Sensor management and fault diagnosis system of wind driven generator

Abstract

The invention relates to a sensor management and fault diagnosis system of a wind driven generator, and aims to solve the problems of low efficiency, insufficient detection precision of a single sensor and the like in the operation and maintenance management of the existing wind driven generator. According to the system, high-precision, low-cost and automatic operation and maintenance management of the wind turbine generator is realized by combining a plurality of detection means such as the sound sensor, the three-axis acceleration sensor and the laser sensor. The system comprises a sensor module, a data acquisition module, a calculation and instruction module, a communication module, an intelligent operation and maintenance system cloud diagnosis platform, an inspection PDA and the like. Through the multi-sensor fusion and intelligent diagnosis technology, fault features can be more accurately captured, the misjudgment rate is reduced, the operation and maintenance efficiency is improved, the operation and maintenance cost is reduced, and wide application prospects are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to an intelligent abnormality detection system for wind turbines based on multi-sensor fusion and sound recognition, which is used to achieve high-precision, low-cost, and automated operation and maintenance management of wind turbines. Background Art

[0002] As a clean energy, wind power generation has developed rapidly around the world and in my country. However, wind turbines are easily affected by the natural environment and malfunction, and the remote location of wind farms makes monitoring and management difficult. Traditional operation and maintenance methods are inefficient and costly, making it difficult to achieve real-time monitoring and accurate fault prediction, and unable to detect and eliminate mechanical failures in a timely manner. Existing single sensor detection methods also have the disadvantages of single data, insufficient detection accuracy and reliability, and cannot meet the needs of wind turbine fault detection. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent abnormality detection system for wind turbines based on multi-sensor fusion and sound recognition, which realizes high-precision, low-cost, and automated operation and maintenance management of wind turbines by combining multiple detection methods such as sound sensors, three-axis acceleration sensors, and laser sensors.

[0004] The technical solution of the present invention includes: a sensor module, including a sound sensor and a traditional sensor (a three-axis acceleration sensor and a laser sensor); a data acquisition module, used to collect sensor data and perform preliminary fusion and preprocessing; a calculation and instruction module, used to process the collected data, run relevant algorithms for fault diagnosis and analysis, and issue instructions; a communication module, used to achieve communication connection between various parts of the system and ensure data transmission; an intelligent operation and maintenance system cloud diagnosis platform, used to receive processed data, perform multi-source heterogeneous data storage, and use integrated artificial intelligence algorithms to perform in-depth diagnosis of the fan status, provide early warnings and maintenance suggestions; and an inspection PDA, used for operation and maintenance personnel to perform inspection operations on site and obtain relevant data and information.

[0005] The beneficial effects of the present invention include: improving fault diagnosis accuracy: through multi-sensor fusion and intelligent diagnosis technology, it is possible to capture fault characteristics more accurately and reduce the misjudgment rate; improving operation and maintenance efficiency: real-time monitoring and automated diagnosis functions enable operation and maintenance personnel to obtain wind turbine status information in a timely manner, warn of faults in advance, avoid sudden shutdowns, reduce downtime, and improve power generation efficiency; reducing operation and maintenance costs: accurately locate fault points, avoid unnecessary repairs and replacement of parts, reduce maintenance costs, extend the service life of wind turbines, and improve the economic benefits of wind power projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] FIG1 is a system architecture diagram of the present invention.

[0007] FIG2 is a flow chart of the algorithm of the present invention.

[0008] FIG. 3 is a diagram of the data storage and diagnosis architecture of the present invention. DETAILED DESCRIPTION

[0009] System composition

[0010] Hardware

[0011] Sound sensor (sound collection module): used to collect sound signals from the fan operation and detect abnormal mechanical sounds. Sound sensors are distributed in key parts of the fan, such as blades, hubs, generators, etc., to monitor the operation status of the fan in all directions.

[0012] Traditional sensors (physical sensing modules): including triaxial accelerometers and laser sensors. The triaxial accelerometer can be installed vertically or horizontally at the vibration point to be measured to monitor the blade vibration signal, with a frequency response of 0.2~5000Hz and an acceleration range of 0-50g; the laser sensor internal gear ring is equipped with a synchronous encoder to calculate the pitch angle, and 2-3 sets of laser distance measuring sensors are installed on the wheel side to collect the bolt end face distance and elongation, with a measurement range of 35±15mm and a repeatability of 10um.

[0013] Data acquisition module: The high-speed collector is responsible for collecting sensor data. It has Ethernet, USB and wireless Wi-Fi interfaces, and can collect data offline. Its CPU central processor uses an ARM core board, embedded with an embedded real-time operating system Linux, and realizes powerful data acquisition and fusion functions through a field programmable gate array FPGA.

[0014] Computing and instruction module: The industrial computer is used to process the collected data, run relevant algorithms for fault diagnosis and analysis, and issue instructions.

[0015] Communication module: The local area network realizes the communication connection between various parts of the system and ensures data transmission.

[0016] Inspection PDA: It is convenient for operation and maintenance personnel to conduct inspection operations on site and obtain relevant data and information.

[0017] Software The DCGAN adversarial network is used to train a generative adversarial network model on abnormal mechanical sound data to reduce noise and increase the number of training samples.

[0018] Deep learning is used to enhance the data of the signals captured by the sound sensor, and adversarial multi-task learning and inverse mapping techniques are used to transform the spectrum representation to enhance the signal.

