A wind turbine variable pitch bearing full life cycle remaining life prediction method and related device

Abstract

The application belongs to the technical field of wind power generation, and discloses a wind turbine variable-pitch bearing full-life-cycle residual life prediction method and related equipment; wherein the wind turbine variable-pitch bearing full-life-cycle residual life prediction method comprises the following steps: inputting multi-source monitoring data into a pre-trained stage classification model, outputting a degradation stage label through the stage classification model, calling a residual life prediction sub-model corresponding to the degradation stage label from a pre-constructed model library according to the degradation stage label, inputting the multi-source monitoring data into the called residual life prediction sub-model, and outputting a residual life prediction result of the wind turbine variable-pitch bearing through the residual life prediction sub-model; and the application can realize accurate and automatic division of a healthy stage, an early degradation stage, a middle degradation stage and an accelerated degradation stage.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a method and related equipment for predicting the remaining lifespan of a wind turbine pitch bearing throughout its entire life cycle. Background Technology

[0002] Pitch bearings are critical core components in wind turbine generators, connecting the blades and the hub. Operating under complex alternating loads and harsh environmental conditions for extended periods, they are highly susceptible to progressive degradation failures such as fatigue wear, pitting, and cracking, directly impacting the safety and power generation efficiency of the wind turbine. Therefore, accurately predicting the remaining lifespan of pitch bearings is of significant engineering importance for enabling condition-based maintenance in wind farms and reducing operation and maintenance costs.

[0003] In existing technologies, methods for predicting the remaining life of wind turbine pitch bearings mainly fall into two categories. One category is prediction methods based on physical models, which estimate the remaining life by establishing physical equations for the propagation of fatigue cracks or the evolution of wear in the bearing. However, the actual operating conditions of pitch bearings are complex and varied, making it difficult for physical models to accurately describe the differences in degradation mechanisms at different stages throughout their entire life cycle. The other category is prediction methods based on data-driven approaches, which extract degradation features using multi-source monitoring signals such as vibration and temperature, and combine them with models such as neural networks and support vector regression for life prediction. However, existing data-driven methods often use a single prediction model to uniformly model the entire life cycle, failing to fully consider the differences in degradation rates and degradation patterns exhibited by pitch bearings in the healthy, early, intermediate, and accelerated degradation stages. This results in a single model struggling to maintain high prediction accuracy across all stages of the entire life cycle.

[0004] Therefore, the existing technology has the problem of not fully considering the differences in degradation rates and degradation patterns of pitch bearings in the healthy stage, early degradation stage, intermediate degradation stage and accelerated degradation stage, which urgently needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method and related equipment for predicting the remaining life of wind turbine pitch bearings throughout their entire life cycle, in order to overcome the problems existing in the prior art. This invention can achieve accurate and automatic division of the healthy stage, early degradation stage, intermediate degradation stage and accelerated degradation stage.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for predicting the remaining life of a wind turbine pitch bearing throughout its entire life cycle, comprising the following steps: Acquire multi-source monitoring data of the wind turbine pitch bearing throughout its entire life cycle. The multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, speed signals, and SCADA system operating parameter data. Multi-source monitoring data is input into a pre-trained stage classification model, and the degradation stage label is output through the stage classification model. Based on the degradation stage label, the remaining lifetime prediction sub-model corresponding to the degradation stage label is called from a pre-built model library; wherein, the model library is built based on several independent training channels; Multi-source monitoring data is input into the called remaining life prediction sub-model, and the remaining life prediction result of the wind turbine pitch bearing is output through the remaining life prediction sub-model.

[0007] In some embodiments, inputting multi-source monitoring data into a pre-trained stage classification model and outputting degradation stage labels through the stage classification model specifically includes: After aligning and normalizing the multi-source monitoring data according to the timestamps, the data is input into the pre-trained stage classification model. The input layer of the stage classification model receives multi-source monitoring data. After several layers of feature extraction, the output layer of the stage classification model calculates the probability vectors of the current monitoring data belonging to the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage respectively through the Softmax activation function. The stage label corresponding to the maximum probability vector is taken as the output degradation stage label.

