A hierarchical progressive wind power transmission chain multi-modal fusion health assessment method

Abstract

The application discloses a kind of hierarchical progressive wind power transmission chain multi-modal fusion health assessment methods, for the problem of insufficient precision of traditional single signal diagnosis, propose a kind of hierarchical progressive diagnostic framework, and by processing multidimensional signal realizes from traditional single-dimensional analysis upgrade to multi-modal intelligent diagnosis.Through system acquisition gear box three months history vibration signal, the threshold of the effective value of vibration signal is calculated, to judge whether the transmission chain is normal;Through the vibration signal analysis of gram angle difference field, the fault feature can be further extracted, to judge whether the transmission chain is in warning state;Through simultaneously evaluating the front and rear temperature difference-power curve of main shaft, oil temperature-power curve, generator front and rear temperature difference-power curve and cabin noise-power curve, the final state is judged.This method improves the accuracy and timeliness of wind turbine condition monitoring by multi-level decision wind turbine transmission chain intelligent health assessment.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine condition monitoring and fault diagnosis technology, focusing on achieving intelligent assessment of equipment health status through multi-source data fusion and hierarchical decision-making. Through a three-level logic of initial screening with vibration thresholds, re-judgment of image features, and cross-validation of multiple parameters, it achieves hierarchical early warning from "normal" to "warning" and then to "danger," effectively reducing the false judgment rate of single signals and improving the accuracy of fault location. Background Technology

[0002] As complex electromechanical systems, wind turbines rely heavily on early fault diagnosis and health status assessment of their core components (such as gearboxes and generators) to ensure reliable operation. Traditional condition monitoring technologies often depend on single vibration signal threshold alarms or manual inspections, which have limitations such as high diagnostic lag and high false alarm rates.

[0003] First, there is insufficient sensitivity to single signals: the static threshold of the effective vibration value (RMS) is easily affected by fluctuations in operating conditions, making it difficult to distinguish between normal load changes and early fault signs. Second, the mining of time-series features is superficial: conventional time-frequency analysis methods (such as FFT and wavelet transform) have limited ability to extract features from non-stationary vibration signals, making it difficult to capture the time-series patterns of weak faults. Finally, there is a lack of multi-parameter collaborative verification: the dynamic correlation between auxiliary parameters such as temperature and noise and the power curve is not quantified by the system, resulting in ambiguity in the location of the root cause of the fault.

[0004] In recent years, intelligent diagnostic technologies based on multi-source data fusion and deep learning image representation have gradually emerged. For example, Gram angle field (GADF) can enhance the texture representation of temporal features and improve the classification accuracy of CNN models by encoding vibration signals into two-dimensional images; while multi-parameter curve collaborative analysis (such as temperature difference-power and noise-power) verifies fault consistency through the coupling relationship between thermodynamics and mechanical state, reducing the risk of misjudgment. However, existing methods still have room for improvement in dynamic threshold calibration, hierarchical decision logic, and adaptability to complex working conditions. Summary of the Invention

[0005] The hierarchical evaluation process proposed in this invention integrates historical data-driven dynamic thresholds, GADF image features, and multi-parameter curve cross-validation, aiming to build a more robust wind turbine condition assessment system and provide technical support for intelligent operation and maintenance.

[0006] This invention provides a hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains, comprising:

[0007] Read the historical vibration data of the front and rear ends of the gearbox output shaft within a preset time period under normal operating conditions of the unit, in order to calculate the effective value and determine the effective value threshold of the vibration;

[0008] Real-time vibration data of the front and rear ends of the gearbox output shaft are acquired to further calculate the real-time effective value.

[0009] If the real-time valid value does not exceed the valid value threshold, the output evaluation result is normal; if the real-time valid value exceeds the valid value threshold, fault characteristics are identified through the Gramian Angular Difference Field (GADF).

[0010] If the fault feature identification is judged to be normal, the evaluation result is a warning; if the fault feature identification is judged to be abnormal, a multi-parameter joint threshold comparison is further performed.

