Intelligent maintenance and fault diagnosis system of wind turbine variable pitch system

Abstract

This invention discloses an intelligent maintenance and fault diagnosis system for wind turbine pitch systems, specifically relating to the field of wind turbines. It includes a multi-sensor data acquisition module, a data preprocessing module, a real-time status monitoring module, a fault feature identification module, an intelligent diagnostic decision-making module, a predictive maintenance strategy module, and a human-machine interface module. Addressing the limitations of existing wind turbine pitch system maintenance methods, such as limited fault diagnosis capabilities, low maintenance efficiency, and difficulty in cost control, this invention employs a modular design and integrates signal processing, machine learning, and visualization technologies to construct an intelligent maintenance and fault diagnosis system. Through the coordinated operation of functional modules such as multi-sensor all-around data acquisition, high-precision fault feature identification, intelligent diagnostic decision-making, and predictive maintenance strategy formulation, it achieves early fault warning and accurate diagnosis of key components of the pitch system, significantly improving equipment operational reliability and maintenance efficiency.

Description

Technical Field

[0001] This invention relates to the field of wind turbine technology, and more specifically, to an intelligent maintenance and fault diagnosis system for a wind turbine pitch system. Background Technology

[0002] Current wind turbine pitch system maintenance primarily employs a combination of periodic maintenance and reactive fault handling. The traditional maintenance process begins by collecting basic operating parameters such as current, voltage, and speed through a SCADA system. Then, maintenance personnel periodically conduct manual inspections of key components like pitch bearings, pitch motors, and reducers, recording the equipment's appearance and basic operating data. Finally, a fixed-cycle maintenance plan is developed based on equipment uptime and manufacturer recommendations. Fault identification relies heavily on experience and simple threshold alarms, with repairs only initiated when obvious fault symptoms appear.

[0003] However, existing technologies have significant shortcomings, mainly in terms of limited fault diagnosis capabilities, low maintenance efficiency, and difficulty in cost control. Traditional threshold alarm methods can only detect obvious faults that have already occurred, failing to provide early warnings or identify progressive faults. This results in equipment not being addressed in its early stages and developing into serious malfunctions. Regular maintenance models lack specificity, potentially leading to over-maintenance and increased costs, or missing optimal maintenance opportunities and causing sudden failures. Furthermore, the reliability of data from single sensors and human experience is insufficient, making it difficult to accurately identify complex fault modes and multiple coupled faults, thus failing to provide a scientific basis for operational decisions. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent maintenance and fault diagnosis system for wind turbine pitch system, which solves the problems mentioned in the background art through the following solutions.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent maintenance and fault diagnosis system for a wind turbine pitch system, comprising:

[0006] Multi-sensor data acquisition module: By deploying sensors in the wind turbine pitch system, it collects real-time operating status data of the pitch system and connects directly to the data preprocessing module;

[0007] Data preprocessing module: Receives raw sensor data transmitted from the multi-sensor data acquisition module, performs filtering, noise reduction, standardization and feature extraction on the data. It adopts a multi-level filtering architecture, firstly performs hardware-level filtering to remove high-frequency noise, and then performs software-level adaptive filtering. The processed data will be transmitted to the real-time status monitoring module and the fault feature identification module at the same time.

[0008] Real-time status monitoring module: Based on the data provided by the data preprocessing module, the module uses sliding window technology to monitor the operating status parameters of each component of the pitch system in real time, establishes a benchmark model of the normal operating status of the pitch system, and judges the health status of the system by calculating the deviation between the current status and the benchmark status in real time.

[0009] Fault Feature Identification Module: Receives feature data from the data preprocessing module and status tags from the real-time status monitoring module, uses machine learning algorithms to identify the characteristic patterns of typical faults in the pitch system, and transmits the identification results to the intelligent diagnostic decision module.

[0010] Intelligent Diagnosis Decision Module: By integrating the status information from the real-time status monitoring module and the feature recognition results from the fault feature recognition module, a multi-layer decision fusion mechanism is adopted to arrive at the final fault diagnosis conclusion. The diagnosis result will be transmitted to the predictive maintenance strategy module and the human-machine interface module at the same time.