[0008] In some embodiments, the step of calling the remaining lifetime prediction sub-model corresponding to the degradation stage label from a pre-built model library specifically includes: The degradation stage label is used as an index identifier and matched with the stage label mapping relationship of each remaining life prediction sub-model in the pre-built model library. The model library stores four remaining life prediction sub-models corresponding to the healthy stage, early degradation stage, intermediate degradation stage and accelerated degradation stage, respectively. Each remaining life prediction sub-model forms a one-to-one mapping relationship with the corresponding degradation stage label. Based on the matching results, the remaining lifetime prediction sub-model corresponding to the degradation stage label is called from the model library.

[0009] In some embodiments, the step of outputting the remaining life prediction result of the wind turbine pitch bearing through the remaining life prediction sub-model specifically includes: The input layer of the remaining lifetime prediction sub-model receives the current multi-source monitoring data. After forward propagation calculation, the output layer of the remaining lifetime prediction sub-model outputs the predicted value of the remaining lifetime of the wind turbine pitch bearing from the current moment to the failure threshold.

[0010] In some embodiments, after the predicted remaining life of the output wind turbine pitch bearing, the method further includes: The remaining life prediction results are compared with preset multi-level early warning thresholds, and different levels of early warning signals are output based on the comparison results. The multi-level early warning thresholds include a first-level early warning threshold, a second-level early warning threshold, and a third-level early warning threshold. The first-level early warning threshold corresponds to a remaining life prediction value of more than 50%, the second-level early warning threshold corresponds to a remaining life prediction value of 20% to 50%, and the third-level early warning threshold corresponds to a remaining life prediction value of less than 20%.

[0011] In some embodiments, the construction steps of the model library specifically include: Obtain the historical monitoring data sample set corresponding to each degradation stage label, and allocate the historical monitoring data sample set of each stage to four independent training channels according to the degradation stage label. In each training channel, the historical monitoring data sample set of the current stage is used as the input feature, and the real degradation trajectory of the subsequent stage is used as the supervision label to train the remaining life prediction sub-model that matches the degradation pattern of the current stage. The four remaining lifetime prediction sub-models that have been trained are packaged according to the degradation stage labels to obtain the constructed model library. Each remaining lifetime prediction sub-model in the model library corresponds one-to-one with each degradation stage label.

[0012] In some embodiments, the training channels include a healthy phase training channel, an early degradation phase training channel, a mid-stage degradation phase training channel, and an accelerated degradation phase training channel. The training channel for the healthy stage is based on a statistical model of the Weibull distribution, the training channels for the early and mid-stage degradation stages are based on deep learning models of long short-term memory networks, and the training channel for the accelerated degradation stage is based on a degradation tracking model of particle filtering, resulting in four sub-models for predicting remaining lifetime after training.

[0013] Secondly, the present invention provides a system for predicting the remaining lifespan of a wind turbine pitch bearing throughout its entire life cycle, comprising: The multi-source monitoring data acquisition module is used to acquire multi-source monitoring data of the wind turbine pitch bearing throughout its entire life cycle. The multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, speed signals, and SCADA system operating parameter data. The stage classification module is used to input multi-source monitoring data into a pre-trained stage classification model and output degradation stage labels through the stage classification model; The remaining lifetime prediction sub-model calling module is used to call the remaining lifetime prediction sub-model corresponding to the degradation stage label from a pre-built model library; wherein, the model library is built based on several independent training channels; The remaining life prediction result output module is used to input multi-source monitoring data into the called remaining life prediction sub-model, and output the remaining life prediction result of the wind turbine pitch bearing through the remaining life prediction sub-model.

[0014] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting the remaining life of the wind turbine pitch bearing throughout its entire life cycle.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the remaining lifespan of a wind turbine pitch bearing throughout its entire life cycle.

[0016] The above technical solution has the following advantages or beneficial effects: Firstly, this invention provides a method for predicting the remaining life of wind turbine pitch bearings throughout their entire lifecycle. By constructing a stage classification model to identify the current monitoring data in real time, it effectively solves the problem of reliance on manual experience and strong subjectivity in the degradation stage division of existing technologies, achieving accurate and automatic division of the healthy stage, early degradation stage, mid-term degradation stage, and accelerated degradation stage. Secondly, for different degradation stages, dedicated remaining life prediction sub-models matching their degradation patterns are constructed, overcoming the shortcomings of existing single models that cannot take into account the different degradation rates of each stage of the entire lifecycle. The healthy stage uses a Weibull statistical model to describe the initial degradation trend, the early and mid-term stages use a long short-term memory network to capture nonlinear degradation evolution, and the accelerated stage uses a particle filter to track the rapid deterioration process. Each model performs its specific function, significantly improving the prediction accuracy of each stage of the entire lifecycle. In addition, through the pre-construction of the model library and the stage label indexing mechanism, in practical applications, only forward calculation of the model is required to obtain the prediction results, without the need for online iterative training, which greatly reduces computational overhead and response latency, meeting the engineering needs of real-time prediction and rapid decision-making in wind farms. In summary, this invention effectively solves the technical problems of strong subjectivity in stage division, insufficient prediction accuracy of single models, and high online calculation costs in the prior art, and provides a reliable and efficient intelligent solution for condition-based maintenance of wind turbine pitch bearings.