[0011] If all parameters are within the corresponding threshold range, the output evaluation result is an alarm; if any parameter is abnormal, the output evaluation result is danger and a danger alarm is triggered.

[0012] Preferably, the effective value X RMS The calculation formula is:

[0013]

[0014] Where x[n] (n=1,2,...,N) is a discrete vibration signal, and N is the number of samplings.

[0015] Preferably, the effective value threshold is determined as follows:

[0016] A box plot was constructed based on the effective values ​​obtained from historical vibration data, and Q1 (25th percentile), Q3 (75th percentile), and IQR (interquartile range) were calculated.

[0017] The effective value threshold is determined based on the box plot.

[0018] Preferably, the formula for calculating the effective value threshold is as follows:

[0019] Q1 (25% quantile)

[0020]

[0021] Q3 (75% quantile)

[0022]

[0023] IQR (interquartile range)

[0024] IQR = Q3 - Q1

[0025] Valid value threshold (maximum number of observations)

[0026] Threshold = Q3 + 1.5 × IQR.

[0027] Preferably, the step of identifying fault features using the Gramian Angular Difference Field (GADF) includes:

[0028] The historical vibration data is preprocessed, and the preprocessed one-dimensional time series data is converted into a two-dimensional image feature matrix in polar coordinates using the Gram angle difference field (GADF).

[0029] The two-dimensional image feature matrix is ​​used to train a deep learning model CNN;

[0030] The real-time vibration data is processed through the same procedure and then input into the trained deep learning model CNN. The diamond decision node is used to determine whether the result deviates from the normal threshold.

[0031] If the fault characteristics do not deviate from the normal threshold, the fault feature identification is judged as normal.

[0032] If the fault deviates from the normal threshold, the fault feature is identified as abnormal.

[0033] Preferably, the step of converting the preprocessed one-dimensional time-series data into a two-dimensional image feature matrix in polar coordinates using the Gram angular difference field (GADF) includes:

[0034] Normalize the unconverted one-dimensional time series data;

[0035] Take the inverse cosine of the normalized value xi′ of time point i to obtain the corresponding angle θi. Then, set the radius to normalize the timestamp i to a preset interval as the radius ri, thereby mapping the one-dimensional time series to the polar coordinate system and preserving the time order and numerical relationship.

[0036] A two-dimensional image feature matrix is ​​constructed using the sine values ​​of the angle differences at various time points as elements.

[0037] Preferably, the formula for calculating the sine value of the angle difference between the two time points i and j is:

[0038] GADF i,j =sin(θ) i -θ j ).

[0039] Preferably, the multi-parameter joint threshold comparison step includes:

[0040] Obtain the unit's raw operating data;

[0041] The raw operating data is preprocessed, and the power-oil temperature curve, power-shaft temperature difference curve, and power-generator front and rear temperature difference curve are plotted.

[0042] Based on a predetermined static threshold, the parameters of the key nodes of the three curves are verified.

[0043] If all parameters are within the corresponding threshold range, the output evaluation result is an alarm; if any parameter is abnormal, the output evaluation result is danger and a danger alarm is triggered.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. Based on a multi-level progressive verification architecture and a multi-source data fusion mechanism, this invention breaks through the limitations of a single vibration threshold criterion. It filters layer by layer through a three-level logic of “vibration effective value → Gram angle field image → multi-parameter power curve”, and triggers cross-verification of images and thermal / acoustic parameters only when vibration is abnormal, thus effectively reducing the false alarm rate.

[0046] 2. This invention introduces correlation analysis between parameters such as spindle temperature difference, oil temperature, and engine compartment noise and the power curve. By identifying synchronous anomalies in thermodynamic and acoustic states (such as sudden changes in the slope of the temperature difference-power curve), the root cause of the fault can be pinpointed, avoiding misjudgment based on a single signal.

