A fault diagnosis and health management system for fans

Abstract

The present application relates to a fault diagnosis and health management system for a wind turbine, including an edge data acquisition module, a big data governance module, an early fault diagnosis module, a performance evaluation and root cause analysis module, a fault impact quantitative assessment module, an intelligent scheduling module, an intelligent spare parts module and an intelligent knowledge base. The present application can integrate various types of multivariate heterogeneous data of the core components of the wind turbine, and take the failure mechanism of the components as a guide, with the goal of monitoring and diagnosing the failure mode of the core components, and realize automatic root cause analysis, failure impact and risk assessment of the diagnosed failure mode, and quantitative assessment of the time of failure expansion, providing a basis for the maintenance and overhaul of the diagnosed failure mode, the time window allowed for processing and the urgency assessment, and at the same time using this information as input for intelligent operation and maintenance scheduling, intelligent optimization of spare parts and knowledge base management, and realizing a two-way closed loop from diagnosis and early warning to maintenance implementation and from operation and maintenance implementation results to diagnosis and early warning.

Description

Technical Field

[0001] The present application relates to a fault diagnosis and health management system for a wind turbine, which is applicable to the technical field of health monitoring of wind power equipment. Background Art

[0002] In wind turbines, rotor diameters and megawatt ratings are increasingly trending toward larger sizes. In addition to the turbine's built-in supervisory control and data acquisition (SCADA) system and drive train vibration monitoring system (CMS), the following factors are driving increasing attention to fault diagnosis, early warning, and health management systems for core turbine components: The nonlinear effects of large turbine structures lead to more complex and sensitive multi-body dynamic responses, increasing operational risks; the cost of unit components increases significantly, making replacement and repair costs higher than for smaller turbines; and the risk of damage and defects arising from manufacturing, transportation, and hoisting increases. To reduce the overall lifecycle cost of wind turbines, it is imperative to transform transmission operation and maintenance models into precise operations through intelligent means, minimizing downtime and maintenance costs, and avoiding major safety incidents such as overhauls, replacements, and even tower collapses.

[0003] Various big data platforms and intelligent diagnostic platforms have been introduced in existing technologies, but most of these platforms are independent and face the dilemma of information silos. On the one hand, they fail to cover all core components of wind turbines, and on the other hand, they fail to cover multiple types of monitoring data from the same wind turbine. Moreover, many platforms or systems are primarily based on one or a few types of data, monitoring the status of one or a few components, and diagnosing abnormalities. They rarely systematically focus on comprehensive core components of wind turbines and provide detailed failure mode-based fault warning and diagnosis.

[0004] Therefore, the existing technology requires a fault diagnosis and health management system for wind turbines, which can integrate multi-dimensional heterogeneous data, and be guided by the failure mechanism of the core components of the wind turbine, with the goal of monitoring and diagnosing the important and high-frequency failure modes of each core component. It can transform the traditional operation and maintenance mode into precise operation and maintenance, and can achieve automation and intelligent effects. Summary of the Invention

[0005] The present application provides a fault diagnosis and health management system for a wind turbine, which can integrate various multi-dimensional heterogeneous data of the core components of the wind turbine, and is guided by the failure mechanism of the components, with the goal of monitoring and diagnosing the failure mode of the core components, and realizes automatic root cause analysis, failure impact and risk assessment of the diagnosed failure mode, and quantitative assessment of the time of failure expansion, providing a basis for the maintenance and overhaul of the diagnosed failure mode, the time window allowed for processing, and the urgency assessment. At the same time, this information is used as input for intelligent operation and maintenance scheduling, intelligent optimization of spare parts, and knowledge base management, and realizes a two-way closed loop from diagnosis and warning to maintenance implementation, and from operation and maintenance implementation results to diagnosis and warning.

[0006] This application relates to a fault diagnosis and health management system for wind turbines, including the following modules:

[0007] Edge data acquisition module, which collects data from wind turbine components and checks data quality to remove interference data;

[0008] The big data governance module is used to aggregate, classify and store various types of heterogeneous data. It uses a built-in data quality screening model to perform quality screening on data of different types and structures.

[0009] An early fault diagnosis module, which is equipped with an algorithmic model of the failure mode of the wind turbine components. The algorithmic model selectively uses the multivariate heterogeneous data in the edge data acquisition module for diagnosis according to the needs of failure diagnosis, and issues early warning prompts based on the diagnosis results;

[0010] The performance evaluation and root cause analysis module evaluates the impact of the failure mode corresponding to the diagnosis results of the early fault diagnosis module on the performance of the wind turbine and analyzes the root causes that may cause the failure mode;

[0011] Fault impact quantitative assessment module, which performs quantitative assessment and safety risk assessment on the failure impact of the diagnosed fault and retrieves the corresponding decision-making plan and maintenance plan;

[0012] The intelligent scheduling module dynamically reads the maintenance tasks submitted based on fault diagnosis and evaluation, as well as the input information used as the scheduling model, and provides the output results of the intelligent scheduling;

[0013] The intelligent spare parts module automatically reads spare parts information and dynamically reads demand information from the task pool of the intelligent scheduling module, optimizes intelligent spare parts management tasks, and outputs management measures;

[0014] Intelligent knowledge base, which provides query and input of professional information, and dynamically accumulates expert knowledge and implementation process experience.

[0015] The algorithm model sustainability evaluation module provides automatic closed-loop and iteration of the algorithm model performance and completeness throughout its life cycle.

