A wind turbine main shaft cooperative diagnosis method and device and electronic equipment

Abstract

The present application relates to the field of wind power operation and maintenance monitoring technology, and discloses a wind turbine main shaft cooperative diagnosis method, device and electronic equipment, the present application obtains the monitoring data of wind turbine main shaft; the monitoring data at least includes bearing geometric parameter, temperature signal and power signal; the working condition decoupling processing is carried out to monitoring data, and angle domain vibration sequence and / or dynamic thermal balance residual error are obtained; the pre-trained neural network model is used to reconstruct angle domain vibration sequence and / or dynamic thermal balance residual error, and abnormal score is obtained; the theoretical fault characteristic order is calculated according to bearing geometric parameter, and the pre-constructed expert knowledge graph is used for time sequence logical reasoning to the theoretical fault characteristic order, so that the fault evolution stage and fault type are determined; according to abnormal score, fault evolution stage and fault type, the main shaft cooperative diagnosis result is generated, and the present application realizes the deep integration of data driving and physical mechanism.

Description

Technical Field

[0001] This invention relates to the field of wind power generation operation and maintenance monitoring technology, specifically to a method, device, and electronic equipment for collaborative diagnosis of wind turbine main shafts. Background Technology

[0002] Currently, the operation and maintenance monitoring of wind turbine main shafts mainly relies on monitoring data and vibration signal-based monitoring methods. With the increasing demand for intelligent wind power operation and maintenance, how to achieve real-time, efficient, and accurate automatic status monitoring and fault early warning of the main shaft system is an important research direction in the field of wind power operation and maintenance monitoring technology.

[0003] However, existing solutions have significant drawbacks in practical applications: First, AI diagnostic models relying solely on pure data-driven approaches have low signal-to-noise ratios under extremely low spindle speeds, making it difficult to extract subtle early fault features and unable to diagnose the physical type of the fault. Second, mechanistic analysis methods based on fixed physical formulas fail when wind speed fluctuations cause frequent changes in rotational speed, easily leading to false alarms and missed alarms. Third, diagnostic accuracy is highly dependent on the engineer's personal experience, making it impossible to achieve real-time, automated fault classification and early warning. Summary of the Invention

[0004] This invention provides a method, device, and electronic equipment for collaborative diagnosis of wind turbine main shafts, in order to solve the problems of being unable to diagnose the physical type of fault, being prone to false alarms and missed alarms, and being unable to perform real-time automated diagnosis due to reliance on human experience.

[0005] In a first aspect, the present invention provides a method for collaborative diagnosis of the main shaft of a wind turbine, comprising: Acquire monitoring data of the wind turbine main shaft; the monitoring data should include at least bearing geometric parameters, temperature signals, and power signals; The monitoring data is decoupled from the operating conditions to obtain the angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and the temperature signal. Anomaly scores are obtained by reconstructing the angular domain vibration sequence and / or dynamic thermal equilibrium residual using a pre-trained neural network model. The theoretical fault characteristic order is calculated based on the bearing geometric parameters, and the fault evolution stage and fault type are determined by using a pre-constructed expert knowledge graph to perform temporal logical reasoning on the theoretical fault characteristic order. Based on the anomaly score, fault evolution stage, and fault type, generate spindle collaborative diagnostic results.

[0006] Beneficial Effects: By acquiring monitoring data including bearing geometric parameters, temperature signals, and power signals, and performing decoupling processing of operating conditions, this approach utilizes a pre-trained neural network model to reconstruct the angular domain vibration sequence and / or dynamic thermal balance residuals to output anomaly scores. Simultaneously, it leverages a pre-constructed expert knowledge graph combined with theoretical fault characteristic orders for temporal logical reasoning, achieving a deep integration of data-driven perception and expert mechanism reasoning. This overcomes the limitations of traditional methods that rely solely on single-dimensional indicators or black-box AI models. It can automatically eliminate operating condition interference under drastic fluctuations in wind speed and load, while also ensuring high diagnostic sensitivity and interpretability, generating spindle-based collaborative diagnostic results containing clearly defined fault physical types and stages. Furthermore, it possesses sensor compatibility and degradation capabilities, allowing it to degrade to mechanism residual-based monitoring even in the absence of vibration sensors, thus improving the engineering applicability of the solution.

[0007] In one optional implementation, the monitoring data also includes vibration signals and the corresponding instantaneous phase angle; The angular domain vibration sequence was obtained, including: By using the instantaneous phase angle to perform order resampling of the vibration signal, an angular domain vibration sequence is obtained.

[0008] Beneficial effects: By introducing vibration signals and corresponding instantaneous phase angles, and using the instantaneous phase angles to perform order resampling of the vibration signals, the non-stationary time-domain vibration signals are mapped to the angular domain with equal angular intervals; the spectral ambiguity caused by extremely low spindle speed and violent fluctuations is eliminated, and the weak fault overtone features are clearly visible on the order spectrum corresponding to the angular domain vibration sequence, which greatly improves the extraction accuracy of early fault features under varying working conditions.

[0009] In one optional implementation, the vibration signal is resampled in order using the instantaneous phase angle to obtain an angular domain vibration sequence, including: Calculate the instantaneous angular velocity and instantaneous angle of the wind turbine main shaft based on the instantaneous phase angle; By using the Shannon-Nyquist sampling theorem or a linear interpolation algorithm, vibration amplitudes at preset intervals are extracted from the vibration signal to obtain an angular domain vibration sequence.

[0010] Beneficial effects: By calculating the instantaneous angular velocity and instantaneous angle of the wind turbine main shaft based on the instantaneous phase angle, and using the Shannon-Nyquist sampling theorem or linear interpolation algorithm for amplitude extraction, it is ensured that discrete time sampling points can be accurately interpolated and mapped into discrete data with equal angular intervals under low speed and non-uniform sampling conditions.

[0011] In one optional implementation, the monitoring data also includes rotational speed signals and wind speed signals, and the temperature signals include ambient temperature and measured temperature of the main bearing. The dynamic thermal equilibrium residuals are obtained, including: The total frictional torque is calculated based on the power signal, speed signal, and temperature signal. The theoretical heat generation power is calculated based on the total frictional torque and speed signal. The dynamic thermal resistance is calculated based on the wind speed signal. The theoretical temperature rise at the current moment is calculated based on the theoretical heat generation power, ambient temperature, dynamic thermal resistance, and the theoretical temperature rise at the previous moment. The difference between the measured temperature of the main bearing and the theoretical temperature rise at the current moment is taken as the dynamic thermal balance residual.

[0012] Beneficial effects: The total friction torque and theoretical heat generation power are calculated by comprehensively considering power signals, speed signals, and temperature signals. Dynamic thermal resistance is calculated using wind speed signals. The theoretical temperature rise at the current moment is calculated iteratively by combining the theoretical temperature rise at the previous moment. A dynamic thermal model that can track seasonal changes, ambient temperature, and high-load power fluctuations in real time has been successfully constructed. The dynamic thermal balance residual obtained by subtracting the measured temperature of the main bearing from the theoretical temperature rise eliminates the temperature rise under normal operating conditions, solving the problem that traditional fixed threshold temperature alarms are prone to false alarms and missed alarms under high temperature or high load conditions.

[0013] In one alternative implementation, the neural network model is trained in the following manner: The wavelet packet decomposition algorithm was used to extract the feature signal corresponding to pitting of the main bearing from the vibration sequence in the angular domain. The feature signal is input into the neural network model to be trained for feature extraction and reconstruction to obtain the reconstructed feature signal; Based on the characteristic signal and the reconstructed characteristic signal, calculate the data reconstruction error term, and construct the physical consistency penalty term according to the preset spindle force balance equation or the preset minimum lubrication film thickness limit criterion. A loss function is constructed based on the data reconstruction error term and the physical consistency penalty term. The loss function is then used to train and validate the autoencoder in the neural network model, resulting in a neural network model containing an autoencoder with embedded physical constraints.

[0014] Beneficial effects: The wavelet packet decomposition algorithm is used to extract the characteristic signals of pitting at specific frequency bands from the vibration sequence in the angular domain, preventing the weak pulses generated by early wear of the spindle at extremely low speed and low energy from being drowned out by background noise; at the same time, a physical consistency penalty term based on the spindle force balance equation or the minimum limit criterion of lubricating film thickness is introduced into the loss function, so that the trained neural network model is strictly constrained by the mechanism law while learning the data pattern, reducing the false reconstruction rate of the pure data-driven model and improving the accuracy and robustness of the anomaly score.

