A method for diagnosing motor and load by current harmonic and vibration feature fusion

Abstract

This invention relates to the field of motor fault diagnosis and condition monitoring technology, and discloses a method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics. The method includes synchronously acquiring three-phase current signals and multi-axial vibration signals of the motor to be diagnosed, and performing signal preprocessing to generate current preprocessed signals and vibration preprocessed signals; based on the current preprocessed signals, performing harmonic feature extraction processing to generate current harmonic feature data; by synchronously acquiring three-phase current signals and multi-axial vibration signals, and extracting current harmonic features and vibration time-frequency domain features respectively, and then performing time-series alignment and matching of the two types of features to form a set of synergistic feature pairs, the electrical characteristics of the current and the mechanical state of the vibration can be correlated and complemented, thereby ensuring the comprehensiveness and synergy of multi-source heterogeneous feature information fusion.

Description

Technical Field

[0001] This invention relates to the field of motor fault diagnosis and condition monitoring technology, specifically a method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics. Background Technology

[0002] An electric motor is an electromagnetic device that converts or transmits electrical energy based on the law of electromagnetic induction. In physics, a load is an electronic component connected to two ends of a circuit with a certain potential difference, used to convert electrical energy into other forms of energy. In electrical engineering, it refers to a device that receives electrical energy in a circuit and is a general term for all kinds of electrical appliances. Common loads include resistors, engines, light bulbs, air conditioners, electric motors, and other components that consume power.

[0003] Currently, in the field of motor and load condition monitoring and fault diagnosis, single current analysis or vibration analysis methods are usually used. Due to the complex operating conditions of motors and the variable loads, it is difficult to achieve real-time coordination and deep correlation between electrical characteristics and mechanical condition characteristics when relying solely on current signals or vibration signals for diagnosis.

[0004] Therefore, a method for diagnosing motors and loads by synergistic integration of current harmonics and vibration characteristics is proposed to solve the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for diagnosing motors and loads by synergistically fusing current harmonics and vibration characteristics, thus solving the problem mentioned in the background art of difficulty in achieving real-time coordination and deep correlation between electrical characteristics and mechanical state characteristics.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics, comprising the following steps: S1. Synchronously acquire the three-phase current signal and multi-axial vibration signal of the motor to be diagnosed, and perform signal preprocessing to generate current preprocessing signal and vibration preprocessing signal. S2. Based on the current preprocessing signal, perform current signal harmonic feature extraction processing to generate current harmonic feature data; S3. Based on the vibration preprocessing signal, perform time-domain and frequency-domain feature extraction processing of the vibration signal to generate vibration time-frequency domain feature data; S4. Perform feature co-processing on the current harmonic feature data and the vibration time-frequency domain feature data to generate a time-aligned set of co-feature pairs; S5. Based on the aforementioned set of collaborative features, perform deep fusion processing of current harmonics and vibration features to construct a high-dimensional fusion feature matrix; S6. Based on the high-dimensional fusion feature matrix and the pre-stored standard fusion feature matrix library of motor and load multi-state, the optimal matching state feature is retrieved through a biomimetic intelligent optimization algorithm to generate preliminary diagnostic state type labels. S7. Based on the high-dimensional fusion feature matrix, the abnormal motor operation is initially determined by the abnormal detection model, and abnormal status identification data is generated. S8. Based on the preliminary diagnostic status type label and the abnormal status identifier data, perform refined classification and diagnosis of fault types, and generate fault diagnosis result data for the target motor and load.

[0007] Preferably, the signal preprocessing in S1 includes the following steps: S11. The original three-phase stator current signal of the motor during operation is synchronously collected in the motor power supply circuit by a high-precision current sensor. At the same time, the original axial, radial and tangential vibration signals of the motor during operation are synchronously collected at specific measuring points on the motor bearing housing and the casing by a multi-axis vibration acceleration sensor. S12. Perform preprocessing on the original three-phase stator current signal, including removing DC components, power frequency notch filtering, and low-pass filtering, to generate a current preprocessed signal. S13. Perform preprocessing on the original multiaxial vibration signal, including detrending term, bandpass filtering, and noise reduction, to generate a vibration preprocessed signal.

[0008] Preferably, the current signal harmonic feature extraction processing in S2 includes the following steps: S21. Perform a fast Fourier transform on the current preprocessed signal to obtain the current signal spectrum; S22. Extract harmonic parameters, including the amplitude and phase of the fundamental, fifth, seventh, eleventh, and thirteenth harmonics, from the current signal spectrum, and calculate the total harmonic distortion rate and the harmonic content characteristics of each harmonic. S23. The extracted harmonic amplitude, phase, total harmonic distortion rate and harmonic content rate are combined to construct a current harmonic feature vector, thereby generating the current harmonic feature data.

