Nonlinear phase difference field multi-scale energy mask drilling winch modal identification method

Abstract

The application discloses a nonlinear phase difference field multi-scale energy mask drilling winch modal identification method, and comprises the following steps: through envelope demodulation and time-varying wave filtering, decoupling and stripping low-frequency sea wave modulation interference, and reconstructing a steady-state response sequence; then the sequence is mapped to a polar coordinate system, and a two-dimensional Gromi phase angle difference field matrix representing mechanical nonlinear stiffness degradation is constructed; on this basis, a double-track analysis strategy is adopted: one track extracts transient impact energy characteristics caused by gear surface or bearing peeling through multi-scale sliding filtering, and the other track analyzes global periodic modulation characteristics caused by macroscopic wear through spatial pooling; finally, a nonlinear cross-correlation mask is constructed, global mechanical white noise is inhibited by using the transient impact characteristics, and the pure characteristics are mapped to a winch health modal index. The application effectively solves the engineering problems that the winch multi-scale faults are prone to be missed and false reported under complex sea conditions.

Description

Technical Field

[0001] This invention belongs to the field of marine heavy machinery equipment operation status monitoring and fault diagnosis technology, and in particular relates to a nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method. Background Technology

[0002] Offshore drilling platform winches are key power equipment used in drilling operations to perform core processes such as tripping drill strings, running casing, and lifting heavy objects. Their transmission system typically consists of a high-power motor, a multi-stage heavy-duty gearbox, a braking mechanism, and a giant drum. Because offshore drilling platforms are isolated on the sea surface, the winches, during long-term operation, not only endure frequent heavy-load starts and stops and rapid acceleration and deceleration, but are also severely affected by the alternating pitching and heave loads caused by the periodic impact of waves.

[0003] The complex dynamic response generated by the coupling of internal and external excitations results in the vibration signal of the winch transmission system exhibiting extremely strong non-stationarity, nonlinearity, and time-varying frequency drift characteristics. When early microscopic damage occurs in the gears or bearings inside the winch (such as pitting on the tooth surface or spalling of the bearing raceway), these weak transient impact characteristics are often completely drowned out by the low-frequency heavy load envelope caused by the rise and fall of ocean waves and the high-frequency mechanical white noise from surrounding multi-source equipment (such as mud pumps and diesel generators).

[0004] To address the aforementioned issues, existing fault diagnosis and condition awareness methods suffer from the following technical bottlenecks. Traditional time-frequency analysis methods (such as Short-Time Fourier Transform (STFT) and wavelet analysis) and existing data-driven models are mostly based on the assumption of "relatively stable system conditions." When directly processing measured signals from marine winches, these methods cannot effectively distinguish between "fundamental frequency fluctuations caused by wave heave" and "frequency modulation caused by mechanical wear," and struggle to isolate interference from non-stationary marine environmental loads. This leads to frequent false alarms and missed alarms in diagnostic systems under severe sea conditions, resulting in a persistently high false alarm rate.

[0005] The failure evolution of winches typically involves two dimensions: first, high-frequency transient impacts caused by microscopic peeling of components (with strong local temporal abrupt changes); and second, low-frequency periodic modulations caused by overall structural wear and stiffness reduction (with global temporal continuity). Existing signal imaging methods (such as Markov transfer fields and conventional grayscale mapping) combined with single-scale convolutional neural networks (CNNs) often suffer from trade-offs: if the focus is on capturing high-frequency transient changes, the global degradation trend is easily lost; if the focus is on extracting low-frequency periodic features, early weak damage signals are smoothed out during feature pooling.

[0006] In recent years, some studies have attempted to extract features at different scales using multi-branch network structures and then perform simple linear splicing or fully connected fusion. However, in the extreme environment of high noise and low signal-to-noise ratio (SNR) at sea, simple feature stacking not only introduces a large number of redundant environmental noise vectors but also causes the "curse of dimensionality," leading to model overfitting and failing to meet the stringent requirements of robustness and generalization ability of diagnostic systems in actual marine engineering.

[0007] In summary, existing monitoring and diagnostic methods have inherent limitations when dealing with large drilling winches in complex sea conditions, such as difficulty in decoupling load interference, poor coordination of multi-scale damage features, and weak feature representation ability under strong noise.

