An elevator fault detection method and system based on artificial intelligence

Abstract

This application provides an artificial intelligence-based elevator fault detection method and system. The method determines the search range for the number of vibration signal decomposition layers based on the differences in the vibration physical characteristics between various mechanical components during elevator operation and the sudden load changes during elevator operation. It determines the candidate range of bandwidth constraints for each mechanical component's vibration with the goal of minimizing the aliasing of vibration physical characteristics between the components. Modal decomposition is performed by traversing different parameter combinations within the search range and the candidate range. Feature components containing fault information of each mechanical component are selected from the decomposition results. All feature components are input into a trained artificial intelligence classification model to determine the fault type of the elevator traction system. Using the solution of this application, the vibration characteristics of each mechanical component of the elevator can be extracted completely and independently under conditions of vibration coupling and sudden load changes.

Description

Technical Field

[0001] This application relates to the field of elevator control technology, and more specifically, to an elevator fault detection method and system based on artificial intelligence. Background Technology

[0002] With the acceleration of urbanization and the widespread use of high-rise buildings, elevators, as core equipment in vertical transportation, are receiving increasing attention for their safe, reliable, and efficient operation. Elevator mechanical systems are constantly subjected to frequent starts and stops, load changes, and high-speed operation. Key components such as traction machines, guide rails, and counterweights are prone to wear, imbalance, or malfunction. If these issues are not detected and addressed promptly, they can lead to elevator entrapment, component damage, or even safety accidents. Traditional maintenance methods rely mainly on periodic inspections and post-incident repairs, lacking real-time perception and intelligent diagnostic capabilities for operational status. Therefore, vibration signal-based fault diagnosis technology is gradually becoming an effective means of elevator health management.

[0003] Currently, there are some signal processing and machine learning-based methods in the field of elevator fault diagnosis. For example, vibration signals are processed and input into artificial intelligence models through time-frequency analysis methods such as wavelet transform and empirical mode decomposition. However, elevator vibration signals are characterized by multi-source coupling, non-stationarity, and strong interference from load abrupt changes. How to accurately and adaptively extract the fault features of each component from complex signals is a key challenge to achieving intelligent diagnosis and precise control. When dealing with the coupled vibration of multiple elevator components, existing technologies usually set fixed parameters or perform global optimization, which is difficult to adapt to the differences in vibration characteristics of components at different stages of elevator operation and the non-stationary changes in signals caused by load abrupt changes. This leads to problems such as mode aliasing, large residual noise, and insufficient extraction of fault features in the decomposition results, affecting the accuracy of diagnosis. Therefore, how to completely and independently extract the vibration characteristics of each mechanical component of the elevator under the condition of vibration coupling and load abrupt changes has become a difficult problem for the industry. Summary of the Invention

[0004] This application provides an elevator fault detection method and system based on artificial intelligence, which can completely and independently extract the vibration characteristics of each mechanical component of the elevator under the condition of vibration coupling and sudden load change.

[0005] Firstly, this application provides an artificial intelligence-based elevator fault detection method for use in an AI-based elevator control system to detect elevator faults. The method includes: Vibration signals of elevator mechanical components are collected during elevator operation; The difference in vibration physical characteristics between various mechanical components during elevator operation is determined based on the spectrum of the vibration signal, and the difference is dynamically expanded according to the load change during elevator operation to obtain the search range of vibration signal decomposition layers. The candidate range of bandwidth constraints for vibration of each mechanical component of the elevator is determined with the goal of minimizing the degree of overlap of vibration physical characteristics among the mechanical components. Modal decomposition is performed by traversing different parameter combinations within the search range and the candidate range. Based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components, feature components containing fault information of each mechanical component of the elevator are selected. Then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components. The fault feature vector set is input into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

[0006] In some embodiments, determining the degree of difference in vibration physical characteristics between various mechanical components during elevator operation based on the spectrum of the vibration signal specifically includes: Based on the vibration signals, determine the vibration characteristic diagrams of each mechanical component during elevator operation; The resonance peaks in the vibration characteristic diagram are extracted to obtain the degree of difference in the vibration physical characteristics between the various mechanical components during elevator operation.

