Elevator fault prediction method and system based on sound and vibration fusion

Abstract

The invention discloses an elevator fault prediction method and system based on sound and vibration fusion, and the method comprises the steps: carrying out the weighted fusion of a vibration signal and a sound signal according to a signal-to-noise ratio, and then combining with a feature matrix composed of a wavelet packet transformation normalized energy value, a time domain and a frequency domain, applying a Sodat swarm algorithm to an optimization support vector machine model, and carrying out the prediction of an elevator fault. And carrying out feature classification on the feature matrix by using the model. According to the invention, the accuracy of fault prediction can be improved, and the risk of sudden faults is reduced, so that the maintenance efficiency is optimized and the fault rate is reduced.

Description

Technical Field

[0001] The present invention relates to elevator fault prediction, and in particular to an elevator fault prediction method and system based on acoustic-vibration fusion. Background Art

[0002] As a key vertical transportation tool, elevators are widely used in rail transit, high-rise buildings, commercial places and other scenarios. The stability and safety of their operation are directly related to daily travel and life safety. Therefore, the importance of elevator fault diagnosis and prediction is self-evident. Existing elevator fault diagnosis technologies mainly rely on the analysis of a single signal source of vibration signals. These methods usually extract features of signals in the time domain or frequency domain and combine them with pattern recognition algorithms for fault detection. Although these methods are effective to a certain extent, due to the complexity and variability of the fault signals generated during the operation of the elevator, the feature extraction of a single signal source is difficult to fully reflect the multi-dimensional characteristics of the fault, which limits the accuracy of the diagnosis.

[0003] Intelligent decision-making and diagnosis of elevator fault diagnosis have always been the core of research in this field. The process involves the use of artificial intelligence and machine learning techniques to automatically identify, analyze and solve potential faults of mechanical equipment. Support vector machine (SVM) can achieve global optimal solutions rather than local extreme values ​​in the optimization process due to its solid theoretical foundation, while effectively avoiding the risks of overfitting and underfitting of the model, thereby improving its generalization ability. However, despite the excellent performance of SVM on small samples and linearly separable problems, its application in elevator fault diagnosis still faces many challenges. The performance of SVM is often limited when dealing with fault data with nonlinearity and complex boundaries. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide an elevator fault prediction method and system, which can integrate the advantages of sound signals and vibration signals, and optimize the support vector machine at the same time, so as to improve the recognition accuracy of elevator fault characteristics and improve the accuracy of fault prediction.

[0005] Technical solution: The elevator fault prediction method based on acoustic-vibration fusion described in the present invention comprises the following steps:

[0006] (1) collecting the sound signal and vibration signal of the elevator, performing weighted summation on the sound signal and the vibration signal to obtain a fusion signal, wherein the weights of the sound signal and the vibration signal are proportional to their signal-to-noise ratios;

[0007] (2) performing wavelet packet transform on the fused signal to obtain sub-signals of several frequency bands, calculating the energy value of each frequency band and normalizing them; extracting the time domain features and frequency domain features of the fused signal; and forming a feature matrix with the normalized energy value, time domain features and frequency domain features of the fused signal;

[0008] (3) The feature matrix is ​​input into a fault prediction model for fault classification to obtain a fault prediction result of the elevator. The fault prediction model is a support vector machine model. The penalty factor and kernel function parameters of the support vector machine model are optimized using a sand cat swarm optimization algorithm.

[0009] Furthermore, in step (1), before weighted summing the sound signal and the vibration signal, the method also includes: processing the sound signal and the vibration signal using a variational mode decomposition method, decomposing them into intrinsic mode functions of several frequency bands, denoising the high-frequency modes, reconstructing the denoised modal components, and obtaining a noise-reduced sound signal and a noise-reduced vibration signal.

[0010] Furthermore, in step (1), the fusion signal is:

[0011] S fused (t) = a audio ·S audio (t)+a vibration ·S vibration (t);

[0012] In the formula, S fused (t) represents the fusion signal, S audio (t) and S vibration (t) represent the sound signal and vibration signal respectively, a audio and a vibration is the weighting coefficient corresponding to the sound signal and the vibration signal, and t is the time.