[0008] Therefore, there is an urgent need for a novel health mode identification method that can adaptively eliminate the interference of variable loads in the marine environment and deeply coordinate local transient impacts and global degradation cycles to ensure the safety and efficiency of offshore drilling operations. Summary of the Invention

[0009] The problem to be solved by this invention is to provide a nonlinear phase difference field multi-scale energy masking method for drilling winch mode identification. This method achieves high reliability perception and diagnosis of early minor damage and macroscopic degradation trend of winch under harsh working conditions with extremely low signal-to-noise ratio and superimposed alternating wave loads by constructing a closed-loop technical system of "environmental load decoupling - nonlinear topology mapping - dual-track feature coordination - cross-correlation masking enhancement".

[0010] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method, comprising the following steps: S1: Obtain the steady-state response sequence decoupled from the heave load; S2: Construct the two-dimensional Grammy phase angle difference field matrix; S3: Analysis of the characteristics of micro-transient shocks and macro-cycle modulation in a dual-track system; S4: Feature collaboration and purification recognition based on cross-correlation mask.

[0011] Furthermore, S1 includes the following steps: S11: Acquire the original dynamic response carrier signal of the winch drive system of the offshore drilling platform; S12: Use variational mode decomposition algorithm to extract the low-frequency reference envelope in the signal that corresponds to the wave rise and fall period; S13: Using the low-frequency reference envelope as the compensation reference, perform time-varying notch filtering on the high-frequency intrinsic mode components to remove the time-varying load modulation interference caused by the ship's pitch and heave, and reconstruct the steady-state response sequence characterizing the state of the winch's mechanical body.

[0012] Furthermore, in S2, the amplitude information of the steady-state response sequence is converted into an anticosine phase angle in polar coordinates. By calculating the sine value of the difference between the phase angles corresponding to any two sampling points on the time series, a two-dimensional Grami phase angle difference field space matrix characterizing the nonlinear stiffness degradation trend of the winch gear and bearing is constructed.

[0013] Furthermore, in S3, a dual-track network architecture is used to perform deep parallel analysis of the spatial matrix, including: In the first analytical track, a local correlation analytical unit containing a multi-scale sliding filter kernel is used to perform a two-dimensional sliding window convolution operation on the spatial matrix to capture high-frequency abrupt pixel clusters caused by pitting of winch teeth or peeling of bearings, and to extract local transient impact energy feature vectors. In the second analytical track, the spatial matrix is ​​downsampled in parallel using spatial pooling to capture global periodic texture changes caused by winch wear and extract global periodic modulation feature maps.

[0014] Furthermore, in the first analytical track, the receptive field size of the multi-scale sliding filter kernel strictly corresponds to the pixel span on the two-dimensional spatial matrix mapped by the mechanical meshing frequencies of the winch's high-speed, medium-speed, and low-speed gears.

[0015] Furthermore, S4 includes the following steps: S41: Using a nonlinear cross-correlation mask function, the amplitude of the local transient impact energy feature vector extracted from the first analytical trajectory is mapped to the [0,1] interval as a spatial attention weight mask; S42: Perform an element-wise Hadamard product operation on the spatial attention weight mask and the global periodic modulation feature map extracted by the second analytical track, and use the local high-energy impact features to adaptively suppress the low-frequency mechanical background white noise in the global features to generate a purified health assessment matrix.

[0016] Furthermore, it also includes S5: Health Modality Nonlinear Mapping: The purified health assessment matrix is ​​input into the nonlinear modality classifier, and the health modality assessment index of the winch is output.