[0016] Among them, the algorithm model sustainability evaluation module includes a feature library and a model library. The feature library stores and automatically evaluates and updates the features related to the early fault diagnosis model; the model library contains all algorithm models related to the system and meets the algorithm model calling requirements of each module. The models in the model library can perform model performance evaluation.

[0017] Among them, the edge data acquisition module includes: a multi-heterogeneous data acquisition unit, which collects data of various categories and different sampling frequencies; an edge data condition acquisition unit, which automatically collects data of a set duration and that meets the set characteristics or conditions based on the algorithm model of the fan operating parameters, environmental conditions, and failure feature recognition function; an edge data quality management unit, which automatically checks the data quality and automatically screens interference data for the collected original data when performing conditional acquisition.

[0018] Among them, the big data governance module performs quality screening on data of different types and structures, specifically including: integrity screening, including screening of null value ratio and resampling ratio; uniqueness screening, including screening of duplication ratio; timeliness screening, including screening of timestamp vacancy ratio; validity screening, including screening of data type anomalies, over-range, and state variable validity; accuracy screening, including screening of sampling rate, noise data ratio, data accuracy, and zero drift; consistency screening, including screening of logical deviation ratio.

[0019] Among them, the early fault diagnosis module independently models various failure modes of each component, and the data read from the big data governance module will be cleaned by calling the data quality algorithm model built into the early fault diagnosis module; the performance evaluation and root cause analysis module will construct a power curve degradation model and a life attenuation model, and output the degraded power curve and the reduced life attenuation curve; Among them, the expert experience corresponding to each failure mode will be precipitated and saved in the intelligent knowledge base through the pre-constructed DFMEA library. When a failure mode is diagnosed, the root cause path tree in the intelligent knowledge base is automatically called, and the root cause analysis model corresponding to each path branch is started. The module locks the root cause that caused the failure mode according to the root cause analysis results, and displays the root cause prompt through the application module.

[0020] Among them, the fault impact quantitative assessment module uses the degraded power curve and the reduced life attenuation curve output by the performance evaluation and root cause analysis module to perform a quantitative assessment of the single-day power generation loss and a quantitative assessment of the time window for the failure to worsen to the next stage based on the input data; the fault impact quantitative assessment module calls the personnel and wind turbine safety impact knowledge base corresponding to the failure mode from the intelligent knowledge base to perform a safety risk assessment; the fault impact quantitative assessment module also includes a decision-making unit, which uses information including power generation loss, operable window time, and safety risk as input for operation and maintenance plan decisions, and calls corresponding decision information, decision-making process, and maintenance plan from the intelligent knowledge base.

[0021] Among them, the intelligent scheduling module will dynamically monitor the location and status information of personnel and vehicles, support the collection and storage of multi-dimensional heterogeneous information, and automatically count working hours information to provide closed-loop feedback for the scheduling model; the final confirmed failure mode and failure cause will be fed back to the early fault diagnosis module and the performance evaluation and root cause analysis module, and the operation link information that is more in line with the actual situation on site during the execution process will be fed back to the intelligent knowledge base to optimize existing plans and processes; at the same time, it will automatically provide troubleshooting prompts for wind turbines or wind farms with similar root cause phenomena.

[0022] Among them, the intelligent spare parts module uses the spare parts information and the demand information as input for spare parts management optimization, reads the failure frequency and proportion of each component failure mode from the intelligent knowledge base, uses the algorithm model for dynamic optimization management of intelligent spare parts, optimizes the intelligent spare parts management tasks for the target, and outputs specific management measures.

[0023] Among them, the evaluation dimensions of the algorithm model sustainability evaluation module include the failure mode coverage completeness dimension and the algorithm model economy dimension; the failure mode coverage completeness dimension is based on the failure types dynamically counted in the knowledge base, and for failure modes that are not covered by the models in the algorithm model library, it automatically prompts to add and develop early failure diagnosis models and performance evaluation models corresponding to the failure mode; the algorithm model economy dimension diagnoses the electricity and labor costs saved, minus the electricity loss and labor cost waste caused by algorithm false alarms, minus the additional downtime electricity loss and labor cost caused by algorithm omissions, and evaluates the algorithm model based on the two economic indicators of total electricity loss and labor cost.

[0024] According to the fault diagnosis and health management system of this application, it is possible to integrate the multivariate heterogeneous data of each component and system of the wind turbine, and take the failure mechanism of the core components of the wind turbine as a guide, with the goal of monitoring and diagnosing the important and high-frequency failure modes of each core component. The monitoring and diagnosis results can further conduct root cause analysis, failure impact and risk assessment, and can be closed-loop applied to operation and maintenance scheduling optimization, spare parts optimization and management, and even professional knowledge accumulation. At the same time, operation and maintenance implementation information can be closed-loop fed back to monitoring and diagnosis, ultimately achieving automation and intelligent effects. In particular, the fault diagnosis and health management system according to this application has the following technical advantages:

[0025] (1) The early fault diagnosis module comprehensively covers the high-frequency and major failure types of each core component of the wind turbine through the algorithm model library, ensuring the comprehensive coverage of the smart health management function;

[0026] (2) Different models in the model library will interact with each other due to the needs of data quality and root cause analysis, thus avoiding the isolated existence of models from the systematic perspective of failure impact;

[0027] (3) Not only did the diagnostic results realize their value in operation and maintenance implementation and spare parts management, but also the algorithm model was continuously iterated and optimized from the two dimensions of completeness and economy through dynamic statistics of failure types and frequencies after implementation.