[0013] Furthermore, in step (2), the fused signal is subjected to wavelet packet transformation using the db4 wavelet basis function to obtain sub-signals of several frequency bands, and the energy value of the sub-signal of each frequency band is extracted and normalized to obtain the normalized energy value of the fused signal.

[0014] Furthermore, in step (2), the time domain characteristics of the fusion signal include mean, variance, peak-to-peak value, kurtosis, skewness and kurtosis.

[0015] Furthermore, in step (2), the frequency domain characteristics of the fusion signal include frequency centroid, frequency mean square and frequency variance.

[0016] Furthermore, in step (2), after the normalized energy value, time domain features and frequency domain features of the fusion signal are combined into a feature matrix, the following steps are further included:

[0017] The singular value decomposition of the feature matrix is ​​performed to obtain the singular values, and the principal component analysis method is used to reduce the dimension of the feature matrix according to the singular values.

[0018] The elevator fault prediction system based on acoustic-vibration fusion of the present invention comprises:

[0019] A signal and fusion acquisition unit, used to collect the sound signal and vibration signal of the elevator, and perform weighted summation of the sound signal and the vibration signal to obtain a fusion signal, wherein the weights of the sound signal and the vibration signal are proportional to their signal-to-noise ratios;

[0020] A feature extraction unit is used to perform wavelet packet transform on the fused signal to obtain sub-signals of several frequency bands, calculate the energy value of each frequency band and normalize it; extract the time domain features and frequency domain features of the fused signal; and form a feature matrix with the normalized energy value, time domain features and frequency domain features of the fused signal;

[0021] The fault prediction unit is used to input the feature matrix into a fault prediction model for fault classification to obtain a fault prediction result of the elevator. The fault prediction model is a support vector machine model, and the penalty factor and kernel function parameters of the support vector machine model are optimized using a sand cat swarm optimization algorithm.

[0022] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and is characterized in that when the computer program is loaded into the processor, the elevator fault prediction method based on acoustic-vibration fusion is implemented.

[0023] The computer-readable storage medium of the present invention stores a computer program, and is characterized in that when the computer program is executed by a processor, the elevator fault prediction method based on acoustic-vibration fusion is implemented.

[0024] Beneficial effects: Compared with the prior art, the advantages of the present invention are: the present invention performs weighted fusion of the collected vibration signals and sound signals according to the signal-to-noise ratio, which improves the signal data for subsequent fault feature extraction; combines the wavelet packet transform to normalize the energy value, the feature matrix composed of the time domain and the frequency domain, and performs feature dimension reduction on the feature matrix to obtain a feature matrix with higher discrimination. Applying the sand cat swarm algorithm to optimize the machine learning support vector machine model, and using the model to perform feature classification on the fault feature matrix, can improve the accuracy of fault prediction, reduce the risk of sudden failures, thereby optimizing maintenance efficiency and reducing the failure rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of the elevator fault prediction method of the present invention;

[0026] Figure 2 It is a schematic diagram of the three-layer wavelet packet transform of the present invention;

[0027] Figure 3 It is a flow chart of the fault diagnosis model of the present invention;

[0028] Figure 4 A multi-type fault classification logic diagram of a fault prediction model according to an embodiment of the present invention;

[0029] Figure 5 is a three-dimensional coordinate diagram of the test function f1 of an embodiment of the present invention;

[0030] Figure 6 is the test function f of the embodiment of the present invention 10 3D coordinate diagram of ;

[0031] Figure 7 It is a fitness iteration line graph of the four algorithms of the embodiment of the present invention on the test function f1;

[0032] Figure 8 The four algorithms of the embodiments of the present invention are tested in the function f 10 Fitness iteration line graph on . DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0034] like Figure 1 As shown, the elevator fault prediction method based on acoustic-vibration fusion includes the following steps.

