Fan rotating part anomaly detection method based on artificial intelligence

Abstract

The invention discloses a fan rotating part anomaly detection method based on artificial intelligence. The fan rotating part anomaly detection method comprises the steps that S1, a fan rotating part sound signal set is constructed; s2, generating a fan rotating part sound signal feature vector set; s3, training to obtain a deep neural network model for detecting the abnormal state of the fan rotating part; s4, a hierarchical reinforcement learning algorithm is utilized to analyze and classify the feature vector set of the fan rotating part, and a high-level strategy judges whether to activate a low-level strategy according to the operating environment and the load condition of the fan and the overall change trend of the sound signal set; according to the low-layer strategy, whether an abnormal state exists or not is further judged according to the fan rotating component feature vector set; s5, denoising and filtering parameters of the sound signal set are adjusted in real time; and S6, when the low-level strategy identifies an abnormal state through analysis of the feature vector set, the system automatically triggers an alarm mechanism. According to the method, the fan anomaly detection cost is reduced, and the detection precision and the intelligent level are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fans, and in particular to an abnormality detection method for rotating parts of a fan based on artificial intelligence. Background Art

[0002] In the prior art, abnormal detection of rotating parts of wind turbines mainly relies on vibration sensors and other mechanical sensing equipment, which monitor the vibration signals of wind turbine blades, rotors and bearings to determine the operating status of the wind turbine. However, the installation and maintenance costs of traditional vibration sensors are high. In the application scenario of large-scale wind farms, their hardware configuration is complex and the overall maintenance cost is high. At the same time, due to the limited detection sensitivity of vibration sensors, traditional detection methods can often only detect abnormal conditions when they are relatively obvious, making it difficult to achieve early warning of faults, resulting in potential small problems gradually evolving into larger equipment failures if not promptly resolved, increasing the difficulty of wind turbine maintenance and the risk of downtime.

[0003] In addition, existing abnormality detection methods make little use of the sound signals of the rotating parts of the fan. They usually only focus on the mechanical vibration signals and ignore the noise characteristics generated by the fan during operation. In fact, the rotating parts of the fan will generate specific acoustic signals during operation. The acoustic signals are closely related to the operating status of the components and can reflect early signs of abnormalities. Therefore, the existing technology fails to fully utilize the sound signals of the fan operation for abnormality detection, resulting in limited detection accuracy and sensitivity.

[0004] In addition, the existing technology has a low level of intelligence and mainly relies on manually set thresholds and simple signal analysis methods, and lacks adaptive capabilities. With the complexity of wind turbine equipment and the diversity of operating environments, it is difficult to cope with the changes of wind turbines under different loads, wind speeds and environmental noise conditions, resulting in common false alarms and missed alarms, which affect the overall operational efficiency of wind farms.

[0005] In summary, the existing technology has the following main shortcomings: first, the installation and maintenance costs of traditional vibration sensors are high, and they are difficult to adapt to the application of large-scale wind farms; second, the existing methods are difficult to capture the early abnormal signals of the rotating parts of the wind turbine, and cannot achieve timely early warning; finally, the existing technology lacks effective utilization of the sound signals generated during the operation of the wind turbine and has a low level of intelligence, and cannot adapt to complex changes in the operating environment. Summary of the invention

[0006] One object of the present invention is to propose a method for detecting abnormalities of rotating parts of a fan based on artificial intelligence. The present invention not only reduces the cost of abnormality detection of the fan, but also improves the detection accuracy and intelligence level.

[0007] According to an embodiment of the present invention, a method for detecting abnormalities of a rotating component of a fan based on artificial intelligence comprises the following steps:

[0008] S1. Arrange multiple acoustic sensor nodes at multiple key locations of the fan to collect fan blade signals, rotor vibration signals, and bearing noise signals generated by the fan rotating parts during operation in real time, and construct a sound signal set of the fan rotating parts;

[0009] S2. Preprocess the collected sound signal set of the fan rotating parts to remove environmental noise and irrelevant signals, and perform spectrum analysis and time-frequency analysis on the sound signal set of the fan rotating parts to obtain the frequency domain characteristics and time domain characteristics of the operation of the fan rotating parts, and generate a feature vector set of the sound signal of the fan rotating parts;

[0010] S3. Based on the historical operation data of the fan and the characteristic vector set of the sound signal of the fan rotating parts, a hierarchical reinforcement learning algorithm is used to train the deep neural network model. The training process includes the high-level strategy judging whether the characteristic vector set needs to be further analyzed through the overall analysis of the sound signal set of the fan rotating parts, and the low-level strategy performing fine-grained classification and analysis on the characteristic vector set under the instruction of the high-level strategy, and training a deep neural network model for detecting abnormal states of the fan rotating parts;

[0011] S4. Input the currently generated feature vector set into the trained deep neural network model, and use the hierarchical reinforcement learning algorithm to analyze and classify the feature vector set of the fan rotating parts. The high-level strategy determines whether to activate the low-level strategy based on the operating environment, load conditions and overall change trend of the sound signal set of the fan. The low-level strategy further determines whether there is an abnormal state based on the feature vector set of the fan rotating parts.

[0012] S5. Dynamically adjust the sensitivity and operating frequency of the acoustic sensor node according to the operating environment parameters of the wind turbine, optimize the collection quality of the sound signal set of the rotating parts of the wind turbine, and adjust the denoising and filtering parameters of the sound signal set in real time;

[0013] S6. When the low-level strategy identifies an abnormal state through analysis of the feature vector set, the system automatically triggers the alarm mechanism, generates a report containing the abnormality type, abnormal location and possible causes, and sends a warning signal to the operation and maintenance team of the wind farm.