[0028] (4) From the edge data acquisition module, big data governance module to the early fault diagnosis module, the quality management and assurance of data in the entire process are achieved, providing a strong foundation for improving the accuracy of the algorithm and reducing the missed alarm rate and false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the system architecture of the fault diagnosis and health management system of this application.

[0030] Figure 2 This is a comparison chart of the power curves of the gearbox high-speed shaft bearing after wear and before damage in the embodiment.

[0031] Figure 3 1 is a comparison chart of the life attenuation curves of the high-speed shaft bearings of the gearbox in the embodiment.

[0032] Figure 4 4 is a schematic diagram of the decision-making process of the fault impact quantitative assessment module in the embodiment. DETAILED DESCRIPTION

[0033] To make the purpose, technical solutions and advantages of this application more clear, the embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of this application can be combined with each other in any way.

[0034] According to the present application, a fault diagnosis and health management system for a wind turbine includes the following modules: an edge data acquisition module, a big data governance module, an early fault diagnosis module, a performance evaluation and root cause analysis module, a fault impact quantitative assessment module, an intelligent scheduling module, an intelligent spare parts module, an intelligent knowledge base and an algorithm model sustainability evaluation module.

[0035] The edge data acquisition system of the present application is located on the wind turbine side. In the case where the only available data is SCADA data, it can also be set up in a wind farm monitoring center or data center near the booster station. The booster station of a wind farm refers to a facility that increases the output voltage of a wind turbine to a higher voltage level and sends it out. Other modules can be centrally set up on one server or dispersed on multiple servers. The server can be located in the wind farm booster station monitoring center, or in a regional centralized control center or a group control center. The data transmission from the edge data acquisition system to the wind farm booster station monitoring center server can be through the wind farm ring network, or through specially set up wired and wireless transmission links. For each module dispersedly set up on different servers, data transmission can be transmitted through a dedicated power line, or through an isolation network switch, through a public network system.

[0036] The intelligent health management system of the present application can be used for intelligent diagnosis, early warning and health management of early faults of core components of wind turbines, for example, it can be used for blades, pitch systems, main shaft systems, gearboxes, generators, couplings, yaw systems, towers and foundations, wind measurement systems, converters and main control cabinets, etc. Sensors are set on the components that are not covered by the existing SCADA system measurement point signals of each wind turbine to collect corresponding signals. Specifically, by setting a dual-axis acceleration sensor on each of the three blades of each wind turbine to collect the vibration signals of the blades during the operation of the wind turbine; by setting a dual-axis acceleration sensor at the pitch bearing position at the root of the three blades to collect the vibration signals of the pitch system, especially the pitch bearing, during the operation of the wind turbine; by installing distance measuring sensors at the risk section position of the blade root and tower and the bolt sub-end face to collect the bolt relaxation deformation data during the operation of the wind turbine; by installing acceleration sensors on the mechanical transmission chain, i.e. the main bearing, the low-speed stage and high-speed stage of the gearbox, the coupling, the generator drive and non-drive shaft bearings, and at the generator base. Install vertical acceleration sensors to collect vibration signals of the transmission chain during operation; install dual-axis acceleration sensors and dual-axis dynamic inclination sensors on the inner wall of the tower top to collect vibration and shaking signals of the tower during wind turbine operation; install static bidirectional inclination sensors on the foundation ring to collect inclination angles caused by uneven settlement of the foundation during wind turbine operation; install video measurement points at appropriate locations in the cabin to collect images and sounds during transmission chain operation; install high-frequency ranging sensor equipment at appropriate locations in the bottom of the cabin to dynamically collect distance signals between the blade tip and the tower during wind turbine operation.

[0037] An edge data collection cabinet is installed at the base of each wind turbine tower or in the nacelle. This cabinet houses an edge data collection system. This system incorporates an intelligent data quality screening model that screens data for common anomalies, including mean value checks, non-null checks, positive and negative ratio checks, upper and lower limits checks, RMS range checks, adjacent point equality checks, and outlier point ratio checks. It also provides targeted screening for signal quality-impacting features such as electromagnetic interference, temperature drift, and environmental noise. The edge data collection system automatically collects targeted, conditional data based on wind turbine operating parameters such as wind speed, rotational speed, blade angle, yaw state, and power level from the SCADA system, acquiring high-quality data with minimal disk space consumption. Furthermore, the edge data collection system synchronously collects this diverse and heterogeneous data in a time-aligned manner and automatically categorizes and stores it. Data collected by each wind turbine's edge data collection module is sliced, allowing high-frequency data to be exchanged and connected to the wind farm ring network at a low cost. This data is then transmitted back to a server at the monitoring center at the substation. The server at the substation collects various data from the wind turbine's SCADA system through a soft gateway and integrates it with data collected and transmitted from the edge intelligent module. Server configuration requires a comprehensive assessment based on multi-dimensional requirements. This includes assessing raw data storage requirements, including sampling frequency, sampling intervals, and aging strategies for all types of data at all measurement points. The operational strategies for all algorithm models, including those for each core component's failure mode diagnosis algorithm, performance evaluation and root cause analysis algorithm, fault impact assessment algorithm, intelligent scheduling algorithm, and intelligent spare parts, must also be considered, along with individual resource requirements and overall hardware resource requirements. The intelligent knowledge base must also include information on failure mechanisms, operation and inspection plans, and other relevant information for each failure mode, as well as the space resources required for subsequent closed-loop sedimentation.