[0017] Furthermore, the present invention also provides a nonlinear phase difference field multi-scale energy mask drilling winch mode recognition system, which, when running the aforementioned nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method, includes: The signal acquisition module is used to acquire the original dynamic response carrier signal of the winch drive system of an offshore drilling platform under actual variable load conditions. The heave load decoupling module is used to perform multi-band analysis on the original dynamic response carrier signal, adaptively extract the low-frequency reference envelope corresponding to the wave heave period, use the low-frequency reference envelope as a compensation reference to perform time-varying notch filtering on the separated high-frequency intrinsic mode components, dynamically remove the time-varying load modulation interference caused by the ship's pitch and heave, and reconstruct a steady-state response sequence that only represents the true degradation state of the winch mechanical body. The Grammy phase angle difference field matrix construction module is used to map the amplitude information of the steady-state response sequence to the polar coordinate system and convert it into an inverse cosine phase angle sequence. By calculating the sine value of the difference between the phase angles corresponding to any two sampling points at different times, a two-dimensional Grammy phase angle difference field space matrix is ​​constructed. The dual-track feature parsing module is used to perform deep parallel parsing of the spatial matrix. It includes a first parsing track unit and a second parsing track unit. The first parsing track unit uses a local correlation parsing unit containing a multi-scale sliding filter kernel to perform a two-dimensional hierarchical sliding window convolution operation on the spatial matrix to extract local transient impact energy feature vectors. The second parsing track unit performs adaptive spatial pooling downsampling on the spatial matrix to extract global periodic modulation feature maps. The cross-correlation mask feature collaborative purification module is used to compress and map the amplitude of the local transient impact energy feature vector to the [0,1] interval to generate a spatial attention weight mask, and to perform element-wise Hadamard product operation on the spatial attention weight mask and the global periodic modulation feature map to generate a purified health assessment matrix. The health modality classification output module is used to input the purified health assessment matrix into the nonlinear modality classifier and output the accurate health modality assessment index of the current operating state of the winch.

[0018] Furthermore, the present invention also provides a computer device, including a memory, a processor, and an algorithm stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the drilling winch mode recognition method as described above.

[0019] Furthermore, the present invention also provides a computer-readable storage medium storing a computer algorithm, which, when executed by a processor, implements the drilling winch mode recognition method.

[0020] The advantages and positive effects of this invention are: 1. This invention has made a breakthrough in solving the problem of masking fault characteristics under complex sea state and variable loads: Existing general diagnostic models are prone to misinterpreting heave loads caused by periodic wave impacts as low-frequency modulation features of mechanical faults when dealing with offshore facilities, resulting in an extremely high false alarm rate. This invention innovatively introduces a "heave load decoupling" mechanism. Through variational mode decomposition and time-varying notch filtering, it eliminates spurious feature interference from hull motion at the physical level, providing a high-fidelity steady-state sequence foundation for subsequent high-dimensional topology mapping and significantly improving the system's anti-interference capability under extreme sea conditions.

[0021] 2. This invention achieves perfect multi-scale synergy between microscopic transient mutations and macroscopic degradation cycles: Early, minor damage to mechanical equipment typically manifests as high-frequency transient impacts, while overall wear exhibits low-frequency periodic modulation. This invention constructs a two-dimensional Grammy phase angle difference field matrix, transforming a one-dimensional non-stationary signal into a two-dimensional topological structure containing temporal geometric information. Based on this, an innovative "dual-track feature analysis architecture" cleverly utilizes a multi-scale sliding filter kernel to accurately capture microscopic abrupt changes in wear, while simultaneously employing spatial pooling to lock in macroscopic wear patterns, successfully overcoming the limitations of traditional single-network approaches in complex feature extraction. Furthermore, by physically binding the receptive field size of the multi-scale sliding filter kernel to the meshing frequencies of the winch's gears at each stage, the feature extraction gains extremely strong mechanical interpretability.

[0022] 3. This invention proposes a high-order feature purification mechanism (cross-correlation mask enhancement) under strong noise environments: To address the "curse of dimensionality" and redundant noise interference that easily arise in multi-source sensor data fusion, this invention completely abandons mainstream feature linear splicing or fully connected fusion strategies, and innovatively proposes a cross-correlation masking mechanism based on Hadamard product. This mechanism simulates the "masking effect" in human acoustic direction finding, that is, using microscopic features containing high-energy information of deterministic damage as "guide light sources" to directionally illuminate the fault evolution region in the macroscopic periodic feature map, thereby adaptively suppressing useless environmental white noise in strong mechanical background noise with extremely low signal-to-noise ratio (SNR). This collaborative purification strategy enables the model to maintain excellent generalization ability and robustness even with limited marine experimental samples. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.

[0024] Figure 2 This is a logical structure diagram of a feature collaborative purification unit based on cross-correlation mask and Hadamard product according to an embodiment of the present invention. Detailed Implementation

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

[0026] The embodiments of the present invention will be further described below with reference to the accompanying drawings: To address the engineering challenges posed by the severe interference of wave heave loads on offshore drilling platform winches during tripping operations, leading to high false alarm rates and difficulty in extracting early, subtle damage at multiple scales using traditional fault diagnosis models, this invention constructs a closed-loop health assessment system encompassing "environmental load decoupling, nonlinear topology mapping, dual-track feature synergy, and cross-correlation mask enhancement." For example... Figure 1 As shown, the nonlinear phase difference field multi-scale energy mask drilling winch mode identification method includes the following steps.

