Crane line fault diagnosis system and method based on multi-source data fusion

Abstract

The present application relates to the technical field of crane line fault diagnosis, in particular to a crane line fault diagnosis system and method based on multi-source data fusion, which comprises a data acquisition and processing unit, a mechanical and electrical coupling feature unit and a feature fusion and fault quantization unit, the present application obtains track three-axis vibration acceleration and three-phase current waveform through the data acquisition and processing unit, the mechanical and electrical coupling feature unit performs three-dimensional vector synthesis and wavelet packet decomposition on the vibration data, constructs a space-time correlation matrix of harmonic distortion and vibration, the feature fusion and fault quantization unit outputs coupling factors by using a bidirectional long short-term memory network and an attention mechanism, after time-frequency domain analysis and sample entropy determination, outputs a fault probability value through a dynamic time warping matching algorithm, realizes space-time correlation modeling and dynamic fault matching of multi-source data, and improves fault positioning accuracy and diagnosis accuracy.

Description

Technical Field

[0001] This invention relates to the field of crane circuit fault diagnosis technology, and more specifically, to a crane circuit fault diagnosis system and method based on multi-source data fusion. Background Technology

[0002] Intelligent fault diagnosis of crane circuits is an important technology. In modern industrial logistics and heavy machinery operations, crane circuit fault diagnosis technology is a core means to ensure the safe operation of crane equipment and prevent major accidents by monitoring the operating status of equipment in real time and locating circuit abnormalities.

[0003] With the development of industrial intelligence and the increasing complexity of cranes, constructing a multi-source data diagnostic system integrating mechanical vibration and electrical parameters is crucial for improving fault identification accuracy and reducing unplanned downtime. Traditional single-parameter analysis-based diagnostic methods are no longer sufficient to address the complex diagnostic needs of mechanical-electrical coupled faults. However, traditional crane circuit fault diagnosis technologies suffer from insufficient mechanical-electrical parameter correlation analysis and low coupled fault identification rates. Existing solutions only independently analyze single parameters such as vibration or current, without establishing a spatiotemporal correlation model between mechanical vibration and electrical parameters. When track deformation causes poor contact, the correlation delay between sudden vibration energy changes and current harmonic distortion exceeds 200ms, causing fault location errors to exceed preset thresholds. The traditional processing method cannot effectively separate the multi-frequency characteristics of mechanical vibration from the harmonic components of electrical parameters. The feature extraction accuracy in the 100-200Hz coupled fault frequency band is low, which increases the misjudgment rate of insulation aging and poor contact. In addition, the traditional system lacks a dynamic fault mode matching mechanism. When faced with fault types such as track deformation, insulation aging and poor contact, it cannot adaptively update the fault template, resulting in a low success rate of identifying new fault types. This lack of multi-source data association and insufficient dynamic matching capability ultimately makes the traditional solution have a low comprehensive fault diagnosis accuracy under complex working conditions, which is difficult to meet the high reliability operation requirements of crane equipment. To solve this problem, we provide a crane line fault diagnosis system and method based on multi-source data fusion. Summary of the Invention

[0004] The purpose of this invention is to provide a crane line fault diagnosis system and method based on multi-source data fusion to solve the problems mentioned in the background art.

[0005] Because traditional mechanical-electrical parameter correlation analysis is insufficient, the correlation between vibration and current parameters has a long identification delay and large positioning error, this case study constructs a spatiotemporal correlation matrix by using mechanical-electrical coupling feature units, which can shorten the correlation delay and improve the fault location accuracy.

[0006] Because traditional technologies have low fault identification rates, low feature extraction accuracy in the 100-200Hz frequency band, and high misjudgment rates, this case study uses feature fusion and fault quantization units to perform feature fusion and fault quantization, which can improve feature extraction accuracy and reduce misjudgment rate.

[0007] To achieve the above objectives, a crane circuit fault diagnosis system based on multi-source data fusion is provided, comprising the following units:

[0008] The data acquisition and processing unit processes the vibration acceleration data of the track along the three axes collected by the triaxial accelerometer to generate a vibration vector sequence. It also collects the three-phase current waveform of the power supply line through the current sensor, extracts its harmonic components synchronously, and calculates the total harmonic distortion rate and the odd-even harmonic energy ratio based on the harmonic components.

[0009] The mechanical-electric coupling feature unit performs three-dimensional vector synthesis on the vibration vector sequence to generate the track deformation energy distribution matrix, and maps it to 32 sub-frequency bands through wavelet packet decomposition. Then, the energy proportion of each sub-frequency band is calculated, and the spatiotemporal correlation matrix between harmonic distortion and vibration is constructed based on the energy proportion and the total harmonic distortion rate.

[0010] The feature fusion and fault quantification unit inputs the spatiotemporal correlation matrix and the odd-even harmonic energy ratio into the bidirectional long short-term memory network, dynamically allocates feature weights through an attention mechanism, outputs a coupling factor, performs time-frequency domain joint analysis on the coupling factor to extract intrinsic mode functions, calculates the sample entropy of each intrinsic mode function, determines whether there are abnormal electrical parameters caused by mechanical vibration based on the sample entropy, and outputs the fault probability value based on the determination result using a dynamic time warping matching algorithm.

[0011] As a further improvement to this technical solution, the method for generating the orbital deformation energy distribution matrix through three-dimensional vector synthesis includes:

[0012] The X, Y, and Z axial vibration acceleration data collected by the triaxial accelerometer are used to calculate the square value of each axial acceleration. The square values ​​are integrated and accumulated with a time window length of 100ms and an overlap rate of 50% to obtain the triaxial energy components. The triaxial energy components are superimposed in time series to generate the orbital deformation energy distribution matrix, and the matrix is ​​normalized.

[0013] As a further improvement to this technical solution, the wavelet packet decomposition method includes:

[0014] The orbital deformation energy distribution matrix is ​​decomposed into 5 levels using wavelet basis functions to generate 32 sub-bands. The energy proportion of each sub-band is calculated and sorted from high to low. When the absolute value of the natural logarithm of the ratio of the energy of a sub-band to that of its adjacent band is greater than 1.5, it is identified as an outlier and removed.

[0015] As a further improvement to this technical solution, the method for constructing the spatiotemporal correlation matrix includes:

[0016] Pearson correlation coefficients are calculated for the energy percentage of the retained sub-bands and the total harmonic distortion rate, respectively, generating a 32×1 dimensional correlation coefficient vector. Weight vectors are calculated based on the energy entropy of each sub-band. The correlation coefficient vector and the weight vector are multiplied to generate a spatiotemporal correlation matrix.