[0007] In some embodiments, the difference is dynamically expanded based on the sudden changes in load during elevator operation, and the search range for the number of vibration signal decomposition layers specifically includes: The uncertainty of load change during elevator operation is determined based on the vibration signal. Based on the uncertainty, the difference is dynamically expanded to obtain the search range for the number of vibration signal decomposition layers.

[0008] In some embodiments, determining the candidate range of bandwidth constraints for vibration of each mechanical component of the elevator, with the goal of minimizing the degree of aliasing of vibration physical characteristics among the mechanical components, specifically includes: The initial range of bandwidth constraint is preset for each mechanical component of the elevator during vibration; The initial range is divided into three equal parts to obtain two internal dividing points; The vibration signal is decomposed at each internal segmentation point to obtain the degree of aliasing between the components in each decomposition result. Based on all degrees of aliasing, the candidate range for bandwidth constraints during vibration of each mechanical component of the elevator is determined.

[0009] In some embodiments, performing mode decomposition by traversing different parameter combinations within the search range and the candidate range specifically includes: Filter out multiple integer parameter combinations within the search range and the candidate range; The vibration signal is modally decomposed using different parameter combinations to obtain multiple modal components and residual signals corresponding to each parameter combination.

[0010] In some embodiments, the feature components containing fault information of various mechanical components of the elevator are selected based on the entropy ratio of the residual energy in each decomposition result and the correlation between adjacent components. Specifically, this includes: Select one decomposition result as the selected decomposition result, and obtain all modal components and residual signals corresponding to the selected decomposition result; Determine the entropy ratio of the residual energy in the vibration signal; Determine the correlation between every two modal components in all modal components; The average relevance is determined based on all relevance scores. Further determine the entropy percentage and average correlation of the residual energy in the remaining decomposition results; Feature components containing fault information of various mechanical components of the elevator were selected by filtering out the entropy ratio of all residual energy and all average correlations.

[0011] In some embodiments, the feature components containing fault information of various mechanical components of the elevator are filtered out by the entropy ratio of all residual energy and all average correlations. This specifically includes the following implementation: The entropy percentage of all residual energy and all average correlations constitute the solution set; Based on the set of solutions, a frontier curve of the non-dominated solution is constructed on a two-dimensional plane with the entropy ratio of residual energy and the average correlation as coordinate axes; The optimal point for adapting to the vibration and load changes of various mechanical components of the elevator is identified on the leading edge curve. All modal components corresponding to the optimal point are obtained, and multiple feature components containing fault information of various mechanical components of the elevator are selected from them based on the vibration and impact during elevator failure.

[0012] Secondly, this application provides an elevator control system based on artificial intelligence, including a fault detection unit, the fault detection unit comprising: The acquisition module is used to collect vibration signals of the elevator's mechanical components during elevator operation; The processing module is used to determine the degree of difference in the vibration physical characteristics between the mechanical components during elevator operation based on the spectrum of the vibration signal, and to dynamically expand the degree of difference according to the load change during elevator operation to obtain the search range of the vibration signal decomposition layer. The processing module is also used to determine the candidate range of bandwidth constraints for vibration of each mechanical component of the elevator with the goal of minimizing the degree of overlap of vibration physical characteristics between each mechanical component. The processing module is also used to perform mode decomposition by traversing different parameter combinations within the search range and the candidate range, and to filter out feature components containing fault information of each mechanical component of the elevator based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components. Then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components. The execution module is used to input the fault feature vector set into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described artificial intelligence-based elevator fault detection method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned artificial intelligence-based elevator fault detection method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The elevator fault detection method and system based on artificial intelligence provided in this application first collects vibration signals of elevator mechanical components during elevator operation; determines the degree of difference in vibration physical characteristics between various mechanical components during elevator operation based on the spectrum of the vibration signals, and dynamically expands the degree of difference according to the load change during elevator operation to obtain the search range of vibration signal decomposition layers; determines the candidate range of bandwidth constraints for vibration of each mechanical component with the goal of minimizing the degree of aliasing of vibration physical characteristics between various mechanical components; performs modal decomposition by traversing different parameter combinations within the search range and the candidate range, and filters feature components containing fault information of each mechanical component based on the entropy ratio of residual energy in each decomposition result and the correlation between adjacent components; then constructs a fault feature vector set characterizing the state of the elevator traction system through the time domain and frequency domain features of all feature components; inputs the fault feature vector set into a trained artificial intelligence classification model for identification and classification, and then determines the fault type of the elevator traction system.