[0014] Optionally, the S1 specifically includes the following steps:

[0015] S11. Arrange acoustic sensor nodes {S1, S2, …, S n}, where n is the number of acoustic sensor nodes, and the acoustic sensor nodes S iUsed to collect fan blade signals, rotor vibration signals and bearing noise signals generated by the fan rotating parts during operation in real time;

[0016] S12, the acoustic sensor node collects the sound signals of the rotating parts of the fan at a fixed time interval Δt to obtain a time series signal:

[0017] {x1(t),x2(t),…,x n (t)};

[0018] Among them, x i (t) represents the acoustic sensor node S at time t i The collected sound signal;

[0019] S13, construct the sound signal set X(t) of the fan rotating parts:

[0020] X(t)=[x1(t),x2(t),…,x n (t)];

[0021] Among them, X(t) is the sound signal set of the rotating parts of the wind turbine at time t, which contains multi-channel sound signals collected by n acoustic sensor nodes;

[0022] S14. For each acoustic sensor node S i The collected signal x i (t extracts the corresponding blade signal, rotor vibration signal and bearing noise signal, which are recorded as

[0023] S15, combining different types of sound signals collected by each acoustic sensor node to construct a multi-channel sound signal set X of the fan rotating parts c (t):

[0024]

[0025] Among them, X c (t) is the combined signal set of the rotating parts of the wind turbine at time t, including different types of sound signals collected by all acoustic sensor nodes;

[0026] S16. Generate a complete sound signal dataset D of the fan rotating parts:

[0027] D={X c (t1),X c (t2),..,X c (t m )}.

[0028] Optionally, the S2 specifically includes the following steps:

[0029] S21, pre-process the sound signal data set D of the fan rotating parts, and apply a bandpass filter to filter each group of sound signals, and the filtering range is [f min ,f max ], where f min is the minimum effective frequency, f max is the highest effective frequency, which is used to remove irrelevant environmental noise and obtain the filtered sound signal data set D of the fan rotating parts f ;

[0030] S22, the sound signal data set D of the fan rotating parts after filtering f Normalization is performed to make the amplitude of each sound signal within the same numerical range, and the normalized sound signal data set D of the fan rotating parts is obtained. n ;

[0031] S23, the normalized sound signal data set D of the fan rotating parts n Perform spectrum analysis and use short-time Fourier transform to convert time domain signals into time-frequency domain signals to obtain frequency domain characteristics of the fan rotating parts;

[0032] S24. Sound signal dataset D for rotating parts of fan n Perform time domain analysis to extract instantaneous energy E t , signal envelope and amplitude variation time domain characteristics;

[0033] S25, the frequency domain characteristic vector V of the fan rotating parts f and the time domain feature vector V t Combined to form the characteristic vector set V of the sound signal of the fan rotating parts c .

[0034] Optionally, S3 specifically includes the following steps:

[0035] S31, using the historical operation data set H of the fan and the characteristic vector set V of the sound signal of the rotating parts of the fan c , construct a dataset D for training deep neural network models train :

[0036]

[0037] in, is the characteristic vector of the sound signal of the i-th fan rotating component, including the characteristic vector of the sound signal of the fan rotating component in the frequency domain and the time domain wind turbine rotating parts sound signal feature vector y (i) is the corresponding fan operation status label, and its value is normal state (y (i)=0) or abnormal state (y (i) =1); N is the number of training samples;

[0038] S32. Establish a deep neural network model M of the hierarchical reinforcement learning algorithm, including the high-level strategy network π h and the low-level policy network π l :

[0039] High-Level Policy Network π h Used to input the characteristic vector of the sound signal of the fan rotating parts For overall evaluation, the high-level policy network π h Receive the characteristic vector of the sound signal of the rotating parts of the fan As input, compute high-level policy actions The probability value of , determines whether further analysis is needed:

[0040]

[0041] Among them, W h is the weight matrix of the high-level strategy network, indicating the influence of each feature on the high-level decision; b h is the bias vector; σ(·) is the activation function; represents the probability that the high-level strategy decides on further analysis;

[0042] when When δ is the preset threshold, the high-level strategy decides to activate the low-level strategy network π l , the characteristic vector of the sound signal of the fan rotating parts Conduct fine-grained classification and analysis;

[0043] Low-level policy network π l When the high-level strategy determines that further analysis is needed, the characteristic vector of the sound signal of the wind turbine rotating parts is analyzed. Perform fine-grained classification and abnormal state identification, and the low-level policy network calculates low-level policy actions Predict specific abnormal conditions of wind turbine rotating parts:

[0044]

[0045] Among them, W l is the weight matrix of the low-level strategy network, indicating the impact of each feature on different abnormal states; b l is the bias vector; is the probability distribution vector of K possible abnormal states of the fan rotating parts;

[0046] S33. Define the overall loss function L, and optimize the parameters of the deep neural network model by combining the losses of high-level and low-level strategies:

[0047]

[0048] Among them, the first term is the cross entropy loss of the high-level strategy, y (i) is the label that indicates whether further analysis is actually needed; λ is the weight coefficient; L l is the loss of the low-level strategy, when Calculation time:

[0049]

[0050] Among them, N' is to satisfy The number of samples; is the true label of the i-th sample in the k-th abnormal state; The probability of the kth abnormal state predicted by the low-level strategy;

[0051] S34, by minimizing the loss function L, use gradient descent to update the high-level strategy network parameters {W h ,b h} and low-level policy network parameters {W l ,b l}, a deep neural network model M is trained to detect abnormal status of rotating parts of the wind turbine.

[0052] Optionally, the S4 specifically includes the following steps:

[0053] S41, the sound signal feature vector set of the fan rotating component generated by the current fan rotating component Input into the trained deep neural network model M;

[0054] S42, High-Level Strategy Network π h Receive the current feature vector And according to the current operating environment parameters E of the fan and the overall change trend ΔV of the sound signal set of the fan rotating parts c Computing high-level policy actions The probability value of is used to determine whether to activate the low-level strategy for further analysis;

[0055] S43, when the high-level strategy output When δ is the preset threshold, the low-level strategy network π is activated l Current feature vector of the fan rotating parts Perform fine-grained analysis and classification to further determine whether the fan's rotating parts are in an abnormal state;

[0056] S44, Low-level strategy network π l Receive the current feature vector And output low-level policy actions Predict the specific state or abnormal state of the wind turbine rotating parts;

[0057] S45, according to the state probability distribution of the low-level strategy network output Determine whether the fan rotating parts are in an abnormal state. If the probability of a certain state satisfy Where ∈ is the recognition threshold of the abnormal state, which means that the rotating parts of the fan are in an abnormal state.

