acquisition unit 3.

[0075] Data processing and analysis center 2 is used to receive dual-source vibration signals sent by reference vibration acquisition unit 1 and hybrid response vibration acquisition unit 3, perform time alignment of dual-source signals, and complete subsequent signal preprocessing, dual-source differential correction, feature construction and sediment category identification processing.

[0076] Assume the triaxial acceleration acquired from the location of the power source is the reference signal. The triaxial acceleration acquired from the position of the working component is a hybrid response signal. .

[0077] II. Signal Preprocessing and Windowing Signal preprocessing was performed on the dual-source triaxial acceleration signals, including detrending and bandpass filtering. Specifically, firstly, linear detrending was applied to the acceleration time series of the reference signal and the mixed response signal in each direction to suppress low-frequency drift and structurally gradual variation terms; then, a fourth-order Butterworth bandpass filter was applied to the detrended signals to suppress high-frequency random interference unrelated to the sediment response. In this embodiment, the filter passband was set to 0.5–500 Hz.

[0078] During engineering operations, the overall vibration process is non-stationary; however, under relatively stable operating parameters, the coupled vibration of the operating component and sediment within a finite time window in the same sedimentary layer can be statistically approximated as a quasi-steady-state process. Therefore, the preprocessed signal is windowed along the operating depth at set operating depth intervals to form multiple time windows for subsequent dual-source differential correction and feature construction.

[0079] III. Dual-source differential correction To separate the propagation vibration transmission component that is highly correlated with the reference signal from the mixed response signal, this embodiment constructs a transfer mapping operator. Modeling the reference signal yields the estimated signal of the body vibration transmission components. :

[0080] In obtaining the estimated signal of the body vibration transmission components Then, it is extracted from the mixed response signal. The two signals are canceled out to obtain the corrected signal. : .

[0081] In this embodiment, the transitive mapping operator An adaptive filter based on the minimum mean square error criterion is used for implementation.

[0082] The above dual-source differential correction can statistically reduce the body vibration interference that is highly correlated with the power source, making the correction signal more prominent in terms of sediment response characteristics.

[0083] IV. Construction of Multidimensional Sediment Vibration Characteristics Multidimensional sediment vibration features were extracted and constructed from the corrected signal within each time window. The extracted multidimensional sediment vibration features included: Time-domain statistical characteristics: root mean square value, peak factor, skewness, kurtosis, etc., are used to characterize the vibration amplitude level and impact. Frequency domain energy distribution characteristics: Based on the power spectral density, the centroid frequency of the power spectrum, the proportion of energy in the frequency band, and the dominant frequency energy are used to characterize the modulation effect of sediments on the vibration energy distribution. Complexity stability characteristics: such as sample entropy, permutation entropy, or equivalent stability statistics, are used to quantify the nonlinearity and stability of the response.

[0084] V. Sediment Category Identification Output The multidimensional sediment vibration characteristics are normalized to form a sample feature vector of uniform scale, which is then input into a pre-trained sediment category recognition model to output the sediment category corresponding to the current working depth. The recognition results are correlated with the working depth or penetration depth to form a sediment profile sequence that changes continuously with the working depth.

[0085] In some implementations, the identification results can be verified by combining borehole core sampling, engineering geological data, or existing stratification records, for adjusting the parameters of the sediment category identification model, or as a reference for updating the sample library.

[0086] The hybrid response vibration acquisition unit 3 of this invention can be rigidly connected in series with the drill pipe in the form of a structural component, and uses the built-in time recorder 4 to realize high-precision time synchronization acquisition of dual-source vibration signals, providing hardware guarantee for high-precision analysis.

[0087] To address the strong mechanical noise interference from drilling rigs, this invention employs a dual-source differential correction model, treating the mixed response signal as a 'body vibration transmission component' and a 'sediment dynamic response component'. Using a reference signal as the reference input and a transfer mapping operator as the core, the model adaptively estimates and dynamically cancels the body vibration transmission component transmitted through the drill pipe within a time window, thereby obtaining an estimate of the sediment dynamic response component and improving the extraction purity of the estimated sediment dynamic response component.

[0088] This invention achieves in-situ identification without increasing sampling disruption (such as additional excavation, coring, or cutting), without altering the construction process, and without applying additional special stimulation to the formation. The identification process is completed as part of the drilling monitoring during construction, without adding extra steps. It belongs to the "construction-as-you-go identification" technology, realizing non-destructive drilling identification throughout the entire process. Without adding extra steps or operation time, it achieves continuous, real-time scanning of seabed sediment types, overcoming the discontinuous defects of traditional discrete sampling, effectively capturing thin, weak interlayers that are easily missed, and significantly improving exploration efficiency.