[0038] This implementation method will configure one server for the management of data from the edge and SCADA, one server for algorithm model operation and update iteration, one server for application display and user configuration management, and one server for the management and storage of intelligent knowledge base information. Each server becomes an organic whole through the integration of the software system. The software system involved in this application includes multiple functional modules: edge data acquisition module, early fault diagnosis module, big data governance module, performance evaluation and root cause analysis module, fault impact quantitative assessment module, intelligent scheduling module, intelligent spare parts module, and intelligent knowledge base. Except for the edge data acquisition module located on the wind turbine side, the remaining modules are all located on the station-side server on the booster station side. The system architecture is as follows: Figure 1 shown.

[0039] Edge data acquisition module

[0040] The edge data acquisition module of the present application has data quality management and screening functions, conditional acquisition functions, and multi-heterogeneous acquisition functions, which realizes the acquisition of data from wind turbine components, including blade vibration, transmission chain vibration, transmission chain sound and image, pitch system vibration, bolt displacement, blade clearance related distance data, tower vibration, tower inclination, foundation inclination from new measuring points; and data from SCADA or main control system: wind speed, wind direction, converter active power, generator speed, wind rotor speed, pitch angle, wind angle, yaw angle, cabin acceleration, ambient temperature, cabin temperature, generator bearing temperature, generator winding temperature, gearbox oil temperature, gearbox bearing temperature, converter temperature, converter machine-side current, main bearing temperature, main shaft temperature, converter DC bus voltage, grid-side current, generator water pump water pressure, pitch motor temperature and current, main control cabinet temperature, gearbox lubrication system oil pressure and other parameters. The above data covers various signals related to high-frequency or major mechanical or electrical failure characteristics related to blades, blade roots and tower connection bolts, pitch control systems, transmission chain main shafts and main bearings, gearboxes, couplings, generators, towers, foundations, converters, main control cabinets, wind measurement systems, yaw systems, etc. Specifically, the edge data acquisition module has the following functional units:

[0041] Multi-dimensional heterogeneous data acquisition unit: This unit is capable of collecting data of various types and sampling frequencies, including but not limited to data collected by the wind turbine's existing data acquisition systems, such as SCADA systems, master control systems, wind power prediction systems, and CMS systems. It also collects various types of data collected by additional sensors, such as sound or noise, oil particle count data, vibration data from core components such as blades, towers, pitch systems, and yaw systems, load data, clearance data, wind turbine wind radar data, visible light or infrared video data from inside and outside the wind turbine, displacement and distance data related to bolt loosening or breakage, tower and foundation inclination data, temperature, and other different types of data. It can also include unstructured data such as maintenance records, technical modification records, scheduled inspection records, and root cause analysis records for daily wind farm operations. The types of data collected can partially or completely cover each core component or subsystem of the wind turbine as needed.

[0042] Edge data condition acquisition unit: The system's edge data acquisition module has a built-in algorithm model that supports the identification of failure or fault characteristics based on fan operating parameters, environmental conditions, and set failure or fault characteristics. It can automatically collect data of a set duration and that meets set characteristics or conditions as needed.

[0043] Edge data quality management unit: The system can automatically check data quality-related features as needed, such as mean check, non-empty check, positive and negative ratio check, upper and lower limit check, root mean square range check, adjacent point same value check, and outlier point ratio check; at the same time, when performing conditional and timed collection, the collected raw data is set with an automatic screening function for set interference scenarios to prevent interference data from flowing into the system from the source.

[0044] Big data governance module

[0045] The Big Data Governance Module of this application is used to aggregate, categorize, and store various types of heterogeneous data. This includes data collected by the edge data acquisition module, data accessed from existing data systems such as SCADA systems, master control systems, wind power forecasting and wind resource assessment systems, various expert knowledge and field personnel experience, maintenance results, meteorological data, and the output results of various algorithmic model analyses. This module also provides data input for all other modules, which read data from it as needed.

[0046] This module automatically provides six-dimensional quality screening for data of different types and structures through a built-in data quality screening model. Specifically, these include completeness screening, including null value ratio and resampling ratio; uniqueness screening, including duplication ratio; timeliness screening, including timestamp vacancy ratio; validity screening, including data type anomalies, over-range, and state variable validity; accuracy screening, including sampling rate, noise data ratio, data precision, and zero-point drift; and consistency screening, including logical deviation ratio. Based on the actual conditions of each type of data, the system selectively configures data quality requirements for at least one dimension. Only data that meets these requirements is allowed to enter the early fault diagnosis module.

[0047] Early fault diagnosis module

[0048] The early fault diagnosis module of the present application is equipped with an algorithmic model that can fully cover the high-frequency and major failure modes of each core component corresponding to the wind turbine component tree displayed in the intelligent knowledge base, such as blades, blade roots and tower connecting bolts, pitch systems, transmission chain main shafts and main bearings, gearboxes, couplings, as well as generators, towers, foundations, converters, main control cabinets, wind measurement systems, and yaw systems. The corresponding high-frequency and major failure mode diagnostic models. Taking the gearbox as an example, it at least includes the diagnostic models corresponding to the failure modes of gearbox gear wear, broken teeth, shaft misalignment and imbalance, and wear and cracking of the inner and outer rings of the bearings, retainers, and rolling elements. The algorithmic model selectively uses the multivariate heterogeneous data collected by the edge data acquisition module for diagnosis according to the needs of failure diagnosis, so as to achieve lightweight storage and computing resources. The diagnostic results will be displayed to the user in the form of early warnings. Taking a gearbox as an example, the early fault diagnosis module automatically reads various data types related to significant and frequent gearbox failure modes, including temperature, vibration, sound, image, fan active power, and lubrication system oil pressure. This multi-dimensional data is used for diagnosis and further analysis and assessment in the root cause analysis module and the failure impact module. The diagnostic results are then presented to users as early warnings through the software system's application modules.