[0011] Predictive maintenance strategy module: Based on the diagnostic results of the intelligent diagnostic decision module and historical maintenance data, predictive maintenance strategies and maintenance plans are formulated. The maintenance strategies are displayed to the operation and maintenance personnel through the human-computer interaction interface module to guide the actual maintenance work.

[0012] Human-computer interaction interface module: Provides system monitoring interface and maintenance management functions for operation and maintenance personnel. It adopts a web architecture design, supports remote access and mobile terminal adaptation, and makes it easy for operation and maintenance personnel to keep track of equipment status at any time.

[0013] The technical effects and advantages of this invention are as follows:

[0014] 1. This invention significantly improves the accuracy of fault diagnosis and early warning capability through the collaborative work of a multi-sensor data acquisition module, a data preprocessing module, and a fault feature identification module. The comprehensive deployment of multiple types of sensors in key parts of the pitch system, combined with advanced signal processing technology and machine learning algorithms, can capture weak fault features and progressive degradation signals that are difficult to detect by traditional methods. By fusing multi-source information through support vector machine multi-classification algorithm and DS evidence theory, it achieves accurate identification of typical faults such as bearing wear, motor winding overheating, and reducer gear damage, effectively solving the problems of lagging fault identification and insufficient accuracy in the prior art.

[0015] 2. The integrated application of the intelligent diagnostic decision-making module and the predictive maintenance strategy module realizes the transformation from passive maintenance to proactive maintenance, which greatly improves maintenance efficiency and the rationality of resource allocation. By predicting the remaining service life of the equipment through the Weibull proportional risk model and combining it with the maintenance cost optimization model to formulate a scientific maintenance plan, maintenance work can be carried out at the optimal time, avoiding the problems of over-maintenance and under-maintenance in the traditional periodic maintenance mode. The real-time status monitoring and trend analysis functions enable maintenance personnel to grasp the health status of the equipment in advance, rationally allocate maintenance resources and spare parts reserves, and significantly improve the pertinence and timeliness of maintenance work.

[0016] 3. The human-computer interaction interface module provides visual monitoring and intelligent decision support functions, which greatly improve the convenience of operation and maintenance management and the scientific nature of decision-making. The remote access capability of the Web architecture and the mobile terminal adaptation design enable operation and maintenance personnel to grasp the equipment status anytime and anywhere and respond to abnormal situations in a timely manner. The three-dimensional visualization interface and multi-level information display methods intuitively present complex diagnostic results and maintenance suggestions, reducing the requirements for the professional skills of operation and maintenance personnel. The automatic alarm and report generation functions further improve the fault response speed and the completeness of maintenance records, effectively solving the problems of untimely information transmission and insufficient decision-making basis in traditional maintenance management. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0018] Figure 2 This is a schematic diagram of the fault feature identification module of the present invention.

[0019] Figure 3 This is a schematic diagram of the intelligent diagnostic decision module structure of the present invention.

[0020] Figure 4 This is a schematic diagram of the human-computer interaction interface module structure of the present invention. Detailed Implementation

[0021] 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, 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 are within the scope of protection of the present invention.

[0022] refer to Figures 1-4 The intelligent maintenance and fault diagnosis system for a wind turbine pitch system shown includes:

[0023] Multi-sensor data acquisition module: By deploying sensors in the wind turbine pitch system, the system's operating status data is collected in real time and directly connected to the data preprocessing module.

[0024] The multi-sensor data acquisition module installs triaxial accelerometers on the outer and inner rings of the pitch bearing of each blade, with a sampling frequency set to 10kHz, to monitor the radial and axial vibration of the bearings; a temperature sensor is installed on the stator winding of the pitch motor to monitor the temperature range of -40°C to 150°C; a current sensor is installed at the output end of the pitch motor to monitor the changes in three-phase current in real time; torque sensors and speed sensors are installed on the input and output shafts of the pitch reducer; and strain gauge sensors are installed at the blade root to monitor the stress state of the blade.

[0025] All sensors are connected to the data acquisition controller via a CAN bus network. The data acquisition controller uses a 32-bit ARM processor and has local data buffering and preliminary filtering functions. The sensors are powered by a 24V DC power supply and use PoE technology to achieve single-cable transmission.