[0015] In one alternative implementation, the bearing geometry parameters include roller diameter, pitch circle diameter, contact angle, and number of rollers; The theoretical fault characteristic order is calculated based on the bearing's geometric parameters, including: Based on the roller diameter, pitch circle diameter, contact angle, and number of rollers, calculate the characteristic order of outer ring damage, inner ring damage, roller self-rotation failure, and cage failure respectively. The damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the fault characteristic order of the roller rotation, and the fault characteristic order of the cage are combined into the theoretical fault characteristic order.

[0016] Beneficial effects: Based on specific bearing geometric parameters such as roller diameter, pitch circle diameter, contact angle, and number of rollers, the characteristic order of outer ring damage, inner ring damage, roller rotation failure, and cage failure is quantitatively calculated and combined into a theoretical fault characteristic order. This provides a theoretical frequency / order benchmark that strictly conforms to the physical structure for subsequent feature matching analysis of the expert knowledge graph, ensuring that the diagnostic results can accurately locate specific physical components.

[0017] In one alternative implementation, the expert knowledge graph is constructed in the following manner: The life cycle of the wind turbine main shaft is divided into multiple stages, including the normal operation stage, the initial lubrication degradation stage, the intermediate friction and vibration stage, the later geometric damage stage, and the final temperature rise runaway stage. Determine the state transition probability matrix between different stages based on historical failure data and prior knowledge; A dynamic Bayesian network is constructed based on the stage and state transition probability matrix, and an expert knowledge graph is built using the dynamic Bayesian network.

[0018] Beneficial effects: The life cycle of the main bearing is divided into five physical stages: normal operation, initial lubrication degradation, mid-term frictional vibration, late-stage geometric damage, and final temperature rise runaway. A dynamic Bayesian network is constructed based on historical failure data and prior knowledge constraints. The irreversible evolution process of main bearing damage is expressed by the state transition probability matrix, which overcomes the shortcomings of traditional static diagnosis and isolated single index. It realizes dynamic temporal probabilistic reasoning of fault evolution trend and current physical stage.

[0019] In one optional implementation, a pre-built expert knowledge graph is used to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type, including: Using expert knowledge graphs, temporal logic reasoning is performed on angular domain vibration sequences and dynamic thermal equilibrium residuals to determine the fault evolution stages; The system health is determined based on the fault evolution stage, and the order spectrum is obtained by performing order spectrum analysis on the angular domain vibration sequence. If the dynamic thermal balance residual does not continue to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be outer ring damage. If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is less than or equal to the preset amplitude threshold, then the fault type is determined to be lubrication failure. If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be severe wear accompanied by thermal damage.

[0020] Beneficial effects: The system health is calculated by temporal logic reasoning, and multi-condition joint threshold judgment is made by using the dynamic thermal balance residual and the amplitude of the order spectrum at any octave of the theoretical fault characteristic order. The three complex fault types, namely simple structural damage, pure lubrication failure and severe wear accompanied by thermal damage, are completely mutually exclusive logically divided, eliminating cross-judgment and giving the diagnostic results clear physical diagnostic significance.

[0021] In one optional implementation, a spindle collaborative diagnostic result is generated based on the anomaly score, fault evolution stage, and fault type, including: The feature matching degree between the order spectrum and the theoretical fault feature order is calculated using the Gaussian kernel window function; Based on the speed signal and system health in the monitoring data, the mechanism weight coefficient is determined by an adaptive weight algorithm. Based on the mechanism weight coefficient, the anomaly score and feature matching degree are weighted and summed to obtain the diagnostic confidence. Based on diagnostic confidence, fault evolution stage, and fault type, generate spindle collaborative diagnostic results.

[0022] Beneficial effects: By using the Gaussian kernel window function to calculate the feature matching degree between the measured order spectrum and the theoretical fault feature order, it can tolerate the drift of the feature spectrum caused by sensor installation deviation, slight speed slippage, or slight change in contact angle under heavy load. At the same time, by using the speed signal and system health status to determine the mechanism weight coefficient through an adaptive weighting algorithm, it realizes the dynamic allocation of anomaly score and feature matching degree under different speeds and different degrees of degradation, ensuring that the generated diagnostic confidence is objective and reliable, and can directly guide automated operation and maintenance decisions.

[0023] Secondly, the present invention provides a wind turbine main shaft collaborative diagnostic device, comprising: The acquisition module is used to acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals, and power signals. The decoupling module is used to decouple the monitoring data under operating conditions to obtain the angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and temperature signal. The reconstruction module is used to reconstruct the angular domain vibration sequence and / or dynamic thermal equilibrium residual using a pre-trained neural network model to obtain anomaly scores. The determination module is used to calculate the order of theoretical fault characteristics based on bearing geometric parameters, and to perform temporal logical reasoning on the order of theoretical fault characteristics using a pre-built expert knowledge graph to determine the fault evolution stage and fault type. The generation module is used to generate spindle collaborative diagnostic results based on anomaly scores, fault evolution stages, and fault types.

[0024] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform a wind turbine main shaft collaborative diagnosis method according to the first aspect or any corresponding embodiment described above.

[0025] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute a wind turbine main shaft collaborative diagnosis method according to the first aspect or any corresponding embodiment described above.

[0026] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute a wind turbine main shaft collaborative diagnosis method according to the first aspect or any corresponding embodiment described above. Attached Figure Description

[0027] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the first step of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention; Figure 2 This is a second flowchart illustrating a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the process of training a neural network model according to an embodiment of the present invention; Figure 4This is a schematic diagram of the third process of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention; Figure 6 This is a structural block diagram of a wind turbine main shaft collaborative diagnostic device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0030] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0031] To address the problems of purely data-driven AI models being black boxes with low signal-to-noise ratios at low speeds, fixed-frequency mechanism models being prone to false alarms and missed alarms under varying wind speeds, and the inability to automate diagnosis due to high reliance on human experience in related technologies, this invention decouples multi-source heterogeneous monitoring data under different operating conditions to obtain angular domain vibration sequences and dynamic thermal balance residuals. Secondly, it uses a neural network model embedded with physical constraints for reconstruction calculations, solving the problem of weak pulses being easily ignored at low speeds and outputting anomaly scores. Finally, combining the theoretical fault characteristic order calculated from bearing geometric parameters, it uses an expert knowledge graph based on a dynamic Bayesian network for temporal logical reasoning, overcoming reliance on human experience and achieving automated collaborative diagnosis of the spindle fault evolution stage and clearly defined physical fault types.

[0032] According to an embodiment of the present invention, a method for collaborative diagnosis of the main shaft of a wind turbine is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] This embodiment provides a method for collaborative diagnosis of wind turbine main shafts, which can be used in the aforementioned electronic equipment. Figure 1This is a schematic flowchart of the first step in a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the process includes the following steps: Step S101: Obtain monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals and power signals.

[0034] Specifically, monitoring data refers to multi-source heterogeneous data collected for comprehensive evaluation of the physical operating status of the wind turbine main shaft. Bearing geometric parameters are fixed parameters characterizing the internal physical dimensions of the wind turbine main bearing, including roller diameter, pitch circle diameter, contact angle, and number of rollers. Temperature signals refer to the measured temperature of the main bearing and the ambient temperature of the wind turbine's environment, collected using temperature sensors. Power signals refer to the instantaneous power during main shaft operation, collected by the wind turbine monitoring system.

[0035] Acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals, and power signals. Connect to the wind turbine's signal acquisition front-end and monitoring system via a data interface to read the bearing geometric parameters, temperature signals, and power signals of the wind turbine main shaft in real time or at scheduled intervals. During main shaft operation, the bearing geometric parameters determine the baseline for the inherent fault characteristic frequency of the main bearing structure; the temperature signal directly reflects the heat accumulation state of the main bearing under frictional power consumption; and the power signal characterizes the load and operating condition of the main shaft. Complete acquisition of the above multi-source data provides fundamental data support for subsequent elimination of operating condition fluctuation interference and the fusion diagnosis of mechanism and data-driven approaches. Additionally, if the main bearing geometric parameters are unknown, a lateral comparison model of other wind turbines in the same field can be used to replace the calculation of specific physical formulas to obtain baseline parameters.

[0036] Step S102: Perform operating condition decoupling processing on the monitoring data to obtain the angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and temperature signal.