[0016] Therefore, this application dynamically combines the differences in vibration physical characteristics between various mechanical components with load uncertainty, i.e., sudden load changes, to enable the number of decomposition layers to adapt to real-time changes in component vibration characteristics and load conditions during elevator operation, thus improving the targeting of feature extraction. Subsequently, bandwidth constraint optimization guided by aliasing degree reduces modal confusion and ensures effective separation of vibration characteristics of each component. A dual-objective screening mechanism of residual energy entropy ratio and component correlation is adopted to enhance component independence while ensuring sufficient decomposition, thereby extracting purer and more representative fault features. Finally, an artificial intelligence model is used to achieve accurate classification of fault types and generate control commands accordingly, realizing a closed-loop intelligent response from fault perception, diagnosis to control, significantly improving the fault early warning capability, operational safety, and maintenance intelligence level of the elevator system. In summary, the solution of this application can completely and independently extract the vibration characteristics of each mechanical component of the elevator under conditions of vibration coupling and sudden load changes. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of an AI-based elevator fault detection method according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the determination of the degree of difference according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of a candidate range according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a fault detection unit according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements an artificial intelligence-based elevator fault detection method according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of an AI-based elevator fault detection method according to some embodiments of this application. The AI-based elevator fault detection method mainly includes the following steps: In step 101, vibration signals of the elevator's mechanical components are collected during elevator operation.

[0020] In specific implementation, the vibration signal of the elevator mechanical components during elevator operation can be collected in the following way: a piezoelectric accelerometer can be installed on the housing of the traction machine bearing seat, and the vibration acceleration time domain signal can be collected synchronously according to the preset sampling frequency and sampling duration. The collected vibration acceleration time domain signal is used as the vibration signal of the elevator mechanical components. The sampling frequency can be set according to the requirements. For example, it is set to 2560Hz in this application. Furthermore, the sampling duration can be preset to the time for the elevator to complete at least 10 uniform speed runs. In other embodiments, it can also be installed on the top of the elevator car.

[0021] In step 102, the difference in vibration physical characteristics between various mechanical components during elevator operation is determined based on the spectrum of the vibration signal, and the difference is dynamically expanded according to the load change during elevator operation to obtain the search range of vibration signal decomposition layers.

[0022] In some embodiments, reference Figure 2 The figure is an exemplary flowchart illustrating the determination of the degree of difference according to some embodiments of this application. The determination of the degree of difference in the vibration physical characteristics between various mechanical components during elevator operation based on the spectrum of the vibration signal in this application can be achieved using the following steps: In step 1021, the vibration characteristic diagram of each mechanical component during elevator operation is determined based on the vibration signal; In step 1022, the resonance peaks in the vibration characteristic diagram are extracted to obtain the degree of difference in the vibration physical characteristics between the various mechanical components during elevator operation.

[0023] In specific implementation, the vibration characteristic diagram of each mechanical component during elevator operation can be determined based on the vibration signal in the following way: the vibration signal is converted from the time domain to the frequency domain, and the signal in the frequency domain is used as the vibration characteristic diagram of each mechanical component during elevator operation. The vibration signal can be converted from the time domain to the frequency domain by the fast Fourier transform in the prior art. In other embodiments, other prior art can also be used to convert the vibration signal from the time domain to the frequency domain, which is not limited here.

[0024] It should be noted that the vibration characteristic diagram in this application refers to a data diagram that visually reflects the distribution of vibration energy of elevator mechanical components at different frequencies, and is used to reveal the main vibration characteristic frequencies of each component.

[0025] In specific implementation, the resonance peaks in the vibration characteristic diagram are extracted to obtain the difference in vibration physical characteristics between various mechanical components during elevator operation. This can be achieved in the following way: First, the vibration characteristic diagram is smoothed to suppress random noise. For example, moving average filtering or Gaussian filtering can be used for smoothing preprocessing to suppress random noise. Then, a peak detection algorithm is used to traverse the smoothed spectrum, identify all local maxima, and screen out resonance peaks with amplitudes higher than the background noise. Finally, the statistical result of the number of screened resonance peaks is directly used as a parameter characterizing the number of independent vibration sources of mechanical components, that is, the difference in vibration physical characteristics between various mechanical components during elevator operation. The peak detection algorithm can adopt a method known in the art, such as a local maximum detection algorithm based on amplitude threshold.