[0026] Data preprocessing module: Receives raw sensor data transmitted from the multi-sensor data acquisition module, performs filtering, noise reduction, standardization, and feature extraction on the data. It adopts a multi-level filtering architecture, first performing hardware-level filtering to remove high-frequency noise, and then performing software-level adaptive filtering. The processed data will be transmitted to the real-time status monitoring module and the fault feature identification module simultaneously.

[0027] The data preprocessing module adopts a three-level processing architecture: the first level is anti-aliasing filtering, the second level is adaptive filtering, and the third level is data standardization processing.

[0028] The anti-aliasing filter uses an 8th-order Butterworth low-pass filter for the vibration signal, with the cutoff frequency set to 40% of the sampling frequency.

[0029] Furthermore, the anti-aliasing filter is implemented using a hardware-level analog filter, completing the filtering process before the signal enters the analog-to-digital converter. The filter uses an 8th-order Butterworth low-pass filter topology, which has the flattest amplitude-frequency response characteristics within the passband. The filter cutoff frequency is set to 40% of the sampling frequency; for a vibration signal with a sampling frequency of 10kHz, the cutoff frequency is set to 4kHz. The Butterworth filter achieves its 8th-order characteristics by cascading four second-order active filter units. Each second-order unit adopts a Sallen-Key topology, using high-precision operational amplifiers and low-temperature drift resistors and capacitors. The filter's stopband attenuation rate is 48dB per octave, effectively suppressing signal components above the Nyquist frequency. The filter's input impedance is designed to be high to reduce the load on the sensor signal, while the output impedance is designed to be low to drive the subsequent sampling circuit. The filter is powered by a dual positive and negative power supply to ensure signal integrity and linearity.

[0030] The adaptive filtering uses the LMS (Least Mean Square) adaptive algorithm to remove environmental noise interference.

[0031] The filter weight update formula for the LMS least mean square adaptive algorithm is as follows:

[0032] w(n+1) = w(n) + μ·e(n)·x(n)

[0033] Where w(n) is the weight vector of the nth iteration, μ is the step size factor with a value of 0.001, e(n) is the error signal, and x(n) is the input signal vector.

[0034] Furthermore, the adaptive filtering is implemented using the LMS (Least Mean Square) adaptive algorithm in digital signal processing. The filtering process is completed in the digital signal processor through software algorithms. The algorithm adopts a finite impulse response filter structure with a filter length of 64th order, which can effectively remove environmental noise interference. The reference signal comes from the relatively stable temperature sensor signal in the system and serves as the benchmark for noise estimation. The LMS algorithm minimizes the mean square value of the output error by continuously adjusting the filter weight coefficients. The weight update adopts the gradient descent method, and the step size factor is set to 0.001 to ensure the convergence and stability of the algorithm. The convergence process of the algorithm is achieved by calculating the error between the desired signal and the filter output signal, and then adjusting the weight vector according to the magnitude and direction of the error. The weight coefficients are updated once at each sampling point, enabling the filter to track the changing characteristics of the signal and noise in real time. Fixed-point arithmetic is used in the algorithm implementation to improve the processing speed, and the adjustment range of the weight coefficients is limited within a reasonable range to prevent divergence. The filtering processing delay is controlled within 10 milliseconds to meet the real-time processing requirements.

[0035] The data standardization process employs the Z-score standardization method.

[0036] In terms of feature extraction, time-domain features and frequency-domain features are extracted from vibration signals. Time-domain features include RMS value, peak value, skewness, and kurtosis. Frequency-domain features include power spectral density and characteristic frequency amplitude. Statistical features and trend features are extracted from temperature and current signals.

[0037] Furthermore, the feature extraction employs corresponding methods for different types of sensor signals. For vibration signals, time-domain feature parameters are first calculated, including RMS value, peak value, skewness, and kurtosis. The RMS value reflects the overall vibration intensity, the peak value represents the instantaneous amplitude peak, and the skewness and kurtosis describe the non-Gaussian nature of the signal amplitude distribution, reflecting early signs of equipment failure. Based on this, frequency-domain features are also extracted, and the power spectral density distribution is calculated using Fast Fourier Transform to identify the characteristic frequencies and their amplitudes related to the fault. For temperature and current signals, the focus is on extracting statistical features such as average value, standard deviation, and extreme values, as well as trend features such as first-order difference and second-order difference, to describe the overall trend of signal change. All feature parameters are standardized to form a feature vector, which serves as the input for subsequent fault diagnosis. The feature extraction algorithm is implemented on a digital signal processor, and combined with optimized calculation methods, it can meet the requirements of real-time processing.