[0027] S1: Obtain the steady-state response sequence decoupled from the heave load.

[0028] Specifically, the original dynamic response carrier signal of the winch drive system of an offshore drilling platform under actual variable load conditions is acquired. The signal is then analyzed using a variational mode decomposition algorithm across multiple frequency bands, adaptively extracting the low-frequency reference envelope corresponding to the height of the wave heave period. Using this low-frequency reference envelope as a compensation reference, time-varying notch filtering is applied to the separated high-frequency intrinsic mode components. This process dynamically removes time-varying load modulation interference caused by hull pitch and heave, eliminates frequency drift artifacts in the signal, and thus reconstructs a steady-state response sequence that only characterizes the true degradation state of the winch mechanism.

[0029] S2: Construct the two-dimensional Grammy phase angle difference field matrix.

[0030] Specifically, the amplitude information of the reconstructed steady-state response sequence is mapped to a polar coordinate system, transforming it into an inverse cosine phase angle sequence. By calculating the sine value of the difference in phase angles between any two sampling points at different times in this time series, the local abrupt changes and global cyclic correlations of the signal over the time span are captured, thereby constructing a two-dimensional Grammy phase angle difference field space matrix that can intuitively characterize the degradation trend of winch gear meshing stiffness and bearing clearance.

[0031] S3: Dual-track system analysis of micro-transient shocks and macro-cycle modulation characteristics.

[0032] Specifically, a dual-track network architecture is used to perform deep parallel analysis on the spatial matrix: in the first analysis track (local microscopic tracking), a local correlation analysis unit containing a multi-scale sliding filter kernel is used to perform a two-dimensional hierarchical sliding window convolution operation on the spatial matrix. This operation aims to accurately capture high-frequency abrupt pixel clusters caused by early pitting, spalling, or local damage to the bearing raceway of the winch tooth surface, thereby extracting a feature vector containing high energy density local transient impact energy.

[0033] Preferably, the receptive field size of the multi-scale sliding filter kernel is strictly corresponding to the pixel span of the mechanical meshing frequencies of the high-speed, medium-speed, and low-speed gears in the marine winch transmission chain, mapped onto the two-dimensional spatial matrix.

[0034] In the second analytical track (global macroscopic evolution), the spatial matrix is ​​subjected to adaptive spatial pooling downsampling in parallel. By expanding the receptive field, the global periodic texture change pattern caused by long-term distributed wear of the winch and loosening of the support structure is captured, thereby extracting the global periodic modulation feature map that characterizes the overall deterioration trend of the equipment.

[0035] S4: Feature collaboration and purification recognition based on cross-correlation mask.

[0036] Specifically, to achieve targeted enhancement of sensitive features in strong mechanical background noise, this step abandons the traditional feature concatenation method and adopts a nonlinear cross-correlation masking mechanism. First, a nonlinear activation function (such as Sigmoid) is used to compress and map the amplitude of the local transient impact energy feature vector extracted from the first analytical track to the [0,1] interval, generating a spatial attention weight mask. Subsequently, the spatial attention weight mask is used to perform an element-wise Hadamard product operation with the global periodic modulation feature map extracted from the second analytical track. This operation utilizes the local high-energy abrupt impact features to adaptively "illuminate" the key evolution regions related to the fault in the global features, while deeply suppressing redundant low-frequency mechanical environment white noise. Finally, a purified health assessment matrix is ​​generated and input into a nonlinear modal classifier to output an accurate health modal assessment index of the winch's current operating state.

[0037] The present invention will now be described in detail with reference to specific embodiments: S1: Heave load decoupling processing based on envelope demodulation.

[0038] In offshore drilling operations, the vibration signals of the winch drive system are often mixed with low-frequency alternating load interference caused by the ship's pitching and heave. Without decoupling, this can easily mask early high-frequency, weak fault impacts in gears or bearings.

[0039] Signal Acquisition and Variational Mode Decomposition: The raw dynamic response carrier signal is acquired by using high-frequency accelerometers deployed on the winch motor drive end and the key bearing housing of the gearbox. The adaptive variational mode decomposition (VMD) algorithm is used to... Decomposed into Each intrinsic mode component (IMF) has a specific center frequency. Number of decomposition layers. and punishment factors The selection is strictly based on the mechanical rotation frequency of the winch input shaft. Compared with the measured main frequency of ocean waves The harmonic relationships (typically between 0.05Hz and 0.2Hz) are dynamically determined. By calculating the cross-correlation coefficients between each IMF component and the dominant wave frequency, the dominant mode component containing wave modulation interference is adaptively locked. .