[0017] As a further improvement to this technical solution, the bidirectional long short-term memory network first expands the 32-dimensional spatiotemporal correlation matrix into a continuous vector by row and concatenates it with the odd-even harmonic energy ratio to form a 33-dimensional input vector. The forward hidden state is calculated step by step in chronological order, and the state at each time step is determined by the hidden output of the previous time step and the current input vector. At the same time, the backward hidden state is calculated step by step in reverse chronological order, and the state at each time step is determined by the hidden output of the next time step and the current input vector. Finally, the forward and backward hidden states at the same time step are concatenated into a 128-dimensional fusion vector.

[0018] As a further improvement to this technical solution, the attention mechanism first extracts the 128-dimensional fusion vector of all time steps within the historical time window, calculates the correlation strength score between the current fusion vector and each historical vector through the weight matrix and activates it with the hyperbolic tangent function, converts the score into probability distribution weights through exponential function and summation operation, and then generates a context vector representing key historical features by weighted summation of the historical vectors according to the weights. Finally, the current fusion vector and the context vector are concatenated and the coupling factor in the range of 0 to 1 is output through the S-shaped growth curve function.

[0019] As a further improvement to this technical solution, the time-frequency domain joint analysis first identifies all the maxima and minima of the coupling factor time series. Cubic spline interpolation is used to connect the maxima to form the upper envelope and connect the minima to form the lower envelope. The point-by-point arithmetic mean of the upper and lower envelopes is calculated as the mean envelope. The mean envelope is subtracted from the original coupling factor and the selection is repeated until the condition that the difference between the number of extreme points and the number of zero-crossing points does not exceed 1 and the mean of the envelope at any point is zero is met to extract the intrinsic mode functions. Finally, each intrinsic mode function is orthogonally transformed and the instantaneous phase angle change rate is calculated as the instantaneous frequency. The instantaneous frequencies of all mode functions are superimposed to generate a time-frequency energy distribution map.

[0020] As a further improvement to this technical solution, the sample entropy determination first constructs a vector sequence by truncating a continuous subsequence of length 3 from the third intrinsic mode function with a time delay of 5 sampling points, counts the number of matching pairs in all vector pairs whose distance is less than 0.2 times the standard deviation of the sequence, and outputs the sample entropy value based on the matching ratio by performing a natural logarithm operation.

[0021] When the sample entropy value is greater than 1.2 for 3 consecutive seconds, and the time-frequency energy integral value in the 100-200Hz frequency band exceeds 0.35 and the kurtosis statistic of the third intrinsic mode function is greater than 4.5, it is determined that there is an electrical abnormality caused by mechanical vibration.

[0022] As a further improvement to this technical solution, the fault probability value generation first presets a physical constraint template, as follows:

[0023] The orbital deformation template is: the coupling factor monotonically increases and the rate of change is consistently higher than 0.3;

[0024] The insulation aging template is defined as follows: sample entropy less than 0.8 and total harmonic distortion rate greater than 18%.

[0025] The template for poor contact is: the peak value of the time-frequency energy integral in the 50-150Hz frequency band exceeds 0.4;

[0026] The minimum cumulative alignment distance between the sequence under test and each template is calculated using a dynamic programming algorithm. This distance is then mapped to a fault probability value of 0 to 1 using an exponential decay function. When the highest fault probability exceeds 0.8 and the condition for determining that mechanical vibration causes electrical abnormalities is met, the corresponding fault type and probability are output.

[0027] The second objective of this invention is to provide a method for implementing a crane line fault diagnosis system based on multi-source data fusion, comprising the following steps:

[0028] S1. Collect the triaxial vibration acceleration and three-phase current waveforms of the track, calculate the square value of the axial acceleration from the vibration data and generate triaxial energy components by integrating according to a 100ms time window and a 50% overlap rate, extract the harmonic components from the current data and calculate the total harmonic distortion rate and the odd-even harmonic energy ratio.

[0029] S2. The orbital deformation energy distribution matrix is ​​generated by superimposing the three-axis energy components. The wavelet basis function is used to perform a 5-level decomposition to obtain 32 sub-bands. Outlier bands with an absolute value of the natural logarithm of the energy ratio of adjacent bands greater than 1.5 are removed. The Pearson correlation coefficient vector between the energy ratio of the retained sub-bands and the total harmonic distortion rate is calculated. The spatiotemporal correlation matrix is ​​generated by weighting the band energy entropy.

[0030] S3. Expand the spatiotemporal correlation matrix and concatenate it with the odd-even harmonic energy ratio to form a 33-dimensional input vector. Extract forward / backward temporal features through a bidirectional long short-term memory network and output a 128-dimensional fusion vector. Generate a coupling factor by dynamically weighting the historical vector based on the attention mechanism. Extract the first 5 intrinsic mode functions from the empirical mode decomposition of the coupling factor, calculate the sample entropy value of the third intrinsic mode function, and determine anomalies by combining the time-frequency energy integral of 100-200Hz band greater than 0.35 and kurtosis greater than 4.5.

[0031] S4. Define a physical template and use dynamic time warping to calculate the minimum cumulative alignment distance. Map the distance to the fault probability using an exponential decay function. When the highest probability is greater than 0.8, output the fault type and update the template library.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] This invention utilizes a multi-source data fusion mechanism to correlate triaxial vibration acceleration and three-phase current waveform data, constructing a spatiotemporal correlation matrix. This addresses the shortcomings of traditional techniques in correlating mechanical and electrical parameters, shortens the correlation delay between vibration energy mutations and current harmonic distortion, and improves fault location accuracy. Wavelet packet decomposition separates multi-frequency band features of mechanical vibration from harmonic components of electrical parameters. Combined with a bidirectional long short-term memory network and attention mechanism, feature fusion improves the feature extraction accuracy of the 100–200 Hz coupled fault frequency band and reduces the misjudgment rate of track deformation, insulation aging, and poor contact. Furthermore, a dynamic time warping matching algorithm adaptively updates the fault template based on a preset physical constraint template, improving the success rate of identifying new fault types and enhancing the overall fault diagnosis accuracy under complex operating conditions. Attached Figure Description

[0034] Figure 1 This is an overall block diagram of the present invention;

[0035] Figure 2 This is the overall flowchart of the present invention.

[0036] The meanings of the labels in the diagram are as follows:

[0037] 1. Data acquisition and processing unit; 2. Mechanical and electrical coupling feature unit; 3. Feature fusion and fault quantification unit. Detailed Implementation

[0038] 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.