[0038] Real-time status monitoring module: Based on the data provided by the data preprocessing module, the module uses sliding window technology to monitor the operating status parameters of each component of the pitch system in real time, establishes a benchmark model of the normal operating status of the pitch system, and judges the health status of the system by calculating the deviation between the current status and the benchmark status in real time.

[0039] The monitoring results of the real-time status monitoring module provide status labels for the fault feature identification module and real-time status information for the intelligent diagnostic decision-making module.

[0040] The real-time status monitoring module adopts the Mahalanobis distance method with multi-parameter fusion. First, a multi-dimensional feature space benchmark model of the normal operating state of the pitch system is established, and the effective value of vibration, temperature, current, torque and speed parameters are selected as state vectors to calculate the Mahalanobis distance.

[0041] The formula for calculating Mahalanobis distance is:

[0042] MD = √[(x-μ)] T ·S -1 ·(x-μ)]

[0043] Where x is the current state vector, μ is the baseline state mean vector, and S is the covariance matrix; the Mahalanobis distance threshold is set to the upper and lower limits determined by the 3σ principle; a sliding window with a length of 100 data points is used to update the state evaluation results once per second.

[0044] The Mahalanobis distance is calculated by taking the square root of the Euclidean distance after the difference between the current state vector and the mean vector of the reference state through the inverse matrix transformation of the covariance matrix. The upper and lower limits of the Mahalanobis distance threshold are determined by the principle of three times the standard deviation.

[0045] The system employs a sliding window technique with 100 data points, updating the state assessment results every second to ensure real-time performance. Simultaneously, a state trend analysis mechanism is established, using exponential smoothing to predict state development trends. The exponential smoothing method calculates the current smoothed value by weighting the current observation value with the smoothed value from the previous time step, with the smoothing coefficient set to 0.3 to control the influence of historical data on the current prediction.

[0046] The formula for calculating the exponential smoothing method is as follows:

[0047] S(t) = α·x(t) + (1-α)·S(t-1)

[0048] Where S(t) is the smoothed value at time t, α is the smoothing coefficient, and x(t) is the observed value at time t.

[0049] The status monitoring results provide status label information to the fault feature identification module and transmit real-time status information to the intelligent diagnostic decision module. When the Mahalanobis distance exceeds the set threshold, the system automatically marks it as an abnormal state and triggers a further fault feature identification process. Under normal conditions, the system continuously monitors the changing trends of various parameters to provide status degradation information for predictive maintenance.

[0050] Fault Feature Identification Module: Receives feature data from the data preprocessing module and status tags from the real-time status monitoring module, uses machine learning algorithms to identify the characteristic patterns of typical faults in the pitch system, and transmits the identification results to the intelligent diagnostic decision module.

[0051] The fault feature identification module uses a support vector machine multi-classification algorithm to automatically identify typical faults in the pitch system; for different fault types, corresponding feature parameters are extracted, including pitch bearing faults, pitch motor faults, and pitch reducer faults.

[0052] The fault feature identification module extracts the envelope spectrum characteristic frequency, impact index, and kurtosis index of the vibration signal as identification features for pitch bearing faults. The characteristic frequency calculation formula for bearing outer ring faults is f_out=0.4×n×Z1×f_r, where n is the rotational speed, Z1 is the number of rolling elements, and f_r is the rotor frequency. For pitch motor faults, the module extracts the harmonic characteristics of the stator current and the temperature change rate as identification criteria. Overheating faults in the motor windings are identified by monitoring the odd-order harmonic components in the stator current and the rate of temperature rise. For pitch reducer faults, the module extracts torque fluctuation characteristics and gear meshing frequency components in the vibration spectrum. The characteristic frequency of gear damage faults is f_gear=Z2×f_shaft, where Z2 is the number of teeth and f_shaft is the shaft rotation frequency.