[0040] Envelope extraction and time-varying notch filtering: for Perform Hilbert transform to extract the low-frequency reference envelope corresponding to the height of the wave rise and fall period. Subsequently, with The instantaneous frequency offset is used as the compensation benchmark to design a time-varying notch filter. The remaining high-frequency intrinsic mode components (containing real mechanical damage information) are passed through the time-varying notch filter to dynamically filter out the time-varying load modulation frequency band caused by hull motion, thereby reconstructing a steady-state response sequence stripped of environmental pseudo-features. .

[0041] S2: Construction of the two-dimensional Grammy phase angle difference field matrix.

[0042] In order to highlight both local transient impacts and global degradation trends under strong background noise, this embodiment maps the one-dimensional time series to a two-dimensional phase space with a topological structure.

[0043] Polar coordinate mapping: First, the steady-state response sequence is mapped... The amplitude is normalized to The interval is obtained Then, it is mapped to polar coordinates, and the normalized amplitude is converted into an inverse cosine phase angle. , in For the first Normalized amplitude of each sampling point.

[0044] Difference matrix generation: Calculate the difference matrix between any two sampling points in a time series. and The sine value of the difference in phase angles, i.e.:

[0045] This constructs a size of Two-dimensional Grami phase difference field (GADF) space matrix .

[0046] The sine operation of phase angle difference is extremely sensitive to high-frequency abrupt changes in the sequence (such as short-term impacts caused by tooth surface peeling), and can amplify weak temporal perturbations into significant diagonal cross textures (i.e. local high-frequency pixel clusters) in two-dimensional images, while preserving the low-frequency periodic background distribution pattern caused by macroscopic wear.

[0047] Step S3: Analyze the characteristics of microscopic transient shocks and macroscopic periodic modulation using a dual-track system.

[0048] The constructed spatial matrix The data is synchronously input into the "dual-track feature parsing network" designed in this invention, such as... Figure 1 The central area shown.

[0049] First analytical trajectory (microscopic transient impact tracking): A local correlation parsing unit is employed, incorporating a multi-scale sliding filter kernel (i.e., the shift window mechanism in the original Swing architecture). In the matrix... Two-dimensional hierarchical sliding window convolution operations are performed to specifically capture high-frequency abrupt pixel clusters caused by local microscopic damage.

[0050] Core parameter binding: Set at least three filter cores with different receptive field sizes. These three sizes strictly correspond to the gear meshing frequencies of the winch's high-speed stage (input), medium-speed stage, and low-speed stage (drum end). The time span of pixels mapped on the matrix. This design ensures that the network can accurately and adaptively lock the transient impact energy at different speed levels and extract the local transient impact energy feature vector. .

[0051] Second analytical trajectory (global macroscopic evolutionary topology): Parallel to Large-step spatial pooling downsampling (i.e., the original CNN branch) is performed. By expanding the receptive field, the global periodic texture variation of the matrix caused by factors such as increased winch bearing clearance and decreased overall stiffness is captured. After multi-level pooling and nonlinear mapping, a global periodic modulation feature map characterizing the overall degradation trend of the equipment is extracted. .

[0052] S4: Feature Coordination and Purification Recognition Based on Cross-Correlation Mask Under strong mechanical background noise at sea (such as vibration from nearby diesel generators), traditional feature splicing can cause the classifier to fall into the "curse of dimensionality" and amplify the noise. This embodiment employs an innovative cross-correlation Hadamard product masking mechanism, such as... Figure 2 As shown.

[0053] Spatial attention weight mask generation: This involves extracting local feature vectors containing deterministic damage high-energy information from the first orbit. By using the Sigmoid nonlinear activation function, its amplitude is compressed and mapped to... Intervals, generating spatial attention weight masks This mask actually represents the "local temporal / spatial location where the fault is most likely to occur".