[0037] Specifically, operating condition decoupling refers to the process of using physical mechanism models and mathematical mapping transformations to eliminate interference from non-faulty operating conditions such as wind speed fluctuations, drastic changes in rotational speed, and high-load operation on the observed data. Angle-domain vibration sequence refers to mapping and transforming non-stationary time-domain vibration signals collected at equal time intervals into a discrete vibration amplitude sequence arranged at equal angular intervals. Dynamic thermal balance residual refers to the difference between the measured temperature of the main bearing and the theoretical temperature rise calculated using theoretical heat generation models and dynamic thermal resistance models.

[0038] The monitoring data undergoes decoupling processing to obtain the angular domain vibration sequence and / or dynamic thermal balance residual. The dynamic thermal balance residual is calculated based on power and temperature signals. During processing, on the one hand, considering the extremely low and drastically fluctuating spindle speed, a pulse generator is used to acquire the instantaneous phase angle of the spindle, mapping the time-domain vibration signal to an angular domain with equal angular intervals. This eliminates the spectral ambiguity caused by speed fluctuations, resulting in the angular domain vibration sequence. On the other hand, a theoretical heat generation model for the main bearing is constructed using expert experience. The theoretical heat generation power and theoretical temperature rise of the main bearing are calculated by combining the power and temperature signals. The difference between the measured temperature of the main bearing and the theoretical temperature rise in the temperature signal is then used to obtain the dynamic thermal balance residual. This eliminates the influence of normal temperature rise caused by seasonal changes and high-load operation in strong winds, achieving decoupling of vibration and thermal characteristics. If there is no CMS (Condition Monitoring System), i.e., vibration sensors, this solution can degenerate into mechanism residual diagnosis based on SCADA (Supervisory Control And Data Acquisition) data, which is identified by the nonlinear offset of the power-temperature rise curve. SCADA data refers to the operating data collected by the conventional SCADA system of the wind farm, mainly including: power signal, temperature signal, speed signal, wind speed signal, etc.

[0039] Step S103: The angular domain vibration sequence and / or dynamic thermal equilibrium residual are reconstructed using a pre-trained neural network model to obtain anomaly scores.

[0040] Specifically, a pre-trained neural network model refers to a deep autoencoder network model that has been pre-trained using historical data and incorporates physical mechanism constraints into its loss function. Anomaly score is a numerical indicator used to quantify the degree to which current measured data deviates from normal physical laws and the degree of reconstructed distribution.

[0041] An anomaly score is obtained by reconstructing the angular domain vibration sequence and / or dynamic thermal balance residual using a pre-trained neural network model. The angular domain vibration sequence and / or dynamic thermal balance residual are input into the pre-trained neural network model, which performs deep feature extraction and reconstruction through its internal neural network architecture. Because the neural network model incorporates physical consistency constraints during pre-training, it prevents weak fault pulses at low speeds from being drowned out by background noise. When the main bearing is in a healthy state, the model's reconstruction error is minimal; however, when the main bearing exhibits early wear or degradation, the measured data disrupts the original physical and data reconstruction patterns, leading to an increased reconstruction error. Based on this, the model calculates and outputs an anomaly score representing the degree of deviation from the current main bearing's health. Furthermore, in practical applications, the neural network model can be replaced by a model scheme combining variational mode decomposition and support vector machines.

[0042] Step S104: Calculate the theoretical fault characteristic order based on the bearing geometric parameters, and use the pre-constructed expert knowledge graph to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type.

[0043] Specifically, the theoretical fault characteristic order refers to the set of theoretical order values ​​calculated based on the specific geometric dimensions of the main bearing, used to characterize the vibration and impact period when different physical components of the bearing experience damage. The pre-constructed expert knowledge graph refers to a temporal causal logic inference graph that integrates the physical evolution laws of bearing failure and is built on a dynamic Bayesian network. Temporal logic reasoning refers to the process of using a dynamic Bayesian network to calculate the probability of occurrence of each deterioration stage of the main bearing based on the evolution trend of observed indicators along the time axis. The fault evolution stage refers to the specific deterioration period in the life cycle of the main bearing. The fault type refers to the specific category of physical damage that occurs in the main bearing.

[0044] The theoretical fault characteristic order is calculated based on the bearing's geometric parameters, and a pre-built expert knowledge graph is used to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type. Based on the obtained bearing geometric parameters, the corresponding theoretical fault characteristic order is automatically calculated using formulas. When an abnormal score indicates an abnormal risk, the pre-built expert knowledge graph is automatically triggered. This expert knowledge graph uses a dynamic Bayesian network to perform temporal causal matching and reasoning with the angular domain vibration sequence and dynamic thermal balance residuals. It assesses the state transition probabilities between the normal operation stage, the initial lubrication degradation stage, the intermediate friction excitation stage, the later geometric damage stage, and the final temperature rise runaway stage, thereby accurately determining the current fault evolution stage and specific fault type of the spindle.

[0045] Step S105: Generate spindle collaborative diagnosis results based on anomaly score, fault evolution stage, and fault type.

[0046] Specifically, the spindle collaborative diagnosis result refers to a comprehensive diagnostic report that integrates the quantitative anomaly score output by the data-driven layer and the physical damage information inferred by the expert mechanism layer, and includes clear operation and maintenance recommendations.

[0047] Based on anomaly scores, fault evolution stages, and fault types, a collaborative diagnostic result for the main shaft is generated. The anomaly scores output from the data-driven loop are weighted and integrated with the fault evolution stages and fault types determined by the expert mechanism loop. By combining anomaly sensitivity with physical interpretability, pre-set automated processing plans from the expert database are automatically matched and invoked to generate a collaborative diagnostic result for the main shaft, including the specific fault location component, the degree of damage, diagnostic confidence level, and maintenance recommendations. This enables real-time, efficient, and highly interpretable automatic status monitoring and fault early warning for the wind turbine main shaft.

[0048] This embodiment provides a collaborative diagnostic method for wind turbine main shafts. By acquiring monitoring data including bearing geometric parameters, temperature signals, and power signals, and performing decoupling processing of operating conditions, it utilizes a pre-trained neural network model to reconstruct the angular domain vibration sequence and / or dynamic thermal balance residuals to output anomaly scores. Simultaneously, it employs a pre-constructed expert knowledge graph combined with theoretical fault characteristic orders for temporal logical reasoning, achieving a deep integration of data-driven perception and expert mechanism reasoning. This method overcomes the limitations of traditional methods that rely solely on single-dimensional indicators or black-box AI models. It can automatically eliminate operating condition interference under drastic fluctuations in wind speed and load, while also ensuring high diagnostic sensitivity and interpretability, generating collaborative diagnostic results for the main shaft that clearly define the physical type and stage of the fault. Furthermore, it possesses sensor compatibility degradation capability, allowing it to degrade to monitoring based on mechanism residuals even in the absence of vibration sensors, thus improving the engineering applicability of the solution.

[0049] This embodiment provides a method for collaborative diagnosis of wind turbine main shafts, which can be used in the aforementioned electronic equipment. Figure 2 This is a schematic diagram of the second process of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention, as shown below. Figure 2 As shown, the process includes the following steps: Step S201: Acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals, and power signals. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0050] Step S202: Perform operating condition decoupling processing on the monitoring data to obtain the angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and temperature signal.

[0051] Specifically, step S202 includes: Step S2021: The vibration signal is resampled by order using the instantaneous phase angle to obtain the angular domain vibration sequence.

[0052] Specifically, the angular domain vibration sequence needs to be obtained based on the vibration signal. The angular domain vibration sequence can only be obtained when the collected monitoring data includes the vibration signal. That is, if there is no vibration sensor (CMS), this embodiment can degenerate into mechanism residual diagnosis based on SCADA data, which is identified by the nonlinear shift of the power-temperature rise curve, that is, identification is only based on the dynamic thermal balance residual.

[0053] Therefore, it can be determined that when the monitoring data contains vibration signals, the angular domain vibration sequence and dynamic thermal balance residual can be obtained simultaneously after the monitoring data is decoupled from the operating conditions. When the monitoring data does not contain vibration signals, only the dynamic thermal balance residual can be obtained after the monitoring data is decoupled from the operating conditions.

[0054] Instantaneous phase angle refers to the value obtained using a pulse generator that characterizes the position of the wind turbine's main shaft at any given moment. The rotation angle characteristic quantity is denoted as The vibration signal refers to the time-domain vibration acceleration signal of the main bearing acquired using a high-frequency vibration sensor, denoted as... Order resampling refers to the process of transforming and mapping vibration signals sampled at equal time intervals into vibration amplitude sequences sampled at equal angle intervals.