[0053] The Support Vector Machine (SVM) classifier uses a radial basis function kernel, with the kernel function expression being K(xi,xj)=exp(-γ||xi-xj|| 2 ), where γ is the kernel parameter, which is optimized and determined in the range of 0.001 to 10 through grid search; the multi-class problem is implemented using a one-to-one strategy. For the k-class fault problem, k(k-1) / 2 binary classifiers are constructed, and each binary classifier is responsible for distinguishing two specific fault types; the classification decision function is f(x)=sign(∑(αiyiK(xi,x)+b)), where αi is the Lagrange multiplier, yi is the class label value of +1 or -1, and b is the bias term.

[0054] The model training data was constructed using historical fault data and normal operation data. The dataset contains 500 bearing fault samples, 300 motor fault samples, 200 reducer fault samples, and 1000 normal state samples, which are divided into training and test sets in a 7:3 ratio. The feature vector dimension is set to 15 dimensions, including 8 time-domain features and 7 frequency-domain features. The training process uses the sequential minimum optimization algorithm to solve the dual problem of the support vector machine, and the convergence criterion is set to a gradient change of less than 0.001. Finally, the training data is normalized, and the feature values ​​are mapped to the interval between 0 and 1 using the maximum and minimum value normalization method.

[0055] Intelligent Diagnosis Decision Module: By integrating the status information from the real-time status monitoring module and the feature recognition results from the fault feature recognition module, a multi-layer decision fusion mechanism is adopted to arrive at the final fault diagnosis conclusion. The diagnosis result will be simultaneously transmitted to the predictive maintenance strategy module and the human-machine interface module.

[0056] The intelligent diagnostic decision-making module uses DS evidence theory to fuse multi-source information and sets a basic probability allocation function. For each possible fault hypothesis, the evidence from the real-time status monitoring module and the evidence from the fault feature identification module are given their respective basic probability allocation values. During the evidence fusion process, all possible combinations of the two evidence sources are calculated. When the two pieces of evidence point to the same fault, their probabilities are multiplied and then added together. When the two pieces of evidence conflict with each other, the probability of the conflicting part is deducted from the denominator.

[0057] A fault diagnosis rule base is established, which adopts a conditional judgment form for diagnosis rules. Each rule includes a condition part and a conclusion part. The condition part describes the specific symptoms, and the conclusion part gives the corresponding fault type and confidence level value. For example, when the effective value of bearing vibration exceeds the set threshold and the bearing characteristic frequency appears in the envelope spectrum, the system determines that it is a bearing outer ring fault and gives a confidence level of 0.8.

[0058] The final diagnostic result is determined using the maximum confidence criterion. The system selects the fault type with the highest confidence as the diagnostic conclusion and outputs the complete probability distribution of all possible faults. The system sets the diagnostic threshold to 0.7. When the highest confidence is lower than this threshold, the system outputs a suspected fault status and prompts that further inspection and confirmation are needed. This implementation method can effectively handle the uncertainty information in the diagnostic process and improve the reliability and accuracy of the diagnostic results.

[0059] Furthermore, given the basic probability assignment function m, for the fault hypothesis Hi, the evidence E1 from state monitoring and the evidence E2 from feature recognition give basic probability assignments m1(Hi) and m2(Hi) respectively. Then the evidence fusion rule is:

[0060]

[0061] Where A is the fused focal element, B and C are the focal elements of the two evidence sources respectively, and ∅ is the empty set.

[0062] Predictive maintenance strategy module: Based on the diagnostic results of the intelligent diagnostic decision module and historical maintenance data, predictive maintenance strategies and maintenance plans are formulated. The maintenance strategies are displayed to the operation and maintenance personnel through the human-computer interaction interface module to guide the actual maintenance work.

[0063] The predictive maintenance strategy module uses a remaining useful life prediction algorithm, combined with equipment degradation model and maintenance cost model, to optimize maintenance timing and methods.

[0064] The remaining useful life prediction adopts the Weibull proportional hazards model. First, a mathematical model of the equipment degradation process is established. The conditional hazard function is defined as the product of the baseline hazard function and the exponential function. The parameters of the exponential function are determined by the inner product of the regression coefficient vector and the covariate vector. The covariate vector includes three main factors: operating time, load, and environmental conditions. The regression coefficient vector is determined by establishing a regression model through historical failure data.