[0054] Feature synergy and purification (Hadamard product): As a "guide light source", it is used in conjunction with the global periodic modulation feature map extracted from the second track. Perform element-wise Hadamard product. Operations:

[0055] This operation utilizes the locally high-energy abrupt shock features (masks) to adaptively amplify the weights of key regions highly correlated with fault evolution in the global feature map. Simultaneously, regions in the mask close to 0 (i.e., stable regions without shocks) directly clear the corresponding global feature values ​​to zero, thereby deeply suppressing redundant low-frequency mechanical environment white noise and ultimately generating a purified health assessment matrix. .

[0056] S5: Health Modal Nonlinear Mapping Purify the health assessment matrix The input is fed into a pre-defined multilayer perceptron (MLP) nonlinear classifier. After passing through a fully connected layer and a softmax activation function, the high-dimensional feature matrix is ​​mapped to a probability distribution sequence of the winch's current possible damage levels. The label corresponding to the maximum probability is selected, and the winch's final health modality index is output.

[0057] The advantages and positive effects of this invention are: 1. This invention has made a breakthrough in solving the problem of masking fault characteristics under complex sea state and variable loads: Existing general diagnostic models are prone to misinterpreting heave loads caused by periodic wave impacts as low-frequency modulation features of mechanical faults when dealing with offshore facilities, resulting in an extremely high false alarm rate. This invention innovatively introduces a "heave load decoupling" mechanism. Through variational mode decomposition and time-varying notch filtering, it eliminates spurious feature interference from hull motion at the physical level, providing a high-fidelity steady-state sequence foundation for subsequent high-dimensional topology mapping and significantly improving the system's anti-interference capability under extreme sea conditions.

[0058] 2. This invention achieves perfect multi-scale synergy between microscopic transient mutations and macroscopic degradation cycles: Early, minor damage to mechanical equipment typically manifests as high-frequency transient impacts, while overall wear exhibits low-frequency periodic modulation. This invention constructs a two-dimensional Grammy phase angle difference field matrix, transforming a one-dimensional non-stationary signal into a two-dimensional topological structure containing temporal geometric information. Based on this, an innovative "dual-track feature analysis architecture" cleverly utilizes a multi-scale sliding filter kernel to accurately capture microscopic abrupt changes in wear, while simultaneously employing spatial pooling to lock in macroscopic wear patterns, successfully overcoming the limitations of traditional single-network approaches in complex feature extraction. Furthermore, by physically binding the receptive field size of the multi-scale sliding filter kernel to the meshing frequencies of the winch's gears at each stage, the feature extraction gains extremely strong mechanical interpretability.

[0059] 3. This invention proposes a high-order feature purification mechanism (cross-correlation mask enhancement) under strong noise environments: To address the "curse of dimensionality" and redundant noise interference that easily arise in multi-source sensor data fusion, this invention completely abandons mainstream feature linear splicing or fully connected fusion strategies, and innovatively proposes a cross-correlation masking mechanism based on Hadamard product. This mechanism simulates the "masking effect" in human acoustic direction finding, that is, using microscopic features containing high-energy information of deterministic damage as "guide light sources" to directionally illuminate the fault evolution region in the macroscopic periodic feature map, thereby adaptively suppressing useless environmental white noise in strong mechanical background noise with extremely low signal-to-noise ratio (SNR). This collaborative purification strategy enables the model to maintain excellent generalization ability and robustness even with limited marine experimental samples.

[0060] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A nonlinear phase difference field multi-scale energy masking method for drilling winch mode recognition, characterized in that: Includes the following steps, S1: Obtain the steady-state response sequence decoupled from the heave load; S2: Construct the two-dimensional Grammy phase angle difference field matrix; S3: Analysis of the characteristics of micro-transient shocks and macro-cycle modulation in a dual-track system; S4: Feature collaboration and purification recognition based on cross-correlation mask.

2. The nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method according to claim 1, characterized in that: S1 includes the following steps: S11: Acquire the original dynamic response carrier signal of the winch drive system of the offshore drilling platform; S12: Use variational mode decomposition algorithm to extract the low-frequency reference envelope in the signal that corresponds to the wave rise and fall period; S13: Using the low-frequency reference envelope as the compensation reference, perform time-varying notch filtering on the high-frequency intrinsic mode components to remove the time-varying load modulation interference caused by the ship's pitch and heave, and reconstruct the steady-state response sequence characterizing the state of the winch's mechanical body.