[0019] Sinc convolution operation is used to filter the mechanical abnormal sound frequency domain signal and capture the frequency characteristics.

[0020] ResNet is added to build a convolutional neural network model for sound feature extraction and deep residual learning for classification and recognition, which solves the gradient vanishing problem and improves noise resistance and computational efficiency.

[0021] Workflow Data collection: Sensors collect real-time data on the equipment operation and production operation of wind turbines, including sound, vibration, displacement and other information.

[0022] Data preprocessing: The data acquisition module collects sensor data through multiple interfaces and uses its powerful processing capabilities to perform preliminary fusion and preprocessing.

[0023] Fault diagnosis: The industrial computer runs a deep learning algorithm based on the improved SincNet to analyze the sound data, and makes a comprehensive judgment based on the traditional sensor data. Through the sound recognition fault and sensor cross-abnormal situation verification, the fan fault is accurately predicted.

[0024] Cloud diagnosis: The cloud diagnosis platform of the intelligent operation and maintenance system receives processed data, stores multi-source heterogeneous data, and uses integrated artificial intelligence algorithms to conduct in-depth diagnosis of the wind turbine status, providing early warnings and maintenance recommendations.

[0025] Operation and maintenance management: Based on the early warning and diagnosis results, operation and maintenance personnel conduct on-site inspections and maintenance through inspection PDA and other equipment to achieve precise maintenance and improve operational efficiency.

Claims

1. A sensor management and fault diagnosis system for a wind turbine, characterized in that: include: The sensor module includes a sound sensor and a traditional sensor. The sound sensor is used to collect the sound signal of the fan operation. The traditional sensor includes a triaxial acceleration sensor and a laser sensor. The triaxial acceleration sensor is used to monitor the blade vibration signal. The laser sensor is used to collect the bolt end face distance and elongation. The data acquisition module is connected to the sensor module and is used to collect sensor data and perform preliminary fusion and preprocessing. The calculation and instruction module is connected to the data acquisition module and is used to process the collected data, run relevant algorithms for fault diagnosis and analysis, and issue instructions. A communication module is connected to the computing and command module to realize communication connection between various parts of the system and ensure data transmission; an intelligent operation and maintenance system cloud diagnosis platform is connected to the computing and command module to receive processed data, store multi-source heterogeneous data, and use an integrated artificial intelligence algorithm to perform in-depth diagnosis of the wind turbine status and provide early warning and maintenance suggestions; The inspection PDA is connected to the computing and command modules and is used by operation and maintenance personnel to perform inspection operations on site and obtain relevant data and information.

2. The system according to claim 1, characterized in that The sound sensors in the sensor module are distributed in key parts of the wind turbine, such as blades, hubs, generators, etc., and are used to monitor the operating status of the wind turbine in all directions.

3. The system according to claim 1, characterized in that The data acquisition module includes a high-speed collector with Ethernet, USB and wireless Wi-Fi interfaces, which can collect data offline. Its CPU central processor adopts an ARM core board, implanted with an embedded real-time operating system Linux, and realizes powerful data acquisition fusion functions through a field programmable gate array FPGA.

4. The system according to claim 1, characterized in that The computing and instruction module includes an industrial computer, which is used to run a deep learning algorithm based on the improved SincNet to analyze the sound data, make a comprehensive judgment based on traditional sensor data, and accurately predict fan failures.

5. The system according to claim 1, characterized in that The cloud diagnosis platform of the intelligent operation and maintenance system includes multi-source heterogeneous data cloud storage, analyzes and stores data collected from different components of the wind turbine, establishes a data center, selects databases based on different data characteristics, and performs normalization and redundant storage.

6. The system according to claim 1, characterized in that The intelligent operation and maintenance system cloud diagnosis platform also includes intelligent cloud diagnosis of wind turbines, which integrates artificial intelligence algorithms, encapsulates service interfaces, and combines data calculations to identify the health of the entire wind turbine, blade failures, etc., to ensure low algorithm coupling and low upgrade costs.

7. The system according to claim 1, characterized in that The communication module uses at least one of Ethernet, USB and wireless Wi-Fi interfaces to ensure stable and efficient data transmission between various parts of the system.

8. The system according to claim 1, characterized in that The inspection PDA has data receiving, displaying and storing functions, and can receive wind turbine status information and maintenance suggestions from the calculation and instruction module, and provide on-site viewing and operation by operation and maintenance personnel.

9. The system according to claim 1, characterized in that The triaxial acceleration sensor can be installed vertically or horizontally at the vibration point to be measured to monitor the blade vibration signal, with a frequency response of 0.2~5000Hz and an acceleration range of 0-50g; the laser sensor inner ring gear is installed with a synchronous encoder to calculate the pitch angle, and 2-3 groups of laser ranging sensors are installed on the wheel side to collect the bolt end face distance and elongation, with a measurement range of 35±15mm and a repeatability of 10um.

10. The system according to claim 1, characterized in that The deep learning algorithm based on the improved SincNet includes using the DCGAN adversarial network to train the mechanical abnormal sound data to generate an adversarial network model, reduce noise and increase the number of training samples; using deep learning to enhance the data of the signal captured by the sound sensor, using adversarial multi-task learning and inverse mapping technology to convert the spectrum to represent the enhanced signal; using Sinc convolution operation to screen the mechanical abnormal sound frequency domain signal and capture the frequency characteristics; ResNet is added to build a convolutional neural network model with deep residual learning for sound feature extraction for classification and recognition.

Images