[0009] Preferably, the time-domain and frequency-domain feature extraction processing of the vibration signal in S3 includes the following steps: S31. Extract time-domain features from the vibration preprocessing signal and calculate time-domain feature parameters including RMS value, peak value, impulse factor, kurtosis index, and waveform factor. S32. The vibration preprocessing signal is subjected to frequency domain feature extraction. The bearing characteristic frequency and the amplitude of its sideband components are obtained by envelope demodulation analysis. The amplitude of the rotation frequency, harmonic frequency and its harmonic components are extracted by spectrum analysis. The spectrum feature parameters including the center of gravity frequency and the mean square frequency are calculated. S33. Combine the time-domain feature parameters and frequency-domain feature parameters along the axial dimension to generate the vibration time-frequency domain feature data.

[0010] Preferably, the feature collaborative preprocessing in S4 includes the following steps: S41. Obtain the current harmonic characteristic data and the vibration time-frequency domain characteristic data, and perform time-stamp alignment processing on the two types of characteristic data based on the same timestamp of signal acquisition; S42. For the time-scaled aligned feature data, the dynamic time warping algorithm is used to perform length warping and feature point matching on the current harmonic feature sequence and the vibration time-frequency domain feature sequence. S43. Pair the current harmonic feature vector and the vibration time-frequency domain feature vector that have completed time-scale alignment and feature point matching to form a time-aligned cooperative feature pair set consisting of multiple current feature and vibration feature pairs.

[0011] Preferably, the deep fusion processing of current harmonics and vibration characteristics in S5 includes the following steps: S51. Obtain the set of cooperative feature pairs, and concatenate the current harmonic feature vector and the vibration time-frequency domain feature vector in each cooperative feature pair to form a primary fusion feature vector. S52. Input the primary fused feature vector into a multilayer perceptron neural network. The multilayer perceptron neural network performs nonlinear transformation and feature dimensionality reduction on the input features to extract deep abstract features that can simultaneously characterize the specificity of current harmonics and the correlation of vibration state. S53. Arrange the deep abstract features output by the multilayer perceptron neural network in time sequence to construct a high-dimensional fusion feature matrix.

[0012] Preferably, the optimal matching state feature retrieval in S6 includes the following steps: S61. Establish a standard fusion feature matrix library for motors and loads in multiple states. The matrix library stores standard high-dimensional fusion feature matrix samples of motors under various known health states, load conditions and fault types. S62. Calculate the similarity between the high-dimensional fusion feature matrix and each standard sample in the standard fusion feature matrix library. The similarity calculation adopts a comprehensive metric method that weights cosine similarity and Euclidean distance. S63. The similarity calculation process is optimized by using the Osprey optimization algorithm to quickly locate the K nearest neighbor standard samples that are most similar to the high-dimensional fusion feature matrix in the standard library. The osprey optimization algorithm simulates the exploration and development behavior of ospreys, iteratively updating the positions of candidate samples in the search space of the standard feature matrix library until the optimal match is found. S64. Statistically analyze the state labels corresponding to the K nearest neighbor standard samples, and output the state label with the highest frequency as the preliminary diagnostic state type label.

[0013] Preferably, the preliminary determination of abnormal motor operation in S7 includes the following steps: S71. Construct a sparse autoencoder as an anomaly detection model, and use high-dimensional fusion feature matrix samples under normal motor operation to perform unsupervised training on the sparse autoencoder so that it learns the feature reconstruction pattern under normal conditions. S72. Input the high-dimensional fusion feature matrix to be diagnosed into the trained sparse autoencoder to obtain its output reconstruction matrix; S73. Calculate the reconstruction error between the high-dimensional fusion feature matrix and the output reconstruction matrix. When the reconstruction error exceeds a preset anomaly threshold, generate the anomaly status identifier data indicating the presence of an anomaly. S74. When the reconstruction error does not exceed the abnormal threshold, generate abnormal state identifier data indicating that the state is normal.

[0014] Preferably, the refined classification and diagnosis of fault types in S8 includes the following steps: S81. Obtain the preliminary diagnosis status type label and the abnormal status identification data. When the abnormal status identification data shows that it is normal, directly output the label in the preliminary diagnosis status type label that represents normal and known load conditions as the target motor and load fault diagnosis result data. S82. When the abnormal status indicator data is displayed as abnormal, the refined classification process is initiated: The high-dimensional fusion feature matrix is ​​input into a pre-trained support vector machine multi-classification model to obtain the specific fault category determination; S83. Logically verify and fuse the refined classification results with the preliminary diagnostic status type labels to generate the final target motor and load fault diagnosis result data, which includes fault type, suspected faulty components, and severity level.

[0015] Preferably, the method further includes: S91. Evaluate the reliability of the fault diagnosis results data of the target motor and load. S92. Encapsulate the target motor and load fault diagnosis result data and their corresponding confidence scores together with the original current harmonic characteristic trend diagram, vibration spectrum diagram, and key node data of the feature fusion process. S93. Visualize and render the packaged data to generate a final motor and load collaborative diagnostic report that includes diagnostic conclusions, credibility, feature maps, and maintenance suggestions. Output and store the report through a human-machine interface.