[0058] Optionally, the S6 specifically includes the following steps:

[0059] S61, when the low-level strategy network π l According to the characteristic vector of the current fan rotating parts When an abnormal state is identified, the system automatically triggers the alarm mechanism and generates an alarm report containing abnormal information. The alarm report includes the following contents:

[0060] Exception type T error : The probability distribution output by the low-level policy network The abnormal state type T identified in error , so that the highest probability corresponding to the abnormal state Where ∈ is the threshold for anomaly identification;

[0061] Abnormal position P error :Determine the abnormal position P according to the characteristics of the abnormal signal source in the sound signal of the fan rotating parts error ;

[0062] Abnormal occurrence time t error : Record the time point t when the anomaly is identified error , assist the operation and maintenance team to track the timeline of abnormal events;

[0063] Possible cause of abnormality error :Based on the feature vector analysis, infer the possible abnormal cause R error ,Combining historical data and low-level strategies to identify abnormal features and provide possible fault cause inferences;

[0064] S62, report the alarm containing abnormal information to R alarm The report is sent to the wind farm operation and maintenance team. The report is transmitted to the operation and maintenance management system through the communication network and notifies relevant personnel in the form of a warning signal;

[0065] S63. After the warning signal is generated, the system records and stores information about the abnormal event, including the abnormal type T error , abnormal position P error , Occurrence time t error And possible reasons errorAs reference data for future wind turbine maintenance and abnormal analysis, the system automatically updates the historical operation data set H of the wind turbine;

[0066] S64. After the operation and maintenance team receives the warning signal and takes relevant maintenance measures, the system continuously monitors the operating status of the rotating parts of the wind turbine. If the abnormal status is repaired, the system will generate a status recovery report and record the status after maintenance. If the abnormal status is not repaired, the system will continue to send reminder signals.

[0067] Optionally, the identification of possible causes of the anomaly includes when the lower layer policy network π l When it is identified that the fan rotating parts are in an abnormal state, the sound signal feature vector of the fan rotating parts in the current abnormal state is extracted:

[0068]

[0069] in, is the frequency domain feature vector of the abnormal state, is the time domain feature vector of the abnormal state;

[0070] Extract the historical abnormal feature vector set from the wind turbine historical operation data set H:

[0071]

[0072] in, is the characteristic vector of the sound signal of the fan rotating component with the hth historical abnormality, and M is the number of historical abnormality records;

[0073] Calculate the current abnormal feature vector With each historical anomaly feature vector The similarity S between h , determine the historical anomaly feature vector with the highest similarity in,

[0074] Extract from historical anomaly records Corresponding abnormal reason As a possible cause of the current exception R error , the inferred possible abnormal cause R error Exception report R recorded alarm middle.

[0075] The beneficial effects of the present invention are:

[0076] (1) The present invention optimizes the deep neural network model by adopting a hierarchical reinforcement learning algorithm, and realizes efficient multi-level decision-making when detecting abnormalities of rotating parts of the fan. The high-level strategy decides in real time whether to activate the low-level strategy for fine-grained analysis according to the overall operating status and environmental conditions of the fan. Under the guidance of the high-level strategy, the low-level strategy accurately judges the abnormal state through fine-grained analysis of the feature vector. The multi-level decision-making mechanism not only improves the detection sensitivity and can timely capture the early abnormal signals of the fan, avoiding the underreporting and false alarm caused by the traditional method relying on a single threshold, but also significantly improves the accuracy and efficiency of fan abnormality detection.

[0077] (2) The present invention innovatively introduces voiceprint recognition technology, and combines it with a deep learning model to perform a refined analysis of the sound signals of the rotating parts of the fan. By extracting the frequency domain and time domain features of the collected sound signals to form a high-dimensional feature vector set, it can not only analyze tiny acoustic changes but also effectively filter out environmental noise interference. The combination of voiceprint recognition and multi-dimensional feature extraction significantly enhances the sensitivity of anomaly detection, enabling the system to warn of potential problems in advance by identifying subtle changes in voiceprint signals in the early stages of fan failure, thereby avoiding the limitation of traditional vibration detection methods that cannot detect anomalies in time at an early stage.

[0078] (3) Based on feature vector analysis combined with historical abnormal data sets, the present invention intelligently infers the possible causes of the current abnormality of the wind turbine rotating components through similarity calculation and generates an abnormality report. By comparing the similarity between the current abnormal characteristics and historical data, the fault cause inference is quickly provided, which reduces the time and effort required for traditional manual diagnosis and improves the efficiency of fault location and processing. The report lists the abnormality type, location, time and cause in detail, which facilitates the operation and maintenance team to quickly take targeted maintenance measures, thereby significantly improving the overall operating efficiency of the wind farm and reducing the downtime and maintenance costs of the wind turbine. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0080] Figure 1 This is a flow chart of a method for detecting abnormalities of rotating parts of a fan based on artificial intelligence proposed by the present invention;

[0081] Figure 2 This is a schematic diagram of the training process of a deep neural network model based on a hierarchical reinforcement learning algorithm in an artificial intelligence-based abnormality detection method for fan rotating parts proposed in the present invention. DETAILED DESCRIPTION

[0082] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0083] refer to Figure 1-2 , a method for detecting abnormalities of rotating parts of a fan based on artificial intelligence, comprising the following steps:

[0084] S1. Arrange multiple acoustic sensor nodes at multiple key locations of the fan to collect fan blade signals, rotor vibration signals, and bearing noise signals generated by the fan rotating parts during operation in real time, and construct a sound signal set of the fan rotating parts;

