CN-120867962-B - Multi-mode fusion health assessment method for hierarchical progressive wind power transmission chain

CN120867962BCN 120867962 BCN120867962 BCN 120867962BCN-120867962-B

Abstract

The invention discloses a multi-mode fusion health assessment method of a hierarchical progressive wind power transmission chain, which aims at the problem of insufficient accuracy of traditional single signal diagnosis, provides a hierarchical progressive diagnosis framework, and realizes upgrading from traditional single-dimensional analysis to multi-mode intelligent diagnosis by processing multi-dimensional signals. The system collects three calendar history vibration signals of the gear box, calculates a threshold value of an effective value of the vibration signal, judges whether a transmission chain is normal or not, can further extract fault characteristics through analysis of the vibration signal by a gram angle difference field, judges whether the transmission chain is in a warning state or not, and judges the final state through simultaneously evaluating a front-rear temperature difference-power curve of a main shaft, an oil temperature-power curve, a front-rear temperature difference-power curve of a generator and a cabin noise-power curve. According to the method, the accuracy and timeliness of the state monitoring of the wind turbine are remarkably improved through intelligent health assessment of the multi-level decision wind turbine transmission chain.

Inventors

WEN XIAOQIANG
WANG JIANGUO
DING MINGHAO
SHEN YANJIE
WU YINGJIE
CHENG SHUN
Yin Xinrui
YANG YANJUN
XIN HONGWEI
ZHU GUOQIANG

Assignees

东北电力大学

Dates

Publication Date: 20260508
Application Date: 20250722

Claims (6)

1. A multi-mode fusion health assessment method of a hierarchical progressive wind power transmission chain is characterized by comprising the following steps of: reading historical vibration data of the front end and the rear end of an output shaft of the gear box of the unit in a preset period under a normal working condition, so as to calculate an effective value and determine an effective value threshold of vibration; Acquiring real-time vibration data of the front end and the rear end of an output shaft of the gearbox, and further performing effective value calculation to obtain a real-time effective value; If the real-time effective value exceeds the effective value threshold, fault feature identification is carried out through a gram angle difference field; If the fault feature identification is judged to be normal, outputting an evaluation result as a warning, and if the fault feature identification is judged to be abnormal, further carrying out multi-parameter joint threshold comparison; if any parameter is abnormal, outputting the evaluation result as danger and triggering a danger alarm; The effective value The calculation formula of (2) is as follows: Wherein x [ N ] (n=1, 2,., N) is the discrete vibration signal, N is the number of samples; The determination mode of the effective value threshold value is as follows: Establishing a box graph according to the effective value obtained by calculation of the historical vibration data, and calculating Q1, Q3 and IQR, wherein Q1 represents 25% quantile, Q3 represents 75% quantile and IQR represents four quantile distances; An effective value threshold is determined from the box plot.
2. The hierarchical progressive wind power transmission chain multimodal fusion health assessment method of claim 1, wherein the effective value threshold is calculated as follows: Wherein, the Representing the maximum observed value of the effective value threshold.
3. The hierarchical progressive wind power transmission chain multimodal fusion health assessment method of claim 1, wherein the step of fault feature identification through a gram angle difference field comprises: preprocessing the historical vibration data, and converting the preprocessed one-dimensional time sequence data into a two-dimensional image feature matrix under a polar coordinate system by using a gram angle difference field; Training a deep learning model CNN by utilizing the two-dimensional image feature matrix; inputting the real-time vibration data into a trained deep learning model CNN after the same processing flow, and judging whether the result deviates from a normal threshold value or not through diamond decision nodes; if the normal threshold value is not deviated, the fault characteristic identification is judged to be normal; if the threshold value deviates from the normal threshold value, the fault feature recognition is judged to be abnormal.
4. The hierarchical progressive wind power transmission chain multi-modal fusion health assessment method according to claim 1, wherein the step of converting the preprocessed one-dimensional time series data into a two-dimensional image feature matrix in a polar coordinate system by using the gram angle difference field comprises: normalizing the unconverted one-dimensional time sequence data; taking inverse cosine of the value normalized by the time point to obtain a corresponding angle, setting a radius, normalizing the time stamp to a preset interval as the radius, and mapping the one-dimensional time sequence to a polar coordinate system, and reserving a time sequence and a numerical relation; and constructing a two-dimensional image feature matrix by taking sine values of the angle differences of the time points as elements.
5. The hierarchical progressive wind power transmission chain multi-modal fusion health assessment method according to claim 1, wherein the calculation formula of the sine value of the angle difference of two time points i and j is: 。
6. the method for multi-modal fusion health assessment of a hierarchical progressive wind power transmission chain according to claim 1, wherein the step of multi-parameter joint threshold comparison comprises: acquiring original operation data of a unit; preprocessing the original operation data, and drawing to obtain a power-oil temperature curve, a power-shaft temperature difference curve and a power-generator front-rear temperature difference curve; According to a predetermined static threshold value, carrying out parameter verification on key nodes of the three curves; If all the parameters do not exceed the corresponding threshold range, the evaluation result is output as an alarm, and if any parameter is abnormal, the evaluation result is output as a danger and a danger alarm is triggered.