3. The nonlinear phase difference field multi-scale energy mask drilling winch mode identification method according to claim 1 or 2, characterized in that: In step S2, the amplitude information of the steady-state response sequence is converted into an anticosine phase angle in polar coordinates. By calculating the sine value of the difference between the phase angles corresponding to any two sampling points on the time series, a two-dimensional Grami phase angle difference field space matrix characterizing the nonlinear stiffness degradation trend of the winch gear and bearing is constructed.

4. The nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method according to claim 3, characterized in that: In S3, a dual-track network architecture is used to perform deep parallel analysis of the spatial matrix, including: In the first analytical track, a local correlation analytical unit containing a multi-scale sliding filter kernel is used to perform a two-dimensional sliding window convolution operation on the spatial matrix to capture high-frequency abrupt pixel clusters caused by pitting of winch teeth or peeling of bearings, and to extract local transient impact energy feature vectors. In the second analytical track, the spatial matrix is ​​downsampled in parallel using spatial pooling to capture global periodic texture changes caused by winch wear and extract global periodic modulation feature maps.

5. The nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method according to claim 4, characterized in that: In the first analytical track, the receptive field size of the multi-scale sliding filter kernel strictly corresponds to the pixel span on the two-dimensional spatial matrix mapped by the mechanical meshing frequencies of the winch's high-speed, medium-speed, and low-speed gears.

6. The nonlinear phase difference field multi-scale energy mask drilling winch mode recognition method according to claim 4, characterized in that: S4 includes the following steps: S41: Using a nonlinear cross-correlation mask function, the amplitude of the local transient impact energy feature vector extracted from the first analytical trajectory is mapped to the [0,1] interval as a spatial attention weight mask; S42: Perform an element-wise Hadamard product operation on the spatial attention weight mask and the global periodic modulation feature map extracted by the second analytical track, and use the local high-energy impact features to adaptively suppress the low-frequency mechanical background white noise in the global features to generate a purified health assessment matrix.

7. The nonlinear phase difference field multi-scale energy mask drilling winch mode identification method according to claim 1 or 2, characterized in that: It also includes S5: Health Modal Nonlinear Mapping: Input the purified health assessment matrix into the nonlinear modal classifier and output the health modal assessment index of the winch.

8. A nonlinear phase difference field multi-scale energy masking drilling winch mode recognition system, characterized in that: The nonlinear phase difference field multi-scale energy mask drilling winch mode identification method according to any one of claims 1 to 7 includes, The signal acquisition module is used to acquire the original dynamic response carrier signal of the winch drive system of an offshore drilling platform under actual variable load conditions. The heave load decoupling module is used to perform multi-band analysis on the original dynamic response carrier signal, adaptively extract the low-frequency reference envelope corresponding to the wave heave period, use the low-frequency reference envelope as a compensation reference to perform time-varying notch filtering on the separated high-frequency intrinsic mode components, dynamically remove the time-varying load modulation interference caused by the ship's pitch and heave, and reconstruct a steady-state response sequence that only represents the true degradation state of the winch mechanical body. The Grammy phase angle difference field matrix construction module is used to map the amplitude information of the steady-state response sequence to the polar coordinate system and convert it into an inverse cosine phase angle sequence. By calculating the sine value of the difference between the phase angles corresponding to any two sampling points at different times, a two-dimensional Grammy phase angle difference field space matrix is ​​constructed. The dual-track feature parsing module is used to perform deep parallel parsing of the spatial matrix. It includes a first parsing track unit and a second parsing track unit. The first parsing track unit uses a local correlation parsing unit containing a multi-scale sliding filter kernel to perform a two-dimensional hierarchical sliding window convolution operation on the spatial matrix to extract local transient impact energy feature vectors. The second parsing track unit performs adaptive spatial pooling downsampling on the spatial matrix to extract global periodic modulation feature maps. The cross-correlation mask feature collaborative purification module is used to compress and map the amplitude of the local transient impact energy feature vector to the [0,1] interval to generate a spatial attention weight mask, and to perform element-wise Hadamard product operation on the spatial attention weight mask and the global periodic modulation feature map to generate a purified health assessment matrix. The health modality classification output module is used to input the purified health assessment matrix into the nonlinear modality classifier and output the accurate health modality assessment index of the current operating state of the winch.

9. A computer device comprising a memory, a processor, and an algorithm stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the drilling winch mode recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer algorithm, characterized in that, When the computer algorithm is executed by the processor, it implements the drilling winch mode recognition method as described in any one of claims 1 to 7.

Images