[0085] S2. Preprocess the collected sound signal set of the fan rotating parts to remove environmental noise and irrelevant signals, and perform spectrum analysis and time-frequency analysis on the sound signal set of the fan rotating parts to obtain the frequency domain characteristics and time domain characteristics of the operation of the fan rotating parts, and generate a feature vector set of the sound signal of the fan rotating parts;

[0086] S3. Based on the historical operation data of the fan and the characteristic vector set of the sound signal of the fan rotating parts, a hierarchical reinforcement learning algorithm is used to train the deep neural network model. The training process includes the high-level strategy judging whether the characteristic vector set needs to be further analyzed through the overall analysis of the sound signal set of the fan rotating parts, and the low-level strategy performing fine-grained classification and analysis on the characteristic vector set under the instruction of the high-level strategy, and training a deep neural network model for detecting abnormal states of the fan rotating parts;

[0087] S4. Input the currently generated feature vector set into the trained deep neural network model, and use the hierarchical reinforcement learning algorithm to analyze and classify the feature vector set of the fan rotating parts. The high-level strategy determines whether to activate the low-level strategy based on the operating environment, load conditions and overall change trend of the sound signal set of the fan. The low-level strategy further determines whether there is an abnormal state based on the feature vector set of the fan rotating parts.

[0088] S5. Dynamically adjust the sensitivity and operating frequency of the acoustic sensor node according to the operating environment parameters of the wind turbine, optimize the collection quality of the sound signal set of the rotating parts of the wind turbine, and adjust the denoising and filtering parameters of the sound signal set in real time;

[0089] S6. When the low-level strategy identifies an abnormal state through analysis of the feature vector set, the system automatically triggers the alarm mechanism, generates a report containing the abnormality type, abnormal location and possible causes, and sends a warning signal to the operation and maintenance team of the wind farm.

[0090] In this implementation, S1 specifically includes the following steps:

[0091] S11. Arrange acoustic sensor nodes {S1, S2, …, S n}, where n is the number of acoustic sensor nodes, and the acoustic sensor nodes S i Used to collect fan blade signals, rotor vibration signals and bearing noise signals generated by the fan rotating parts during operation in real time;

[0092] S12, the acoustic sensor node collects the sound signals of the rotating parts of the fan at a fixed time interval Δt to obtain a time series signal:

[0093] {x1(t),x2(t),…,x n (t)};

[0094] Among them, x i (t) represents the acoustic sensor node S at time t i The collected sound signal;

[0095] S13, construct the sound signal set X(t) of the fan rotating parts:

[0096] X(t)=[x1(t),x2(t),…,x n (t)];

[0097] Among them, X(t) is the sound signal set of the rotating parts of the wind turbine at time t, which contains multi-channel sound signals collected by n acoustic sensor nodes;

[0098] S14. For each acoustic sensor node S i The collected signal x i (t extracts the corresponding blade signal, rotor vibration signal and bearing noise signal, which are recorded as

[0099] S15, combining different types of sound signals collected by each acoustic sensor node to construct a multi-channel sound signal set X of the fan rotating parts c (t):

[0100]

[0101] Among them, X c (t) is the combined signal set of the rotating parts of the wind turbine at time t, including different types of sound signals collected by all acoustic sensor nodes;

[0102] S16. Generate a complete sound signal dataset D of the fan rotating parts:

[0103] D={X c (t1),X c (t2),…,X c(t m )}.

[0104] In this implementation, S2 specifically includes the following steps:

[0105] S21, pre-process the sound signal data set D of the fan rotating parts, and apply a bandpass filter to filter each group of sound signals, and the filtering range is [f min ,f max ], where f min is the minimum effective frequency, f max is the highest effective frequency, which is used to remove irrelevant environmental noise and obtain the filtered sound signal data set D of the fan rotating parts f ;

[0106] S22, the sound signal data set D of the fan rotating parts after filtering f Normalization is performed to make the amplitude of each sound signal within the same numerical range, and the normalized sound signal data set D of the fan rotating parts is obtained. n ;

[0107] S23, the normalized sound signal data set D of the fan rotating parts n Perform spectrum analysis and use short-time Fourier transform to convert time domain signals into time-frequency domain signals to obtain frequency domain characteristics of the fan rotating parts;

[0108] S24. Sound signal dataset D for rotating parts of fan n Perform time domain analysis to extract instantaneous energy E t , signal envelope and amplitude variation time domain characteristics;

[0109] S25, the frequency domain characteristic vector V of the fan rotating parts f and the time domain feature vector V t Combined to form the characteristic vector set V of the sound signal of the fan rotating parts c .

[0110] In this implementation, S3 specifically includes the following steps:

[0111] S31, using the historical operation data set H of the fan and the characteristic vector set V of the sound signal of the rotating parts of the fan c , construct a dataset D for training deep neural network models train :

[0112]

[0113] in, is the characteristic vector of the sound signal of the i-th fan rotating component, including the characteristic vector of the sound signal of the fan rotating component in the frequency domain and the time domain wind turbine rotating parts sound signal feature vector y (i) is the corresponding fan operation status label, and its value is normal state (y (i) =0) or abnormal state (y (i) =1); N is the number of training samples;

[0114] S32. Establish a deep neural network model M of the hierarchical reinforcement learning algorithm, including the high-level strategy network π h and the low-level policy network π l :

[0115] High-Level Policy Network π h Used to input the characteristic vector of the sound signal of the fan rotating parts For overall evaluation, the high-level policy network π h Receive the characteristic vector of the sound signal of the rotating parts of the fan As input, compute high-level policy actions The probability value of , determines whether further analysis is needed:

[0116]

[0117] Among them, W h is the weight matrix of the high-level strategy network, indicating the influence of each feature on the high-level decision; b h is the bias vector; σ(·) is the activation function; represents the probability that the high-level strategy decides on further analysis;