[0016] Compared with the prior art, the present invention provides a method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics, which has the following beneficial effects: 1. In this invention, by synchronously acquiring three-phase current signals and multi-axial vibration signals, and extracting current harmonic features and vibration time-frequency domain features respectively, and then performing time-series alignment and matching of the two types of features to form a set of synergistic feature pairs, the electrical characteristics of the current and the mechanical state of the vibration can be correlated and complemented, thereby ensuring the comprehensiveness and synergy of multi-source heterogeneous feature information fusion.

[0017] 2. In this invention, the Osprey optimization algorithm is used to perform optimal matching retrieval in the standard feature library to obtain preliminary state labels, and a sparse autoencoder is used for anomaly detection. Then, a support vector machine model is combined for refined fault classification, so that the diagnosis process takes into account retrieval efficiency, anomaly sensitivity and classification accuracy, thereby ensuring the speed, reliability and accuracy of the diagnosis results.

[0018] 3. In this invention, by assessing the credibility of the diagnostic results and encapsulating and rendering them together with the original feature map and key data from the fusion process into a visual diagnostic report, the entire diagnostic process forms a complete automated link from data acquisition, feature fusion, intelligent diagnosis to result output, thereby ensuring the interpretability and practicality of the diagnostic results and providing a basis for subsequent decision-making. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics, according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] For specific implementation examples, please refer to: Figure 1 A method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics includes the following steps: S1. Synchronously acquire the three-phase current signal and multi-axial vibration signal of the motor to be diagnosed, and perform signal preprocessing to generate current preprocessing signal and vibration preprocessing signal. S2. Based on the current preprocessing signal, perform harmonic feature extraction processing of the current signal to generate current harmonic feature data; S3. Based on the vibration preprocessing signal, perform time-domain and frequency-domain feature extraction processing of the vibration signal to generate vibration time-frequency domain feature data; S4. Perform feature co-processing on the current harmonic feature data and the vibration time-frequency domain feature data to generate a time-aligned set of co-feature pairs; S5. Based on the set of collaborative features, perform deep fusion processing of current harmonics and vibration features to construct a high-dimensional fusion feature matrix; S6. Based on the high-dimensional fusion feature matrix and the pre-stored standard fusion feature matrix library of motor and load multi-state, the optimal matching state feature retrieval is performed through the biomimetic intelligent optimization algorithm to generate preliminary diagnostic state type labels. S7. Based on the high-dimensional fusion feature matrix, the abnormality detection model is used to make a preliminary judgment on the abnormality of motor operation and generate abnormality status identification data. S8. Based on the preliminary diagnostic status type label and abnormal status identifier data, perform refined classification and diagnosis of fault types, and generate fault diagnosis result data for the target motor and load.

[0022] Signal preprocessing in S1 includes the following steps: S11. The original three-phase stator current signal of the motor during operation is synchronously collected in the motor power supply circuit by a high-precision current sensor. At the same time, the original axial, radial and tangential vibration signals of the motor during operation are synchronously collected at specific measuring points on the motor bearing housing and the casing by a multi-axis vibration acceleration sensor. S12. Perform preprocessing on the original three-phase stator current signal, including removing DC components, power frequency notch filtering, and low-pass filtering, to generate a current preprocessed signal. In practice, the DC component is first removed from the synchronously acquired raw three-phase stator current signal; then, the average signal is calculated for a length of... The formula for calculating the DC component of a discrete signal sequence is: ; in This refers to the DC component in the current signal. The total length of the discrete signal sequence. For the first The original signal values ​​of the three-phase stator current at each sampling point The sampling point number; Subtracting the DC component from the original signal yields the signal after DC removal. Subsequently, in order to eliminate power grid frequency and harmonic interference, [the following measures were taken]. For power frequency notch filtering, a second-order IIR notch filter is used, whose transfer function in the Z-domain can be expressed as: ; in For notch filters in The transfer function of a domain. This is the digital angular frequency corresponding to the center frequency of the notch filter. To control the pole radius parameter of the notch depth, For the complex variables in the Z-transform; finally, to preserve the characteristic harmonic frequency bands related to motor faults and suppress high-frequency noise, the notch-filtered signal is low-pass filtered, with a cutoff frequency of . The analog domain amplitude response formula for the Butterworth low-pass filter is: ; in For low-pass filters at frequency The amplitude response at that point, The filter order is used; through the above steps, the current preprocessed signal is finally generated. S13. Perform preprocessing on the original multi-axial vibration signal, including detrending term, bandpass filtering, and noise reduction, to generate a vibration preprocessed signal; In practice, for the synchronously acquired multi-axial vibration raw acceleration signals, the trend terms introduced by sensor temperature drift and baseline drift are first eliminated; then, a polynomial fitting method is used... polynomial of order Fit the slow-changing trend in the signal, and then subtract the trend term from the original signal to obtain the detrended signal: ; in Signals after trend reversal This is the original acceleration signal of multi-axial vibration; Subsequently, based on the characteristic frequency range of the motor bearings and rotor, the lower and upper cutoff frequencies of the bandpass filter are set, and the bandpass filter is applied. Bandpass filtering is performed to retain the frequency band where the fault characteristics are located and to filter out low-frequency vibrations and high-frequency noise. Finally, to further suppress in-band noise, the bandpass-filtered signal is denoised. Wavelet thresholding is used to decompose the signal by selecting an appropriate wavelet basis and decomposition level. The high-frequency coefficients obtained by decomposition are subjected to soft thresholding to suppress noise. Wavelet reconstruction is then performed to finally generate the vibration preprocessed signal.