[0049] The algorithm model of this module is established using a strategy that integrates mechanism and data-driven fusion. It primarily captures the key failure characteristics of various failure modes through multiple dimensions and approaches under the guidance of failure mechanisms. It then extracts algorithmic features through a variety of data-driven strategies, including multiple regression analysis, neural networks, EMD decomposition, support vector machines, random forests, and YOLO. Taking the failure diagnosis of high-speed shaft bearings in gearboxes as an example, through analysis of failure mechanisms and failure phenomena, it is found that the failure of high-speed shaft bearings in gearboxes is often accompanied by abnormal diagnostic features, primarily manifested in: setting sideband modulation signals of the failure characteristic frequency and its multiples, and the fact that the failure process of high-speed shaft bearings is accompanied by an increase in noise levels and an abnormal increase in nearby temperature values relative to the ambient temperature. By integrating these features, multiple regression analysis and indicator analysis can be used to simultaneously track the abnormal characteristics of vibration, sound, and temperature relative to operating data, thereby enabling the diagnosis and early warning of bearing failures.

[0050] Unlike traditional condition monitoring, the early fault diagnosis module of this application focuses on specific fault diagnosis and independently models various failure modes of each component, that is, one failure mode corresponds to one diagnostic model. Taking the failure of the high-speed shaft of the gearbox as an example, the module will establish independent failure diagnosis models such as gearbox inner ring wear, outer ring wear, rolling three-dimensional wear, and cage cracking to accurately diagnose the failure type. Before reading in data for diagnosis, the module will also specifically consider some set data quality issues, such as frequency multiplication interference related to sound signals, environmental noise interference, etc. The data read in from the big data governance module will be cleaned by calling the data quality algorithm model built into the early fault diagnosis module.

[0051] Performance evaluation and root cause analysis module

[0052] The Performance Evaluation and Root Cause Analysis module evaluates the impact of failure modes corresponding to the fault diagnosis module's diagnostic results on wind turbine performance, such as their impact on power generation performance and component structural load-bearing capacity. This module constructs a power curve degradation model and a life decay model, outputting a degraded power curve and a reduced life decay curve. Figure 2 and Figure 3 A comparison of the power curve and life decay curve of the gearbox high-speed shaft bearing after wear and before damage is presented in one embodiment. This shows that the wind turbine suffers from power curve degradation, but there is no significant power curve degradation before and after the failure mode itself. However, the life decay curve shows a significant impact before and after the damage. Based on multiple root cause analysis sub-models, the primary root cause of the damage was ultimately identified as rotor imbalance, which caused increased drive train vibration levels and accelerated fatigue aging of the high-speed shaft bearing. Other possible causes, such as high-speed shaft imbalance and improper lubrication, were ruled out.

[0053] This module also analyzes the root causes of the diagnosed failure modes. For each failure mode, expert experience is accumulated and stored in the intelligent knowledge base through a pre-built DFMEA library. When a failure mode is diagnosed, this module automatically calls the root cause path tree in the intelligent knowledge base and initiates the root cause analysis model corresponding to each path branch. This root cause analysis model can be a standalone model or directly linked to the relevant failure mode diagnosis results of other components. Based on the root cause analysis results, the module identifies the root cause that triggered the failure mode and displays the root cause prompt through the application module.

[0054] This application provides a root cause analysis function, the results of which can be used to fundamentally prevent the occurrence of subsequent similar failures, solving the problem that only diagnosis and early warning can often only deal with the failure symptoms but cannot fundamentally solve the problem.

[0055] Fault impact quantitative assessment module

[0056] The Fault Impact Quantification Assessment module reads wind turbine status parameters from SCADA, such as power curtailment and shutdown, short-term and long-term wind power forecast data, meteorological data displaying weather and temperature information, and intelligent knowledge base data to support a quantitative assessment of the failure impact of the diagnosed fault. Using this data as input, the module uses the degraded power curve and reduced life decay curve output by the Performance Evaluation and Root Cause Analysis module, combined with wind power forecast data and meteorological data, to quantitatively assess single-day power generation losses and the time window for the failure to progress to the next severity level. For gearbox high-speed shaft bearing damage, only the remaining time window for the failure to progress to the next severity level is evaluated. The assessment results show that the window for the damage to progress to the next severity level is, for example, 58 days. Therefore, the optimal treatment for the current damage is within 58 days. The diagnosed root cause typically also corresponds to a failure mode, and the same process can be used for safety and economic evaluation. The recommended latest execution node value can be determined based on the smaller of the current damage treatment window and the root cause treatment window, whichever is greater, and is then pushed into the decision process.

[0057] This module also uses the intelligent knowledge base to access the corresponding personnel and wind turbine safety impact knowledge base for the failure mode to conduct a safety risk assessment. For the root cause identified, the module automatically triggers an impact assessment on the wind farm unit, preventing further expansion of the problem at the source. It also supports manual verification of the potential impact on other wind farm turbines of the same model.