[0026] It should be noted that, in this application, the difference degree is used to characterize the estimated number of mutually independent mechanical vibration sources in the vibration signal.

[0027] In some embodiments, dynamically expanding the difference degree based on the sudden changes in load during elevator operation to obtain the search range for the number of vibration signal decomposition layers can be achieved through the following steps: The uncertainty of load change during elevator operation is determined based on the vibration signal. Based on the uncertainty, the difference is dynamically expanded to obtain the search range for the number of vibration signal decomposition layers.

[0028] In practice, the uncertainty of load change during elevator operation can be determined based on the vibration signal in the following way: the information entropy of the vibration signal can be used as the uncertainty of load change during elevator operation.

[0029] It should be noted that, in this application, uncertainty is a parameter value used to quantify the complexity and randomness of the vibration signal due to the influence of working conditions such as load changes.

[0030] In specific implementation, the search range for the number of vibration signal decomposition layers can be obtained by dynamically expanding the difference based on the uncertainty, as follows: the uncertainty is used as the width of the dynamic expansion to expand the difference. Specifically, the difference is used as the lower limit, the sum of the difference and the uncertainty is used as the upper limit, and the interval formed by the lower limit and the upper limit is used as the search range for the number of vibration signal decomposition layers.

[0031] It should be noted that the search range in this application refers to a range of decomposition layer values ​​preset for subsequent parameter optimization.

[0032] Furthermore, it should be noted that step 102 in this application, targeting the specific working condition of frequent vibration coupling of multiple components and load changes in elevator vibration signals, combines the estimation of the number of independent vibration sources of components (difference) with the signal complexity (uncertainty) introduced by load changes to dynamically determine the search range of the signal decomposition layer. Specifically, the number of spectral resonance peaks is extracted to quantify the independent vibration characteristics of mechanical components, ensuring that the decomposition layer can basically cover the main vibration sources. At the same time, the information entropy of the vibration signal is introduced to measure the random impact of load changes and is used as the expansion width, so that the upper limit of the search range can adaptively accommodate additional transient components or nonlinear effects that may be excited by load changes. The advantage of this method is that it specifically solves the problem of poor adaptability of fixed decomposition parameters caused by the diversity of components and the dynamic nature of loads in elevator scenarios, providing an optimized starting point for subsequent feature extraction that reflects the essence of the mechanical structure and accommodates changes in operating state, thereby improving the accuracy of fault feature separation and the overall robustness of the diagnostic system.

[0033] In step 103, the candidate range of bandwidth constraints for vibration of each mechanical component of the elevator is determined with the goal of minimizing the degree of overlap of vibration physical characteristics between each mechanical component.

[0034] In some embodiments, reference Figure 3 The figure is an exemplary flowchart illustrating the determination of candidate ranges according to some embodiments of this application. In this application, the candidate ranges for determining the bandwidth constraints of each mechanical component of an elevator when it vibrates, with the goal of minimizing the degree of overlap of vibration physical characteristics between the mechanical components, can be achieved by the following steps: In step 1031, the initial range of bandwidth constraint is preset when each mechanical component of the elevator vibrates; In step 1032, the initial range is divided into three equal parts to obtain two internal dividing points; In step 1033, the vibration signal is decomposed through each internal segmentation point to obtain the degree of aliasing between each component in each decomposition result. In step 1034, candidate ranges for bandwidth constraints during vibration of each mechanical component of the elevator are determined based on all degrees of aliasing.

[0035] It should be noted that the initial range in this application is an empirical range preset based on prior research on the characteristics of elevator vibration signals. This range aims to cover most of the reasonable values ​​of the bandwidth constraint parameters required to effectively separate the vibration characteristics of different mechanical components of the elevator traction system under typical operating conditions. Its specific boundary values ​​are derived from previous experimental research literature in related fields and are the conventional starting search space for parameter optimization when those skilled in the art perform such signal decomposition. Based on the typical fault characteristic frequency range and sampling frequency of the main mechanical components of the elevator traction system (such as the traction machine, guide wheel, brake, etc.), an initial range covering 0.5 to 5 times the highest frequency of interest is preset, for example, [100, 2500] Hz. This range can be preliminarily determined through spectral analysis of historical normal and fault vibration signals.