[0035] (1) A sound collector and a vibration sensor are installed on the elevator to conduct fault experiments to collect signal data in normal state and fault state. Since the elevator operating environment is often accompanied by noise, in order to improve the quality of the signal, the collected sound signal and vibration signal need to be denoised. To this end, the variational mode decomposition (VMD) method is used to process the signal. VMD decomposes the signal into several intrinsic mode functions (IMFs) of different frequency bands, and the formula is:

[0036]

[0037] Among them, u k (t) is the kth mode function, ω k is the corresponding center frequency.

[0038] Noise is usually concentrated in high-frequency components. By filtering or other denoising processes on these high-frequency modes, the impact of noise can be effectively reduced. After processing, the denoised modal components are reconstructed to obtain a clearer and more accurate signal.

[0039] (2) In elevator fault diagnosis, in order to more accurately identify fault features and improve the accuracy of fault prediction, first, in order to quantify the quality and reliability of acoustic and vibration signals, the signal-to-noise ratio (SNR) of each signal is calculated. The calculation formula of the signal-to-noise ratio is:

[0040]

[0041] In the formula, Signal Power refers to the RMS power of the signal, and Noise Power refers to the RMS power of the background noise.

[0042] The value of the signal-to-noise ratio can reflect the clarity of the signal and the degree of noise interference. The higher the signal-to-noise ratio, the better the signal quality. Then, based on the obtained signal-to-noise ratio, the weighting coefficient a of the sound signal and the vibration signal is determined. i The determination of the weighting coefficient is proportional to the signal-to-noise ratio. Signals with higher signal-to-noise ratios will be given greater weights. This method can ensure that the signal with higher information quality contributes to the final result as much as possible during the fusion process. The calculated weighting coefficient is used to perform weighted fusion of the acoustic signal and the vibration signal. The formula for weighted fusion is:

[0043] S fused (t) = a audio ·S audio (t)+a vibration ·S vibration (t);

[0044] In the formula, S fused (t) represents the signal after weighted fusion, S audio (t) and S vibration (t) represent the acoustic signal and vibration signal respectively, a audio and a vibration is the corresponding weighting coefficient.

[0045] Through weighted fusion, the advantages of the two signals can be combined to obtain a more representative and reliable fusion signal. This weighted fusion method not only combines the complementary information of the acoustic signal and the vibration signal, but also effectively improves the overall quality of the signal, making subsequent fault analysis and prediction more accurate. Through this fusion method, the operating status of the elevator can be captured more comprehensively and the performance of the fault diagnosis system can be significantly enhanced.

[0046] (3) Figure 2 As shown in the figure, when fault features are extracted from the fused signal, the "db4" wavelet basis function is first used to perform a three-layer wavelet packet transform on the fused signal. Through the wavelet packet transform, the signal is decomposed into multiple sub-signals of different frequency bands, and a total of 8 frequency band signals including high frequency and low frequency are obtained, so that the energy value of each frequency band can be extracted and normalized. Figure 2 Where S is the fusion signal data, A is the low-frequency signal of the first transformation, D is the high-frequency signal of the first transformation, AAA, AAD, ADA, ADD, DAA, DAD, DDA, DDD are the sub-band signals after the three-time decomposition of the wavelet packet. The calculation formula of the frequency band energy is:

[0047]

[0048] Where E i is the energy value of each frequency band of the i-th frequency band decomposed by wavelet packet, T is the time length of the signal, t1 and t2 are the time range of the signal, and the three-layer wavelet packet transform can obtain 8 frequency band signals and 8 energy value features.