[0118] when When δ is the preset threshold, the high-level strategy decides to activate the low-level strategy network π l , the characteristic vector of the sound signal of the fan rotating parts Conduct fine-grained classification and analysis;

[0119] Low-level policy network π l When the high-level strategy determines that further analysis is needed, the characteristic vector of the sound signal of the wind turbine rotating parts is analyzed. Perform fine-grained classification and abnormal state identification, and the low-level policy network calculates low-level policy actions Predict specific abnormal conditions of wind turbine rotating parts:

[0120]

[0121] Among them, W l is the weight matrix of the low-level strategy network, indicating the impact of each feature on different abnormal states; b l is the bias vector; is the probability distribution vector of K possible abnormal states of the fan rotating parts;

[0122] S33. Define the overall loss function L, and optimize the parameters of the deep neural network model by combining the losses of high-level and low-level strategies:

[0123]

[0124] Among them, the first term is the cross entropy loss of the high-level strategy, y (i) is the label that indicates whether further analysis is actually needed; λ is the weight coefficient; L l is the loss of the low-level strategy, when Calculation time:

[0125]

[0126] Among them, N' is to satisfy The number of samples; is the true label of the i-th sample in the k-th abnormal state; The probability of the kth abnormal state predicted by the low-level strategy;

[0127] S34, by minimizing the loss function L, use gradient descent to update the high-level strategy network parameters {W h ,b h} and low-level policy network parameters {W l ,b l}, a deep neural network model M is trained to detect abnormal status of rotating parts of the wind turbine.

[0128] In this implementation, S4 specifically includes the following steps:

[0129] S41, the sound signal feature vector set of the fan rotating component generated by the current fan rotating component Input into the trained deep neural network model M;

[0130] S42, High-Level Strategy Network π h Receive the current feature vector And according to the current operating environment parameters E of the fan and the overall change trend ΔV of the sound signal set of the fan rotating parts c Computing high-level policy actions The probability value of is used to determine whether to activate the low-level strategy for further analysis;

[0131] S43, when the high-level strategy output When δ is the preset threshold, the low-level strategy network π is activated l Current feature vector of the fan rotating parts Perform fine-grained analysis and classification to further determine whether the fan's rotating parts are in an abnormal state;

[0132] S44, Low-level strategy network π l Receive the current feature vector And output low-level policy actions Predict the specific state or abnormal state of the wind turbine rotating parts;

[0133] S45, according to the state probability distribution of the low-level strategy network output Determine whether the fan rotating parts are in an abnormal state. If the probability of a certain state satisfy Where ∈ is the recognition threshold of the abnormal state, which means that the rotating parts of the fan are in an abnormal state.

[0134] In this implementation, S6 specifically includes the following steps:

[0135] S61, when the low-level strategy network π l According to the characteristic vector of the current fan rotating parts When an abnormal state is identified, the system automatically triggers the alarm mechanism and generates an alarm report containing abnormal information. The alarm report includes the following:

[0136] Exception type T error : The probability distribution output by the low-level policy network The abnormal state type T identified in error , so that the highest probability corresponding to the abnormal state Where ∈ is the threshold for anomaly identification;

[0137] Abnormal position P error :Determine the abnormal position P according to the characteristics of the abnormal signal source in the sound signal of the fan rotating parts error ;

[0138] Abnormal occurrence time t error : Record the time point t when the anomaly is identified error , assist the operation and maintenance team to track the timeline of abnormal events;

[0139] Possible cause of abnormality error :Based on the feature vector analysis, infer the possible abnormal cause R error ,Combining historical data and low-level strategies to identify abnormal features and provide possible fault cause inferences;

[0140] S62, report the alarm containing abnormal information to R alarm The report is sent to the wind farm operation and maintenance team. The report is transmitted to the operation and maintenance management system through the communication network and notifies relevant personnel in the form of a warning signal;

[0141] S63. After the warning signal is generated, the system records and stores information about the abnormal event, including the abnormal type Terror , abnormal position P error , Occurrence time t error And possible reasons error As reference data for future wind turbine maintenance and abnormal analysis, the system automatically updates the historical operation data set H of the wind turbine;

[0142] S64. After the operation and maintenance team receives the warning signal and takes relevant maintenance measures, the system continuously monitors the operating status of the rotating parts of the wind turbine. If the abnormal status is repaired, the system will generate a status recovery report and record the status after maintenance. If the abnormal status is not repaired, the system will continue to send reminder signals.

[0143] In this embodiment, the identification of possible causes of abnormality includes when the lower layer strategy network π l When it is identified that the fan rotating parts are in an abnormal state, the sound signal feature vector of the fan rotating parts in the current abnormal state is extracted:

[0144]

[0145] in, is the frequency domain feature vector of the abnormal state, is the time domain feature vector of the abnormal state;

[0146] Extract the historical abnormal feature vector set from the wind turbine historical operation data set H:

[0147]

[0148] in, is the characteristic vector of the sound signal of the fan rotating component with the hth historical abnormality, and M is the number of historical abnormality records;

[0149] Calculate the current abnormal feature vector With each historical anomaly feature vector The similarity S between h , determine the historical anomaly feature vector with the highest similarity in,

[0150] Extract from historical anomaly records Corresponding abnormal reason As a possible cause of the current exception R error , the inferred possible abnormal cause R error Exception report R recorded alarm middle.

[0151] Embodiment 1:

[0152] In order to verify the feasibility and effectiveness of the present invention in the abnormality detection of wind turbine rotating parts, the experimenters used two real wind turbine operation data sets: WindFarm-Dataset-A and WindFarm-Dataset-B. The two data sets recorded the operating conditions of wind turbines in two different wind farms, respectively. WindFarm-Dataset-A contains wind turbine operation data from January 2021 to December 2021, including wind speed, load, vibration signal and sound signal information, a total of 50 wind turbines. WindFarm-Dataset-B contains the operation data of 100 wind turbines from July 2020 to July 2021, mainly recording the sound signals and operating environment information of wind turbine rotating parts. The experimenters divided the two data sets into training set and test set in a ratio of 8:2.