[0036] In specific implementation, dividing the initial range into three equal parts to obtain two internal dividing points can be achieved in the following way: using an equal interval search strategy, assuming the initial range is [A,B], first calculate the interval length L=BA, then divide the interval length into three equal parts to obtain the step size Δ=L / 3, and then obtain the two internal dividing points a1 and a2 by calculating A+Δ and A+2Δ.

[0037] It should be noted that, in this application, the internal dividing point is a specific parameter value point set according to mathematical division rules within the initial range for testing the effect of specific bandwidth constraint parameters in order to execute the equal division interval search strategy.

[0038] In specific implementation, the vibration signal is decomposed through each internal segmentation point to obtain the degree of aliasing between components in each decomposition result. This can be achieved in the following way: using internal segmentation points a1 and a2 as bandwidth constraint parameters and the lower limit of the search range of the decomposition level as the decomposition level, variational mode decomposition is performed on the vibration signal. Specifically, in the iterative solution of the variational problem in variational mode decomposition, the bandwidth constraint parameters are directly applied to the update formula for the frequency band estimation of each modal component in each iteration. That is, the bandwidth constraint parameters are used as the penalty factor in the formula for the spectrum estimation value in the iterative solution process, thereby obtaining multiple modal components corresponding to each internal segmentation point. Subsequently, the permutation entropy of all modal components corresponding to each segmentation point is calculated, and the reciprocal of the permutation entropy is used as the degree of aliasing between components in the decomposition result corresponding to each segmentation point.

[0039] It should be noted that the degree of aliasing in this application is a parameter value used to quantitatively evaluate the frequency component confusion and overlap between the modal components obtained by signal decomposition under a selected bandwidth constraint.

[0040] In practice, the candidate range for bandwidth constraints of each mechanical component of the elevator during vibration can be determined based on all the aliasing levels as follows: compare the aliasing levels obtained after decomposition using internal segmentation points a1 and a2 respectively, and take the interval formed by the internal segmentation point corresponding to the smaller aliasing level and the closer boundary in the initial range as the candidate range for bandwidth constraints of each mechanical component of the elevator during vibration. For example, if the aliasing level obtained after decomposition of internal segmentation point a1 is smaller, and a1 is closer to the lower limit in the initial range, then [the lower limit of the initial range, a1] is taken as the candidate range for bandwidth constraints of each mechanical component of the elevator during vibration.

[0041] It should be noted that the candidate range in this application is the bandwidth constraint parameter preference range determined after preliminary screening by comparing the degree of aliasing corresponding to different internal segmentation points, and is used for the final parameter fine optimization in subsequent steps.

[0042] Furthermore, it should be noted that step 103 in this application addresses the challenge of potential overlap in the frequency bands of vibration characteristics of multiple components in elevator vibration signals. It employs a rapid interval screening strategy based on aliasing degree assessment to determine the candidate range of bandwidth constraint parameters. This method does not blindly traverse the entire initial range, but rather selects two key test points by dividing the signal into three equal parts. It utilizes the reciprocal of the permutation entropy, an indicator that effectively quantifies the degree of frequency confusion between components, to rapidly evaluate and compare the decomposition effect of the test points. By selecting test points with lower aliasing and their adjacent initial boundaries, the search range can be efficiently converged to a sub-interval that is more likely to contain the optimal solution. The advantage of this approach is that it is specifically suited to the dual requirements of real-time performance and decomposition quality in elevator scenarios: it significantly reduces the parameter space for subsequent fine-grained optimization with extremely low initial computational cost (requiring only two decomposition tests), thereby reducing the overall computational complexity of the optimization process. This provides key technical support for achieving efficient online adaptive signal decomposition and fault feature extraction in elevator operation monitoring systems.

[0043] In step 104, different parameter combinations are traversed within the search range and the candidate range to perform mode decomposition. Based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components, feature components containing fault information of each mechanical component of the elevator are selected. Then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components.

[0044] In some embodiments, performing mode decomposition by traversing different parameter combinations within the search range and the candidate range can be achieved using the following steps: Filter out multiple integer parameter combinations within the search range and the candidate range; The vibration signal is modally decomposed using different parameter combinations to obtain multiple modal components and residual signals corresponding to each parameter combination.