[0023] The process of extracting harmonic features of the current signal in S2 includes the following steps: S21. Perform a fast Fourier transform on the current preprocessed signal to obtain the current signal spectrum; The formula for Fast Fourier Transform is:

[0024] ; in It is the first The complex spectrum of each frequency component It is a discrete sampling sequence of the current preprocessed signal. It is the imaginary unit. The number of data points to perform the Fourier transform. For time-domain indexing; S22. Extract harmonic parameters, including the amplitude and phase of the fundamental, fifth, seventh, eleventh, and thirteenth harmonics, from the current signal spectrum, and calculate the total harmonic distortion rate and the harmonic content characteristics of each order. Total harmonic distortion: ; in Total harmonic distortion (THD) The amplitude of the fundamental current. For the first The amplitude of the second harmonic current. For harmonic order, The highest order of the harmonics under consideration; Harmonic content characteristics: ; in For the first Subharmonic content; S23. Combine the extracted harmonic amplitude, phase, total harmonic distortion rate and the content rate of each harmonic to construct a current harmonic feature vector, and generate current harmonic feature data.

[0025] The time-domain and frequency-domain feature extraction and processing of vibration signals in S3 includes the following steps: S31. Extract time-domain features from the vibration preprocessing signal and calculate time-domain feature parameters including RMS value, peak value, impulse factor, kurtosis index, and waveform factor. Valid values: ; in For valid values, For the first Vibration preprocessing signal values ​​at each sampling point This represents the total number of signal sampling points. peak value of the signal The absolute maximum value of the signal; impulse factor: ; in For pulse factor, The peak value of the signal; Kurtosis index: ; in As a kurtosis index, This represents the average value of the vibration preprocessing signal; Calculate waveform factor: ; in Waveform factor; S32. Extract frequency domain features from the vibration preprocessing signal, obtain the bearing characteristic frequency and the amplitude of its sideband components through envelope demodulation analysis, extract the amplitude of the rotation frequency, harmonic frequency and its harmonic components through spectrum analysis, and calculate the spectrum characteristic parameters including the center of gravity frequency and mean square frequency. Center of gravity frequency: ; in For the center of gravity frequency, For the first in the spectrum The frequency values ​​corresponding to each spectral line For frequency The corresponding spectral amplitude at that location, This represents the total number of spectral lines in the spectrum. Mean square frequency: ; in The mean square frequency; S33. Combine the time-domain characteristic parameters and frequency-domain characteristic parameters along the axial dimension to generate vibration time-frequency domain characteristic data.

[0026] Feature co-processing in S4 includes the following steps: S41. Obtain current harmonic characteristic data and vibration time-frequency domain characteristic data, and perform time-stamp alignment processing on the two types of characteristic data based on the same timestamp of signal acquisition; S42. For the time-aligned feature data, the dynamic time warping algorithm is used to warp the length of the current harmonic feature sequence and the vibration time-frequency domain feature sequence and match the feature points so that the features within the same physical time period have a corresponding relationship. In practice, the dynamic time warping algorithm aims to find an optimal warping path. This minimizes the cumulative distance between the two sequences; the distance metric is typically Euclidean distance, and the cumulative distance is calculated recursively using dynamic programming. ; in To accumulate distance, Current eigenvector With vibration eigenvectors Local distance between them , For time point indexing. By backtracking the optimal path, a one-to-one matching relationship between feature points of two sequences is obtained. Based on this matching relationship, the feature sequences are interpolated and sampled to make them reach the same effective length, and finally a set of time-aligned collaborative feature pairs is generated. S43. Pair the current harmonic feature vector and the vibration time-frequency domain feature vector that have completed time-scale alignment and feature point matching to form a time-aligned cooperative feature pair set consisting of multiple current feature and vibration feature pairs.

[0027] The deep integration processing of current harmonics and vibration characteristics in S5 includes the following steps: S51. Obtain the set of cooperative feature pairs, and concatenate the current harmonic feature vector and the vibration time-frequency domain feature vector in each cooperative feature pair to form a primary fusion feature vector. S52. Input the primary fusion feature vector into the multilayer perceptron neural network. The multilayer perceptron neural network performs nonlinear transformation and feature dimensionality reduction on the input features to extract deep abstract features that can simultaneously characterize the specificity of current harmonics and the correlation of vibration state. In practice, the concatenated primary fusion feature vector is used as input and fed into a multilayer perceptron neural network. This network consists of an input layer, at least one hidden layer, and an output layer. The hidden layer uses the ReLU activation function to perform a nonlinear transformation on the input features, uncovering the deep coupling relationship between current and vibration features. By designing the number of neurons in the hidden layer to be less than that in the input layer, the network learns during training to compress and refine the high-dimensional primary fusion features into a low-dimensional but more information-dense representation, achieving feature dimensionality reduction and extraction of deep abstract features. This process can be represented as: ; in The extracted deep abstract feature vector, This is the initial fusion feature vector. and These are the network's weights and bias parameters. Represents the forward propagation function of the MLP; arrange the deep abstract feature vectors of all time steps to construct a high-dimensional fusion feature matrix; S53. Arrange the deep abstract features output by the multilayer perceptron neural network in time series order to construct a high-dimensional fusion feature matrix.