[0065] The equipment degradation process is modeled as follows:

[0066] h(t|x)=h0(t)·exp(β T x)

[0067] Where h(t|x) is the conditional risk function, h0(t) is the baseline risk function, β is the regression coefficient vector, and x is the covariate vector.

[0068] The baseline risk function is modeled using the Weibull distribution. The scale parameter and shape parameter of the Weibull distribution are determined by fitting historical failure data. The scale parameter reflects the characteristic lifespan of the equipment, and the shape parameter reflects the change of the failure rate over time. The parameters are estimated using the maximum likelihood estimation method to establish the Weibull distribution model.

[0069] The survival function of the Weibull distribution is:

[0070] S(t) = exp[-(t / η)^k]

[0071] Where η is the scale parameter, k is the shape parameter, and t represents the operating time of the equipment.

[0072] The survival function is calculated using the cumulative distribution function of the Weibull distribution, representing the probability that the device will still be able to operate normally after time t. The remaining useful life prediction is achieved by integrating the survival function over the time interval from the current time to infinity, and the integral result is divided by the survival function value at the current time for conditional probability correction.

[0073] The formula for predicting the remaining useful life is:

[0074] RUL=∫[tc to ∞] S(t|x) / S(tc|x) dt

[0075] Where tc is the current time, x is the covariate vector including runtime, load, and environmental conditions, and S(t|x) is the conditional survival function.

[0076] The integral represents the remaining usable lifespan of the equipment from the current moment. It is divided by the current survival probability for conditional probability correction. In the actual calculation process, the infinity time is set to 3 times the equipment's design lifespan. The numerical integration method is used to solve the problem, with the integration step size set to 1 hour. The integral value is calculated using the trapezoidal integration method. When the calculated remaining lifespan is less than the preset maintenance lead time, the system automatically triggers maintenance suggestions. The prediction results are updated every 24 hours to ensure the timeliness of maintenance decisions.

[0077] The intelligent diagnostic decision-making module uses a minimum maintenance cost model to optimize maintenance strategies and establishes a total cost function model. The total cost of maintenance decisions is decomposed into three components: failure cost, maintenance cost, and downtime loss cost. Failure cost is calculated by multiplying the failure probability by the economic loss caused by a single failure. The failure probability is determined based on the remaining service life prediction results and historical failure statistics.

[0078] The total cost function model is as follows:

[0079] C = Cp·P(failure) + Cm + Cd·D

[0080] Where Cp is the economic loss (yuan) caused by a single failure, P (failure) is the failure probability (dimensionless), Cm is the maintenance cost (yuan), Cd is the downtime loss coefficient (yuan / hour), and D is the downtime (hours).

[0081] In this embodiment, it is necessary to specifically explain that the downtime loss coefficient Cd is determined through the power generation revenue loss of the wind turbine: First, the rated installed capacity of the wind turbine is obtained, in megawatts as the basic power parameter. Combined with the average wind speed data and wind turbine power curve of the wind turbine's location, the average capacity factor of the wind turbine is calculated, which is the ratio of the actual power generation to the rated power. Then, using the local grid's on-grid electricity price data, the revenue per kilowatt-hour is determined according to the wind power on-grid electricity price standard. By multiplying the annual power generation hours of the wind turbine by the average capacity factor, the effective power generation hours of the wind turbine are obtained. The effective power generation hours are divided by 8760 hours to calculate the proportion of the wind turbine's hourly power generation revenue to the annual revenue. The downtime loss coefficient Cd is equal to the wind turbine's rated power multiplied by the average capacity factor and then multiplied by the on-grid electricity price to obtain the hourly power generation revenue loss amount. Considering that wind turbine downtime will also affect the overall power generation plan of the wind farm and grid dispatch, an indirect loss coefficient of 20% is added to the basic loss. The final downtime loss coefficient is in yuan per hour and is directly used for the calculation of the maintenance cost optimization model.

[0082] Human-computer interaction interface module: Provides system monitoring interface and maintenance management functions for operation and maintenance personnel. It adopts a web architecture design, supports remote access and mobile terminal adaptation, and makes it easy for operation and maintenance personnel to keep track of equipment status at any time.