[0045] In a specific implementation, the parameter combination for filtering multiple integers within the search range and the candidate range can be implemented in the following way: for each integer X within the search range and each integer Y within the candidate range, any data pair consisting of (X, Y) is used as a parameter combination.

[0046] In specific implementation, different parameter combinations are used to perform modal decomposition on the vibration signal to obtain multiple modal components corresponding to each parameter combination. This can be achieved in the following way: for each parameter combination (X,Y), X is used as the decomposition level and Y is used as the penalty factor to establish and solve the corresponding constrained variational problem. Through iterative optimization using the alternating direction multiplier method, X intrinsic modal components with different center frequencies and a residual signal corresponding to each parameter combination are finally obtained.

[0047] In some embodiments, the feature components containing fault information of various mechanical components of the elevator can be screened based on the entropy ratio of the residual energy in each decomposition result and the correlation between adjacent components. This can be achieved by the following steps: Select one decomposition result as the selected decomposition result, and obtain all modal components and residual signals corresponding to the selected decomposition result; Determine the entropy ratio of the residual energy in the vibration signal; Determine the correlation between every two modal components in all modal components; The average relevance is determined based on all relevance scores. Further determine the entropy percentage and average correlation of the residual energy in the remaining decomposition results; Feature components containing fault information of various mechanical components of the elevator were selected by filtering out the entropy ratio of all residual energy and all average correlations.

[0048] In specific implementation, the entropy ratio of the residual energy in the residual signal to the entropy ratio of the vibration signal can be determined in the following way: First, calculate the total energy of the vibration signal and the total energy of the residual signal, and calculate the ratio of the total energy of the residual signal to the total energy of the vibration signal. Then, substitute this ratio into the Shannon entropy formula, that is, H=-Rlog(R), where H is the Shannon entropy, R is the ratio of the total energy of the residual signal to the total energy of the vibration signal, and log represents the logarithm to the base 2. Finally, the obtained Shannon entropy is used as the entropy ratio of the residual energy in the residual signal to the entropy ratio of the vibration signal.

[0049] It should be noted that the entropy ratio in this application is a parameter value used to evaluate how much vibrational energy that was not successfully separated is contained in the residual part of the signal after modal decomposition.

[0050] Furthermore, it should be noted that in this application, the proportion of residual energy is regarded as a probabilistic event, thereby quantifying the uncertainty of the information it contains. A simple energy ratio can only reflect the proportion of the residual part in terms of energy size, but it cannot distinguish whether this part of energy is random noise without any rules or hides regular fault components. By using Shannon entropy, any potential, unseparated regular fault information in the residual signal can be detected and punished more sensitively.

[0051] In practice, the correlation between every two modal components in all modal components can be determined as follows: for every two modal components in all acquired modal components, calculate the Pearson correlation coefficient between the two modal components, and use the Pearson correlation coefficient as the correlation between the two modal components, thereby obtaining the correlation between every two modal components.

[0052] It should be noted that in this application, correlation is a parameter value used to measure the similarity of any two modal components in the time domain waveform. Its physical meaning is to determine whether the two modal components may originate from the vibration of the same mechanical component.

[0053] In practice, the average relevance can be determined based on all relevance values ​​by taking the average value of all relevance values ​​as the average relevance value.

[0054] It should be noted that the average correlation in this application is a parameter value used to comprehensively evaluate the overall independence among all modal components obtained from a certain decomposition.

[0055] In some embodiments, the feature components containing fault information of each mechanical component of the elevator can be screened by using the entropy ratio of all residual energy and all average correlations, which can be achieved by the following steps: The entropy percentage of all residual energy and all average correlations constitute the solution set; Based on the set of solutions, a frontier curve of the non-dominated solution is constructed on a two-dimensional plane with the entropy ratio of residual energy and the average correlation as coordinate axes; The optimal point for adapting to the vibration and load changes of various mechanical components of the elevator is identified on the leading edge curve. All modal components corresponding to the optimal point are obtained, and multiple feature components containing fault information of various mechanical components of the elevator are selected from them based on the vibration and impact during elevator failure.