[0089] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A method for in-situ identification of marine sediments based on dual-source differential vibration signatures, characterized in that, The method Includes the following: Vibration signals were simultaneously collected at the power source location and the working component location of the engineering equipment. The vibration signal collected at the power source location was used as the reference signal, and the vibration signal collected at the working component location was recorded as the hybrid response signal. Construct a dual-source differential correction model: Simultaneously construct a transfer mapping operator with the reference signal and the hybrid response signal as inputs. After processing the reference signal with the transfer mapping operator, obtain the estimated signal of the body vibration transfer component. Subtract the hybrid response signal from the estimated signal of the body vibration transfer component to obtain the correction signal. Extract multidimensional sediment vibration features from the correction signal to characterize the dynamic response of sediments; Based on the multidimensional sediment vibration characteristics, the sediment category identification model is used to identify the category of seabed sediments in situ.

2. The method for in-situ identification of seafloor sediments based on dual-source differential vibration signatures according to claim 1, characterized in that, The transfer mapping operator is implemented using an adaptive filter based on the minimum mean square error criterion, a normalized adaptive filter, correlation modeling, or statistical weighting. In the specific implementation process: the time window is divided according to the operation depth interval, and the time window corresponding to the current operation depth interval is used as a local update range; within the time window, the transfer mapping relationship is recursively updated using the real-time data of each discrete sampling point.

3. The in-situ identification method for seabed sediments based on dual-source differential vibration characteristics according to claim 1, characterized in that, The multidimensional sediment vibration characteristics include: time-domain statistical characteristics characterizing vibration amplitude and stability, frequency-domain characteristics characterizing vibration energy distribution characteristics, and complexity stability characteristics characterizing the complexity or stability of vibration response.

4. The in-situ identification method for seabed sediments based on dual-source differential vibration characteristics according to claim 3, characterized in that, The time-domain statistical features include root mean square value, peak factor, skewness, and kurtosis; the frequency-domain features include power spectrum centroid frequency, frequency band energy proportion, and dominant frequency energy; the complexity stability features include sample entropy and permutation entropy.

5. The in-situ identification method for seabed sediments based on dual-source differential vibration characteristics according to claim 1, characterized in that, Samples and labels for constructing a sediment category identification model: During seabed engineering surveys, penetration or piling operations, the reference signal and mixed response signal are windowed according to a preset operating depth interval to form a time window sequence; dual-source differential correction is performed within each time window to obtain the corresponding correction signal sequence; multidimensional sediment vibration features are extracted from the correction signal sequence within each time window, and the multidimensional sediment vibration features are spliced ​​together to form a sample feature vector; Simultaneously obtain the sediment category label corresponding to the feature vector of each sample; The training set is composed of all sample feature vectors and their corresponding sediment category labels.

6. A system for in-situ identification of seabed sediments based on dual-source differential vibration characteristics, characterized in that, The system performs the method of claim 1, including: The data acquisition module is used to simultaneously acquire dual-source vibration signals from the power source location and the working component location of the engineering equipment; The signal preprocessing and windowing module is used to preprocess the acquired dual-source vibration signals and window the dual-source vibration signals along the working depth to determine the corresponding time window for each working depth section during continuous drilling. The dual-source differential correction module is used to output correction signals for the dual-source vibration signals using a dual-source differential correction model within each time window; The feature construction module is used to extract and construct multidimensional sediment vibration features based on the correction signal; The sediment category identification module is used to predict and output the identification result of the sediment category based on the multidimensional sediment vibration characteristics as input.

7. The in-situ identification system for seabed sediments based on dual-source differential vibration characteristics according to claim 6, characterized in that, The data acquisition module includes a reference vibration acquisition unit and a hybrid response vibration acquisition unit. The reference vibration acquisition unit is located at the power source of the engineering equipment, and the hybrid response vibration acquisition unit is located at the working component. The hybrid response vibration acquisition unit has a short section structure. The main body of the hybrid response vibration acquisition unit is a section of the drill rod. The upper and lower ends are rigidly connected in series with the drill rod and the lower working component, respectively, and are arranged coaxially with the drill rod. The hybrid response vibration acquisition unit has a sealed pressure chamber inside, which integrates a time recorder, a triaxial acceleration sensor, a data storage unit, and a battery unit. The time recorder is used to record the acquired hybrid response signal according to time, the triaxial acceleration sensor is used to acquire the triaxial acceleration vibration signal at the location of the working component, and the data storage unit is used to locally cache the data according to time in abnormal conditions.

Images