[0055] The vibration signal is resampled in order using the instantaneous phase angle to obtain an angular domain vibration sequence. By receiving the instantaneous phase angle generated by the pulse generator, the non-stationary vibration signal collected in the time domain is resampled and its amplitude mapped according to the equal angular intervals of the spindle rotation to obtain an angular domain vibration sequence that eliminates the influence of rotational speed fluctuations.

[0056] Specifically, step S2021 above includes: Step a1: Calculate the instantaneous angular velocity and instantaneous angle of the wind turbine main shaft based on the instantaneous phase angle.

[0057] Specifically, instantaneous angular velocity refers to the velocity of the wind turbine's main shaft at a given time, calculated based on the time difference between adjacent pulses. The rotational angular velocity, denoted as The instantaneous angle refers to the real-time rotation angle of the principal axis calculated by interpolating the instantaneous angular velocity over time, denoted as . .

[0058] The instantaneous angular velocity and instantaneous angle of the wind turbine's main shaft are calculated based on the instantaneous phase angle. The instantaneous phase angle is acquired using a pulse generator. By calculating the time difference between the arrival of adjacent pulses, the instantaneous angular velocity of the spindle at each moment can be deduced. Based on this, the speed at any given time point is calculated using real-time rotational speed through integration and interpolation. The actual instantaneous angle rotated by the spindle .

[0059] Step a2: Use the Shannon-Nyquist sampling theorem or a linear interpolation algorithm to extract the vibration amplitude at preset intervals from the vibration signal to obtain the angular domain vibration sequence.

[0060] Specifically, the preset interval time refers to the set of specific time points corresponding to the original time axis, calculated by reversing back from equal angular intervals, denoted as... The Shannon-Nyquist sampling theorem refers to band-limited interpolation resampling algorithms (such as the Shannon resampling algorithm) used to avoid spectral aliasing when resampling signals.

[0061] The vibration amplitude at preset time intervals is extracted from the vibration signal using the Shannon-Nyquist sampling theorem or a linear interpolation algorithm to obtain the angular domain vibration sequence. Given the instantaneous angle... Based on this, we return to the original vibration signal sequence collected at equal time intervals. In this process, the Shannon-Nyquist sampling theorem or linear interpolation algorithms are used to extract the time values ​​at a specific moment. The corresponding vibration amplitude is used to convert the vibration data into a discrete sequence with angle as the horizontal axis. This yields the angular domain vibration sequence, eliminating the spectral ambiguity caused by drastic changes in spindle speed due to wind speed fluctuations.

[0062] Step S2022: Calculate the total frictional torque based on the power signal, speed signal, and temperature signal; calculate the theoretical heat generation power based on the total frictional torque and speed signal; and calculate the dynamic thermal resistance based on the wind speed signal.

[0063] Specifically, the total frictional torque refers to the sum of mechanical torques that impede the relative motion of the main bearing during operation, denoted as . Theoretical heat generation power refers to the total theoretical heat generation power inside the bearing calculated based on the principles of physical mechanics and thermodynamics, denoted as . Dynamic thermal resistance refers to the system's equivalent thermal resistance, which changes in real time due to variations in the convective heat transfer coefficient caused by changes in the wind speed at the top of the tower. It is denoted as... .

[0064] The total frictional torque is calculated based on the power, speed, and temperature signals. The theoretical heat generation power is calculated based on the total frictional torque and speed signals. The dynamic thermal resistance is calculated based on the wind speed signal. (Total frictional torque of the main bearing) It consists of four linearly superimposed parts, as shown in formula (1): (1) Among them, rolling friction torque refers to the torque component generated between the raceway and the rolling elements due to inlet shear and deformation, denoted as . The calculation formula is shown in formula (2): (2) In the formula, To reduce the inlet shear heating coefficient, The coefficient for reducing kinematic reflux / starvation. The rolling friction geometric load factor is determined by the bearing type, structural parameters, and equivalent load. The kinematic viscosity of the lubricating oil (grease) at the operating temperature is dynamically corrected from the measured temperature of the main bearing in the temperature signal using the Vogel-Cameron relation. The instantaneous spindle speed is obtained in real time from SCADA. Let be the bearing pitch circle diameter. Sliding friction torque refers to the torque component generated between contact surfaces due to relative sliding, denoted as . The calculation formula is shown in formula (3): (3) In the formula, The sliding friction geometric load factor, The coefficient of sliding friction is denoted as . Lubricant drag torque refers to the torque component generated by the shear drag resistance of the lubricating grease, denoted as . The calculation formula is shown in formula (4): (4) In the formula, This is the damping coefficient related to the grease filling amount and viscosity of the bearing cavity. The sealing friction torque refers to the fixed torque component generated by the friction of the seal lip, denoted as... Its calculation formula is shown in formula (5): (5) In the formula, , , These are fixed constants related to the sealing material and lip design. The total frictional torque is calculated based on... Then, based on the physical equation for the conversion of frictional power consumption into heat during the operation of the main bearing, and combined with the instantaneous angular velocity, the theoretical heat generation power was calculated. As shown in formula (6): (6) Meanwhile, based on the tower top wind speed in the wind speed signal Calculate the dynamic thermal resistance of the system using the thermal resistance function. As shown in formula (7): (7) In the formula, , and is the fitting constant for convective heat transfer.

[0065] Step S2023: Calculate the theoretical temperature rise at the current moment based on the theoretical heat generation power, ambient temperature, dynamic thermal resistance, and the theoretical temperature rise at the previous moment.

[0066] Specifically, the theoretical temperature rise refers to the theoretically predicted temperature of the main bearing calculated using the backward difference method based on the continuous differential equation, denoted as... Ambient temperature refers to the ambient temperature at the top of the wind turbine tower obtained from SCADA, denoted as . .

[0067] The theoretical temperature rise at the current moment is calculated based on the theoretical heat generation power, ambient temperature, dynamic thermal resistance, and the theoretical temperature rise at the previous moment. The continuous differential equation for the conversion of mechanical energy to thermal energy is applied using a time step of [missing information]. Discretize using the backdifference method to obtain the current time step. Theoretical temperature rise The calculation formula is shown in formula (8): (8) In the formula, for The ambient temperature at any given time This is the system heat capacity.

[0068] Step S2024: The difference between the measured temperature of the main bearing and the theoretical temperature rise at the current moment is taken as the dynamic thermal balance residual.

[0069] Specifically, the measured temperature of the main bearing refers to the actual temperature collected in real time by a temperature sensor installed on the main bearing housing, denoted as . .

[0070] The difference between the measured temperature of the main bearing and the theoretical temperature rise at the current moment is taken as the dynamic thermal balance residual. The measured temperature of the main bearing... With theoretical temperature rise The thermal characteristic residual index is obtained by subtraction. It is used to eliminate the impact of normal temperature rise caused by seasonal changes and high-load operation in strong winds.

[0071] Step S203: Reconstruct the angular domain vibration sequence and / or dynamic thermal equilibrium residuals using a pre-trained neural network model to obtain anomaly scores. For details, please refer to [link to relevant documentation]. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0072] Step S204: Calculate the theoretical fault characteristic order based on the bearing geometric parameters, and use a pre-constructed expert knowledge graph to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type. For details, please refer to [link to relevant documentation]. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0073] Step S205: Generate spindle collaborative diagnostic results based on anomaly score, fault evolution stage, and fault type. For details, please refer to [link to relevant documentation]. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0074] This embodiment provides a collaborative diagnostic method for wind turbine main shafts, which makes subtle fault overtone characteristics clearly visible on the order spectrum corresponding to the angular domain vibration sequence, significantly improving the extraction accuracy of early fault characteristics under varying operating conditions. Furthermore, a dynamic thermal model capable of tracking seasonal changes, ambient temperature, and high-load power fluctuations in real time is constructed. The dynamic thermal balance residual obtained by subtracting the measured temperature from the theoretical temperature rise of the main bearing eliminates the temperature rise under normal operating conditions, solving the problem of traditional fixed-threshold temperature alarms being prone to false alarms and missed alarms under high-temperature or high-load conditions.

[0075] This embodiment provides a method for collaborative diagnosis of wind turbine main shafts, which can be used in the aforementioned electronic equipment. Figure 4 This is a schematic diagram of the third process of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention, as shown below. Figure 4 As shown, the process includes the following steps: Step S401: Acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals, and power signals. For details, please refer to [link to relevant documentation]. Figure 2 Step S201 of the illustrated embodiment will not be described again here.