[0049] Based on the calculation of the energy value feature, the time domain and frequency domain features of the fused signal are calculated. The time domain features of the fused signal include the mean value of the signal Variance(X var ), peak-to-peak value (X pp ), Kurtosis (K v ), skewness (S k ) and kurtosis (K u ) and other characteristic values, which can reflect the overall statistical characteristics and distribution characteristics of the signal. Perform short-time Fourier transform on the fused signal to obtain frequency domain characteristics such as frequency center of gravity (FC), mean square frequency (MSF) and frequency variance (VF) of the fused signal, which are used to describe the distribution and energy concentration of the signal in the frequency domain. These features are combined into a comprehensive feature matrix, which includes the normalized energy values ​​of all frequency bands. As well as time domain and frequency domain features, a comprehensive feature description is formed. The feature matrix is ​​in the following form:

[0050]

[0051] (4) Since the feature matrix contains a large amount of feature information and the number of feature categories is still large, the redundant features will cause the fault diagnosis accuracy to fluctuate, and it is necessary to reduce the feature dimension of the feature matrix. Perform singular value decomposition (SVD) on the feature matrix X of the fusion signal. SVD decomposes the matrix X into the product of the left singular matrix U, the diagonal matrix S, and the right singular matrix V, where S contains singular values ​​arranged in size, which reflect the relative importance of the features.

[0052] X=U·S·V T ;

[0053] On this basis, further dimension reduction is performed through principal component analysis (PCA). PCA calculates the covariance matrix of the feature matrix and extracts the main eigenvectors (principal components) through eigenvalue decomposition. These principal components are sorted according to the size of the eigenvalues, indicating the main direction of variation in the data. In order to maintain the representativeness of the features, principal components with a cumulative contribution rate of more than 95% are selected. These principal components retain most of the variation information in the feature data. Using the selected principal components, the original feature matrix is ​​linearly transformed to obtain the reduced-dimensional feature matrix X reduce .

[0054] X reduce =X·P;

[0055] Among them, P is the projection matrix containing the main components. This dimensionality reduction process not only reduces the feature dimension, but also enhances the representation ability of the feature, so that the data retains important information while reducing complexity. This reduced feature matrix is ​​more concise and efficient, providing an accurate and effective data foundation for subsequent fault classification and prediction.

[0056] (5) After feature dimensionality reduction, the feature matrix X reduce It is input into the support vector machine (SVM) model optimized by the improved sand cat swarm optimization algorithm for fault classification. i ,y i ),i=1,2,3…n,x i ∈R n ,y i ∈{+1,-1}, where n is the training vector x i Dimension,y i Represents the category label of the required classification. Then its classification surface equation is:

[0057] y(x)=ω*x+b;

[0058] Where: ω is the normal vector of the hyperplane, and b is the constant term of the hyperplane.

[0059] The SVM model achieves classification by finding the optimal hyperplane. Its optimization goal is to maximize the classification interval and minimize the classification error. The objective function can be expressed as:

[0060]

[0061] Where C is the penalty factor, ξ i is a slack variable.

[0062] Using the Lagrange multiplier method and the dual problem, the optimal w and b are determined, and the optimal classification surface function is obtained as follows:

[0063]

[0064] where a i is the Lagrange multiplier. For the case of linear inseparability, the radial basis kernel function (RBF) is used to map the original classified data to a higher-dimensional space to achieve linear separability in the high-dimensional space. The kernel function is:

[0065] K(x i ,x j ) = exp(-γ||x i -x j || 2 );

[0066] By introducing the kernel function into the spatial transformation, the classification surface function becomes:

[0067]

[0068] (6) Since SVM has limited performance when processing nonlinear and complex boundary fault data, the selection of kernel function becomes particularly critical because it directly affects the performance of the model. In addition, the parameter settings of SVM, such as penalty factors and kernel parameters, also have a significant impact on the diagnostic accuracy. Manually setting these parameters may lead to overfitting or underfitting, thereby affecting the reliability of the model. Therefore, in-depth exploration of effective parameter optimization strategies and kernel function selection methods is crucial to improving the performance of SVM in elevator fault diagnosis. Figure 3 As shown in the figure, in order to optimize the penalty factor C and kernel function parameter γ of the SVM model, the sand cat swarm optimization algorithm is used, and the accuracy of the five-fold cross-validation model is used as the fitness value, with the goal of minimizing the misclassification error rate.