[0017] Secondly, this invention provides a system for predicting the remaining lifespan of wind turbine pitch bearings throughout their entire lifecycle. Through the coordinated operation of a multi-source monitoring data acquisition module, a stage classification module, a remaining lifespan prediction sub-model calling module, and a remaining lifespan prediction result output module, it achieves fully automated execution of the entire process from data acquisition, stage identification, model calling, to lifespan prediction. The stage classification module automatically and accurately divides the degradation stages, overcoming the subjectivity inherent in manual experience-based classification. The remaining lifespan prediction sub-model calling module calls the corresponding dedicated sub-model from the model library based on the stage label index, ensuring that each stage uses the optimal prediction model that matches its degradation pattern, significantly improving the prediction accuracy and robustness of each stage throughout the entire lifecycle. The data flow between modules is clear, and the logical connections are tight, collectively forming a complete, real-time deployable intelligent remaining lifespan prediction system, providing an efficient and reliable engineering solution for condition-based maintenance of wind turbine pitch bearings.

[0018] Thirdly, the present invention provides a computer device that, through a processor executing a specific computer program, can efficiently implement the steps of the method of the present invention. When performing data processing tasks, the computer device can accurately perform numerical calculations and logical judgments, avoiding errors caused by human factors. At the same time, since the computer program has high stability and reliability, it can ensure the accuracy and consistency of the data processing results.

[0019] Fourthly, the present invention provides a computer-readable storage medium. By programming the steps of the method of the present invention into a computer program and storing it on the computer-readable storage medium, users can easily load these programs onto any compatible computer device and execute them without rewriting or converting the code, which greatly improves the convenience and flexibility of program execution. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the method for predicting the remaining life of a wind turbine pitch bearing throughout its entire life cycle, as shown in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a computer device as shown in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention.

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] Pitch bearings are subjected to complex alternating loads during wind turbine operation, and their degradation process exhibits distinct stage characteristics, including a healthy stage, an early degradation stage, a mid-term degradation stage, and an accelerated degradation stage. Existing remaining life prediction methods mostly use a single prediction model to uniformly model the entire life cycle, without fully considering the differences in degradation rates and degradation patterns at different stages. This makes it difficult for a single model to maintain high prediction accuracy at all stages of the entire life cycle.

[0025] Based on the above background, this application provides a method for predicting the remaining life of wind turbine pitch bearings throughout their entire life cycle. It uses a stage classification model to identify the current monitoring data in real time, and calls the corresponding remaining life prediction sub-model from a pre-built model library based on the identified degradation stage label. This model library is built based on several independent training channels, and each sub-model matches the degradation patterns of different degradation stages. This enables high-precision remaining life prediction at all stages of the entire life cycle, effectively solving the problem of insufficient prediction accuracy of a single model and meeting the engineering needs of real-time prediction and rapid decision-making at wind farm sites.

[0026] Example: This embodiment provides a method for predicting the remaining lifespan of a wind turbine pitch bearing throughout its entire life cycle. (See also...) Figure 1 This includes the following steps: Step 1: Obtain multi-source monitoring data of the wind turbine pitch bearing throughout its entire life cycle.

[0027] In some embodiments, the multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, rotational speed signals, and SCADA system operating parameter data. The vibration signals reflect the amplitude and frequency characteristics of vibration generated during bearing operation; the temperature signals reflect changes in bearing operating temperature; the pitch torque signals reflect changes in torque transmitted to the bearing during blade pitch control; the rotational speed signals reflect the rotational speed of the wind turbine main shaft and pitch bearing; and the SCADA system operating parameter data includes various operating status parameters collected by the wind turbine control system, such as power, wind speed, and pitch angle. This multi-source monitoring data can comprehensively reflect the actual operating status of the pitch bearing from multiple dimensions, providing a data foundation for subsequent degradation stage identification and remaining life prediction.