[0058] The fault impact quantitative assessment module also includes a decision-making unit, which uses information about power generation loss, operational window time, safety risks, etc. as input for operation and maintenance plan decisions, and calls corresponding decision-making information, decision-making processes, and maintenance plans for diagnosed failures and root causes from the knowledge base. According to the provisions of the decision-making process, different decision-making approval nodes will be selected based on different severity and risk levels. For failure modes and root causes with higher severity and risk levels, higher-level technical and business approver nodes will be selected; for failure modes and root causes with lower severity and risk levels, fewer and lower-level technical or business approver nodes will be selected; after the approval decision is completed, the module will automatically push the maintenance plan to the intelligent scheduling module. Regarding the approver nodes of the decision-making process, the module supports users with business management roles to select them themselves, while users with business execution roles only have viewing and execution permissions. Taking the gearbox high-speed shaft bearing damage as an example, the latest recommended time window for the failure mode and root cause treatment is no later than 58 days from the current day. This time window, together with the treatment plan for the gearbox high-speed shaft bearing damage and aerodynamic imbalance problem, is automatically submitted to the decision-making process. The decision-making process path is as follows: Figure 4As shown in the figure, the final process decision maker chose to execute the gearbox high-speed shaft bearing damage treatment task within 7 days and the aerodynamic imbalance problem treatment task within 40 days. The relevant plans for these two tasks were automatically pushed to the intelligent scheduling module.

[0059] Intelligent Scheduling Module

[0060] The intelligent scheduling module of this application will dynamically read the maintenance tasks submitted based on fault diagnosis and evaluation. It will also automatically read the operation inspection and fault handling tasks pushed by SCADA or third-party monitoring platforms, as well as manually imported annual inspection, maintenance, technical improvement and other tasks formulated offline according to the needs of wind farm management. Various tasks are used as task targets for dynamic intelligent scheduling to build a task pool. Each task will at least include task execution process, task operation manual, safety specifications, vehicles, materials and tools, personnel requirements, estimated working hours and other information. Taking the damage of the high-speed shaft bearing of the gearbox as an example, the processing task of the damage generated by the system of this application can be: 2 on-site operation and maintenance personnel with gearbox maintenance skills, carrying lubricating oil filters and lubricating oil, driving a pickup truck, go to wind turbine No. X to add lubricating oil and replace the filter element, and at the same time extract lubricating oil samples for subsequent testing, with an estimated working time of 4 hours; simultaneously start coordinating third-party aerodynamic imbalance detection and calibration resources to ensure entry within 40 days.

[0061] The module will further read spare parts information, meteorological information, wind function forecast information, vehicle and personnel information, wind farm maps, road conditions or routes, personnel information, wind turbine status information, etc. as input for the scheduling model operation. The module sets up an algorithm model for intelligent scheduling. The model will automatically seek multiple objectives based on the above inputs, such as minimizing power loss due to downtime, minimizing energy consumption, minimizing total working hours, minimizing safety risks, etc.; and automatically give the intelligent scheduling output results based on chronological order, including materials, tools, vehicles, personnel arrangements that match the tasks, operation and maintenance manuals, routes, task sequences, etc., which can be directly used as work order information, and displayed to users through the application. The scheduling optimization results are divided into long, medium and short terms, and can be adjusted and displayed according to user needs. The scheduling of this module also allows for special needs or additional input by personnel, and dynamically updates the scheduling route or operation sequence in real time based on manual input.

[0062] This module also features a closed-loop function, dynamically monitoring the location and status of personnel and vehicles. It supports the collection and storage of diverse and heterogeneous information, including photos, videos, measurement data, and text descriptions of the inspection, maintenance, troubleshooting, and technical upgrade processes. It also automatically compiles work time information to provide closed-loop feedback for the scheduling model. The final identified failure modes and causes are then fed back into the intelligent diagnosis module and the performance evaluation and root cause analysis module. This feedback loop incorporates operational aspects that better reflect actual field conditions into the intelligent knowledge base, optimizing existing solutions and processes. For the gearbox high-speed shaft bearing damage example mentioned above, maintenance personnel uploaded photos of the gearbox lubricant taken on-site and the oil sample test results report to the system. The results showed that the lubricant contained no metal impurities exceeding the standard, necessitating no replacement at this time. This closed-loop confirmation confirms that the bearing damage is currently in its early stages. Meanwhile, a third-party on-site personnel conducted a pneumatic imbalance test, which revealed an imbalance angle of 3.2°. This result was uploaded to the system, verifying the accuracy of the root cause analysis.

[0063] Smart spare parts module

[0064] This module dynamically reads spare parts information of a wind farm or region, such as quantity, validity period, storage conditions and cost, calibration requirements and historical information, unit price, matching models, etc. It also dynamically reads long-, medium- and short-term demand information on tools, spare parts and spare products from the task pool of the intelligent scheduling module as input for spare parts management optimization. It also reads the failure frequency and proportion of each component failure mode / fault from the intelligent knowledge base.

[0065] This module sets up an algorithm model for dynamic optimization management of intelligent spare parts. Based on the above inputs, the model will meet the wind farm's operation and maintenance and technical transformation needs with the lowest inventory cost and highest response efficiency, optimize the intelligent spare parts management tasks for the goal, and output specific management measures, including procurement, multi-library linkage allocation time and objects, calibration or maintenance, etc. Taking the gearbox high-speed shaft bearing damage as an example, after the scheduling requirement is put forward, the module model automatically detects that the inventory of the gearbox lubrication coarse filter deviates too much from the recently updated gearbox-related failure ratio. Therefore, before the gearbox bearing damage task is executed, it automatically triggers the purchase of 10 additional coarse filters and triggers the hydraulic pump tool calibration process. The algorithm model for spare parts optimization management includes spare parts classification model, demand forecasting model, inventory management model, inventory optimization model, etc.