[0039] This invention provides a crane circuit fault diagnosis system based on multi-source data fusion. Please refer to [link / reference]. Figure 1 As shown, it includes the following units:

[0040] The data acquisition and processing unit 1 processes the vibration acceleration data of the track along the three axes acquired by the triaxial accelerometer to generate a vibration vector sequence. It also acquires the three-phase current waveform of the power supply line through the current sensor, extracts its harmonic components synchronously, and calculates the total harmonic distortion rate and the odd-even harmonic energy ratio based on the harmonic components.

[0041] Mechanical-electric coupling feature unit 2 performs three-dimensional vector synthesis on the vibration vector sequence to generate the track deformation energy distribution matrix, and maps it to 32 sub-frequency bands through wavelet packet decomposition. Then, the energy proportion of each sub-frequency band is calculated, and the spatiotemporal correlation matrix between harmonic distortion and vibration is constructed based on the energy proportion and the total harmonic distortion rate.

[0042] To convert triaxial vibration acceleration data into energy characteristics that reflect track deformation, the mechanical-electrical coupling feature unit 2 needs to perform three-dimensional vector synthesis on the vibration data to generate a track deformation energy distribution matrix. The specific implementation method is as follows:

[0043] Methods for generating orbital deformation energy distribution matrices through three-dimensional vector synthesis include:

[0044] Since the square of vibration acceleration is positively correlated with energy, the square of the X, Y, and Z axial vibration acceleration data collected by the triaxial accelerometer is calculated separately. The squares are integrated and accumulated using a sliding window with a time window length of 100 ms and an overlap rate of 50%. Each time window contains 100 milliseconds of squared data (assuming a sampling frequency of 1000 Hz, each window contains 100 data points). The sum of the squares within each window yields the energy component for that window. When the next window slides, the overlap time is 50 milliseconds to ensure the continuity of the energy characteristics. Through this operation, the energy components of the three axes are obtained, transforming the time-domain vibration signal into energy characteristics, making it easier to identify energy abrupt changes caused by track deformation. To comprehensively reflect the deformation energy distribution of the track in three-dimensional space, the energy components of the three axes are superimposed in a time sequence; specifically, time is the row dimension and the axes are the column dimension. The system constructs a two-dimensional matrix. For example, at a certain moment, the energy component on the X-axis is 0.8 g²·ms, on the Y-axis it is 0.6 g²·ms, and on the Z-axis it is 1.0 g²·ms. The row vector corresponding to this time point in the matrix is ​​[0.8, 0.6, 1.0]. As time progresses, the matrix is ​​filled row by row to generate the orbital deformation energy distribution matrix. The number of rows in this matrix corresponds to the number of time windows, and the number of columns is 3, corresponding to the three axes. This visualizes the spatiotemporal characteristics of vibration energy. The matrix is ​​then normalized. Since the acceleration ranges of different axes may differ, the generated orbital deformation energy distribution matrix is ​​normalized to eliminate the influence of dimensions and facilitate subsequent analysis. The specific method is to calculate the maximum value of all elements in the matrix and divide each element by this maximum value, so that the value range of the matrix elements is mapped to the interval [0,1]. After normalization, the energy characteristics of each axis are comparable, and it is convenient to improve the accuracy of subsequent operations such as wavelet packet decomposition.

[0045] In order to separate the key frequency band features characterizing mechanical faults from the track deformation energy distribution matrix, the mechanical-electric coupling feature unit 2 needs to perform wavelet packet decomposition on the matrix and remove outlier frequency bands. The specific implementation method is as follows:

[0046] Wavelet packet decomposition methods include:

[0047] Since wavelet packet decomposition can perform full-band subdivision of signals, a suitable wavelet basis function for mechanical vibration signal analysis is first selected. For each row vector of the track deformation energy distribution matrix, corresponding to the triaxial energy components of each time window, the wavelet basis function is introduced to perform a 5-level decomposition of the track deformation energy distribution matrix, generating 32 sub-bands. The first level of decomposition decomposes the energy signal of each row vector into low-frequency approximate components and high-frequency detail components, corresponding to the overall trend and local changes of the signal, respectively. The second to fifth levels of decomposition continue to recursively decompose the low-frequency and high-frequency components of each level until the fifth level generates 32 sub-bands. For example, the second level decomposes each of the two components of the first level into two components, resulting in four sub-bands. The third level further decomposes them into eight, and so on, until the fifth level forms 32 equal-width bands, each corresponding to a specific frequency range. This decomposition process separates the frequency domain characteristics of the original energy signal layer by layer through multi-resolution analysis. After the decomposition is completed, the energy of each sub-band is calculated. The energy ratio is used to measure importance. The sum of squares of all energy data points within each sub-band is calculated to obtain the energy value of that band. This value is then divided by the total energy of all sub-bands to obtain the energy ratio. The 32 sub-bands are then sorted from highest to lowest energy value, highlighting the main bands with concentrated energy. The total energy ratio of the top 10 high-frequency sub-bands typically exceeds 60%, providing key analysis targets for subsequent fault feature extraction. To eliminate interference from noise or irrelevant bands, outlier bands are removed using an energy ratio threshold. For each sorted sub-band, the energy ratio between each band and its adjacent bands is calculated, and the absolute value of the natural logarithm is taken. When the absolute value of the natural logarithm of the energy ratio of a sub-band to its adjacent bands is greater than 1.5, it is identified as an outlier and removed. This mechanism effectively filters out abnormal bands caused by environmental interference or sensor noise, improving the signal-to-noise ratio of subsequent feature extraction and providing a reliable frequency domain feature basis for the joint diagnosis of mechanical and electrical faults.

[0048] To construct a feature matrix that can characterize the spatiotemporal correlation between mechanical vibration and electrical parameters, the mechanical-electrical coupling feature unit 2 needs to generate a spatiotemporal correlation matrix based on the retained sub-band energy proportion and total harmonic distortion rate through Pearson correlation coefficient analysis and energy entropy weighting. The specific implementation method is as follows:

[0049] Methods for constructing spatiotemporal correlation matrices include:

[0050] Pearson correlation coefficients were calculated for the energy proportion of the retained sub-bands and the total harmonic distortion rate (THD), generating a 32×1 dimensional correlation coefficient vector. The Pearson correlation coefficient is a statistic that measures the degree of linear correlation between two variables, ranging from -1 to 1. A value closer to 1 indicates a stronger positive correlation, closer to -1 indicates a stronger negative correlation, and closer to 0 indicates a weak linear correlation. In this invention, this coefficient is used to quantify the degree of linear correlation between the energy proportion of the retained sub-bands and the THD, revealing the potential link between mechanical vibration energy distribution and abnormal electrical parameters. For each of the 32 retained sub-bands after wavelet packet decomposition, their energy proportions were calculated separately. The Pearson correlation coefficient between the energy proportion sequence and the total harmonic distortion (THD) sequence is calculated. Specifically, taking a sub-band as an example, energy proportion data and corresponding THD data are obtained within multiple time windows. The correlation coefficient between the sub-band and THD is obtained by calculating the ratio of their covariance to standard deviation. Finally, a 32×1 dimensional correlation coefficient vector is generated. This step utilizes existing technology to perform correlation analysis of mechanical and electrical parameters, quantifying the coupling relationship between vibration energy and current harmonics, and providing a correlation feature basis for the subsequent spatiotemporal correlation matrix. To measure the uncertainty of energy distribution in each sub-band, the energy entropy of each sub-band is calculated and a weight vector is generated. The energy entropy is calculated as follows: For the energy proportion sequence of a certain sub-band, first calculate the probability distribution of each energy proportion value, and then obtain the energy entropy of the frequency band by weighted sum of negative log probabilities. The larger the energy entropy value, the more complex the energy distribution of the frequency band, and the richer the fault information it may contain, so it should be given a higher weight. This method generates a 32×1 dimensional weight vector with weight values ​​ranging from 0 to 1, giving higher weights to sub-bands with high energy entropy and improving the ability to represent fault characteristics. The correlation coefficient vector and the weight vector are multiplied to generate a spatiotemporal correlation matrix. The 32×1 dimensional correlation coefficient vector and the weight vector are then multiplied to obtain each... The weighted correlation coefficients corresponding to the sub-frequency bands ultimately generate a 32×1 dimensional spatiotemporal correlation matrix. Each element of this matrix represents the weighted correlation between the energy proportion of the corresponding sub-frequency band and the total harmonic distortion rate, reflecting the spatiotemporal correlation strength between mechanical vibration energy and electrical parameter anomalies in the frequency domain. This needs to be analyzed in detail in subsequent fault diagnosis. By fusing correlation features and energy entropy weights, the spatiotemporal correlation matrix cross-domain correlates the frequency domain features of mechanical vibration with the time domain features of electrical parameters. Compared with single correlation coefficient analysis, it provides a more comprehensive characterization dimension for mechanical-electrical coupling faults and provides structured input data for subsequent feature fusion of bidirectional long short-term memory networks.

[0051] The feature fusion and fault quantification unit 3 inputs the spatiotemporal correlation matrix and the odd-even harmonic energy ratio into the bidirectional long short-term memory network, dynamically allocates feature weights through an attention mechanism, outputs a coupling factor, performs time-frequency domain joint analysis on the coupling factor to extract intrinsic mode functions, calculates the sample entropy of each intrinsic mode function, determines whether there are abnormal electrical parameters caused by mechanical vibration based on the sample entropy, and outputs the fault probability value based on the determination result using a dynamic time warping matching algorithm.

[0052] To extract temporal features from the spatiotemporal correlation matrix and the odd-even harmonic energy ratio, feature fusion and fault quantization unit 3 employs a bidirectional long short-term memory network for temporal modeling. The specific implementation method is as follows:

[0053] Since the bidirectional long short-term memory network requires a one-dimensional continuous vector as input, it first expands the 32-dimensional spatiotemporal correlation matrix into a one-dimensional continuous vector. Each row of this matrix corresponds to a weighted correlation coefficient of a retained sub-band. After expansion, a feature sequence reflecting the frequency domain correlation strength is formed. This sequence is then concatenated with the odd-even harmonic energy ratio to form a 33-dimensional input vector. The odd-even harmonic energy ratio is a one-dimensional feature, representing the ratio of odd-order harmonic energy to even-order harmonic energy. The odd-even harmonic energy ratio reflects the distortion characteristics of the current waveform (e.g., insulation aging leads to an increase in the proportion of odd-order harmonics), complementing the mechanical vibration frequency domain features of the spatiotemporal correlation matrix. After concatenation, the feature fusion of "mechanical frequency domain - electrical time domain" is achieved. The 33-dimensional vector design balances feature integrity and computational efficiency. Compared to directly inputting the matrix (32×1) and processing odd-even harmonics separately, it improves feature interactivity. The 33-dimensional input vector sequence is processed in chronological order. The forward hidden state calculation logic for each time step is as follows:

[0054] During network initialization, the forward hidden state is a zero-dimensional vector with a dimension of 64. The forward hidden state is calculated sequentially over time. The state at each time step is determined by the hidden output of the previous time step and the current input vector. For the input vector at time step t, the forward hidden state is determined by the forward hidden output of the previous time step (t-1) and the current input vector. Specifically, information filtering and updating are performed through forget gates, input gates, and output gates within the network. Historical information related to fault features is retained, while irrelevant noise is forgotten. The output of the forward hidden state at each time step is a 64-dimensional vector to ensure the dimensionality consistency of temporal features. The input sequence is processed synchronously in reverse chronological order to capture the influence of future time steps on the current state. The specific steps are as follows: Starting from the last time step, the backward hidden state is also initially a zero-dimensional vector with a dimension of 64. The backward hidden state at time step t is determined by the backward hidden output of the next time step (t+1) and the current input vector. The output of the backward hidden state at each time step is a 64-dimensional vector, matching the dimension of the forward hidden state. Simultaneously, the forward hidden state is processed sequentially in reverse chronological order. The backward hidden state is calculated step by step. The state at each time step is determined by the hidden output of the next time step and the current input vector. Finally, the forward and backward hidden states at the same time step are concatenated into a 128-dimensional fusion vector. The forward and backward hidden states at the same time step are directly concatenated along the dimensional direction to generate a 128-dimensional fusion vector. This vector contains both the mechanical-electrical correlation features of the past time step and the trend features of the future time step. For example, the combination of the track deformation frequency band features in the forward hidden state and the odd and even harmonic rising trend in the backward hidden state can predict the development direction of the poor contact fault. The concatenation operation preserves the independence of the bidirectional temporal features. Compared with simple weighted summation, it improves the retention rate of the temporal evolution details of the fault features. This bidirectional long short-term memory network processing flow forms a technical closed loop with the spatiotemporal correlation matrix construction mentioned above: through bidirectional temporal feature extraction, the cross-domain correlation features of mechanical vibration and electrical parameters are transformed into a fusion vector containing temporal context, providing rich temporal information for the feature weight allocation of the subsequent attention mechanism and laying the foundation for the accurate output of the coupling factor.