[0028] Step 2: Input the multi-source monitoring data into the pre-trained stage classification model, and output the degradation stage label through the stage classification model.

[0029] In some embodiments, step 2 specifically includes: Step 2.1: Align and normalize the multi-source monitoring data according to timestamps, and then input them into the pre-trained stage classification model.

[0030] Step 2.2: The input layer of the stage classification model receives multi-source monitoring data. After several layers of feature extraction, the output layer of the stage classification model calculates the probability vectors of the current monitoring data belonging to the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage respectively through the Softmax activation function. The stage label corresponding to the maximum probability vector is taken as the output degradation stage label.

[0031] Specifically, the input layer of the stage classification model receives the preprocessed multi-source monitoring data as the initial input. Several layers of feature extraction networks within the model perform layer-by-layer feature transformation and abstraction on the input data, extracting deep feature representations related to the bearing degradation stage. After feature extraction, the output layer of the stage classification model uses the Softmax activation function to calculate the extracted features, converting the model output into a probability distribution. This probability distribution is a probability value vector containing four probability values, representing the probability that the current monitoring data belongs to the healthy stage, the early degradation stage, the intermediate degradation stage, and the accelerated degradation stage, respectively. The sum of all components of the probability value vector is 1. After obtaining the four probability values, the stage category corresponding to the highest probability value is selected, and the stage label corresponding to this stage category is used as the degradation stage label output by the stage classification model. For example, if the probability value belonging to the accelerated degradation stage is the highest in the Softmax output probability value vector, then the output degradation stage label is "accelerated degradation stage". The model output is converted into a probability vector by using the Softmax activation function, so that the probability of belonging to each candidate stage has a clear probabilistic meaning and comparability.

[0032] In this embodiment, preprocessing of multi-source monitoring data by aligning and normalizing by timestamps ensures that monitoring data from different sources and with different dimensions can be input into the stage classification model in a unified format, eliminating the impact of data inconsistency on classification accuracy. Simultaneously, by using the Softmax activation function to output a probability vector and taking the stage label corresponding to the highest probability, the current degradation stage of the monitoring data is automatically and quantitatively determined from multiple candidate degradation stages, ensuring the objectivity and accuracy of the degradation stage label output.

[0033] In some embodiments, the determination of the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage in this application is automatically completed by a pre-trained stage classification model, without relying on human experience or fixed thresholds. Specifically, multi-source monitoring data, after timestamp alignment and normalization, is input into the stage classification model. The model internally uses several layers of feature extraction networks to perform deep feature abstraction of the degradation trends in the data. Finally, the output layer uses a Softmax activation function to calculate the probability vectors of the current monitoring data belonging to each of the four stages, and the stage label corresponding to the highest probability value is taken as the output degradation stage label. During the training phase, the stage classification model has learned from a large number of historical monitoring data samples labeled with each degradation stage, mastering the feature patterns corresponding to different stages. Therefore, it can automatically determine the current stage based on the input data during actual prediction, achieving objectivity, quantification, and automation of degradation stage division.

[0034] In some embodiments, the health stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage are four pre-defined degradation state categories based on the objective laws governing the degradation trajectory of the pitch bearing throughout its entire life cycle. Each stage has distinct degradation rates and characteristics: the health stage corresponds to the bearing's normal operating condition, with no obvious degradation trend in the monitoring data; the early degradation stage corresponds to the bearing beginning to show slight signs of degradation, with a low degradation rate; the intermediate degradation stage corresponds to a significant acceleration in bearing degradation, with monitoring indicators showing a trend change; and the accelerated degradation stage corresponds to the bearing approaching its failure threshold, with a sharp increase in the degradation rate. The stage classification model calculates the probability values ​​of the current multi-source monitoring data belonging to the above four stages using the Softmax activation function, and takes the stage category corresponding to the highest probability value as the output degradation stage label. This determination process has a clear mathematical basis and is repeatable.