[0069] The inspiration for the sand cat swarm intelligent optimization algorithm comes from the sand cat's ability to detect low-frequency noise. Sand cats can find prey with their unique ability whether on the ground or underground. For a d-dimensional optimization problem, each sand cat represents a 1×d array of the problem solution, and each sand cat is located in the upper and lower bounds. Depending on the size of the problem, a candidate matrix is ​​created using the sand cat population:

[0070]

[0071] Where: X i is the i-th sand cat; X ij represents the j-th dimension coordinate of the i-th sand cat. The initialization formula of the sand cat is:

[0072] X i = l b +a×(u b -l b );

[0073] Where: l b is the lower bound of the parameter to be optimized; u b is the upper bound of the parameter to be optimized; a is a random number between 0 and 1.

[0074] After the initialization of the sand cat group is completed, according to the conversion value To control the transition between the exploration phase and the development phase, the calculation formula is:

[0075]

[0076] Where: It is a general sensitivity coefficient that decreases linearly from 2 to 0. Its calculation formula is:

[0077]

[0078] Where: iter c It is the current iteration number; iter max is the maximum number of iterations; s M is the hearing coefficient of the sand cat. The prey search mechanism of the sand cat group relies on low-frequency noise emission. Sand cats can perceive low-frequency sounds below 2kHz. In the calculation, s M Set to 2.

[0079] When |R|>1, the sand cat swarm enters the search phase. The sand cat swarm adjusts its position to explore potential better solutions based on the current optimal solution, current position, and hearing sensitivity range, and its position update is between the current position and the target position. While introducing randomness to enhance search diversity, the algorithm maintains low complexity and running cost, thereby effectively discovering local optimal solutions in new search areas. Search formula for the sand cat swarm:

[0080]

[0081] in is the current optimal position; is the current location; Represents the sensitivity of each sand cat. To avoid falling into the local optimum, the sensitivity range of each sand cat is different, and the calculation formula is as follows:

[0082]

[0083] When |R|≤1, the sand cat group enters the attack hunting stage. First, the sand cats and current location Generate a random location At the same time, assume that the sensitivity range of the sand cat is a circle, the moving direction of the sand cat is determined by the angle θ between 0° and 360° at the center of the circle, and the distance moved each time is

[0084]

[0085] The fault prediction method of the present invention is further described below through specific embodiments.

[0086] In this embodiment, the spatial structure and key components of the elevator are analyzed in detail. From the perspective of spatial structure, the entire elevator system can be divided into four parts, namely, the machine room part, the shaft part and the pit part, the car part, and the floor station part. Among them, the main components that affect the normal operation of the elevator include brakes, speed limiters, guide rails, wire ropes, anti-ropes, etc.; here, the brakes, speed limiters, wire ropes, and anti-ropes are specifically analyzed. ① As the power system of the elevator, the brake is mainly responsible for dragging the car to the designated floor; due to the wear and tear caused by long-term movement and damage to the components, the brake may have abnormal noise, rotor failure, bearing damage, gear wear and other faults. ② The elevator speed limiter is an important safety device in the elevator system. Its main function is to prevent the elevator car from overspeeding during operation and ensure the safe operation of the elevator; wear of the speed limiter or entry of foreign objects, loose tensioning pulleys will cause the speed limiter to produce abnormal noise. ③ As an important part of the elevator, the wire rope is responsible for connecting the car and the traction machine to ensure safe operation. Due to long-term use and load, the wire rope may break, wear, corrode and twist. These faults are manifested as abnormal noise, abnormal vibration and reduced carrying capacity, which may lead to serious safety hazards. ④ The anti-ropes wheel changes the direction of the wire rope to ensure the reasonable arrangement of the wire rope in the elevator system, so that it can realize complex path planning in a limited space; the anti-ropes wheel will wear the wheel surface due to long-term contact, and the wire rope will be partially out of the groove.

[0087] In order to accurately diagnose the fault of the elevator, this embodiment installs the sound collector and vibration sensor at the above-mentioned key parts of the elevator, and conducts a fault experiment to collect signal data under normal state and four key component failures. The sampling time is set to 10 minutes, and 500 signal data samples are divided for each state, and each data sample contains 1000 data points.