[0076] Step S402 involves decoupling the monitoring data to obtain the angular domain vibration sequence and / or dynamic thermal balance residuals; wherein the dynamic thermal balance residuals are calculated based on the power and temperature signals. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0077] Step S403: The angular domain vibration sequence and / or dynamic thermal equilibrium residual are reconstructed using a pre-trained neural network model to obtain anomaly scores.

[0078] In one alternative implementation, Figure 3 This is a schematic diagram of the process of training a neural network model according to an embodiment of the present invention, such as... Figure 3 As shown, the neural network model is trained in the following way: Step b1: Use wavelet packet decomposition algorithm to extract the feature signal corresponding to pitting of the main bearing from the angular domain vibration sequence.

[0079] Specifically, wavelet packet decomposition algorithm refers to a multi-scale signal analysis algorithm capable of finely dividing high-frequency bands into equal-width segments. Feature signals refer to specific frequency band characteristic signals extracted from wavelet packet decomposition, containing the impact energy of early pitting corrosion in main bearings.

[0080] Wavelet packet decomposition was used to extract the characteristic signals corresponding to pitting wear of the main bearing from the angular domain vibration sequence. Wavelet packet decomposition can subdivide the high-frequency band, thereby refining the specific frequency band characteristic signals that may lead to early pitting wear of the main bearing. These signals are then used as input to the neural network model, ensuring that the model does not ignore weak pulses due to low spindle speed and low energy during learning.

[0081] Step b2: Input the feature signal into the neural network model to be trained for feature extraction and reconstruction to obtain the reconstructed feature signal.

[0082] Specifically, reconstructing a feature signal refers to the output signal generated after feature signal input into a neural network model and subsequent feature extraction, encoding, and decoding. This is denoted as... .

[0083] The feature signal is input into the neural network model to be trained for feature extraction and reconstruction, resulting in a reconstructed feature signal. The wavelet packet-processed feature signal is then input into an autoencoder network containing an encoder and decoder, where deep feature extraction is performed to reconstruct the feature signal. .

[0084] Step b3: Calculate the data reconstruction error term based on the characteristic signal and the reconstructed characteristic signal, and construct a physical consistency penalty term based on the preset spindle force balance equation or the preset minimum lubrication film thickness limit criterion.

[0085] Specifically, the data reconstruction error term refers to the loss function term used to quantify the mean square error between the input feature signal and the reconstructed feature signal, denoted as... The physical consistency penalty term refers to a penalty function term constructed based on physical criteria to constrain the updating of model parameters, denoted as... The preset spindle force balance equation or preset minimum lubrication film thickness limit criterion is constructed by calculating the degree to which the input signal deviates from the physical criteria. The penalty constraint is then formed by multiplying it by a weight.

[0086] Based on the characteristic signal and the reconstructed characteristic signal, the data reconstruction error term is calculated, and a physical consistency penalty term is constructed according to the preset spindle force balance equation or the preset minimum lubrication film thickness limit criterion. This is achieved by calculating the input characteristic signal. With reconstructed feature signals The squared L2 norm between the two norms is used to construct the data reconstruction error term, while the signal is calculated according to the physical criteria. For example, the degree of violation of the spindle force balance equation or the minimum limit of lubricating film thickness can be used to construct a physical consistency penalty term.

[0087] Step b4: Construct a loss function based on the data reconstruction error term and the physical consistency penalty term. Use the loss function to train and verify the autoencoder in the neural network model to obtain a neural network model containing an autoencoder with embedded physical constraints.

[0088] Specifically, the loss function refers to the overall objective function that guides the optimization of neural network parameters, denoted as... .

[0089] A loss function is constructed based on the data reconstruction error term and the physical consistency penalty term. This loss function is then used to train and validate the autoencoder in the neural network model, resulting in a neural network model that includes an autoencoder with embedded physical constraints. The total loss function is... The calculation formula is shown in formula (9): (9) In the formula, For the input characteristic signal, The reconstructed feature signal is output. Reconstructing error terms for data , Represents physical consistency penalty item , and These are the corresponding weight coefficients. Training and validation are performed using this loss function to obtain an autoencoder neural network model that incorporates embedded physical constraints.

[0090] Step S404: Calculate the theoretical fault characteristic order based on the bearing geometric parameters, and use the pre-constructed expert knowledge graph to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type.

[0091] Specifically, step S404 includes: Step S4041: Calculate the damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller self-rotation fault, and the cage fault characteristic order based on the roller diameter, pitch circle diameter, contact angle, and number of rollers.

[0092] Specifically, the roller diameter refers to the outer diameter of the rolling elements of the main bearing, denoted as . The pitch circle diameter refers to the diameter of the distribution circle around the centerline of the rolling elements of the main bearing, denoted as . The contact angle is the angle between the line of action of the contact force between the rolling element and the raceway and the radial plane of the bearing, denoted as . The number of rollers refers to the total number of rolling elements inside the main bearing, denoted as . The outer ring damage characteristic order refers to the inherent impact order corresponding to localized damage to the raceway of the outer ring of the main bearing, denoted as... The inner ring damage characteristic order refers to the inherent impact order corresponding to localized damage to the raceway of the inner ring of the main bearing, denoted as... The characteristic order of roller rotation failure refers to the inherent rotational impact order corresponding to local damage on the surface of the rolling element, denoted as... The cage failure characteristic order refers to the inherent operating order corresponding to the failure of the bearing cage, denoted as... .

[0093] Based on the roller diameter, pitch circle diameter, contact angle, and number of rollers, the characteristic order of outer ring damage, inner ring damage, roller rotation failure, and cage failure are calculated. Based on the obtained bearing geometric parameters, the corresponding four key failure characteristic orders are calculated using the following geometrical-physical formulas: Outer ring damage characteristic order The calculation formula is shown in formula (10): (10) Inner ring damage characteristic order The calculation formula is shown in formula (11): (11) Roller self-rotation fault characteristic order The calculation formula is shown in formula (12): (12) Cage Failure Characteristic Order The calculation formula is shown in formula (13): (13) Step S4042: Combine the outer ring damage characteristic order, inner ring damage characteristic order, roller self-rotation fault characteristic order, and cage fault characteristic order into a theoretical fault characteristic order.

[0094] Specifically, the damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller self-rotation fault characteristic order, and the cage fault characteristic order are combined into a theoretical fault characteristic order, denoted as the set of order eigenvalues. .

[0095] In one alternative implementation, the expert knowledge graph is constructed in the following manner: Step c1 divides the life cycle of the wind turbine main shaft into multiple stages, including the normal operation stage, the initial lubrication degradation stage, the intermediate friction and vibration stage, the later geometric damage stage, and the final temperature rise runaway stage.

[0096] Specifically, multiple stages refer to the stages used to describe the main bearing failure over time. The state variables of the non-evolving stage are denoted as: The normal operation phase is denoted as... The initial lubrication degradation stage is denoted as This means that the performance of the lubricating grease has deteriorated, resulting in a residual temperature rise. Weak fluctuations but no order of impact; the intermediate friction-induced vibration stage is denoted as This refers to the weak pitting corrosion of the raceway, where the energy rises at high frequencies but the temperature remains normal; the later geometric damage stage is denoted as... This refers to visible fatigue spalling, with a significant increase in characteristic order and its harmonic energy; the final stage of uncontrolled temperature rise is denoted as... This refers to large-area peeling accompanied by thermal damage and residual temperature rise. A sharp increase.

[0097] The life cycle of the wind turbine main shaft is divided into multiple stages, namely: , , , as well as Five physical evolutionary periods.

[0098] Step c2: Determine the state transition probability matrix between different stages based on historical failure data and prior knowledge.

[0099] Specifically, the state transition probability matrix is ​​a matrix used to describe the probability of the main bearing transitioning from its current state to the next state between adjacent time steps, denoted as matrix. , of which elements Because the bearing damage is irreversible, the transfer matrix... The constraint is an upper triangular matrix, that is, when hour, As shown in formula (14): (14) Determine the state transition probability matrix between different stages based on historical failure data and prior knowledge. Prior knowledge refers to expert prior experience data established through bearing design life, Weibull distribution curves, etc. The method for obtaining the state transition probability matrix involves using discretized SCADA and CMS monitoring data throughout the entire historical failure lifecycle, employing MLE (Maximum Likelihood Estimation) or EM (Expectation-Maximization Algorithm) algorithms to calculate the average duration between each state. The transition probability is calculated as shown in formula (15): (15) When insufficient samples are available, the system is configured and updated online using prior knowledge. By statistically analyzing historical failure data of the same type of unit and combining this with prior knowledge, the system is revised to determine the state transition probability matrix with physical triangular constraints. .