[0035] In some embodiments, the training steps of the stage classification model specifically include: A historical monitoring data sample set covering the entire lifecycle of the pitch bearing is obtained. Each set of multi-source monitoring data in the historical monitoring data sample set is pre-labeled with a corresponding degradation stage label, including healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage. The sample set is divided into a training set and a validation set according to a preset ratio. The multi-source monitoring data in the training set is used as the input features of the stage classification model, and the corresponding degradation stage label is used as the supervision label. The difference between the model's predicted probability distribution and the true label is calculated using the cross-entropy loss function. The network parameters of the stage classification model are iteratively optimized using the backpropagation algorithm until the model's classification accuracy on the validation set meets the preset requirements or the loss function value converges, resulting in a trained stage classification model. This training process ensures that the stage classification model has the ability to accurately identify the stage and output degradation stage labels for the current monitoring data.

[0036] Step 3: Based on the degradation stage label, call the remaining lifetime prediction sub-model corresponding to the degradation stage label from the pre-built model library; wherein, the model library is built based on several independent training channels.

[0037] In some embodiments, step 3 specifically includes: Step 3.1: Use the degradation stage label as an index identifier and match it with the stage label mapping relationship of each remaining life prediction sub-model in the pre-built model library. The model library stores four remaining life prediction sub-models corresponding to the healthy stage, early degradation stage, intermediate degradation stage and accelerated degradation stage, respectively. Each remaining life prediction sub-model forms a one-to-one mapping relationship with the corresponding degradation stage label.

[0038] Step 3.2: Based on the matching results, call the remaining lifetime prediction sub-model corresponding to the degradation stage label from the model library.

[0039] In some embodiments, the construction steps of the model library specifically include: Obtain the historical monitoring data sample set corresponding to each degradation stage label, and allocate the historical monitoring data sample set of each stage to four independent training channels according to the degradation stage label; in each training channel, use the historical monitoring data sample set of the current stage as the input feature and the subsequent real degradation trajectory of the stage as the supervision label to train the remaining life prediction sub-model that matches the degradation pattern of the current stage; encapsulate the four trained remaining life prediction sub-models according to the degradation stage label to obtain the constructed model library, in which each remaining life prediction sub-model corresponds one-to-one with each degradation stage label.

[0040] Specifically, the training channels include a healthy phase training channel, an early degradation phase training channel, a mid-degradation phase training channel, and an accelerated degradation phase training channel. The healthy phase training channel is based on a statistical model of the Weibull distribution, the early degradation phase training channel and the mid-degradation phase training channel are based on deep learning models of long short-term memory networks, and the accelerated degradation phase training channel is based on a degradation tracking model of particle filtering, resulting in four sub-models for predicting remaining lifetime that have been trained.

[0041] This invention achieves automatic connection from degradation stage identification to dedicated prediction model invocation by matching degradation stage labels as index identifiers with the stage label mapping relationship of each sub-model in the model library, and calling the corresponding remaining lifetime prediction sub-model based on the matching results. This ensures that the prediction sub-model matching the degradation law of each stage can be called in different degradation stages, thereby providing model-level guarantee for the accuracy of subsequent remaining lifetime prediction.

[0042] Step 4: Input the multi-source monitoring data into the called remaining life prediction sub-model, and output the remaining life prediction result of the wind turbine pitch bearing through the remaining life prediction sub-model.

[0043] In some embodiments, step 4 specifically includes: The input layer of the remaining lifetime prediction sub-model receives the current multi-source monitoring data. After forward propagation calculation, the output layer of the remaining lifetime prediction sub-model outputs the predicted value of the remaining lifetime of the wind turbine pitch bearing from the current moment to the failure threshold.

[0044] Specifically, for example, in a sub-model based on a Long Short-Term Memory (LSTM) network, the input data is sequentially passed through the gating units and hidden layers of the LSM network for temporal feature transfer and transformation, and finally mapped to the predicted value by the fully connected layer. For a statistical model based on a Weibull distribution, the input data is used as the basis for estimating the parameters of the statistical distribution, calculating the distribution parameters, and extrapolating the expected lifetime. For a degradation tracking model based on particle filtering, the input data is used as the observed value, and the remaining lifetime prediction value is obtained recursively through particle weight updates and state estimation. After forward propagation calculation, the output layer of the remaining lifetime prediction sub-model outputs a specific numerical value, which represents the predicted remaining lifetime of the wind turbine pitch bearing from the current moment to the failure threshold. The unit can be hours, days, or other time measurement units. The failure threshold refers to a pre-set standard limit value used to determine if the bearing has reached a failure state. For example, when the bearing vibration amplitude exceeds a set limit or the degradation characteristic index reaches a preset threshold, the bearing is determined to have reached a failure state.