[0066] Intelligent Knowledge Base

[0067] This module primarily serves two functions: providing professional information query and input for other modules, and dynamically accumulating expert knowledge and implementation experience. This module includes a wind turbine component tree, corresponding fault types for each component, the latest fault frequency, the corresponding safety impact of the fault, the root cause path tree, fault handling solutions and processes, and the resources required for troubleshooting. The wind turbine component tree comprehensively covers every system and component of the wind turbine.

[0068] This module has built-in expert knowledge and maintenance information of various types: the functions of the core components of the wind turbine, failure types, failure mechanisms, safety risks of failure modes, DEFMEA libraries for each failure mode, solutions to failure modes or root causes of failure, wind turbine models and parameters, and a decision-making process for operation and maintenance scheduling that comprehensively considers power generation losses, operable window times, and safety risks. Among them, the DEFMEA library for each failure mode is associated with models related to the root cause analysis process. This process has a built-in decision algorithm model and supports manual adjustment, ultimately outputting approval nodes for operation and maintenance plans for various failure modes. Taking the above-mentioned gearbox high-speed shaft bearing damage as an example, the model mainly refers to the operable window time of 87 days, and gives the latest recommended start time for operation and maintenance as the 58th day based on the decision coefficient of 2 / 3; and submits it together with the maintenance plan to the manual approval process for approval.

[0069] This module provides a closed-loop and sedimentation mechanism, offering feedback loops and updates to the various types of information already included in the intelligent knowledge base. For failure types not already included in the intelligent knowledge base, an interface is provided to support the addition of new functionality. For the aforementioned gearbox high-speed shaft bearing damage example, this module incorporates actions to identify and locate oil leaks in the gearbox lubrication system during early damage inspections.

[0070] For closed-loop feedback and newly added items, the module verifies and approves them through a combination of open scoring and expert review. For example, the newly added item regarding gearbox oil leak inspection was submitted to a built-in scoring process and open to all maintenance personnel at the ten wind farms in the user's area. Upon receiving a 95% approval or favorable rating, the result, along with the newly added item, was submitted to the expert review process, ultimately passed, and officially updated into the intelligent knowledge base.

[0071] This module also provides dynamic statistics of failure or malfunction modes of each component according to failure frequency and failure impact, and displays them through application software to help users dynamically identify new high-frequency or major failure modes that require special attention and closure, and serve as input for intelligent spare parts demand forecasting models.

[0072] Algorithm model sustainable evaluation module

[0073] This module implements automated closed-loop iteration throughout the entire lifecycle of algorithm model performance and completeness, primarily consisting of a feature library and a model library. The feature library primarily stores features related to early fault diagnosis models and automatically evaluates and updates their performance. The model library encompasses all algorithm models relevant to the entire system: early failure diagnosis models, data quality models, remaining life prediction models, performance degradation assessment models, safety risk assessment models, and root cause analysis models. The model library meets the algorithm model call requirements of each module, and the models in the library automatically undergo model performance evaluation.

[0074] Evaluation dimensions include completeness of failure mode coverage and algorithm model cost-effectiveness. Algorithm models are evaluated based on two economic indicators: total power loss and labor costs. Model optimization and iteration are required if they do not meet the evaluation criteria. Models in the model library are not isolated; they interact with each other for reasons such as data quality control and root cause analysis. The completeness of failure mode coverage dimension is based on the failure types dynamically counted in the knowledge base. If a significant, high-frequency failure mode is not covered by the models in the algorithm model library, algorithm personnel are automatically notified to add and develop new early failure diagnosis models and performance evaluation models corresponding to that failure mode. The cost-effectiveness dimension evaluates algorithm models based on the energy and labor costs saved by accurate algorithm diagnosis, minus the energy and labor costs wasted due to false positives, and minus the additional downtime energy and labor costs caused by false negatives. Model optimization and iteration are required if they do not meet the evaluation criteria. Models in the model library are not isolated; they interact with each other for reasons such as data quality control and root cause analysis.

[0075] This application implements data quality management capabilities throughout the entire process by configuring data quality strategies for edge data collection, big data governance, and algorithms, providing the necessary premise and foundation for accurate early warnings in early fault diagnosis modules to avoid false alarms or missed reports. The root cause analysis, failure impact assessment, and time window evaluation functions provided by this application connect the last intelligent link in the implementation of operations and maintenance, from diagnostic warnings to intelligent operations and maintenance and intelligent spare parts management. This solves the problem of manual decision-making over whether and when diagnostic warning results enter the operations and maintenance execution process, and enables automatic decision-making.

[0076] Although the embodiments disclosed in this application are as described above, the contents described are merely embodiments adopted to facilitate understanding of this application and are not intended to limit this application. Any person skilled in the art of the art to which this application belongs may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application. However, the scope of patent protection of this application shall still be based on the scope defined by the attached claims.