[0088] The collected sample signals are decomposed and reconstructed using variational mode decomposition (VMD) to achieve signal noise reduction. The signal after noise reduction enters the signal fusion stage. In this stage, the noise-reduced vibration signal and the sound signal are weighted and fused according to the signal-to-noise ratio weight to obtain comprehensive signal features. The fusion process includes aligning and standardizing the vibration signal and the sound signal, and calculating the weighted fusion value or obtaining a unified data set through feature fusion technology.

[0089] After signal fusion is completed, the feature extraction stage begins. The fused signal is subjected to a three-layer wavelet packet transform, and after decomposition, a total of 8 frequency band signals including high frequency and low frequency are obtained. The energy value of each frequency band is calculated and normalized as the feature matrix of the signal. The time domain features of the fused signal are calculated, including the mean, variance, peak-to-peak value, kurtosis, skewness, and kurtosis. The frequency domain features of the signal are calculated using short-time Fourier transform, including the frequency center of gravity, frequency mean square, and frequency variance. A total of 17 features including energy value, time domain, and frequency domain form the feature matrix of the fused signal, forming a total of 5 500*17 feature matrices.

[0090] After feature extraction, feature dimensionality reduction is performed to reduce data dimensions and redundant information. First, singular value decomposition (SVD) is used to decompose the feature matrix into a 500-order left singular vector matrix, a 500*17 singular value diagonal matrix, and a 17-order right singular vector matrix transposed and multiplied. The principal component analysis (PCA) method selects the first k principal components with a cumulative contribution rate of more than 95%, selects the first k eigenvectors as the principal components, and performs a linear transformation on the original feature matrix to obtain the reduced-dimensional feature matrix X reduce , the size of the feature matrix after dimensionality reduction is 500*k.

[0091] Finally, the parameters of the support vector machine (SVM) are optimized using the Sand Cat Swarm Optimization algorithm (SCSO). The SCSO algorithm is used to optimize the penalty parameter C and kernel function parameter γ of the SVM. SCSO improves the global search and local development capabilities by introducing search and attack strategies, thereby optimizing the performance of the SVM. The optimized SVM model is used for classification, and the five 500*k feature matrices are spliced ​​along the column direction to form a new feature matrix with a dimension of 500*5k. The label of each sample is matched with its corresponding feature vector, and the merged feature matrix and label vector are divided into training set and test set in a ratio of 7:3. The support vector machine (SVM) model is trained using the training set, and the accuracy of the model is evaluated using five-fold cross validation. The accuracy of the model will be used as the fitness value of the SCSO-SVM model. Figure 4 As shown in the figure, the test data is input into the elevator fault prediction model for four binary classifications, and finally the classification accuracy is obtained, thereby verifying the accuracy and generalization ability of the fault prediction model in the classification of different fault types.

[0092] The trained diagnostic model is directly deployed in the server, and data is collected in real time through the sound collector and vibration sensor, and uploaded to the server for further processing. The data uploaded to the server is processed by the elevator diagnostic prediction model, including noise reduction, fusion, feature extraction and fault diagnosis. If a fault occurs, the server will feedback the diagnosed fault results to the user in the form of a graphical interface, alarm system or report. The alarm system reminds users of critical faults through instant pop-up windows and sound prompts, and provides operational suggestions. At the same time, the generated detailed diagnostic report, including fault description, data analysis and maintenance suggestions, will be stored in the database for long-term tracking and trend analysis. In order to ensure the accuracy and adaptability of the model in actual applications, the server will regularly collect data in new states and iteratively update the model. This update process includes retraining of new data and optimization of model parameters to better reflect changes in equipment performance and new failure modes. After each update, the predictive ability and robustness of the model will be enhanced, enabling it to adapt to changing working environments and conditions.