[0100] Step c3: Construct a dynamic Bayesian network based on the stage and state transition probability matrix, and use the dynamic Bayesian network to construct an expert knowledge graph.

[0101] Specifically, dynamic Bayesian networks refer to networks used to describe spindle faults over time. A DBN (Dynamic Bayesian Network) for evolutionary probabilistic processes. Observation nodes refer to... Set of physical feature indicators extracted at each moment ,in These are lubrication anomaly factors (originating from the analysis of subtle trends in oil particle size or temperature). It is the vibration anomaly factor (derived from the reconstruction anomaly score of the physical constraint autoencoder). The degree of matching of the order spectrum features (derived from the degree of matching between the measured vibration order and the physically calculated order K).

[0102] A dynamic Bayesian network is constructed based on the stage and state transition probability matrix, and an expert knowledge graph is built using the dynamic Bayesian network. Five discrete state nodes are used as state variables, combined with observation nodes... and the state transition probability matrix We construct dynamic Bayesian networks and then build an expert knowledge graph with time-series causal inference capabilities.

[0103] Step S4043: Use expert knowledge graph to perform temporal logic reasoning on the angular domain vibration sequence and dynamic thermal balance residual to determine the fault evolution stage.

[0104] Specifically, an expert knowledge graph is used to perform temporal logical reasoning on the angular domain vibration sequence and dynamic thermal equilibrium residuals to determine the fault evolution stage. The probability of being in each evolution stage at the current moment is calculated on the time axis using the expert knowledge graph, thus identifying the most likely fault evolution stage.

[0105] Step S4044: Determine the system health based on the fault evolution stage, and perform order spectrum analysis on the angular domain vibration sequence to obtain the order spectrum.

[0106] Specifically, system health refers to the comprehensive probability, estimated in real time using a dynamic Bayesian network, that the main bearing is in normal operation or in a state of initial slight degradation, denoted as As shown in formula (16): (16) In the formula This refers to the set of observation nodes from historical data to the current moment. The order spectrum refers to the discrete order spectral lines obtained after performing spectral analysis on the angular domain vibration sequence, denoted as... .

[0107] The system health is determined based on the fault evolution stages, and the order spectrum is obtained by performing order spectrum analysis on the angular domain vibration sequence. The system health is then calculated by summing the stage probabilities derived from the dynamic Bayesian network. Simultaneously, for the angular domain vibration sequence Fourier transform spectral analysis was performed to obtain the order spectrum. .

[0108] Step S4045: If the dynamic thermal balance residual does not continue to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be outer ring damage.

[0109] Specifically, the preset amplitude threshold refers to the decision threshold defined based on the significance test of the noise baseline amplitude and energy.

[0110] If the dynamic thermal equilibrium residual does not continue to increase, and the order spectrum is at the theoretical fault characteristic order, such as the outer ring damage characteristic order... If the amplitude at any harmonic frequency is greater than the preset amplitude threshold, the fault type is determined to be outer ring damage.

[0111] Step S4046: If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is less than or equal to the preset amplitude threshold, then the fault type is determined to be lubrication failure.

[0112] Specifically, if the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is less than or equal to the preset amplitude threshold, then the fault type is determined to be lubrication failure.

[0113] Step S4047: If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be severe wear accompanied by thermal damage.

[0114] Specifically, if the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be severe wear accompanied by thermal damage.

[0115] Step S405: Generate spindle collaborative diagnostic results based on anomaly score, fault evolution stage, and fault type.

[0116] Specifically, step S405 includes: Step S4051: Calculate the feature matching degree between the order spectrum and the theoretical fault feature order using the Gaussian kernel window function.

[0117] Specifically, the Gaussian kernel window function is an attenuation window function used to quantify the degree of attenuation of the deviation between the measured order peak and the theoretical order harmonics. The feature matching degree is a numerical index that evaluates the degree of agreement between the significant characteristic peaks in the measured order spectrum and the theoretically calculated order, denoted as... .

[0118] The characteristic matching degree between the order spectrum and the theoretical fault characteristic order is calculated using a Gaussian kernel window function. First, the measured order spectrum... Peak retrieval is performed, and the set of significant characteristic peaks in the measured order spectrum is denoted as... Extracting the first The set of salient characteristic peaks with the largest amplitude; targeting the specific fault characteristic order to be diagnosed. (like Before construction Second harmonic set ,in, Typically, 3 to 5 are chosen. For each characteristic order in the theoretical set... In the measured set Find the measured peak value that is closest to it, as shown in formula (17): (17) Calculate the first using the Gaussian kernel window function Matching score of the first harmonic As shown in formula (18): (18) In the formula, This is the tolerance bandwidth, and its value range is... , These are the frequency doubling weighting coefficients, such as Introducing the energy significance coefficient The test is performed as shown in formula (19): Right now: (19) In the formula, The amplitude of the measured characteristic peak. This represents the average noise amplitude of the order spectrum within the adjacent non-fault frequency band. When At that time, it is determined that the characteristic peak truly exists. The energy significance coefficient is... If it's too low, it means it's just background noise. Approaching 0. The final weighted average yields the comprehensive feature matching degree. As shown in formula (20): (20) Step S4052: Based on the speed signal and system health in the monitoring data, the mechanism weight coefficient is determined using an adaptive weighting algorithm. Based on the mechanism weight coefficient, the anomaly score and feature matching degree are weighted and summed to obtain the diagnostic confidence.

[0119] Specifically, the adaptive weighting algorithm refers to an algorithm that dynamically allocates the weight of deep learning scores and mechanism scores based on the real-time spindle speed and system health. The mechanism weight coefficient is a weighting factor adaptively assigned to the feature matching degree, denoted as... The deep learning weight coefficients are denoted as... Diagnostic confidence score refers to the final judgment score after combining the data-driven anomaly score and the mechanism matching score, denoted as... .

[0120] Based on the speed signal and system health status in the monitoring data, the mechanism weight coefficient is determined using an adaptive weighting algorithm. Based on the mechanism weight coefficient, the anomaly score and feature matching degree are weighted and summed to obtain the diagnostic confidence level. The weight coefficients must satisfy the normalization constraint condition, as shown in formula (21): (twenty one) Utilizing rotational speed With system health The adaptive weight calculation formula is constructed using the following formulas, as shown in formulas (22) and (23): (twenty two) (twenty three) In the formula, The weighting coefficient for the basic mechanism is typically set to a value of... If take , The recommended threshold for spindle speed division is set to [value missing]. , The speed sensitivity adjustment coefficient can be set to... Based on the calculated mechanism weighting coefficients With deep learning weight coefficients For abnormal scores and feature matching degree The diagnostic confidence level is obtained by performing a weighted summation. As shown in formula (24): (twenty four) Step S4053: Generate spindle collaborative diagnostic results based on diagnostic confidence, fault evolution stage, and fault type.

[0121] Specifically, based on diagnostic confidence level, fault evolution stage, and fault type, a spindle collaborative diagnostic result is generated. The system will then calculate the diagnostic confidence level. The system performs weighted fusion and comprehensive output based on the fault evolution stage and fault type, and automatically retrieves the corresponding handling plan from the expert database to generate a spindle collaborative diagnosis result that includes fault location components, severity, and handling suggestions.

[0122] This embodiment provides a collaborative diagnostic method for wind turbine main shafts, constructing a theoretical fault feature order. This provides a rigorous theoretical frequency / order benchmark that conforms to the physical structure for subsequent feature matching analysis using expert knowledge graphs, ensuring that diagnostic results can accurately pinpoint specific physical components. A dynamic Bayesian network is constructed based on historical failure data and prior knowledge constraints. The irreversible evolution process of main bearing damage is expressed using a state transition probability matrix, overcoming the shortcomings of traditional static diagnostics that rely on isolated single indicators. This achieves dynamic temporal probabilistic reasoning regarding fault evolution trends and current physical stages. Furthermore, three complex fault types—simple structural damage, pure lubrication failure, and severe wear accompanied by thermal damage—are logically divided into completely mutually exclusive categories, eliminating cross-judgment and giving the diagnostic results clear physical diagnostic significance. By using the Gaussian kernel window function to calculate the feature matching degree between the measured order spectrum and the theoretical fault feature order, it can tolerate feature spectral line drift caused by sensor installation deviation, slight speed slippage, or slight changes in contact angle under heavy load. At the same time, by using the speed signal and system health status to determine the mechanism weight coefficient through an adaptive weighting algorithm, it realizes the dynamic allocation of anomaly score and feature matching degree under different speeds and different degrees of degradation, ensuring that the generated diagnostic confidence is objective and reliable, and can directly guide automated operation and maintenance decisions.