[0047] 3. This invention achieves end-to-end diagnosis from "anomaly detection" to "fault attribution" through a progressive logic of "threshold initial screening → image re-judgment → multi-curve further analysis", which is significantly better than traditional single-dimensional threshold alarms or manual experience judgment. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains in an embodiment of the present invention.

[0049] Figure 2 This is a comprehensive judgment logic diagram of steps S1 and S2 in an embodiment of the present invention;

[0050] Figure 3 This is a flowchart illustrating fault feature identification using Gram angular difference field images in an embodiment of the present invention;

[0051] Figure 4 This is a comprehensive judgment logic diagram for step S3 in an embodiment of the present invention;

[0052] Figure 5 This is a comprehensive judgment logic diagram for step S4 in an embodiment of the present invention. Detailed Implementation

[0053] It should be noted that the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features described herein are detailed descriptions of the technical solution of the present invention, not limitations thereof. Where there is no conflict, the embodiments and technical features described herein can be combined with each other. The term "and / or" merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0054] To make the purpose, technical solution and advantages of this invention patent clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0055] Combination Figure 1 This embodiment provides a hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains, which includes:

[0056] Step S1: Read and analyze the historical vibration data of the gearbox output shaft under normal operating conditions for three months of the same unit to obtain the threshold value of the effective vibration value;

[0057] Step S2: Real-time acquisition of bearing vibration signals, calculation of effective values, and first-step judgment;

[0058] Step S3: Perform Gram angular field encoding on the real-time acquired bearing vibration signal to output an image of the Gram angular difference field for the second step of judgment;

[0059] Step S4: The judgment in the third step is made by fitting the corresponding power data to the spindle front and rear temperature difference-power curve, oil temperature-power curve, generator front and rear temperature difference-power curve, and engine room noise-power curve.

[0060] Step S5: Based on the three-level judgment, the final output is the result corresponding to different working conditions, namely normal, warning, alarm and danger.

[0061] This multimodal method, by fusing one-dimensional vibration signals, two-dimensional image data (feature maps generated by GADF), and two-dimensional curve analysis, can capture the health status characteristics of the wind turbine drivetrain from different dimensions. One-dimensional data directly reflects the temporal dynamic characteristics of mechanical vibration, while two-dimensional images enhance the characterization of nonlinear and periodic fault modes through color and texture. Curve analysis reveals the long-term trend and anomalous changes in signal evolution. This complementary multi-source information not only improves the comprehensiveness of feature extraction but also achieves joint mining of spatiotemporal features through deep learning models. This significantly enhances the sensitivity and diagnostic accuracy for early, subtle faults (such as gear pitting and bearing wear), while improving noise resistance and model generalization under complex operating conditions. This provides reliable technical support for reducing operation and maintenance costs and achieving predictive maintenance.

[0062] Step S3 describes the Gram Angular Difference Field (GADF), which transforms one-dimensional time-series vibration data into a two-dimensional image. This, combined with a CNN, enables further judgment of fault characteristics. Its core principle is to convert one-dimensional data into a Gram matrix, calculate the cosine difference between each element to match corresponding color patches, and thus output the image. The basic formula is as follows:

[0063] GADF i,j =sin(θ) i -θ j )

[0064] Each element of the matrix represents the sine value of the angular difference between two time points i and j.

[0065] Furthermore, each step will be explained in detail:

[0066] Step S1: Analyze the historical vibration data of the gearbox output shaft under normal operating conditions for three months from the same unit to determine the threshold value of the effective vibration value. First, calculate the effective value of the data under normal operating conditions. The formula for calculating the effective value is as follows:

[0067]

[0068] Discrete vibration signal x[n] (n=1,2,...,N), where N is the number of samples,

[0069] A box plot is constructed for the calculated RMS values, and Q1 (25th percentile), Q3 (75th percentile), and IQR (interquartile range) are calculated. The threshold values ​​for the RMS values ​​are determined using the box plot. Since anomalies in vibration data typically increase, the lower threshold can be ignored, and only the upper threshold is considered and calculated. The parameters and threshold calculation formulas are as follows:

[0070] Q1 (25% quantile)

[0071]

[0072] Q3 (75% quantile)

[0073]

[0074] IQR (interquartile range)

[0075] IQR = Q3 - Q1

[0076] Threshold (maximum number of observations)

[0077] Threshold = Q³ + 1.5 × IQR

[0078] Step S2: Deploy vibration sensors for the front and rear bearings of the gearbox output shaft to acquire bearing vibration signals in real time, calculate effective values, and perform the first judgment. The test data is first used to calculate the effective value, then compared with the threshold calculated in step S1. If the effective value exceeds the threshold, proceed to the next judgment step; otherwise, output the evaluation result "normal". The combined judgment logic diagram of steps S1 and S2 is shown below. Figure 2 As shown.

[0079] Step S3: Further judgment based on fault analysis, as the fault characteristic frequency requires analysis of parameters such as the number of gear teeth Z, rotational speed f, and transmission ratio. Threshold calculation usually relies on empirical settings and lacks universal standards. Different equipment or fault types may require different thresholds. Furthermore, early fault signals are weak and may be submerged by noise, and spectral analysis requires sufficiently high fault frequency energy for detection. Therefore, Gram Angular Difference Field (GADF) is used for image encoding. GADF is a method that converts time-series data into a two-dimensional image to enhance the feature representation capability of time-series data. By visualizing the signal, GADF overcomes the limitations of traditional fault frequency analysis on parameters, noise, and periodicity. The advantages of GADF encoding are as follows:

[0080] Table 1. Advantages of Gram corner field coding

[0081]

[0082] The following are the simplified steps for fault feature recognition based on Gram angle field encoding and Gram angle difference field images: Figure 3 As shown.

[0083] The detailed steps are as follows:

[0084] 1. Time series normalization

[0085] Objective: To scale the original data to a specific range [-1,1] to facilitate subsequent polar coordinate transformation.

[0086] Here, we use minimum-maximum normalization:

[0087]

[0088] 2. Polar coordinate transformation

[0089] Objective: To map a one-dimensional time series to a polar coordinate system while preserving the temporal order and numerical relationships.

[0090] First, take the inverse cosine of the normalized value xi′ to obtain the angle θ. i Then, the timestamp i is normalized to the interval [0,1] and used as the radius r. i .

[0091] 3. Generate Gram Angular Difference Field (GADF)

[0092] Principle: By calculating the sine value of the angle difference, a Gram matrix is ​​constructed to capture the dynamic changes between time points.

[0093] GADF i,j =sin(θ) i -θ j )

[0094] Each element of the matrix represents the sine value of the angular difference between two time points i and j.

[0095] Overall flowchart as follows Figure 4 As shown.

[0096] This embodiment constructs a further time-series data anomaly detection method using a dual-channel architecture. The left channel focuses on offline model training. By inputting historical data and performing wavelet transform preprocessing, the one-dimensional time-series data is converted into a two-dimensional image feature matrix in polar coordinates using Gram Angular Difference Field (GADF), thereby training a deep learning model CNN to generate a detection model with feature extraction and pattern recognition capabilities. The right channel is responsible for online real-time monitoring. The test data is input into the trained model after undergoing the same preprocessing process for inference. The diamond decision node determines whether the result deviates from the normal threshold. If it is abnormal (N), it proceeds to the "next step judgment" stage, while if it is normal (Y), it outputs a "warning".

[0097] Step S4: Based on the multi-parameter joint threshold comparison mechanism, the principle is to monitor the dynamic relationship between power and key temperature parameters during equipment operation in real time, and combine expert experience to set safety boundaries to achieve quantitative assessment of fault risk.