[0093] In order to verify the superiority of the Sand Cat Swarm Algorithm, this embodiment selects the Grey Wolf Optimization Algorithm (GWO) and the Whale Optimization Algorithm (WOA) as well as the classic particle swarm optimization algorithm (PSO) to compare with the Sand Cat Swarm Algorithm (SCSO). Two benchmark test functions are used for testing: f1 is a unimodal function to verify the convergence accuracy of the algorithm; f 10 It is a multi-peak function, which verifies whether the algorithm can escape from the local optimal solution. Figure 5 and Figure 6 f1 and f 10 3D coordinate diagram of .

[0094]

[0095] The search range of f1 is [-100, 100], and the theoretical optimal value is 0; the search range of f3 is [-32, 32], and the theoretical optimal value is 0.

[0096] The algorithm results are statistically shown as follows: Figure 7 and Figure 8 As shown in Figure 2. For the single-peak function f1, the SCSO algorithm has significantly higher optimization accuracy than GWO, PSO, and WOA, and its iteration speed is also better than the other three algorithms. 10 In the results, the iteration accuracy of the SCSO algorithm is slightly higher than that of the GWO algorithm, and the other three algorithms all obtain the same optimal solution, which verifies that the SCSO algorithm has higher stability.

[0097] Before fault classification, the feature vector after dimension reduction needs to be labeled according to the running status of the elevator. The calibrated feature data set is divided into training set and test set in proportion, usually 70% for training and 30% for testing. This division method ensures the independence of the training and testing processes and effectively prevents overfitting. The training set is input into the ISCSO-SVM model for training. Four binary classification fault diagnoses are performed during the model training process, and the accuracy of the model is evaluated by five-fold cross validation. The accuracy of the elevator fault prediction model is used as the fitness value of the SCSO-SVM algorithm, which aims to learn and identify the characteristics of different fault types. After training, the test set is used to evaluate the elevator fault prediction model, and the performance indicators such as accuracy, precision, and recall of the classification results are calculated to verify the performance of the model in practical applications. These evaluation indicators not only reflect the prediction ability of the model on new data, but also provide a basis for further optimization.

[0098] In practical applications, elevator fault diagnosis prediction models may have missed reports and false positives, which usually stems from the model's inability to effectively handle new data types and patterns. In order to solve this problem and improve the accuracy of fault diagnosis, the model needs to be retrained and updated. The specific steps are as follows: First, the operating status of the elevator is continuously monitored, the raw data of vibration and sound signals in the new environment are collected, and noise reduction is performed to improve data quality. Then, data from different sources are fused to ensure the consistency of the new data set with historical data. Next, the model is retrained using a data set containing new features and its parameters are updated to adapt to the new data pattern. After retraining, the performance of the model is evaluated through the test set and compared with the previous version to ensure significant performance improvement. Based on the evaluation results, the model structure and algorithm hyperparameters are dynamically adjusted to further optimize performance. Finally, the trained elevator fault prediction model will output the prediction results of the fault category through the new feature data, thereby achieving effective fault identification and early warning of the elevator. This update mechanism ensures that the model always reflects the current operating status and fault risk, enhancing its effectiveness and reliability in practical applications.

[0099] The elevator fault prediction system based on acoustic-vibration fusion of the present invention comprises:

[0100] A signal and fusion acquisition unit, used to collect the sound signal and vibration signal of the elevator, and perform weighted summation of the sound signal and the vibration signal to obtain a fusion signal, wherein the weights of the sound signal and the vibration signal are proportional to their signal-to-noise ratios;

[0101] A feature extraction unit is used to perform wavelet transform on the fused signal to obtain sub-signals of several frequency bands, calculate the energy value of each frequency band and normalize it; extract the time domain features and frequency domain features of the fused signal; and form a feature matrix with the normalized energy value, time domain features and frequency domain features of the fused signal;

[0102] The fault prediction unit is used to input the feature matrix into a fault prediction model for fault classification to obtain a fault prediction result of the elevator. The fault prediction model is a support vector machine model, and the penalty factor and kernel function parameters of the support vector machine model are optimized using a sand cat swarm optimization algorithm.

[0103] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and is characterized in that when the computer program is loaded into the processor, the elevator fault prediction method based on acoustic-vibration fusion is implemented.