[0083] The human-computer interaction interface module integrates the status display of the real-time status monitoring module, the fault diagnosis results of the intelligent diagnosis and decision-making module, and the maintenance suggestion information of the predictive maintenance strategy module.

[0084] The human-computer interaction interface module is designed using a B / S architecture. The front-end uses HTML5 to build the page structure, CSS3 for style design and responsive layout, and JavaScript for dynamic interaction. The back-end server is built using Node.js technology. The entire interface system includes four main functional pages: a real-time monitoring screen, a fault diagnosis page, a maintenance management page, and a historical data query page. The real-time monitoring screen displays a 3D model of the wind turbine in the center of the interface. The various components of the pitch system in the model are visually represented by color coding: green indicates normal operation, yellow indicates a warning state requiring attention, and red indicates a fault requiring immediate action. The fault diagnosis page is specifically designed to display the analysis results of the intelligent diagnostic system. The system displays the final diagnosis results at the top of the page, a visual representation of the diagnostic confidence level and progress bar in the middle, and the specific location of the fault and detailed handling suggestions at the bottom. The maintenance management page centrally displays predictive maintenance information. The left side shows a list of maintenance plans sorted by time, the middle area displays a trend curve of the remaining service life prediction, and the right side displays detailed information on historical maintenance records. The historical data query page provides data retrieval and analysis functions, allowing users to query required historical operating data by selecting time range, equipment type, and parameter type. The system's data communication mechanism uses the WebSocket protocol to achieve real-time bidirectional communication between the server and client, ensuring that monitoring data is updated to the user interface at a refresh rate of 1 second. When the system detects a fault, the alarm function is automatically activated, displaying an alarm prompt window on the user interface, and the system automatically sends a fault notification SMS to the preset maintenance personnel's mobile phone number. To meet maintenance management needs, the system has a built-in data export function, supporting the export of monitoring data and analysis reports as CSV data files and standardized PDF report documents, facilitating subsequent data analysis and archiving management.

[0085] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0086] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent maintenance and fault diagnosis system for a wind turbine pitch system, characterized in that, include: Multi-sensor data acquisition module: By deploying sensors in the wind turbine pitch system, it collects real-time operating status data of the pitch system and connects directly to the data preprocessing module; Data preprocessing module: Receives raw sensor data transmitted from the multi-sensor data acquisition module, performs filtering, noise reduction, standardization and feature extraction on the data. It adopts a multi-level filtering architecture, firstly performs hardware-level filtering to remove high-frequency noise, and then performs software-level adaptive filtering. The processed data will be transmitted to the real-time status monitoring module and the fault feature identification module at the same time. Real-time status monitoring module: Based on the data provided by the data preprocessing module, the module uses sliding window technology to monitor the operating status parameters of each component of the pitch system in real time, establishes a benchmark model of the normal operating status of the pitch system, and judges the health status of the system by calculating the deviation between the current status and the benchmark status in real time. Fault Feature Identification Module: Receives feature data from the data preprocessing module and status tags from the real-time status monitoring module, uses machine learning algorithms to identify the characteristic patterns of typical faults in the pitch system, and transmits the identification results to the intelligent diagnostic decision module. Intelligent Diagnosis Decision Module: By integrating the status information from the real-time status monitoring module and the feature recognition results from the fault feature recognition module, a multi-layer decision fusion mechanism is adopted to arrive at the final fault diagnosis conclusion. The diagnosis result will be transmitted to the predictive maintenance strategy module and the human-machine interface module at the same time. Predictive maintenance strategy module: Based on the diagnostic results of the intelligent diagnostic decision module and historical maintenance data, predictive maintenance strategies and maintenance plans are formulated. The maintenance strategies are displayed to the operation and maintenance personnel through the human-computer interaction interface module to guide the actual maintenance work. Human-computer interaction interface module: Provides system monitoring interface and maintenance management functions for operation and maintenance personnel. It adopts a web architecture design, supports remote access and mobile terminal adaptation, and makes it easy for operation and maintenance personnel to keep track of equipment status at any time.

2. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The multi-sensor data acquisition module installs triaxial accelerometers on the outer and inner rings of the pitch bearing of each blade, with a sampling frequency set to 10kHz, to monitor the radial and axial vibration of the bearings; a temperature sensor is installed on the stator winding of the pitch motor to monitor the temperature range of -40°C to 150°C; a current sensor is installed at the output end of the pitch motor to monitor the changes in three-phase current in real time; torque sensors and speed sensors are installed on the input and output shafts of the pitch reducer; and strain gauge sensors are installed at the blade root to monitor the stress state of the blade.

3. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The data preprocessing module adopts a three-level processing architecture: the first level is anti-aliasing filtering, the second level is adaptive filtering, and the third level is data standardization processing. The anti-aliasing filter uses an 8th-order Butterworth low-pass filter for the vibration signal, with the cutoff frequency set to 40% of the sampling frequency; The adaptive filtering uses the LMS least mean square adaptive algorithm to remove environmental noise interference. The data standardization process employs the Z-score standardization method.

4. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The real-time status monitoring module adopts the multi-parameter fusion Mahalanobis distance method. First, a multi-dimensional feature space benchmark model of the normal operating state of the pitch system is established, and the vibration effective value, temperature, current, torque and speed parameters are selected as state vectors to calculate the Mahalanobis distance. The system uses a sliding window technique with a length of 100 data points to update the state assessment results once per second; at the same time, it establishes a state trend analysis mechanism and uses exponential smoothing to predict the state development trend.

5. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The fault feature recognition module uses a support vector machine multi-classification algorithm to automatically identify typical faults in the pitch system; for different fault types, corresponding feature parameters are extracted, including pitch bearing faults, pitch motor faults, and pitch reducer faults. Multi-class classification employs a one-to-one strategy, constructing k(k-1) / 2 binary classifiers for k-class problems; The model was trained using historical fault data and normal operation data, with the dataset divided into training and test sets in a 7:3 ratio.

6. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The intelligent diagnostic decision-making module uses DS evidence theory to fuse multi-source information and sets a basic probability allocation function. For each possible fault hypothesis, the evidence from the real-time status monitoring module and the evidence from the fault feature identification module are given their respective basic probability allocation values. During the evidence fusion process, all possible combinations of the two evidence sources are calculated. When the two pieces of evidence point to the same fault, their probabilities are multiplied and then added together. When the two pieces of evidence conflict with each other, the probability of the conflicting part is deducted from the denominator. The final diagnosis is determined using the maximum confidence criterion. The system selects the fault type with the highest confidence as the diagnosis conclusion and outputs the complete probability distribution of all possible faults. The system sets the diagnosis threshold to 0.

7. When the highest confidence is lower than this threshold, the system outputs a suspected fault status and prompts that further inspection and confirmation are needed.

7. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The predictive maintenance strategy module uses a remaining useful life prediction algorithm, combined with an equipment degradation model and a maintenance cost model, to optimize maintenance timing and methods. The remaining useful life prediction adopts the Weibull proportional hazards model. First, a mathematical model of the equipment degradation process is established. The conditional risk function is defined as the product of the benchmark risk function and the exponential function. The parameters of the exponential function are determined by the inner product of the regression coefficient vector and the covariate vector. The covariate vector includes three main factors: operating time, load, and environmental conditions. The regression coefficient vector is determined by establishing a regression model through historical failure data. The baseline risk function is modeled using the Weibull distribution. The scale parameter and shape parameter of the Weibull distribution are determined by fitting historical failure data. The scale parameter reflects the characteristic lifespan of the equipment, and the shape parameter reflects the change of the failure rate over time. The parameters are estimated using the maximum likelihood estimation method to establish the Weibull distribution model.

8. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 7, characterized in that: The intelligent diagnostic decision-making module uses a minimum maintenance cost model to optimize maintenance strategies and establishes a total cost function model. The total cost of maintenance decisions is decomposed into three components: failure cost, maintenance cost, and downtime loss cost. Failure cost is calculated by multiplying the failure probability by the economic loss caused by a single failure. The failure probability is determined based on the remaining service life prediction results and historical failure statistics.

9. The intelligent maintenance and fault diagnosis system for a wind turbine pitch system according to claim 1, characterized in that: The human-computer interaction interface module integrates the status display of the real-time status monitoring module, the fault diagnosis results of the intelligent diagnosis and decision module, and the maintenance suggestion information of the predictive maintenance strategy module.

Images