[0153] In terms of sound signal feature extraction, the experimenters used a pre-trained ResNet50 model to extract features from the voiceprint signal when the fan was running, and the generated embedding dimension was 2048. In order to adapt to the deep learning model of the present invention, the experimenters introduced a 2D convolutional layer to convert the dimension of the voiceprint feature from 2048 to 256, and combined the results of frequency domain feature extraction to construct the final feature vector set.

[0154] In terms of text data, including the fan operation log and environmental parameter records (wind speed, temperature and load information), the experimenters used the pre-trained BERT model to extract features from the operation log, and the extracted text embedding dimension was 768.

[0155] The deep learning model of the present invention is based on a hierarchical reinforcement learning algorithm. The high-level policy network of the model is responsible for analyzing the overall operating status of the wind turbine and determining whether the feature vector needs further analysis. The low-level policy network performs a fine-grained analysis of the feature vector after the high-level policy is activated, and finally outputs the anomaly detection result. The model of the experimenter is trained using the Adam optimizer for a total of 100 epochs, with an initial learning rate of 0.001 and a batch size of 128.

[0156] In order to verify the detection effect of the present invention, the experimenters first selected a wind turbine that had been running for many years in the WindFarm-Dataset-A dataset as the experimental object. In the training set, the method of the present invention combined the characteristics of sound signals and vibration signals to perform multimodal feature extraction and classification training. In the test set, the wind turbine was operated under different wind speeds and load conditions. After 2 months of continuous monitoring, the detection model of the present invention successfully detected an abnormal bearing wear event during the operation of the wind turbine rotor. The specific process is as follows:

[0157] In October 2021, when the fan was running under high load, the system collected abnormal rotor vibration signals and sound signals, the vibration amplitude and sound frequency peak fluctuated, and the instantaneous energy value increased significantly. The high-level strategy of hierarchical reinforcement learning determined that the anomaly required further analysis, activated the low-level strategy to perform fine-grained classification of the feature vector set, and identified signal features that were highly similar to bearing wear by comparing the historical anomaly data set model. Finally, an anomaly report was generated and it was inferred that the bearing may have excessive wear and needed to be lubricated or replaced immediately.

[0158] The alarm report generated by the system was pushed to the wind farm operation and maintenance team within 2 hours after the abnormality was detected. The team carried out maintenance in a timely manner and ultimately avoided the long-term shutdown of the wind turbine. In contrast, traditional vibration sensors only issued an alarm signal when a failure was about to occur. The operation and maintenance team failed to respond in time, resulting in another wind turbine monitored using traditional methods being shut down for 8 hours.

[0159] In order to demonstrate the superiority of the method of the present invention, the experimenters compared the detection performance of the method of the present invention with that of the traditional vibration sensor method on two data sets, WindFarm-Dataset-A and WindFarm-Dataset-B. The experimenters used accuracy, precision, recall and weighted F1 score as evaluation indicators.

[0160] Table 1 Comparison results between the proposed method and the traditional method on WindFarm-Dataset-A

[0161] Monitoring methods Accuracy Accuracy Recall Weighted F1 score Traditional Vibration Sensor Approach 84.5％ 81.2％ 76.8％ 78.9％ Method of the present invention 93.2％ 91.5％ 90.8％ 91.1％

[0162] Table 2 Comparison results between the proposed method and the traditional method on WindFarm-Dataset-B

[0163] Monitoring methods Accuracy Accuracy Recall Weighted F1 score Traditional Vibration Sensor Approach 82.1％ 79.7％ 74.3％ 76.9％ Method of the present invention 94.6％ 92.8％ 91.3％ 92.0％

[0164] It can be seen from Table 1 and Table 2 that the method of the present invention is superior to the traditional vibration sensor method on both data sets, especially in terms of recall rate and weighted F1 score. This shows that the present invention can not only detect the abnormality of the rotating parts of the fan more accurately but also reduce missed alarms and false alarms, greatly improving the reliability and effectiveness of the system.

[0165] The above experiments have verified that the method of the present invention is significantly better than the traditional vibration sensor method in the abnormality detection effect in the actual wind turbine operation. The present invention can not only effectively capture the slight abnormality in the wind turbine operation through multimodal feature extraction, but also realize more intelligent multi-level decision-making through hierarchical reinforcement learning algorithm, thereby improving the sensitivity and accuracy of detection. In data experiments on multiple real wind farms, the method of the present invention has shown good detection performance, effectively reducing the downtime of wind turbines and reducing maintenance costs.

[0166] Taking WindFarm-Dataset-A as an example, by applying the method of the present invention, the operation and maintenance team of the wind farm successfully identified and handled four abnormal events of wind turbine rotating parts within two months, avoiding at least 24 hours of wind turbine downtime. Compared with traditional methods, the maintenance efficiency was improved and the operation and maintenance cost was reduced by about 15%. The effectiveness of the present invention has been fully demonstrated in practical applications, providing strong technical support for the safe and efficient operation of wind farms.

[0167] The present invention optimizes the deep neural network model by adopting a hierarchical reinforcement learning algorithm, and realizes efficient multi-level decision-making when detecting abnormalities of rotating parts of the fan. The high-level strategy decides in real time whether to activate the low-level strategy for fine-grained analysis according to the overall operating status and environmental conditions of the fan. Under the guidance of the high-level strategy, the low-level strategy accurately judges the abnormal state through fine-grained analysis of the feature vector. The multi-level decision-making mechanism not only improves the detection sensitivity, but also can capture the early abnormal signals of the fan in time, avoiding the underreporting and false alarm caused by the traditional method relying on a single threshold, and at the same time significantly improves the accuracy and efficiency of fan abnormality detection.