[0045] In some embodiments, the training steps of the remaining life expectancy prediction sub-model specifically include: Historical monitoring data sample sets corresponding to each degradation stage label are obtained. These sample sets are then distributed to four independent training channels based on the degradation stage label. These four channels correspond to the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage, respectively. In each training channel, the historical monitoring data sample set for the current stage is used as input features, and the actual degradation trajectory after the stage is used as supervision labels. Sub-models for predicting remaining lifespan that match the degradation pattern of the current stage are trained, enabling each sub-model to learn the degradation evolution pattern of its corresponding stage. The four trained sub-models for predicting remaining lifespan are then packaged according to the degradation stage label to obtain a constructed model library. Each sub-model in the model library corresponds one-to-one with each degradation stage label. This independent training channel construction method ensures that each sub-model is trained only on data from a specific degradation stage, thereby fully fitting the degradation pattern of that stage.

[0046] Step 5: Compare the remaining life prediction results with the preset multi-level early warning thresholds, and output early warning signals of different levels based on the comparison results.

[0047] In some embodiments, the multi-level early warning thresholds include a first-level early warning threshold, a second-level early warning threshold, and a third-level early warning threshold. The first-level early warning threshold corresponds to a remaining lifespan prediction value greater than 50%. That is, when the remaining lifespan prediction value accounts for more than 50% of the total lifespan, it is determined that the bearing is currently in a first-level early warning state, and a first-level early warning signal is output. The first-level early warning signal is a low-risk indication, meaning that the bearing is still in the early stage of its lifespan, with ample remaining lifespan, and only routine monitoring is required.

[0048] In some embodiments, the secondary warning threshold corresponds to a remaining life prediction value that is 20% to 50% of the total lifespan. That is, when the remaining lifespan prediction value accounts for 20% to 50% of the total lifespan, the system is determined to be in a secondary warning state, and a secondary warning signal is output. The secondary warning signal indicates a moderate risk, signifying that the bearing has entered a degradation phase, its remaining lifespan is significantly shortened, and that increased monitoring frequency and attention to degradation trends are necessary in subsequent operation.

[0049] In some embodiments, the Level 3 warning threshold corresponds to a remaining lifespan prediction value that is less than 20%. That is, when the remaining lifespan prediction value accounts for less than 20% of the total lifespan, the system is determined to be in a Level 3 warning state, and a Level 3 warning signal is output. The Level 3 warning signal is a high-risk indication, suggesting that the bearing's remaining lifespan is severely insufficient, requiring immediate shutdown for maintenance or replacement to prevent sudden failure.

[0050] In one optional embodiment of this example, the total lifespan is a reference total lifespan value determined in advance based on historical statistical data or bearing design lifespan. The percentage of remaining lifespan prediction is calculated by dividing the remaining lifespan prediction by the total lifespan. In another optional embodiment, the output form of the warning signal may include one or more combinations of audible and visual alarm signals, pop-up notifications from a remote monitoring platform, or SMS notifications sent to maintenance personnel.

[0051] The remaining life prediction method for wind turbine pitch bearings provided in this embodiment uses a stage classification model to identify the current monitoring data in real time. This effectively solves the problem of existing technologies relying on manual experience and being highly subjective in classifying degradation stages, achieving accurate and automatic classification of healthy, early degradation, mid-term degradation, and accelerated degradation stages. Simultaneously, it calls dedicated remaining life prediction sub-models that match the degradation patterns of different degradation stages, overcoming the shortcomings of existing single models that cannot adequately consider the different degradation rates at each stage of the entire life cycle. This significantly improves the prediction accuracy and robustness of each stage of the entire life cycle, providing a reliable intelligent solution for condition-based maintenance of wind turbine pitch bearings.

[0052] In one embodiment of the present invention, a system for predicting the remaining life of a wind turbine pitch bearing throughout its entire life cycle is provided, including a multi-source monitoring data acquisition module, a stage classification module, a remaining life prediction sub-model calling module, and a remaining life prediction result output module.

[0053] The multi-source monitoring data acquisition module is used to acquire multi-source monitoring data throughout the entire lifecycle of the wind turbine pitch bearing. This multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, speed signals, and SCADA system operating parameters. The module communicates with the vibration sensors, temperature sensors, torque sensors, and speed sensors located on the pitch bearing, as well as the data interface of the wind turbine's SCADA system. It collects these various monitoring data in real time, aggregates and organizes the data according to timestamps, and forms a unified format of multi-source monitoring data records for use by subsequent modules.