[0123] Figure 5 This is a schematic diagram of a wind turbine main shaft collaborative diagnosis method according to an embodiment of the present invention, as shown below. Figure 5 As shown, Figure 5It includes a multi-source heterogeneous data acquisition layer, a physical space mapping and feature construction layer, a deep perception layer, a cognitive fusion layer, and a decision output layer.

[0124] Multi-source heterogeneous data acquisition layer: The monitoring data of the wind turbine main shaft is acquired through the data acquisition front end and monitoring interface. The monitoring data includes at least bearing geometric parameters, temperature signals and power signals. When sensor support is available, vibration signals and corresponding instantaneous phase angle, speed signals and wind speed signals are further acquired.

[0125] Physical space mapping and feature construction layer: This layer decouples the monitoring data from operating conditions. On one hand, it uses instantaneous phase angles to perform order resampling of the vibration signal, mapping the non-stationary time-domain vibration signal to an angularly spaced domain to obtain an angular-domain vibration sequence, thus eliminating spectral ambiguity caused by low-speed fluctuations. On the other hand, based on power, speed, wind speed, and temperature signals, it calculates the theoretical heat generation power based on the total friction torque, calculates the dynamic thermal resistance using wind speed, iterates the theoretical temperature rise at the current moment using the backward differential method, and obtains the dynamic thermal balance residual by subtracting the measured temperature of the main bearing from the theoretical temperature rise, thus eliminating interference from normal operating conditions and ambient temperature changes on the temperature rise.

[0126] Deep perception layer: The wavelet packet decomposition algorithm is used to extract the feature signal corresponding to the pitting of the main bearing from the vibration sequence in the angular domain. The extracted feature signal and the dynamic thermal balance residual are input into the pre-trained neural network model. During the training process, the neural network model adopts a physical consistency penalty term that embeds the main shaft force balance equation or the minimum limit criterion of lubricating film thickness to reconstruct and verify the input features, and outputs an anomaly score to quantify the current degree of health deviation based on the reconstruction error.

[0127] Cognitive Fusion Layer: Based on the bearing's geometric parameters, including roller diameter, pitch circle diameter, contact angle, and number of rollers, the layer calculates the damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller rotation fault characteristic order, and the cage fault characteristic order, combining them to obtain the theoretical fault characteristic order. A pre-constructed expert knowledge graph is used to perform temporal logical reasoning on the theoretical fault characteristic order. This expert knowledge graph, built on a dynamic Bayesian network, describes the irreversible transition process between each stage through a state transition probability matrix, determining the fault evolution stage and its corresponding health level. Simultaneously, based on the changing trend of the dynamic thermal balance residual and whether the amplitude of the order spectrum at each harmonic frequency of the theoretical fault characteristic order exceeds a preset threshold, a joint threshold judgment is performed to determine the fault type.

[0128] Decision output layer: The Gaussian kernel window function is used to calculate the feature matching degree between the order spectrum and the theoretical fault feature order. Based on the speed signal and health status, the mechanism weight coefficient is dynamically determined through an adaptive weighting algorithm. The anomaly score and feature matching degree are weighted and summed to obtain the diagnostic confidence. The diagnostic confidence, fault evolution stage and fault type are combined to automatically match the corresponding treatment plan in the expert database and generate a spindle collaborative diagnosis result that includes fault location components, severity and maintenance suggestions. In addition, when the monitoring data lacks vibration signals, this solution can automatically degenerate into a mechanism residual diagnosis method based on SCADA monitoring data, which is identified by the nonlinear offset of the power-temperature rise curve to ensure engineering applicability and robustness.

[0129] This embodiment also provides a wind turbine main shaft collaborative diagnostic device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0130] This embodiment provides a wind turbine main shaft collaborative diagnostic device, such as... Figure 6 As shown, it includes: The acquisition module 601 is used to acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals and power signals; The decoupling module 602 is used to perform operating condition decoupling processing on the monitoring data to obtain the angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and the temperature signal; Reconstruction module 603 is used to reconstruct the angular domain vibration sequence and / or dynamic thermal equilibrium residual using a pre-trained neural network model to obtain anomaly scores; The determination module 604 is used to calculate the theoretical fault characteristic order based on the bearing geometric parameters, and to use a pre-built expert knowledge graph to perform temporal logical reasoning on the theoretical fault characteristic order to determine the fault evolution stage and fault type. The generation module 605 is used to generate spindle collaborative diagnostic results based on the anomaly score, fault evolution stage, and fault type.

[0131] In some alternative implementations, the decoupling module 602 includes: The vibration signal resampling unit is used to perform order resampling of the vibration signal using the instantaneous phase angle to obtain the angular domain vibration sequence.

[0132] The dynamic thermal resistance calculation unit is used to calculate the total friction torque based on the power signal, speed signal, and temperature signal, and to calculate the theoretical heat generation power based on the total friction torque and speed signal, and to calculate the dynamic thermal resistance based on the wind speed signal.

[0133] The theoretical temperature rise calculation unit is used to calculate the theoretical temperature rise at the current moment based on the theoretical heat generation power, ambient temperature, dynamic thermal resistance, and the theoretical temperature rise at the previous moment.

[0134] The dynamic thermal balance residual calculation unit is used to take the difference between the measured temperature of the main bearing and the theoretical temperature rise at the current moment as the dynamic thermal balance residual.

[0135] In some alternative implementations, the vibration signal resampling unit includes: The instantaneous angular velocity and instantaneous angle calculation subunit is used to calculate the instantaneous angular velocity and instantaneous angle of the wind turbine main shaft based on the instantaneous phase angle.

[0136] The sub-units obtained from the angular domain vibration sequence are used to extract the vibration amplitude at preset intervals from the vibration signal using the Shannon-Nyquist sampling theorem or a linear interpolation algorithm, thus obtaining the angular domain vibration sequence.

[0137] In some optional embodiments, a wind turbine main shaft collaborative diagnostic device further includes: The feature signal extraction unit is used to extract the feature signal corresponding to the pitting of the main bearing from the vibration sequence in the angular domain using the wavelet packet decomposition algorithm. The feature signal reconstruction unit is used to input the feature signal into the neural network model to be trained for feature extraction and reconstruction to obtain the reconstructed feature signal; The physical consistency penalty term construction unit is used to calculate the data reconstruction error term based on the characteristic signal and the reconstructed characteristic signal, and to construct the physical consistency penalty term according to the preset spindle force balance equation or the preset minimum lubrication film thickness limit criterion. The neural network model obtains units, which are used to construct a loss function based on the data reconstruction error term and physical consistency penalty term. The loss function is then used to train and verify the autoencoder in the neural network model, resulting in a neural network model containing an autoencoder with embedded physical constraints.

[0138] In some alternative implementations, the determining module 604 includes: The order calculation unit is used to calculate the damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller self-rotation fault characteristic order, and the cage fault characteristic order based on the roller diameter, pitch circle diameter, contact angle, and number of rollers.

[0139] The theoretical fault characteristic order combination unit is used to combine the outer ring damage characteristic order, inner ring damage characteristic order, roller self-rotation fault characteristic order, and cage fault characteristic order into a theoretical fault characteristic order.

[0140] The fault evolution stage determination unit is used to perform temporal logic reasoning on the angular domain vibration sequence and dynamic thermal balance residual using an expert knowledge graph to determine the fault evolution stage.

[0141] The order spectrum is obtained from the unit, which is used to determine the system health based on the fault evolution stage, and to perform order spectrum analysis on the angular domain vibration sequence to obtain the order spectrum.

[0142] The outer ring damage determination unit is used to determine the fault type as outer ring damage if the dynamic thermal balance residual does not continue to increase and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold.

[0143] The lubrication failure confirmation unit is used to determine the fault type as lubrication failure if the dynamic thermal balance residual continues to increase and the amplitude of the order spectrum at any octave of the theoretical fault characteristic order is less than or equal to a preset amplitude threshold.