[0028] The optimal matching state feature retrieval in S6 includes the following steps: S61. Establish a standard fusion feature matrix library for motors and loads in multiple states. The matrix library stores standard high-dimensional fusion feature matrix samples of motors under various known health states, load conditions and fault types. In practice, the operating data of the motor under various standard states are collected in the laboratory and under known working conditions. The data under each standard state is processed synchronously according to the complete process from S1 to S5, and finally one or more representative high-dimensional fusion feature matrices are generated for each state. These feature matrices are associated with their corresponding strictly calibrated state labels and stored together to form a multi-state standard fusion feature matrix library for motors and loads. S62. Calculate the similarity between the high-dimensional fusion feature matrix and each standard sample in the standard fusion feature matrix library. The similarity calculation adopts a comprehensive metric method that weights cosine similarity and Euclidean distance. Cosine similarity: ; in The feature matrix to be tested With standard feature matrix Cosine similarity between them To make the matrix The feature vector obtained after vectorization To make the matrix The feature vector obtained after vectorization; Euclidean distance: ; in The feature matrix to be tested With standard feature matrix The Euclidean distance between them; After normalization, the two are weighted... and Combined, a comprehensive similarity score is obtained: ; in The overall similarity score is calculated. S63. The Osprey optimization algorithm is used to optimize the similarity calculation process, so as to quickly locate the K nearest neighbor standard samples that are most similar to the high-dimensional fusion feature matrix in the standard library. The Osprey optimization algorithm simulates the exploration and development behavior of the Osprey, iteratively updating the positions of candidate samples in the search space of the standard feature matrix library until the optimal match is found; The Osprey location update process involves an exploration phase and a development phase. The location update formula for the exploration phase is as follows: ; in Indicates the first The osprey's new location during the exploration phase Its current location, This indicates the position of the best individual in the current population. To explore factors, Represents a random number within the interval [0,1]. S64. Calculate the state labels corresponding to the K nearest neighbor standard samples, and output the state label with the highest frequency as the preliminary diagnostic state type label.

[0029] Preliminary determination of abnormal motor operation in S7 includes the following steps: S71. Construct a sparse autoencoder as an anomaly detection model, and use high-dimensional fusion feature matrix samples under normal motor operation to train the sparse autoencoder in unsupervised training so that it learns the feature reconstruction pattern under normal conditions. In practice, a sparse autoencoder is a special type of three-layer neural network, consisting of an input layer, hidden layers, and an output layer. Its goal is to reconstruct the input as accurately as possible from the output, while introducing sparsity constraints in the hidden layers. During construction, the number of neurons in both the input and output layers is equal to the dimension q of the vectorized high-dimensional fused feature matrix, while the number of neurons in the hidden layer is typically less than q. Its encoding process converts the input... Mapping to hidden representation The decoding process will Structure as output : ; ; in To hide the representation, The reconstructed vector output by the decoder. and For activation function, , , , These are the network parameters; during training, a large number of high-dimensional fused feature matrices under normal conditions are used as training samples, and the reconstruction error is minimized. The network parameters are optimized by incorporating a sparsity penalty term for hidden layer activation; after training, the sparse autoencoder learns the distribution pattern of the normal state data. S72. Input the high-dimensional fusion feature matrix to be diagnosed into the trained sparse autoencoder and obtain its output reconstruction matrix. S73. Calculate the reconstruction error between the high-dimensional fusion feature matrix and the output reconstruction matrix. When the reconstruction error exceeds the preset anomaly threshold, generate anomaly status identification data indicating the presence of anomalies. The formula for calculating reconstruction error is: ; in For reconstruction error, This represents the first element in the high-dimensional fusion feature matrix of the input. 1 eigenvector To reconstruct the corresponding vector of the output from the sparse autoencoder. The number of feature vectors; S74. When the reconstruction error does not exceed the abnormal threshold, generate abnormal status identifier data indicating that the status is normal.