[0054] The stage classification module is used to input multi-source monitoring data into a pre-trained stage classification model, and output a degradation stage label through the stage classification model. The stage classification module is data-connected to the multi-source monitoring data acquisition module, receives the multi-source monitoring data output by the multi-source monitoring data acquisition module, inputs it into the internally stored pre-trained stage classification model, and outputs the degradation stage label corresponding to the current monitoring data after model calculation. The degradation stage label is one of the following: healthy stage, early degradation stage, intermediate degradation stage, or accelerated degradation stage. The input end of the stage classification module is data-connected to the output end of the multi-source monitoring data acquisition module, and the output end of the stage classification module is data-connected to the input end of the remaining lifespan prediction sub-model calling module.

[0055] The remaining lifespan prediction sub-model invocation module is used to invoke the remaining lifespan prediction sub-model corresponding to the degradation stage label from a pre-built model library. The model library is built based on several independent training channels and stores four remaining lifespan prediction sub-models corresponding to the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage, respectively. Each remaining lifespan prediction sub-model has a one-to-one mapping relationship with its corresponding degradation stage label. The remaining lifespan prediction sub-model invocation module is data-connected to the stage classification module, receives the degradation stage label output by the stage classification module, uses the degradation stage label as an index to perform a matching query in the model library, invokes the remaining lifespan prediction sub-model corresponding to the current degradation stage label, and loads the invoked sub-model into the prediction execution environment.

[0056] The remaining life prediction result output module is used to input multi-source monitoring data into the invoked remaining life prediction sub-model, and output the remaining life prediction result of the wind turbine pitch bearing through the remaining life prediction sub-model. The remaining life prediction result output module is data-connected to both the multi-source monitoring data acquisition module and the remaining life prediction sub-model invoking module. On the one hand, it receives the current multi-source monitoring data output by the multi-source monitoring data acquisition module, and on the other hand, it receives the remaining life prediction sub-model invoked by the remaining life prediction sub-model invoking module. It inputs the current multi-source monitoring data into the remaining life prediction sub-model to perform forward calculation, and outputs the predicted value of the remaining life duration of the wind turbine pitch bearing from the current moment to the failure threshold from the sub-model's output terminal.

[0057] The wind turbine pitch bearing full life cycle remaining life prediction system provided in this embodiment achieves a complete prediction chain from multi-source data acquisition, automatic stage identification, dedicated model invocation, to remaining life output through the coordinated cooperation of various functional modules. The multi-source monitoring data acquisition module provides multi-dimensional data support, the stage classification module realizes accurate automatic division of degradation stages, the remaining life prediction sub-model invocation module ensures that the optimal prediction model is matched for each stage, and the remaining life prediction result output module provides quantified life prediction values. The data flow between the modules is clear and the interfaces are well-defined, together forming a complete, real-time deployable engineering remaining life prediction system that can effectively support the rapid decision-making needs of on-site condition-based maintenance in wind farms.

[0058] See Figure 2 This embodiment provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor is the computing and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in a computer storage medium to realize a corresponding method flow or corresponding function. The processor described in this embodiment can be used to perform related operations of the wind turbine pitch bearing full life cycle remaining life prediction method.

[0059] This embodiment provides a computer-readable storage medium, which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space containing the terminal's operating system; and this storage space also contains one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the wind turbine pitch bearing full life cycle remaining life prediction method in this embodiment.

[0060] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0061] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the remaining life of a wind turbine pitch bearing throughout its entire life cycle, characterized in that, Includes the following steps: Acquire multi-source monitoring data of the wind turbine pitch bearing throughout its entire life cycle. The multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, speed signals, and SCADA system operating parameter data. Multi-source monitoring data is input into a pre-trained stage classification model, and the degradation stage label is output through the stage classification model. Based on the degradation stage label, the remaining lifetime prediction sub-model corresponding to the degradation stage label is called from a pre-built model library; wherein, the model library is built based on several independent training channels; Multi-source monitoring data is input into the called remaining life prediction sub-model, and the remaining life prediction result of the wind turbine pitch bearing is output through the remaining life prediction sub-model.

2. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 1, characterized in that, The step of inputting multi-source monitoring data into a pre-trained stage classification model and outputting degradation stage labels through the stage classification model specifically includes: After aligning and normalizing the multi-source monitoring data according to the timestamps, the data is input into the pre-trained stage classification model. The input layer of the stage classification model receives multi-source monitoring data. After several layers of feature extraction, the output layer of the stage classification model calculates the probability vectors of the current monitoring data belonging to the healthy stage, early degradation stage, intermediate degradation stage, and accelerated degradation stage respectively through the Softmax activation function. The stage label corresponding to the maximum probability vector is taken as the output degradation stage label.

3. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 1, characterized in that, The step of calling the remaining lifetime prediction sub-model corresponding to the degradation stage label from a pre-built model library specifically includes: The degradation stage label is used as an index identifier and matched with the stage label mapping relationship of each remaining life prediction sub-model in the pre-built model library. The model library stores four remaining life prediction sub-models corresponding to the healthy stage, early degradation stage, intermediate degradation stage and accelerated degradation stage, respectively. Each remaining life prediction sub-model forms a one-to-one mapping relationship with the corresponding degradation stage label. Based on the matching results, the remaining lifetime prediction sub-model corresponding to the degradation stage label is called from the model library.

4. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 1, characterized in that, The remaining life prediction results of the wind turbine pitch bearings are output through the remaining life prediction sub-model, specifically including: The input layer of the remaining lifetime prediction sub-model receives the current multi-source monitoring data. After forward propagation calculation, the output layer of the remaining lifetime prediction sub-model outputs the predicted value of the remaining lifetime of the wind turbine pitch bearing from the current moment to the failure threshold.

5. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 1, characterized in that, Following the predicted remaining life of the output wind turbine pitch bearings are also: The remaining life prediction results are compared with preset multi-level early warning thresholds, and different levels of early warning signals are output based on the comparison results. The multi-level early warning thresholds include a first-level early warning threshold, a second-level early warning threshold, and a third-level early warning threshold. The first-level early warning threshold corresponds to a remaining life prediction value of more than 50%, the second-level early warning threshold corresponds to a remaining life prediction value of 20% to 50%, and the third-level early warning threshold corresponds to a remaining life prediction value of less than 20%.

6. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 1, characterized in that, The construction steps of the model library specifically include: Obtain the historical monitoring data sample set corresponding to each degradation stage label, and allocate the historical monitoring data sample set of each stage to four independent training channels according to the degradation stage label. In each training channel, the historical monitoring data sample set of the current stage is used as the input feature, and the real degradation trajectory of the subsequent stage is used as the supervision label to train the remaining life prediction sub-model that matches the degradation pattern of the current stage. The four remaining lifetime prediction sub-models that have been trained are packaged according to the degradation stage labels to obtain the constructed model library. Each remaining lifetime prediction sub-model in the model library corresponds one-to-one with each degradation stage label.

7. The method for predicting the remaining lifespan of a wind turbine pitch bearing according to claim 6, characterized in that, The training channels include a healthy phase training channel, an early deterioration phase training channel, a mid-term deterioration phase training channel, and an accelerated deterioration phase training channel. The training channel for the healthy stage is based on a statistical model of the Weibull distribution, the training channels for the early and mid-stage degradation stages are based on deep learning models of long short-term memory networks, and the training channel for the accelerated degradation stage is based on a degradation tracking model of particle filtering, resulting in four sub-models for predicting remaining lifetime after training.

8. A system for predicting the remaining lifespan of a wind turbine pitch bearing throughout its entire life cycle, characterized in that, The method for predicting the remaining life of a wind turbine pitch bearing according to any one of claims 1-7 includes: The multi-source monitoring data acquisition module is used to acquire multi-source monitoring data of the wind turbine pitch bearing throughout its entire life cycle. The multi-source monitoring data includes vibration signals, temperature signals, pitch torque signals, speed signals, and SCADA system operating parameter data. The stage classification module is used to input multi-source monitoring data into a pre-trained stage classification model and output degradation stage labels through the stage classification model; The remaining lifetime prediction sub-model calling module is used to call the remaining lifetime prediction sub-model corresponding to the degradation stage label from a pre-built model library; wherein, the model library is built based on several independent training channels; The remaining life prediction result output module is used to input multi-source monitoring data into the called remaining life prediction sub-model, and output the remaining life prediction result of the wind turbine pitch bearing through the remaining life prediction sub-model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting the remaining life of the wind turbine pitch bearing throughout its entire life cycle as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting the remaining life of the wind turbine pitch bearing throughout its entire life cycle as described in any one of claims 1-7.

Images