[0144] The severe wear accompanied by thermal damage determination unit is used to determine the fault type as severe wear accompanied by thermal damage if the dynamic thermal balance residual continues to increase and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than a preset amplitude threshold.

[0145] In some optional embodiments, a wind turbine main shaft collaborative diagnostic device further includes: The phase division unit is used to divide the life cycle of the wind turbine main shaft into multiple phases, including the normal operation phase, the initial lubrication degradation phase, the intermediate friction and vibration phase, the later geometric damage phase, and the final temperature rise runaway phase. The state transition probability matrix confirmation unit is used to determine the state transition probability matrix between different stages based on historical failure data and prior knowledge. The expert knowledge graph construction unit is used to construct a dynamic Bayesian network based on the stage and state transition probability matrix, and then use the dynamic Bayesian network to construct an expert knowledge graph.

[0146] In some alternative implementations, the generation module 605 includes: The feature matching degree calculation unit is used to calculate the feature matching degree of the order spectrum and the theoretical fault feature order using the Gaussian kernel window function.

[0147] The diagnostic confidence calculation unit is used to determine the mechanism weight coefficient based on the speed signal and system health in the monitoring data using an adaptive weighting algorithm. Based on the mechanism weight coefficient, the anomaly score and feature matching degree are weighted and summed to obtain the diagnostic confidence.

[0148] The spindle collaborative diagnostic result generation unit is used to generate spindle collaborative diagnostic results based on diagnostic confidence, fault evolution stage, and fault type.

[0149] The wind turbine main shaft collaborative diagnostic device provided in this embodiment of the invention can execute the wind turbine main shaft collaborative diagnostic method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments, and will not be repeated here.

[0150] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0151] The following is a detailed reference. Figure 7 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0152] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0153] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the wind turbine main shaft collaborative diagnostic method of an embodiment of the present invention.

[0154] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0155] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, it implements the wind turbine main shaft collaborative diagnostic method shown in the above embodiments.

[0156] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0157] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for collaborative diagnosis of the main shaft of a wind turbine generator, characterized in that, The method includes: Acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals, and power signals; The monitoring data is subjected to operating condition decoupling processing to obtain an angular domain vibration sequence and / or dynamic thermal balance residual; wherein, the dynamic thermal balance residual is calculated based on the power signal and temperature signal; Anomaly scores are obtained by reconstructing the angular domain vibration sequence and / or the dynamic thermal equilibrium residual using a pre-trained neural network model. The theoretical fault characteristic order is calculated based on the bearing geometric parameters, and the theoretical fault characteristic order is used to perform temporal logical reasoning on the pre-constructed expert knowledge graph to determine the fault evolution stage and fault type. Based on the anomaly score, the fault evolution stage, and the fault type, a spindle collaborative diagnostic result is generated.

2. The method according to claim 1, characterized in that, The monitoring data also includes vibration signals and corresponding instantaneous phase angles; The obtained angular domain vibration sequence includes: The vibration signal is resampled in order using the instantaneous phase angle to obtain an angular domain vibration sequence.

3. The method according to claim 2, characterized in that, The step of resampling the vibration signal by order using the instantaneous phase angle to obtain an angular domain vibration sequence includes: The instantaneous angular velocity and instantaneous angle of the wind turbine main shaft are calculated based on the instantaneous phase angle. The vibration amplitude at preset intervals is extracted from the vibration signal using the Shannon-Nyquist sampling theorem or a linear interpolation algorithm to obtain an angular domain vibration sequence.

4. The method according to claim 1, characterized in that, The monitoring data also includes rotational speed signals and wind speed signals, and the temperature signals include ambient temperature and measured temperature of the main bearing. The dynamic thermal equilibrium residuals are obtained, including: The total frictional torque is calculated based on the power signal, the rotational speed signal, and the temperature signal; the theoretical heat generation power is calculated based on the total frictional torque and the rotational speed signal; and the dynamic thermal resistance is calculated based on the wind speed signal. The theoretical temperature rise at the current moment is calculated based on the theoretical heat generation power, the ambient temperature, the dynamic thermal resistance, and the theoretical temperature rise at the previous moment. The difference between the measured temperature of the main bearing and the theoretical temperature rise at the current moment is taken as the dynamic thermal balance residual.

5. The method according to claim 1, characterized in that, The neural network model is trained in the following manner: The wavelet packet decomposition algorithm was used to extract the feature signal corresponding to the pitting of the main bearing from the angular domain vibration sequence. The feature signal is input into the neural network model to be trained for feature extraction and reconstruction to obtain the reconstructed feature signal; Based on the characteristic signal and the reconstructed characteristic signal, calculate the data reconstruction error term, and construct a physical consistency penalty term based on the preset spindle force balance equation or the preset minimum lubrication film thickness limit criterion. A loss function is constructed based on the data reconstruction error term and the physical consistency penalty term. The loss function is then used to train and verify the autoencoder in the neural network model, resulting in a neural network model containing an autoencoder with embedded physical constraints.

6. The method according to claim 1, characterized in that, The bearing geometry parameters include roller diameter, pitch circle diameter, contact angle, and number of rollers; The calculation of the theoretical fault characteristic order based on the bearing geometric parameters includes: Based on the roller diameter, the pitch circle diameter, the contact angle, and the number of rollers, the damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller rotation failure characteristic order, and the cage failure characteristic order are calculated respectively. The damage characteristic order of the outer ring, the damage characteristic order of the inner ring, the roller self-rotation fault characteristic order, and the cage fault characteristic order are combined into a theoretical fault characteristic order.

7. The method according to claim 1, characterized in that, The expert knowledge graph is constructed in the following manner: The life cycle of the wind turbine main shaft is divided into multiple stages, including the normal operation stage, the initial lubrication degradation stage, the intermediate friction and vibration stage, the later geometric damage stage, and the final temperature rise runaway stage. Determine the state transition probability matrix between different stages based on historical failure data and prior knowledge; A dynamic Bayesian network is constructed based on the stage and the state transition probability matrix, and an expert knowledge graph is constructed using the dynamic Bayesian network.

8. The method according to claim 6, characterized in that, The step of using a pre-constructed expert knowledge graph to perform temporal logical reasoning on the theoretical fault feature order to determine the fault evolution stage and fault type includes: The fault evolution stage is determined by using the expert knowledge graph to perform temporal logic reasoning on the angular domain vibration sequence and the dynamic thermal balance residual. The system health is determined based on the fault evolution stage, and the order spectrum is obtained by performing order spectrum analysis on the angular domain vibration sequence. If the dynamic thermal balance residual does not continue to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be outer ring damage. If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is less than or equal to a preset amplitude threshold, then the fault type is determined to be lubrication failure. If the dynamic thermal balance residual continues to increase, and the amplitude of the order spectrum at any harmonic of the theoretical fault characteristic order is greater than the preset amplitude threshold, then the fault type is determined to be severe wear accompanied by thermal damage.

9. The method according to claim 8, characterized in that, The step of generating a spindle collaborative diagnostic result based on the anomaly score, the fault evolution stage, and the fault type includes: The feature matching degree between the order spectrum and the theoretical fault feature order is calculated using a Gaussian kernel window function. Based on the rotational speed signal and system health in the monitoring data, an adaptive weighting algorithm is used to determine the mechanism weighting coefficient. Based on the mechanism weighting coefficient, the anomaly score and the feature matching degree are weighted and summed to obtain the diagnostic confidence. Based on the diagnostic confidence level, the fault evolution stage, and the fault type, a spindle collaborative diagnostic result is generated.

10. A wind turbine main shaft collaborative diagnostic device, characterized in that, The device includes: The acquisition module is used to acquire monitoring data of the wind turbine main shaft; the monitoring data includes at least bearing geometric parameters, temperature signals and power signals; A decoupling module is used to perform operating condition decoupling processing on the monitoring data to obtain an angular domain vibration sequence and / or dynamic thermal balance residual; wherein the dynamic thermal balance residual is calculated based on the power signal and temperature signal; The reconstruction module is used to reconstruct the angular domain vibration sequence and / or the dynamic thermal equilibrium residual using a pre-trained neural network model to obtain anomaly scores. The determination module is used to calculate the theoretical fault characteristic order based on the bearing geometric parameters, and to perform temporal logical reasoning on the theoretical fault characteristic order using a pre-built expert knowledge graph to determine the fault evolution stage and fault type. The generation module is used to generate spindle collaborative diagnostic results based on the anomaly score, the fault evolution stage, and the fault type.

Images