[0104] The computer-readable storage medium of the present invention stores a computer program, and is characterized in that when the computer program is executed by a processor, the elevator fault prediction method based on acoustic-vibration fusion is implemented.

[0105] The computer-readable storage medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, flash memory or any other medium that can be used to store the desired program code in the form of instructions or data structures and can be accessed by the computer. The processor is used to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiments.

Claims

1. An elevator fault prediction method based on acoustic-vibration fusion, characterized in that: The steps include: (1) collecting the sound signal and vibration signal of the elevator, performing weighted summation on the sound signal and the vibration signal to obtain a fusion signal, wherein the weights of the sound signal and the vibration signal are proportional to their signal-to-noise ratios; (2) performing wavelet packet transform on the fused signal to obtain sub-signals of several frequency bands, calculating the energy value of each frequency band and normalizing it; extracting the time domain features and frequency domain features of the fused signal; The normalized energy value, time domain features and frequency domain features of the fusion signal are combined into a feature matrix; (3) The feature matrix is ​​input into a fault prediction model for fault classification to obtain a fault prediction result of the elevator. The fault prediction model is a support vector machine model. The penalty factor and kernel function parameters of the support vector machine model are optimized using a sand cat swarm optimization algorithm.

2. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1 is characterized in that: In step (1), before weighted summing of the sound signal and the vibration signal, the method also includes: processing the sound signal and the vibration signal by using a variational mode decomposition method, decomposing them into intrinsic mode functions of several frequency bands, denoising the high-frequency mode, reconstructing the denoised modal components, and obtaining a noise-reduced sound signal and a noise-reduced vibration signal.

3. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1, characterized in that: In step (1), the fusion signal is: S fused (t)=a audio ·S audio (t)+a vibration ·S vibration (t); In the formula, S fused (t) represents the fusion signal, a audio (t) and S vibration (t) represent the sound signal and vibration signal respectively, a audin and a vibration is the weighting coefficient corresponding to the sound signal and the vibration signal, and t is the time.

4. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1, characterized in that: In step (2), the fused signal is subjected to wavelet packet transformation using the db4 wavelet basis function to obtain sub-signals of several frequency bands, and the energy value of the sub-signal of each frequency band is extracted and normalized to obtain the normalized energy value of the fused signal.

5. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1, characterized in that: In step (2), the time domain characteristics of the fusion signal include mean, variance, peak-to-peak value, kurtosis, skewness and kurtosis.

6. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1, characterized in that: In step (2), the frequency domain features of the fusion signal include frequency centroid, frequency mean square and frequency variance.

7. The elevator fault prediction method based on acoustic-vibration fusion according to claim 1, characterized in that: In step (2), after the normalized energy value, time domain features and frequency domain features of the fusion signal are formed into a feature matrix, the following steps are further included: The singular value decomposition of the feature matrix is ​​performed to obtain the singular values, and the principal component analysis method is used to reduce the dimension of the feature matrix according to the singular values.

8. An elevator fault prediction system based on acoustic-vibration fusion, characterized in that: include: A signal and fusion acquisition unit, used to collect the sound signal and vibration signal of the elevator, and perform weighted summation of the sound signal and the vibration signal to obtain a fusion signal, wherein the weights of the sound signal and the vibration signal are proportional to their signal-to-noise ratios; A feature extraction unit, used to perform wavelet packet transform on the fused signal to obtain sub-signals of several frequency bands, calculate the energy value of each frequency band and normalize it; and extract the time domain features and frequency domain features of the fused signal; The normalized energy value, time domain features and frequency domain features of the fusion signal are combined into a feature matrix; The fault prediction unit is used to input the feature matrix into a fault prediction model for fault classification to obtain a fault prediction result of the elevator. The fault prediction model is a support vector machine model, and the penalty factor and kernel function parameters of the support vector machine model are optimized using a sand cat swarm optimization algorithm.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the elevator fault prediction method based on acoustic-vibration fusion according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the elevator fault prediction method based on acoustic-vibration fusion according to any one of claims 1 to 7 is implemented.

Images