[0055] To filter key historical features from the 128-dimensional fusion vector output by the bidirectional long short-term memory network, the feature fusion and fault quantification unit 3 uses an attention mechanism to dynamically allocate feature weights and generate a coupling factor characterizing the degree of mechanical-electrical coupling. The specific implementation is as follows:

[0056] First, the length of the historical time window is determined. The attention mechanism extracts 128-dimensional fusion vectors from all time steps within the historical time window, forming a historical feature set. This historical feature set serves as the memory of the attention mechanism, storing historical fault clues related to the current state, providing a data foundation for calculating the association strength. A trainable weight matrix is ​​used to map the current fusion vector to each historical fusion vector. Specifically, the current vector and historical vectors are concatenated column-wise to form a 256-dimensional vector. Multiplication by the weight matrix yields the raw association strength score between each historical vector and the current vector. This raw score is then input into a hyperbolic tangent function for activation. This function's output range is -1 to 1, mapping the score to a symmetrical interval, thus enhancing the ability to distinguish weakly associated features.The physical meaning of the activation process is as follows: when the original score is positive, the larger the value after activation, the stronger the positive correlation between the current feature and historical features; when the original score is negative, the smaller the value after activation, the stronger the negative correlation. The nonlinearity of the function can amplify the differences in strong correlation scores and compress the range of weak correlation scores, making the weight allocation of key features more prominent. The score is transformed into a probability distribution weight through an exponential function and summation operation. Then, based on this weight, a weighted summation of the historical vector is generated to produce a context vector representing key historical features. The activated correlation strength score is then expressed through an exponential function. The function is converted to a non-negative value, and then the exponential scores of all historical vectors are summed. Each exponential score is divided by the sum to obtain the probability distribution weight. This weight reflects the proportion of each historical vector's contribution to the current state, and its value ranges from 0 to 1, with the sum of all weights being 1. The historical fusion vectors are then weighted and summed according to the probability distribution weights to obtain a 64-dimensional context vector. This vector aggregates key features within the historical time window. For example, if strong correlation features in the 100-200Hz frequency band appear multiple times in the historical time window, corresponding to orbital deformation, the context vector will reinforce that frequency band. The feature representation uses a weighted summation operation, which is equivalent to selectively memorizing historical features. Features with high weights are retained, while those with low weights are weakened. Finally, the current fusion vector and the context vector are concatenated, and the coupling factor in the range of 0 to 1 is output through an S-curve growth function. The current fusion vector (128-dimensional) and the context vector (64-dimensional) are concatenated along the dimensional direction to form a 192-dimensional vector, which is then mapped to 1 dimension through a fully connected layer and input into the S-curve growth function (logic sigmoid function) to compress the output value to the range of 0 to 1, generating the final vector. The coupling factor, whose physical meaning is that the closer the value is to 1, the higher the coupling degree between mechanical vibration and electrical parameters, and the more likely there is a related fault; the closer the value is to 0, the lower the coupling degree and the normal operating state of the system. This attention mechanism processing flow is linked with the bidirectional long short-term memory network forming technology mentioned earlier: by dynamically weighting historical time-series features, it highlights the key clues of mechanical-electrical coupling faults, and improves the sensitivity of the coupling factor to coupling faults in the 100-200Hz frequency band, providing more accurate input features for subsequent time-frequency domain analysis and fault probability calculation.

[0057] In order to extract the essential features reflecting mechanical-electrical coupling faults from the coupling factors, the feature fusion and fault quantification unit 3 needs to perform joint time-frequency domain analysis on the coupling factors, extract intrinsic mode functions, and generate time-frequency energy distribution maps. The specific implementation method is as follows:

[0058] Since the distribution of extreme points of the coupling factor reflects the fluctuation characteristics of the signal, the joint time-frequency domain analysis first identifies all the maxima and minima of the coupling factor's time series. Then, cubic spline interpolation is used to connect the maxima to form the upper envelope and the minima to form the lower envelope. The maxima are then fitted with a cubic spline interpolation function to form a smooth upper envelope, which reflects the upper boundary fluctuation trend of the signal. Similarly, cubic spline interpolation is performed on the minima to generate the lower envelope, characterizing the lower boundary fluctuation characteristics of the signal. The smoothness of cubic spline interpolation avoids sawtooth fluctuations in the envelope. Compared to linear interpolation, the envelope more accurately reflects the true edge of the signal. The reduced fitting error of the boundary provides an accurate basis for subsequent mean envelope calculation. The point-by-point arithmetic mean of the upper and lower envelopes is calculated as the mean envelope. The mean envelope is subtracted from the original coupling factor to obtain the initially screened signal. To ensure that the screened signal meets the conditions of the intrinsic mode function (the difference between the number of extreme points and the number of zero-crossing points does not exceed 1, and the mean of the envelope at any point is zero), the screening needs to be repeated until the conditions of the difference between the number of extreme points and the number of zero-crossing points not exceeding 1 and the mean of the envelope at any point are met, thus extracting the intrinsic mode function. This iterative process can effectively separate different frequency components in the signal, making the physical meaning of the intrinsic mode function clearer. The extraction of numbers employs an improved empirical mode decomposition (EMD) method. Iterative screening ensures that each modal function reflects the vibration characteristics of a single frequency component. Unlike traditional EMD methods, this approach introduces an energy threshold constraint during the screening process, considering the mechanical-electrical coupling characteristics of the coupling factor: if the signal energy attenuation exceeds 30% of the current total energy in a screening step, screening is forcibly stopped to avoid over-decomposition leading to mode aliasing. Finally, each intrinsic mode function undergoes an orthogonal transformation, and the instantaneous phase angle change rate is calculated as the instantaneous frequency. To eliminate the correlation between intrinsic mode functions and improve the accuracy of time-frequency analysis, each intrinsic mode function... The orthogonalization transformation of the modal functions involves the following steps: sorting the intrinsic modal functions by energy proportion from high to low, prioritizing the processing of high-energy modes, which correspond to the main fault characteristics; for the i-th modal function, calculating its inner product with the previous i-1 orthogonalized modal functions, and eliminating correlation through linear combination; after orthogonalization, normalizing the amplitude of each modal function to ensure consistent energy scale across modes. Compared to traditional orthogonalization, this method introduces adaptive step size adjustment (the step size dynamically changes with the modal energy proportion) to address the non-stationary characteristics of mechanical-electrical coupling signals, reducing the correlation coefficient between orthogonalized modal functions from 0.3 to 0.Below 1, the independence of time-frequency features is significantly improved. A time-frequency energy distribution map is generated by superimposing the instantaneous frequencies of all modal functions. For each orthogonalized intrinsic modal function, the rate of change of its instantaneous phase angle is calculated as the instantaneous frequency. Specifically, the instantaneous phase of the modal function is obtained through Hilbert transform, and the instantaneous frequency is obtained by differentiating the phase. This frequency value reflects the local vibration frequency of the modal function at each time point. The instantaneous frequencies of all intrinsic modal functions are superimposed in a time series to generate a time-frequency energy distribution map. The horizontal axis represents time, and the vertical axis represents frequency. The color intensity represents energy intensity, which is determined by the square of the modal function's amplitude. For example, a concentrated energy region in the 100-200Hz frequency band within a certain time period, indicated by a darker color, suggests the presence of characteristic frequency components of mechanical-electrical coupling faults during that period. This distribution map visualizes the time-frequency characteristics of the coupling factor, improving the ability to present the time-frequency evolution of non-steady-state faults compared to single spectrum analysis, providing intuitive feature basis for subsequent sample entropy determination and fault probability calculation.