[0030] The refined classification and diagnosis of fault types in S8 includes the following steps: S81. Obtain the preliminary diagnosis status type label and abnormal status identification data. When the abnormal status identification data is normal, directly output the label in the preliminary diagnosis status type label that represents normal and known load conditions as the target motor and load fault diagnosis result data. S82. When the abnormal status indicator data is displayed as abnormal, the refined classification process is initiated; The high-dimensional fused feature matrix is ​​input into a pre-trained support vector machine multi-classification model to obtain the specific fault category determination; this process can be summarized as follows: ; in This represents the fault type classification result vector output by the model. For a pre-trained support vector machine multi-class classification model, The input is a high-dimensional fusion feature matrix. The set of parameters obtained from training the fault sample dataset; S83. Logically verify and merge the refined classification results with the preliminary diagnostic status type labels to generate the final target motor and load fault diagnosis result data. The result data includes fault type, suspected faulty components, and severity level. In practice, the initial diagnostic status type labels and the refined classification results from the support vector machine are obtained; the following logical verification and fusion are then performed: Consistency check: If the preliminary diagnostic status type label is a specific fault type and is consistent with the main classification result in the vector machine fine classification result, then the result is directly adopted and the overall credibility of the final result is improved; Conflict arbitration: When the two are inconsistent, compare the confidence of the refined classification result of the support vector machine with the similarity of the preliminary diagnostic state type label; Information completion: The fault type selected by arbitration is fused with the suspected component and severity level information provided in the support vector machine fine classification results to generate structured final diagnostic result data.

[0031] The method also includes: S91. Conduct a reliability assessment of the fault diagnosis results data of the target motor and load. The credibility assessment formula is: ; in For credibility score, This represents the highest similarity score among the matches. This represents the normalized reconstruction error value. This represents the probability confidence level of the support vector machine classification. Here are the weighting coefficients for each indicator, and ; S92. Encapsulate the target motor and load fault diagnosis results data and their corresponding confidence scores together with the original current harmonic characteristic trend diagram, vibration spectrum diagram, and key node data of the feature fusion process. S93. Visualize and render the packaged data to generate a final motor and load collaborative diagnostic report that includes diagnostic conclusions, credibility, feature maps, and maintenance suggestions. Output and store the report through a human-machine interface.

[0032] The operational steps of a method for diagnosing motors and loads by synergistic integration of current harmonics and vibration characteristics are as follows: Step 1: Synchronous Signal Acquisition and Preprocessing Using a high-precision current sensor and a multi-axis vibration accelerometer, the raw three-phase stator current signal of the motor under diagnosis, as well as the raw multi-axis vibration signals in the axial, radial, and tangential directions, are simultaneously acquired. Subsequently, the current signal is processed by removing the DC component, applying power frequency notch filtering, and low-pass filtering to generate a current preprocessed signal; the vibration signal is processed by removing the trend term, applying bandpass filtering, and noise reduction to generate a vibration preprocessed signal.

[0033] Step 2: Extraction of Current Harmonic Features A Fast Fourier Transform (FFT) is performed on the preprocessed current signal to obtain its spectrum. The amplitude and phase parameters of the fundamental, fifth, seventh, eleventh, and thirteenth harmonics are extracted from the spectrum, and the total harmonic distortion (THD) and the content of each harmonic are calculated. These parameters are then combined to construct a current harmonic feature vector, i.e., current harmonic feature data.

[0034] Step 3: Extraction of vibration time-frequency domain features The vibration preprocessing signal underwent time-domain and frequency-domain feature extraction. In the time domain, characteristic parameters including RMS value, peak value, impulse factor, kurtosis index, and waveform factor were calculated. In the frequency domain, the bearing characteristic frequency and the amplitude of its sideband components were obtained through envelope demodulation analysis, and the amplitude of the rotational frequency, harmonics, and their harmonic components were extracted through spectral analysis. The center-of-gravity frequency and mean square frequency were also calculated as spectral characteristic parameters. Finally, the time-domain and frequency-domain characteristic parameters for each axis were combined to generate vibration time-frequency domain characteristic data.

[0035] Step 4: Feature Collaborative Preprocessing Based on the same timestamp from signal acquisition, the current harmonic feature data and vibration time-frequency domain feature data generated in steps two and three are time-stamp aligned. Next, a dynamic time warping algorithm is used to warp the lengths of the two feature sequences and match feature points, ensuring a correspondence between features within the same time period. Finally, the matched current and vibration feature vectors are paired to form a time-aligned set of cooperative feature pairs.

[0036] Step 5: Deep Feature Fusion Processing A set of collaborative feature pairs is obtained, and the current harmonic feature vector and the vibration time-frequency domain feature vector in each pair are concatenated end-to-end to form a primary fused feature vector. This primary fused feature vector is input into a multilayer perceptron neural network, and through its nonlinear transformation and feature dimensionality reduction capabilities, deep abstract features that can simultaneously characterize the specificity of current harmonics and the correlation of vibration state are extracted. These deep abstract features are arranged in chronological order to construct a high-dimensional fused feature matrix.

[0037] Step Six: Preliminary Status Diagnosis Based on Optimized Search First, a standard high-dimensional fusion feature matrix library is established, containing various known health states, load conditions, and fault types of the motor. Then, the high-dimensional fusion feature matrix obtained in step five is compared with each sample in the standard library to calculate a comprehensive similarity based on cosine similarity and Euclidean distance weighting. During this process, the Osprey optimization algorithm is used to optimize this calculation and search process, quickly locating the K nearest neighbor standard samples most similar to the feature to be diagnosed. Finally, the state labels corresponding to these K nearest neighbor samples are counted, and the label with the highest frequency is output as the preliminary diagnostic state type label.