Claims

1. A fault diagnosis and health management system for a fan, characterized in that: Includes the following modules: Edge data acquisition module, which collects data from wind turbine components and checks data quality to remove interference data; The big data governance module is used to aggregate, classify and store various types of heterogeneous data. It uses a built-in data quality screening model to perform quality screening on data of different types and structures. An early fault diagnosis module, which is equipped with an algorithmic model of the failure mode of the wind turbine components. The algorithmic model selectively uses the multivariate heterogeneous data in the edge data acquisition module for diagnosis according to the needs of failure diagnosis, and issues early warning prompts based on the diagnosis results; The performance evaluation and root cause analysis module evaluates the impact of the failure mode corresponding to the diagnosis results of the early fault diagnosis module on the performance of the wind turbine and analyzes the root causes that may cause the failure mode; Fault impact quantitative assessment module, which performs quantitative assessment and safety risk assessment on the failure impact of the diagnosed fault and retrieves the corresponding decision-making plan and maintenance plan; The intelligent scheduling module dynamically reads the maintenance tasks submitted based on fault diagnosis and evaluation, as well as the input information used as the scheduling model, and provides the output results of the intelligent scheduling; The intelligent spare parts module automatically reads spare parts information and dynamically reads demand information from the task pool of the intelligent scheduling module, optimizes intelligent spare parts management tasks, and outputs management measures; Intelligent knowledge base, which provides query and input of professional information and dynamically accumulates expert knowledge and implementation process experience; Algorithm model sustainability evaluation module, which provides automatic closed-loop and iteration of algorithm model performance and completeness throughout its life cycle; The performance evaluation and root cause analysis module will construct a power curve degradation model and a component life attenuation model, and output a degraded power curve and a reduced life attenuation curve; The pre-built DFMEA library is used to accumulate and store expert experience corresponding to each failure mode in an intelligent knowledge base. When a failure mode is diagnosed, the root cause path tree in the intelligent knowledge base is automatically called to start the root cause analysis model corresponding to each path branch. Based on the root cause analysis results, this module identifies the root cause that caused the failure mode and displays the root cause prompt through the application module. The fault impact quantitative assessment module uses the degraded power curve and the reduced life decay curve output by the performance evaluation and root cause analysis module according to the input data to perform a quantitative assessment of the single-day power generation loss and a quantitative assessment of the time window for the failure to worsen to the next stage; the fault impact quantitative assessment module calls the personnel and wind turbine safety impact knowledge base corresponding to the failure mode from the intelligent knowledge base to perform a safety risk assessment.

2. The fault diagnosis and health management system according to claim 1, characterized in that: The algorithm model sustainability evaluation module includes a feature library and a model library. The feature library stores and automatically evaluates and updates features related to the early fault diagnosis model; the model library contains all system-related algorithm models and meets the algorithm model calling requirements of each module. The models in the model library can perform model performance evaluation.

3. The fault diagnosis and health management system according to claim 2, characterized in that: The edge data acquisition module includes: Multivariate heterogeneous data acquisition unit, which collects data of different categories and sampling frequencies; The edge data condition acquisition unit automatically collects data that meets the set characteristics or conditions for a set period of time based on the algorithm model of wind turbine operating parameters, environmental conditions, and failure feature recognition function; The edge data quality management unit automatically checks the data quality and automatically screens the interference data from the collected raw data during conditional collection.

4. The fault diagnosis and health management system according to claim 3, characterized in that: The big data governance module performs quality screening on data of different types and structures, specifically including: Completeness screening, including screening of null value ratio and resampling ratio; Uniqueness screening, including screening of duplicate proportions; Timeliness screening, including screening of time-stamped vacancy ratios; Validity screening, including data type anomalies, over-range, and status variable validity screening; Accuracy screening, including sampling rate, noise data ratio, data accuracy, and zero drift screening; Consistency screening, including screening for logical deviation ratios.

5. The fault diagnosis and health management system according to claim 4, characterized in that: The early fault diagnosis module independently models various failure modes of each component, and the data read from the big data governance module will be cleaned by calling the data quality algorithm model built into the early fault diagnosis module.

6. The fault diagnosis and health management system according to claim 1, characterized in that: The fault impact quantitative assessment module also includes a decision-making unit, which uses information including power generation loss, operational window time, and safety risks as input for operation and maintenance plan decisions, and calls corresponding decision information, decision-making processes, and maintenance plans from the intelligent knowledge base.

7. The fault diagnosis and health management system according to any one of claims 3 to 6, characterized in that: The intelligent scheduling module will dynamically monitor the location and status information of personnel and vehicles, support the collection and storage of multi-dimensional heterogeneous information, and automatically count working hours to provide closed-loop feedback for the scheduling model; the final confirmed failure mode and failure cause will be fed back to the early fault diagnosis module and the performance evaluation and root cause analysis module, and the operation link information that is more in line with the actual situation on site during the execution process will be fed back to the intelligent knowledge base to optimize existing plans and processes; at the same time, it will automatically provide troubleshooting prompts for wind turbines or wind farms with similar root cause phenomena.

8. The fault diagnosis and health management system according to any one of claims 2 to 6, characterized in that: The intelligent spare parts module uses the spare parts information and the demand information as input for spare parts management optimization, reads the failure frequency and proportion of each component failure mode from the intelligent knowledge base, uses the algorithm model for dynamic optimization management of intelligent spare parts, optimizes the intelligent spare parts management task for the target, and outputs specific management measures.

9. The fault diagnosis and health management system according to claim 8, characterized in that: The evaluation dimensions of the algorithm model sustainability evaluation module include the failure mode coverage completeness dimension and the algorithm model economy dimension; wherein, The failure mode coverage completeness dimension is based on the failure types dynamically counted in the knowledge base. For failure modes that are not covered by the models in the algorithm model library, it automatically prompts the development of new early failure diagnosis models and performance evaluation models corresponding to the failure modes. The algorithm model is evaluated based on the two economic indicators of total power loss and labor cost, which are the power loss and labor cost saved by the economic dimension diagnosis of the algorithm, minus the power loss and labor cost waste caused by the algorithm's false alarms, and minus the additional downtime power loss and labor cost caused by the algorithm's missed alarms.

Images