[0059] To accurately determine whether mechanical vibration causes abnormal electrical parameters, the feature fusion and fault quantification unit 3 needs to make a joint determination based on the sample entropy value of the third intrinsic mode function, the time-frequency energy integral, and the kurtosis statistics. The specific implementation method is as follows:

[0060] Since the third eigenmode function typically corresponds to the characteristic frequency components of mechanical-electrical coupling faults, the sample entropy determination first constructs a vector sequence by truncating a continuous subsequence of length 3 from the third eigenmode function with a time delay of 5 sampling points. This delayed sampling method avoids the high correlation between adjacent subsequences, and compared to zero-delay sampling, the independence of sample entropy calculation is improved. The distance between all vector pairs in the vector sequence is calculated, and the number of matches where the distance is less than 0.2 times the sequence standard deviation is counted. The vector distance is calculated using Euclidean distance, i.e., the square root of the sum of the squares of the differences in each dimension, and the standard deviation is the standard deviation of the entire vector sequence. Subsequently, the natural logarithm is applied based on the matching ratio (the ratio of the number of matches to the total number of vector pairs). The calculation outputs the sample entropy value. The larger the sample entropy value, the higher the complexity and irregularity of the signal, and the more likely a fault is to exist. Based on the matching ratio, the sample entropy value is output through natural logarithm calculation. When the sample entropy value is greater than 1.2 for 3 seconds, it indicates that the signal complexity of the third intrinsic mode function is abnormally increased. The duration of 3 seconds is set based on the development cycle of typical crane faults, such as the evolution of track deformation from slight to severe, which usually exceeds 3 seconds, to avoid misjudgment caused by instantaneous interference. The threshold of 1.2 is determined by analyzing 100 sets of normal and faulty operating condition data. The time-frequency energy in the 100-200Hz frequency band in the time-frequency energy distribution map is integrated to obtain the energy integral value of this frequency band. The integration method is: for each time on the time axis... The average energy integral value is obtained by summing the energy values ​​in the 100-200Hz frequency band and dividing by the time length. When this value exceeds 0.35, it indicates that the energy of mechanical vibration is abnormally concentrated in this fault characteristic frequency band. The threshold of 0.35 is determined through calibration experiments. Kurtosis statistics are used to measure the steepness of the signal amplitude distribution. The ratio of the fourth central moment to the square of the variance of all data points in the third intrinsic mode function is calculated. When the kurtosis statistics of the third intrinsic mode function is greater than 4.5, it indicates that there is a significant impulse component in the signal (the kurtosis of normal signals is mostly close to 3), indicating that there is an electrical anomaly caused by mechanical vibration. When the sample entropy value is greater than 1.2 for 3 seconds and the time-frequency energy integral in the 100-200Hz frequency band exceeds 0.35, it indicates that there is an electrical anomaly caused by mechanical vibration. Furthermore, when all three conditions are met simultaneously—the kurtosis of the third intrinsic mode function being greater than 4.5—an electrical parameter anomaly caused by mechanical vibration is determined to exist. This joint determination mechanism avoids the risk of misjudgment based on a single feature. The sample entropy value reflects the signal complexity and eliminates the interference of stationary noise. Time-frequency energy integration locates the fault characteristic frequency band, ensuring that the anomaly is related to mechanical vibration. The kurtosis statistic captures the impact component and verifies the abrupt change characteristics of the electrical parameters. This sample entropy joint determination process forms a technical closed loop with the time-frequency domain analysis mentioned above: through multi-dimensional analysis of the complexity of the third intrinsic mode function, frequency domain energy distribution, and amplitude characteristics, the coupling relationship between mechanical vibration and electrical anomaly is quantified, providing a reliable basis for subsequent fault probability calculation.

[0061] To transform the mechanical-electrical coupling characteristics into specific fault probability values, the feature fusion and fault quantification unit 3 needs to calculate and output the fault probability based on a preset physical constraint template, using a dynamic programming algorithm and an exponential decay function. The specific implementation method is as follows:

[0062] The generation of fault probability values ​​begins with a preset physical constraint template, as follows:

[0063] The track deformation template is as follows: the coupling factor increases monotonically and the rate of change is consistently higher than 0.3. The monotonically increasing coupling factor reflects the development process of track deformation from slight to severe, and the rate of change is higher than 0.3, indicating a faster deformation rate. The physical basis of this template is that track deformation will cause the vibration energy of the crane to gradually increase during operation, which in turn will lead to a continuous increase in the mechanical-electrical coupling degree. The monotonically increasing characteristic of the coupling factor is highly consistent with the development law of track deformation.

[0064] The insulation aging template is defined as follows: sample entropy less than 0.8 and total harmonic distortion rate (THD) greater than 18%. Sample entropy less than 0.8 indicates low signal complexity (i.e., the current waveform tends to be regular), while THD greater than 18% reflects an abnormal increase in current harmonic content. The physical significance of this template is that insulation aging leads to changes in line impedance, increases the odd harmonic components in the current waveform, and reduces signal irregularity. This has been verified by 100 sets of insulation aging fault data.

[0065] The template for poor contact is: the peak value of the time-frequency energy integral in the 50-150Hz frequency band exceeds 0.4. This frequency band corresponds to the characteristic frequency of the crane contact resistance fluctuation (such as the mechanical vibration frequency caused by the intermittent disconnection of the contact point). The peak value of the energy integral exceeding 0.4 indicates that there is significant energy concentration in this frequency band. Physically, poor contact will cause mechanical impact at the moment of current switching, causing a sudden change in vibration energy in the 50-150Hz frequency band.