[0038] Step 7: Determining the running status based on the anomaly detection model A sparse autoencoder is constructed as an anomaly detection model, and it is trained unsupervised using high-dimensional fusion feature matrix samples under normal motor conditions to learn the feature reconstruction pattern of normal conditions. The high-dimensional fusion feature matrix to be diagnosed is input into the trained sparse autoencoder to obtain its output reconstruction matrix. The reconstruction error between the original matrix and the reconstruction matrix is ​​calculated. If the error exceeds a preset anomaly threshold, an anomaly status label is generated to indicate the presence of an anomaly; otherwise, a label is generated to indicate that the state is normal.

[0039] Step 8: Refined classification of fault types and generation of diagnostic results Based on the preliminary diagnostic status type labels generated in step six and the abnormal status identification data generated in step seven, a judgment is made. If the abnormal status identification data shows normal, the label representing normal or known load conditions in the preliminary diagnostic status type labels is directly output as the final diagnostic result. If it shows abnormal, the refined classification process is initiated: the high-dimensional fusion feature matrix is ​​input into a pre-trained support vector machine multi-classification model to obtain the specific fault category determination. This refined classification result is logically verified and fused with the preliminary diagnostic status type labels to generate the final target motor and load fault diagnosis result data, which includes fault type, suspected faulty components, and severity level.

[0040] Step Nine: Reliability Assessment and Data Encapsulation of Diagnostic Results The reliability of the target motor and load fault diagnosis results data generated in step eight is evaluated to obtain the corresponding reliability score. Subsequently, the diagnosis results data, reliability score, and the original current harmonic characteristic trend map, vibration spectrum map, and key node data of the feature fusion process are packaged together.

[0041] Step 10: Generation and Output of Visual Diagnostic Reports The data encapsulated in step nine is visualized and rendered to generate a final motor and load collaborative diagnostic report containing diagnostic conclusions, credibility, feature maps, and maintenance suggestions. This report is then output and stored through a human-machine interface.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics, characterized in that: Includes the following steps: S1. Synchronously acquire the three-phase current signal and multi-axial vibration signal of the motor to be diagnosed, and perform signal preprocessing to generate current preprocessing signal and vibration preprocessing signal. S2. Based on the current preprocessing signal, perform current signal harmonic feature extraction processing to generate current harmonic feature data; S3. Based on the vibration preprocessing signal, perform time-domain and frequency-domain feature extraction processing of the vibration signal to generate vibration time-frequency domain feature data; S4. Perform feature co-processing on the current harmonic feature data and the vibration time-frequency domain feature data to generate a time-aligned set of co-feature pairs; S5. Based on the aforementioned set of collaborative features, perform deep fusion processing of current harmonics and vibration features to construct a high-dimensional fusion feature matrix; S6. Based on the high-dimensional fusion feature matrix and the pre-stored standard fusion feature matrix library of motor and load multi-state, the optimal matching state feature is retrieved through a biomimetic intelligent optimization algorithm to generate preliminary diagnostic state type labels. S7. Based on the high-dimensional fusion feature matrix, the abnormal motor operation is initially determined by the abnormal detection model, and abnormal status identification data is generated. S8. Based on the preliminary diagnostic status type label and the abnormal status identifier data, perform refined classification and diagnosis of fault types, and generate fault diagnosis result data for the target motor and load.

2. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 1, characterized in that: The signal preprocessing in S1 includes the following steps: S11. The original three-phase stator current signal of the motor during operation is synchronously collected in the motor power supply circuit by a high-precision current sensor. At the same time, the original axial, radial and tangential vibration signals of the motor during operation are synchronously collected at specific measuring points on the motor bearing housing and the casing by a multi-axis vibration acceleration sensor. S12. Perform preprocessing on the original three-phase stator current signal, including removing DC components, power frequency notch filtering, and low-pass filtering, to generate a current preprocessed signal. S13. Perform preprocessing on the original multi-axial vibration signal, including detrending term, bandpass filtering, and noise reduction, to generate a vibration preprocessed signal.

3. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 2, characterized in that: The current signal harmonic feature extraction and processing in S2 includes the following steps: S21. Perform a fast Fourier transform on the current preprocessed signal to obtain the current signal spectrum; S22. Extract harmonic parameters, including the amplitude and phase of the fundamental, fifth, seventh, eleventh, and thirteenth harmonics, from the current signal spectrum, and calculate the total harmonic distortion rate and the harmonic content characteristics of each harmonic. S23. The extracted harmonic amplitude, phase, total harmonic distortion rate and harmonic content rate are combined to construct a current harmonic feature vector, thereby generating the current harmonic feature data.

4. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 3, characterized in that: The time-domain and frequency-domain feature extraction and processing of the vibration signal in S3 includes the following steps: S31. Extract time-domain features from the vibration preprocessing signal and calculate time-domain feature parameters including RMS value, peak value, impulse factor, kurtosis index, and waveform factor. S32. The vibration preprocessing signal is subjected to frequency domain feature extraction. The bearing characteristic frequency and the amplitude of its sideband components are obtained by envelope demodulation analysis. The amplitude of the rotation frequency, harmonic frequency and its harmonic components are extracted by spectrum analysis. The spectrum feature parameters including the center of gravity frequency and the mean square frequency are calculated. S33. Combine the time-domain feature parameters and frequency-domain feature parameters along the axial dimension to generate the vibration time-frequency domain feature data.

5. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 4, characterized in that: The feature collaborative preprocessing in S4 includes the following steps: S41. Obtain the current harmonic characteristic data and the vibration time-frequency domain characteristic data, and perform time-stamp alignment processing on the two types of characteristic data based on the same timestamp of signal acquisition; S42. For the time-scaled aligned feature data, the dynamic time warping algorithm is used to perform length warping and feature point matching on the current harmonic feature sequence and the vibration time-frequency domain feature sequence. S43. Pair the current harmonic feature vector and the vibration time-frequency domain feature vector that have completed time-scale alignment and feature point matching to form a time-aligned cooperative feature pair set consisting of multiple current feature and vibration feature pairs.

6. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 5, characterized in that: The deep integration processing of current harmonics and vibration characteristics in S5 includes the following steps: S51. Obtain the set of cooperative feature pairs, and concatenate the current harmonic feature vector and the vibration time-frequency domain feature vector in each cooperative feature pair to form a primary fusion feature vector. S52. Input the primary fused feature vector into a multilayer perceptron neural network. The multilayer perceptron neural network performs nonlinear transformation and feature dimensionality reduction on the input features to extract deep abstract features that can simultaneously characterize the specificity of current harmonics and the correlation of vibration state. S53. Arrange the deep abstract features output by the multilayer perceptron neural network in time sequence to construct a high-dimensional fusion feature matrix.

7. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 6, characterized in that: The optimal matching state feature retrieval in S6 includes the following steps: S61. Establish a standard fusion feature matrix library for motors and loads in multiple states. The matrix library stores standard high-dimensional fusion feature matrix samples of motors under various known health states, load conditions and fault types. S62. Calculate the similarity between the high-dimensional fusion feature matrix and each standard sample in the standard fusion feature matrix library. The similarity calculation adopts a comprehensive metric method that weights cosine similarity and Euclidean distance. S63. The similarity calculation process is optimized by using the Osprey optimization algorithm to quickly locate the K nearest neighbor standard samples that are most similar to the high-dimensional fusion feature matrix in the standard library. The osprey optimization algorithm simulates the exploration and development behavior of ospreys, iteratively updating the positions of candidate samples in the search space of the standard feature matrix library until the optimal match is found. S64. Statistically analyze the state labels corresponding to the K nearest neighbor standard samples, and output the state label with the highest frequency as the preliminary diagnostic state type label.

8. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 7, characterized in that: The preliminary determination of abnormal motor operation in S7 includes the following steps: S71. Construct a sparse autoencoder as an anomaly detection model, and use high-dimensional fusion feature matrix samples under normal motor operation to perform unsupervised training on the sparse autoencoder so that it learns the feature reconstruction pattern under normal conditions. S72. Input the high-dimensional fusion feature matrix to be diagnosed into the trained sparse autoencoder to obtain its output reconstruction matrix; S73. Calculate the reconstruction error between the high-dimensional fusion feature matrix and the output reconstruction matrix. When the reconstruction error exceeds a preset anomaly threshold, generate anomaly status identification data indicating the presence of an anomaly. S74. When the reconstruction error does not exceed the abnormal threshold, generate the abnormal state identifier data indicating that the state is normal.

9. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 8, characterized in that: The refined classification and diagnosis of fault types in S8 includes the following steps: S81. Obtain the preliminary diagnosis status type label and the abnormal status identification data. When the abnormal status identification data shows that it is normal, directly output the label in the preliminary diagnosis status type label that represents normal and known load conditions as the target motor and load fault diagnosis result data. S82. When the abnormal status indicator data is displayed as abnormal, the refined classification process is initiated: The high-dimensional fusion feature matrix is ​​input into a pre-trained support vector machine multi-classification model to obtain the specific fault category determination; S83. Logically verify and fuse the refined classification results with the preliminary diagnostic status type labels to generate the final target motor and load fault diagnosis result data, which includes fault type, suspected faulty components, and severity level.

10. The method for diagnosing motors and loads by synergistic fusion of current harmonics and vibration characteristics according to claim 9, characterized in that: The method further includes: S91. Evaluate the reliability of the fault diagnosis results data of the target motor and load. S92. Encapsulate the target motor and load fault diagnosis result data and their corresponding confidence scores together with the original current harmonic characteristic trend diagram, vibration spectrum diagram, and key node data of the feature fusion process. S93. Visualize and render the packaged data to generate a final motor and load collaborative diagnostic report that includes diagnostic conclusions, credibility, feature maps, and maintenance suggestions. Output and store the report through a human-machine interface.

Images