[0168] The present invention innovatively introduces voiceprint recognition technology, combines it with a deep learning model to perform refined analysis of the sound signals of the rotating parts of the fan, and forms a high-dimensional feature vector set by extracting the frequency domain and time domain features of the collected sound signals. It can not only analyze tiny acoustic changes but also effectively filter out environmental noise interference. The combination of voiceprint recognition and multi-dimensional feature extraction significantly enhances the sensitivity of anomaly detection, enabling the system to provide early warning of potential problems in the early stages of fan failure by identifying subtle changes in voiceprint signals, thereby avoiding the limitation of traditional vibration detection methods that are unable to detect anomalies in a timely manner at an early stage.

[0169] Based on feature vector analysis combined with historical abnormal data sets, the present invention intelligently infers the possible causes of the current abnormality of the wind turbine rotating components through similarity calculation and generates an abnormality report. By comparing the similarity between the current abnormal characteristics and historical data, the fault cause inference is quickly provided, which reduces the time and effort required for traditional manual diagnosis and improves the efficiency of fault location and processing. The report lists the abnormality type, location, time and cause in detail, which facilitates the operation and maintenance team to quickly take targeted maintenance measures, thereby significantly improving the overall operating efficiency of the wind farm and reducing the downtime and maintenance costs of the wind turbine.

[0170] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for detecting abnormalities of rotating parts of a fan based on artificial intelligence, characterized in that: The steps include: S1. Arrange multiple acoustic sensor nodes at multiple key locations of the fan to collect fan blade signals, rotor vibration signals, and bearing noise signals generated by the fan rotating parts during operation in real time, and construct a sound signal set of the fan rotating parts; S2. Preprocess the collected sound signal set of the fan rotating parts to remove environmental noise and irrelevant signals, and perform spectrum analysis and time-frequency analysis on the sound signal set of the fan rotating parts to obtain the frequency domain characteristics and time domain characteristics of the operation of the fan rotating parts, and generate a feature vector set of the sound signal of the fan rotating parts; S3. Based on the historical operation data of the fan and the characteristic vector set of the sound signal of the fan rotating parts, a hierarchical reinforcement learning algorithm is used to train the deep neural network model. The training process includes the high-level strategy judging whether the characteristic vector set needs to be further analyzed through the overall analysis of the sound signal set of the fan rotating parts, and the low-level strategy performing fine-grained classification and analysis on the characteristic vector set under the instruction of the high-level strategy, and training a deep neural network model for detecting abnormal states of the fan rotating parts; S4. Input the currently generated feature vector set into the trained deep neural network model, and use the hierarchical reinforcement learning algorithm to analyze and classify the feature vector set of the fan rotating parts. The high-level strategy determines whether to activate the low-level strategy based on the operating environment, load conditions and overall change trend of the sound signal set of the fan. The low-level strategy further determines whether there is an abnormal state based on the feature vector set of the fan rotating parts. S5. Dynamically adjust the sensitivity and operating frequency of the acoustic sensor node according to the operating environment parameters of the wind turbine, optimize the collection quality of the sound signal set of the rotating parts of the wind turbine, and adjust the denoising and filtering parameters of the sound signal set in real time; S6. When the low-level strategy identifies an abnormal state through analysis of the feature vector set, the system automatically triggers the alarm mechanism, generates a report containing the abnormality type, abnormal location and possible causes, and sends a warning signal to the operation and maintenance team of the wind farm.

2. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The S1 specifically includes the following steps: S11. Arrange acoustic sensor nodes {S1, S2, …, S n }, where n is the number of acoustic sensor nodes, and the acoustic sensor nodes S i Used to collect fan blade signals, rotor vibration signals and bearing noise signals generated by the fan rotating parts during operation in real time; S12, the acoustic sensor node collects the sound signals of the rotating parts of the fan at a fixed time interval Δt to obtain a time series signal: {x1(t),x2(t),…,x n (t)}; Among them, x i (t) represents the acoustic sensor node S at time t i The collected sound signal; S13, construct the sound signal set X(t) of the fan rotating parts: X(t)=[x1(t),x2(t),…,x n (t)]; Among them, X(t) is the sound signal set of the rotating parts of the wind turbine at time t, which contains multi-channel sound signals collected by n acoustic sensor nodes; S14. For each acoustic sensor node S i The collected signal x i (t extracts the corresponding blade signal, rotor vibration signal and bearing noise signal, which are recorded as S15, combining different types of sound signals collected by each acoustic sensor node to construct a multi-channel sound signal set X of the fan rotating parts c (t): Among them, X c (t) is the combined signal set of the rotating parts of the wind turbine at time t, including different types of sound signals collected by all acoustic sensor nodes; S16. Generate a complete sound signal dataset D of the fan rotating parts: D={X c (t1),X c (t2),…,X c (t m )}。 3. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The S2 specifically includes the following steps: S21, pre-process the sound signal data set D of the fan rotating parts, and apply a bandpass filter to filter each group of sound signals, and the filtering range is [f min ,f max ], where f min is the minimum effective frequency, f max is the highest effective frequency, which is used to remove irrelevant environmental noise and obtain the filtered sound signal data set D of the fan rotating parts f ; S22, the sound signal data set D of the fan rotating parts after filtering f Normalization is performed to make the amplitude of each sound signal within the same numerical range, and the normalized sound signal data set D of the fan rotating parts is obtained. n ; S23, the normalized sound signal data set D of the fan rotating parts n Perform spectrum analysis and use short-time Fourier transform to convert time domain signals into time-frequency domain signals to obtain frequency domain characteristics of the fan rotating parts; S24. Sound signal dataset D for rotating parts of fan n Perform time domain analysis to extract instantaneous energy E t , signal envelope and amplitude variation time domain characteristics; S25, the frequency domain characteristic vector V of the fan rotating parts f and the time domain feature vector V t Combined to form the characteristic vector set V of the sound signal of the fan rotating parts c .

4. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The S3 specifically includes the following steps: S31, using the historical operation data set H of the fan and the characteristic vector set V of the sound signal of the rotating parts of the fan c , construct a dataset D for training deep neural network models train : in, is the characteristic vector of the sound signal of the i-th fan rotating component, including the characteristic vector of the sound signal of the fan rotating component in the frequency domain and the time domain wind turbine rotating parts sound signal feature vector y (i) is the corresponding fan operation status label, and its value is normal state (y (i) =0) or abnormal state (y (i) =1); N is the number of training samples; S32. Establish a deep neural network model M of the hierarchical reinforcement learning algorithm, including the high-level strategy network π h and the low-level policy network π l : High-Level Policy Network π h Used to input the characteristic vector of the sound signal of the fan rotating parts For overall evaluation, the high-level policy network π h Receive the characteristic vector of the sound signal of the rotating parts of the fan As input, compute high-level policy actions The probability value of , determines whether further analysis is needed: Among them, W h is the weight matrix of the high-level strategy network, indicating the influence of each feature on the high-level decision; b h is the bias vector; σ(·) is the activation function; represents the probability that the high-level strategy decides on further analysis; when When δ is the preset threshold, the high-level strategy decides to activate the low-level strategy network π l , the characteristic vector of the sound signal of the fan rotating parts Conduct fine-grained classification and analysis; Low-level policy network π l When the high-level strategy determines that further analysis is needed, the characteristic vector of the sound signal of the wind turbine rotating parts is analyzed. Perform fine-grained classification and abnormal state identification, and the low-level policy network calculates low-level policy actions Predict specific abnormal conditions of the wind turbine's rotating parts: Among them, W l is the weight matrix of the low-level strategy network, indicating the impact of each feature on different abnormal states; b l is the bias vector; is the probability distribution vector of K possible abnormal states of the fan rotating parts; S33. Define the overall loss function L, and optimize the parameters of the deep neural network model by combining the losses of high-level and low-level strategies: Among them, the first term is the cross entropy loss of the high-level strategy, y (i) is the label that indicates whether further analysis is actually needed; λ is the weight coefficient; L l is the loss of the low-level strategy, when Calculation time: Among them, N' is to satisfy The number of samples; is the true label of the i-th sample in the k-th abnormal state; The probability of the kth abnormal state predicted by the low-level strategy; S34, by minimizing the loss function L, use gradient descent to update the high-level strategy network parameters {W h ,b h } and low-level policy network parameters {W l ,b l }, a deep neural network model M is trained to detect abnormal status of rotating parts of the wind turbine.

5. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The S4 specifically comprises the following steps: S41, the sound signal feature vector set of the fan rotating component generated by the current fan rotating component Input into the trained deep neural network model M; S42, High-Level Strategy Network π h Receive the current feature vector And according to the current operating environment parameters E of the fan and the overall change trend ΔV of the sound signal set of the fan rotating parts c Computing high-level policy actions The probability value of is used to determine whether to activate the low-level strategy for further analysis; S43, when the high-level strategy output When δ is the preset threshold, the low-level strategy network π is activated l Current feature vector of the fan rotating parts Perform fine-grained analysis and classification to further determine whether the fan's rotating parts are in an abnormal state; S44, Low-level strategy network π l Receive the current feature vector And output low-level policy actions Predict the specific state or abnormal state of the wind turbine rotating parts; S45, according to the state probability distribution of the low-level strategy network output Determine whether the fan rotating parts are in an abnormal state. If the probability of a certain state satisfy Where ∈ is the recognition threshold of the abnormal state, which means that the rotating parts of the fan are in an abnormal state.

6. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The S6 specifically comprises the following steps: S61, when the low-level strategy network π l According to the characteristic vector of the current fan rotating parts When an abnormal state is identified, the system automatically triggers the alarm mechanism and generates an alarm report containing abnormal information. The alarm report includes the following contents: Exception type T error : The probability distribution output by the low-level policy network The abnormal state type T identified in error , so that the highest probability corresponding to the abnormal state Where ∈ is the threshold for anomaly identification; Abnormal position P error :Determine the abnormal position P according to the characteristics of the abnormal signal source in the sound signal of the fan rotating parts error ; Abnormal occurrence time t error : Record the time point t when the anomaly is identified error , assist the operation and maintenance team to track the timeline of abnormal events; Possible cause of abnormality error :Based on the feature vector analysis, infer the possible abnormal cause R error ,Combining historical data and low-level strategies to identify abnormal features and provide possible fault cause inferences; S62, report the alarm containing abnormal information to R alarm The report is sent to the wind farm operation and maintenance team. The report is transmitted to the operation and maintenance management system through the communication network and notifies relevant personnel in the form of a warning signal; S63. After the warning signal is generated, the system records and stores information about the abnormal event, including the abnormal type T error , abnormal position P error , Occurrence time t error And possible reasons error As reference data for future wind turbine maintenance and abnormal analysis, the system automatically updates the historical operation data set H of the wind turbine; S64. After the operation and maintenance team receives the warning signal and takes relevant maintenance measures, the system continuously monitors the operating status of the rotating parts of the wind turbine. If the abnormal status is repaired, the system will generate a status recovery report and record the status after maintenance. If the abnormal status is not repaired, the system will continue to send reminder signals.

7. The method for detecting abnormalities of rotating parts of a fan based on artificial intelligence according to claim 1, characterized in that: The identification of possible causes of the anomaly includes when the lower-level policy network π l When it is identified that the fan rotating parts are in an abnormal state, the sound signal feature vector of the fan rotating parts in the current abnormal state is extracted: in, is the frequency domain feature vector of the abnormal state, is the time domain feature vector of the abnormal state; Extract the historical abnormal feature vector set from the wind turbine historical operation data set H: in, is the characteristic vector of the sound signal of the fan rotating component with the hth historical abnormality, and M is the number of historical abnormality records; Calculate the current abnormal feature vector With each historical anomaly feature vector The similarity S between h , determine the historical anomaly feature vector with the highest similarity in, Extract from historical anomaly records Corresponding abnormal reason As a possible cause of the current exception R error , the inferred possible abnormal cause R error Exception report R recorded alarm middle.

Images