[0066] A dynamic programming algorithm is used to calculate the minimum cumulative alignment distance between the test sequence and each template. This distance is then mapped to a fault probability value between 0 and 1 using an exponential decay function. When the highest fault probability exceeds 0.8 and the condition of mechanical vibration causing electrical anomalies is met, the corresponding fault type and probability are output. The test feature sequence (such as the coupling factor sequence of the current 30 seconds) and the template sequence (preset fault feature sequence) are unified in terms of time dimension. Linear interpolation is used to adjust the number of time points in the test sequence and the template sequence to be consistent, ensuring the feasibility of subsequent alignment calculations. A two-dimensional dynamic programming matrix is ​​constructed, where the rows of the matrix correspond to the time points of the template sequence, and the columns correspond to the time points of the test sequence. In this context, the matrix elements represent the cumulative alignment distance between the first i time points of the template sequence and the first j time points of the test sequence. Initially, the top-left element of the matrix is ​​zero, and the remaining boundary elements are infinity. During the iterative calculation, for each matrix element, the distance between the current template point and the test point is calculated, and the minimum value of the left, top, or top-left elements is added as the cumulative distance of the current element. After iteration, the bottom-right element of the matrix represents the minimum cumulative alignment distance between the template sequence and the test sequence. This distance reflects the degree of matching between the test feature and the template feature; a smaller distance indicates a higher degree of matching. The minimum cumulative distance is mapped to a fault probability value between 0 and 1 using an exponential decay function, as shown in the formula. ,in The attenuation coefficient, with its exponential decay characteristic, allows even small changes in distance to significantly affect the probability value, improving fault discrimination. The fault probability values ​​for the test sequence are calculated for three preset templates, yielding the probability of track deformation, insulation aging, and poor contact. When the highest fault probability exceeds 0.8 and meets the abnormal conditions for sample entropy judgment (sample entropy greater than 1.2 for 3 seconds, energy integral of 100-200Hz exceeding 0.35, and kurtosis greater than 4.5), the corresponding fault type and probability are output. When the number of valid diagnoses for a fault type exceeds 50, the system automatically updates the corresponding template with new typical fault data. The update method is to calculate the mean of the feature sequences of that fault type in 50 diagnoses, using it as the new template sequence. This fault probability value generation process forms a technical closed loop with the sample entropy judgment described earlier. By physically constraining the template, abstract mechanical-electrical characteristics are transformed into specific fault probabilities. Combined with the abnormal judgment conditions of sample entropy, the diagnostic results of crane line faults have clear engineering guidance significance. Compared with traditional threshold alarms, this improves the quantitative accuracy of fault probability output, providing a more precise basis for equipment maintenance decisions.

[0067] This invention acquires the triaxial vibration acceleration and three-phase current waveforms of the track through a data acquisition and processing unit 1. A mechanical-electrical coupling feature unit 2 performs three-dimensional vector synthesis and wavelet packet decomposition on the vibration data to construct a spatiotemporal correlation matrix between harmonic distortion and vibration. A feature fusion and fault quantification unit 3 outputs coupling factors using a bidirectional long short-term memory network and an attention mechanism. After time-frequency domain analysis and sample entropy determination, a fault probability value is output through a dynamic time warping matching algorithm. This invention realizes spatiotemporal correlation modeling and dynamic fault matching of multi-source data, improving fault location accuracy and diagnostic accuracy.

[0068] Please see Figure 2 As shown, the second objective of this invention is to provide a method for implementing a crane line fault diagnosis system based on multi-source data fusion, comprising the following steps:

[0069] S1. Collect the triaxial vibration acceleration and three-phase current waveforms of the track, calculate the square value of the axial acceleration from the vibration data and generate triaxial energy components by integrating according to a 100ms time window and a 50% overlap rate, extract the harmonic components from the current data and calculate the total harmonic distortion rate and the odd-even harmonic energy ratio.

[0070] S2. The orbital deformation energy distribution matrix is ​​generated by superimposing the three-axis energy components. The wavelet basis function is used to perform a 5-level decomposition to obtain 32 sub-bands. Outlier bands with an absolute value of the natural logarithm of the energy ratio of adjacent bands greater than 1.5 are removed. The Pearson correlation coefficient vector between the energy ratio of the retained sub-bands and the total harmonic distortion rate is calculated. The spatiotemporal correlation matrix is ​​generated by weighting the band energy entropy.

[0071] S3. Expand the spatiotemporal correlation matrix and concatenate it with the odd-even harmonic energy ratio to form a 33-dimensional input vector. Extract forward / backward temporal features through a bidirectional long short-term memory network and output a 128-dimensional fusion vector. Generate a coupling factor by dynamically weighting the historical vector based on the attention mechanism. Extract the first 5 intrinsic mode functions from the empirical mode decomposition of the coupling factor, calculate the sample entropy value of the third intrinsic mode function, and determine anomalies by combining the time-frequency energy integral of 100-200Hz band greater than 0.35 and kurtosis greater than 4.5.

[0072] S4. Define a physical template and use dynamic time warping to calculate the minimum cumulative alignment distance. Map the distance to the fault probability using an exponential decay function. When the highest probability is greater than 0.8, output the fault type and update the template library.

[0073] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A crane line fault diagnosis system based on multi-source data fusion, characterized in that, Includes the following units: The data acquisition and processing unit (1) processes the vibration acceleration data of the track along the three axes collected by the triaxial accelerometer to generate a vibration vector sequence, collects the three-phase current waveform of the power supply line through the current sensor, extracts its harmonic components synchronously, and calculates the total harmonic distortion rate and the odd-even harmonic energy ratio based on the harmonic components. The mechanical-electric coupling feature unit (2) performs three-dimensional vector synthesis on the vibration vector sequence to generate the track deformation energy distribution matrix, and maps it to 32 sub-bands through wavelet packet decomposition. Then, it calculates the energy ratio of each sub-band and constructs the spatiotemporal correlation matrix between harmonic distortion and vibration based on the energy ratio and the total harmonic distortion rate. The feature fusion and fault quantification unit (3) inputs the spatiotemporal correlation matrix and the odd-even harmonic energy ratio into the bidirectional long short-term memory network, dynamically allocates feature weights through the attention mechanism, outputs coupling factors, performs time-frequency domain joint analysis on the coupling factors to extract intrinsic mode functions, calculates the sample entropy of each intrinsic mode function, determines whether there are abnormal electrical parameters caused by mechanical vibration based on the sample entropy, and outputs the fault probability value using the dynamic time warping matching algorithm based on the determination result.

2. The overhead crane line fault diagnosis system based on multi-source data fusion according to claim 1, characterized in that: The method for generating the orbital deformation energy distribution matrix by three-dimensional vector synthesis includes: The X, Y, and Z axial vibration acceleration data collected by the triaxial accelerometer are used to calculate the square value of each axial acceleration. The square values ​​are integrated and accumulated with a time window length of 100ms and an overlap rate of 50% to obtain the triaxial energy components. The triaxial energy components are superimposed in time series to generate the orbital deformation energy distribution matrix, and the matrix is ​​normalized.