Description

Multi-mode fusion health assessment method for hierarchical progressive wind power transmission chain Technical Field The invention belongs to the technical field of wind turbine generator system state monitoring and fault diagnosis, and focuses on realizing intelligent evaluation of equipment health state through multi-source data fusion and hierarchical decision. Through the three-level logic of vibration threshold primary screening, image characteristic re-judgment and multi-parameter cross verification, the grading early warning from normal to warning to dangerous is realized, the single signal misjudgment rate is effectively reduced, and the fault positioning accuracy is improved. Background As a complex electromechanical system, early fault diagnosis and health state assessment of core components (such as a gearbox and a generator) of the wind turbine generator are key for guaranteeing reliable operation of the wind turbine generator. The traditional state monitoring technology mostly depends on single vibration signal threshold value alarm or manual inspection, and has the limitations of high diagnosis hysteresis, high false alarm rate and the like. Firstly, single signal sensitivity is insufficient, a static threshold value of a vibration effective value (RMS) is easy to be interfered by working condition fluctuation, normal load change and early fault symptoms are difficult to distinguish, secondly, time sequence feature mining is shallow, a conventional time-frequency analysis method (such as FFT (fast Fourier transform) and wavelet transform) is limited in feature extraction capability of a non-stationary vibration signal, a time sequence mode of weak faults is difficult to capture, and finally, multi-parameter collaborative verification is missing, namely, dynamic relevance of auxiliary parameters such as temperature, noise and the like and a power curve is not quantized by a system, so that a fault root is fuzzy due to positioning. In recent years, intelligent diagnosis technology based on multi-source data fusion and deep learning image characterization is gradually rising. For example, the gram angle field (GADF) can strengthen the texture expression of time sequence characteristics by encoding vibration signals into two-dimensional images, so that the classification precision of a CNN model is improved, and the multi-parameter curve collaborative analysis (such as temperature difference-power and noise-power) can verify the consistency of faults through the coupling relation between thermodynamics and mechanical states, so that the risk of misjudgment is reduced. However, the existing method still has an optimization space in terms of dynamic threshold calibration, hierarchical decision logic and complex working condition adaptability. Disclosure of Invention The hierarchical evaluation flow provided by the invention integrates the dynamic threshold driven by historical data, GADF image features and multi-parameter curve cross verification, and aims to construct a more robust wind turbine generator state evaluation system and provide technical support for intelligent operation and maintenance. The invention provides a multi-mode fusion health assessment method of a hierarchical progressive wind power transmission chain, which comprises the following steps: reading historical vibration data of the front end and the rear end of an output shaft of the gear box of the unit in a preset period under a normal working condition, so as to calculate an effective value and determine an effective value threshold of vibration; Acquiring real-time vibration data of the front end and the rear end of an output shaft of the gearbox, and further performing effective value calculation to obtain a real-time effective value; If the real-time effective value does not exceed the effective value threshold, outputting an evaluation result as normal; if the real-time effective value exceeds the effective value threshold, performing fault feature identification through a gram angle difference field (Gramian Angular DIFFERENCE FIELD, GADF); If the fault feature identification is judged to be normal, outputting an evaluation result as a warning, and if the fault feature identification is judged to be abnormal, further carrying out multi-parameter joint threshold comparison; If all the parameters do not exceed the corresponding threshold range, the evaluation result is output as an alarm, and if any parameter is abnormal, the evaluation result is output as a danger and a danger alarm is triggered. Preferably, the calculation formula of the effective value X RMS is as follows: where x [ N ] (n=1, 2,., N) is the discrete vibration signal and N is the number of samples. Preferably, the effective value threshold is determined in the following manner: establishing a box graph according to the effective value obtained by calculation of the historical vibration data, and calculating Q1 (25% quantile), Q3 (75% quantile) and IQR