[0098] The specific logic is as follows: First, the raw input operating data (power, oil temperature, shaft temperature difference, etc.) is preprocessed to ensure data reliability. Then, power-oil temperature curves, power-shaft temperature difference curves, and power-generator front and rear temperature difference curves are plotted to transform the equipment's temperature response under different loads into a visualized time-series relationship. Next, threshold verification is performed—static thresholds derived using expert experience are used to verify the key nodes of the three curves. If all parameters are within the threshold range (Y), the equipment is considered normal and the result is output; otherwise, if any parameter is abnormal (N), a "danger" alarm is triggered. The flowchart is as follows: Figure 5 As shown.

[0099] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains, characterized in that, include: Read the historical vibration data of the front and rear ends of the gearbox output shaft within a preset time period under normal operating conditions of the unit, in order to calculate the effective value and determine the effective value threshold of the vibration; Real-time vibration data of the front and rear ends of the gearbox output shaft are acquired to further calculate the real-time effective value. If the real-time valid value does not exceed the valid value threshold, the output evaluation result is normal; If the real-time effective value exceeds the effective value threshold, fault characteristics are identified using the Gram angle difference field. If the fault feature identification is judged to be normal, the evaluation result is a warning; if the fault feature identification is judged to be abnormal, a multi-parameter joint threshold comparison is further performed. If all parameters are within the corresponding threshold range, the output evaluation result is an alarm; if any parameter is abnormal, the output evaluation result is danger and a danger alarm is triggered. The effective value The calculation formula is: Where x[n] (n=1,2,...,N) is a discrete vibration signal, and N is the number of samplings; The effective value threshold is determined as follows: A box plot is constructed based on the effective values ​​obtained from historical vibration data, and Q1, Q3, and IQR are calculated. Q1 represents the 25th percentile, Q3 represents the 75th percentile, and IQR represents the interquartile range. The effective value threshold is determined based on the box plot.

2. The hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains according to claim 1, characterized in that, The formula for calculating the effective value threshold is as follows: in, This represents the maximum observed value at the effective value threshold.

3. The hierarchical progressive wind power transmission chain multimodal fusion health assessment method according to claim 1, characterized in that, The steps for fault feature identification via Gram angle difference field include: The historical vibration data is preprocessed, and the preprocessed one-dimensional time series data is converted into a two-dimensional image feature matrix in polar coordinates using the Gram angle difference field. The two-dimensional image feature matrix is ​​used to train a deep learning model CNN; The real-time vibration data is processed through the same procedure and then input into a trained deep learning model CNN. The diamond decision node is used to determine whether the result deviates from the normal threshold. If the fault characteristics do not deviate from the normal threshold, the fault feature identification is judged as normal. If the fault deviates from the normal threshold, the fault feature is identified as abnormal.

4. The hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains according to claim 1, characterized in that, The steps of converting preprocessed one-dimensional time-series data into a two-dimensional image feature matrix in polar coordinates using the Gram difference field include: Normalize the unconverted one-dimensional time series data; The inverse cosine of the normalized time value is taken to obtain the corresponding angle. Then, the radius is set to normalize the timestamp to a preset interval as the radius, thereby mapping the one-dimensional time series to the polar coordinate system and preserving the time order and numerical relationship. A two-dimensional image feature matrix is ​​constructed using the sine values ​​of the angle differences at various time points as elements.

5. The hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains according to claim 1, characterized in that, The formula for calculating the sine value of the angle difference between two time points i and j is: 。 6. The hierarchical, progressive, multimodal fusion health assessment method for wind power transmission chains according to claim 1, characterized in that, The steps of the multi-parameter joint threshold comparison include: Obtain the unit's raw operating data; The raw operating data is preprocessed, and the power-oil temperature curve, power-shaft temperature difference curve, and power-generator front and rear temperature difference curve are plotted. Based on a predetermined static threshold, the parameters of the key nodes of the three curves are verified. If all parameters are within the corresponding threshold range, the output evaluation result is an alarm; if any parameter is abnormal, the output evaluation result is danger and a danger alarm is triggered.

Images