3. The overhead crane line fault diagnosis system based on multi-source data fusion according to claim 1, characterized in that, The wavelet packet decomposition method includes: The orbital deformation energy distribution matrix is ​​decomposed into 5 levels using wavelet basis functions to generate 32 sub-bands. The energy proportion of each sub-band is calculated and sorted from high to low. When the absolute value of the natural logarithm of the ratio of the energy of a sub-band to that of its adjacent band is greater than 1.5, it is identified as an outlier and removed.

4. The hoist line fault diagnosis system based on multi-source data fusion according to claim 3, characterized in that: The method for constructing the spatiotemporal correlation matrix includes: Pearson correlation coefficients are calculated for the energy percentage of the retained sub-bands and the total harmonic distortion rate, respectively, generating a 32×1 dimensional correlation coefficient vector. Weight vectors are calculated based on the energy entropy of each sub-band. The correlation coefficient vector and the weight vector are multiplied to generate a spatiotemporal correlation matrix.

5. The hoist line fault diagnostic system based on multi-source data fusion according to claim 4, characterized in that: The bidirectional long short-term memory network first expands the 32-dimensional spatiotemporal correlation matrix into a continuous vector by rows and concatenates it with the odd-even harmonic energy ratio to form a 33-dimensional input vector. The forward hidden state is calculated step by step in chronological order, and the state at each time step is determined by the hidden output of the previous time step and the current input vector. At the same time, the backward hidden state is calculated step by step in reverse chronological order, and the state at each time step is determined by the hidden output of the next time step and the current input vector. Finally, the forward and backward hidden states at the same time step are concatenated into a 128-dimensional fusion vector.

6. The crane circuit fault diagnosis system based on multi-source data fusion according to claim 5, characterized in that: The attention mechanism first extracts 128-dimensional fusion vectors from all time steps within the historical time window. It then calculates the correlation strength score between the current fusion vector and each historical vector using a weight matrix and activates it with a hyperbolic tangent function. The score is then converted into probability distribution weights using an exponential function and a summation operation. Based on these weights, the historical vectors are weighted and summed to generate a context vector representing key historical features. Finally, the current fusion vector and the context vector are concatenated and a coupling factor in the range of 0 to 1 is output using an S-shaped growth curve function.

7. The crane circuit fault diagnosis system based on multi-source data fusion according to claim 6, characterized in that: The joint time-frequency domain analysis first identifies all maxima and minima of the coupling factor time series. Cubic spline interpolation is used to connect the maxima to form the upper envelope and the minima to form the lower envelope. The arithmetic mean of the upper and lower envelopes is calculated as the mean envelope. The mean envelope is subtracted from the original coupling factor and the selection is repeated until the condition that the difference between the number of extreme points and the number of zero-crossing points does not exceed 1 and the mean of the envelope at any point is zero is met to extract the intrinsic mode functions. Finally, each intrinsic mode function is orthogonalized and the instantaneous phase angle change rate is calculated as the instantaneous frequency. The instantaneous frequencies of all mode functions are superimposed to generate a time-frequency energy distribution map.

8. The crane circuit fault diagnosis system based on multi-source data fusion according to claim 7, characterized in that: The sample entropy determination first extracts a continuous subsequence of length 3 from the third intrinsic mode function with a time delay of 5 sampling points to construct a vector sequence. Then, it counts the number of matching pairs whose distance is less than 0.2 times the standard deviation of the sequence. Based on the matching ratio, it outputs the sample entropy value by performing a natural logarithm operation. When the sample entropy value is greater than 1.2 for 3 consecutive seconds, and the time-frequency energy integral value in the 100-200Hz frequency band exceeds 0.35 and the kurtosis statistic of the third intrinsic mode function is greater than 4.5, it is determined that there is an electrical abnormality caused by mechanical vibration.

9. The crane circuit fault diagnosis system based on multi-source data fusion according to claim 8, characterized in that: The generation of the fault probability value first presets a physical constraint template, as follows: The orbital deformation template is: the coupling factor monotonically increases and the rate of change is consistently higher than 0.3; The insulation aging template is defined as follows: sample entropy less than 0.8 and total harmonic distortion rate greater than 18%. The template for poor contact is: the peak value of the time-frequency energy integral in the 50-150Hz frequency band exceeds 0.4; The minimum cumulative alignment distance between the sequence under test and each template is calculated using a dynamic programming algorithm. This distance is then mapped to a fault probability value of 0 to 1 using an exponential decay function. When the highest fault probability exceeds 0.8 and the condition for determining that mechanical vibration causes electrical abnormalities is met, the corresponding fault type and probability are output.

10. A method for implementing a crane line fault diagnosis system based on multi-source data fusion as described in any one of claims 1-9, characterized in that: Includes the following steps: S1. Collect the triaxial vibration acceleration and three-phase current waveforms of the track, calculate the square value of the axial acceleration from the vibration data and generate triaxial energy components by integrating according to a 100ms time window and a 50% overlap rate, extract the harmonic components from the current data and calculate the total harmonic distortion rate and the odd-even harmonic energy ratio. S2. The orbital deformation energy distribution matrix is ​​generated by superimposing the three-axis energy components. The wavelet basis function is used to perform a 5-level decomposition to obtain 32 sub-bands. Outlier bands with an absolute value of the natural logarithm of the energy ratio of adjacent bands greater than 1.5 are removed. The Pearson correlation coefficient vector between the energy ratio of the retained sub-bands and the total harmonic distortion rate is calculated. The spatiotemporal correlation matrix is ​​generated by weighting the band energy entropy. S3. Expand the spatiotemporal correlation matrix and concatenate it with the odd-even harmonic energy ratio to form a 33-dimensional input vector. Extract forward / backward temporal features through a bidirectional long short-term memory network and output a 128-dimensional fusion vector. Generate a coupling factor by dynamically weighting the historical vector based on the attention mechanism. Extract the first 5 intrinsic mode functions from the empirical mode decomposition of the coupling factor, calculate the sample entropy value of the third intrinsic mode function, and determine anomalies by combining the time-frequency energy integral of 100-200Hz band greater than 0.35 and kurtosis greater than 4.

5. S4. Define a physical template and use dynamic time warping to calculate the minimum cumulative alignment distance. Map the distance to the fault probability using an exponential decay function. When the highest probability is greater than 0.8, output the fault type and update the template library.

Images