[0056] In practice, the entropy ratio of all residual energy and all average correlations can be used to form the solution set in the following way: For the sum and average correlation corresponding to each decomposition result, the sum and average correlation corresponding to each decomposition result are combined into a set of data pairs. Then, all the data pairs are constructed into a two-dimensional array, and this two-dimensional array is used as the solution set.

[0057] It should be noted that, in this application, the set to be solved refers to the set of alternative solutions consisting of all candidate parameter combinations and their corresponding performance evaluation indicators during the parameter optimization process.

[0058] In specific implementation, the construction of the frontier curve of non-dominated solutions on a two-dimensional plane with the entropy ratio of residual energy and average correlation as coordinate axes based on the solution set can be achieved in the following way: multiple non-dominated solutions can be screened using the Pareto optimal solution screening algorithm. Specifically, each data pair in the solution set is compared with other data pairs one by one. If a data pair i satisfies a preset condition, i is marked as a non-dominated solution. All non-dominated solutions are plotted on a two-dimensional plane with the entropy ratio of residual energy and average correlation as coordinate axes. Then, all the plotted points are connected in order of the entropy ratio of residual energy. Finally, the curve obtained by the connection is taken as the frontier curve. The preset condition is that there is no other data pair j, the entropy ratio of residual energy corresponding to j is not greater than the entropy ratio of residual energy of i, and the average correlation of j is not greater than the average correlation of i. In addition, at least one of these two indicators, the value of j is less than the value of i.

[0059] It should be noted that the frontier curve in this application is a boundary curve formed by connecting a series of non-dominated solutions.

[0060] In specific implementation, identifying the optimal point on the Pareto front curve that is suitable for the vibration and load changes of various mechanical components of the elevator can be achieved in the following way: Define a decision criterion for selecting the final compromise solution from the Pareto front curve. A typical known criterion is to calculate the Euclidean distance from each solution on the front curve to the ideal optimal point. The origin (0,0) can be selected as the ideal optimal point, representing the lowest residual signal energy in the decomposition result, that is, the vibration physical characteristics of each mechanical component of the elevator are completely extracted, and the degree of aliasing between any two modal components is minimal, that is, each modal component can best highlight the vibration physical characteristics of a mechanical component of the elevator. Then, select the data pair with the closest Euclidean distance as the optimal point. Furthermore, as a preferred embodiment, according to the emphasis on the sufficiency of decomposition and the independence of components in elevator diagnosis, normalized weights can be assigned to the two indicators in the data pair, the weighted sum of each data pair on the front curve can be calculated, and the data pair with the smallest weighted sum can be selected as the optimal point.

[0061] It should be noted that the "most advantageous point" in this application refers to the point corresponding to the optimal combination of parameters when analyzing the vibration signal of the target elevator.

[0062] In specific implementation, the following method is used to select multiple feature components containing fault information of various mechanical components of the elevator based on the vibration and impact during elevator failure: obtain all modal components corresponding to the optimal parameter combination, calculate the kurtosis value of each modal component to quantify its impact characteristics, and sort all modal components in descending order of kurtosis value. Then, calculate the energy of each modal component, sort all modal components in descending order of kurtosis value, and accumulate the energy starting from the component with the highest kurtosis. Plot the curve of accumulated energy changing with the number of components until the curve reaches an inflection point, i.e., where the slope changes significantly. Finally, all modal components participating in this accumulation are used as feature components containing fault information of various mechanical components of the elevator.

[0063] It should be noted that the characteristic components in this application are the key components in the vibration signal that carry fault information of various mechanical components of the elevator.

[0064] In some embodiments, constructing a fault feature vector set characterizing the state of the elevator traction system using the time and frequency domain features of all feature components can be achieved through the following steps: Select a feature component as the selected feature component, and determine multiple time-domain statistical features of the selected feature component; After converting the selected feature components to the frequency domain, multiple frequency domain features of the selected feature components are extracted. All time-domain statistical features and all frequency-domain features of the selected feature components are sequentially concatenated to form a fault feature vector characterizing the state of the elevator traction system; The fault feature vectors corresponding to the remaining feature components are then determined, thereby obtaining a set of fault feature vectors characterizing the state of the elevator traction system.

[0065] In practice, determining multiple time-domain statistical characteristics of a selected feature component can be achieved by calculating a series of standard time-domain statistical characteristics of the selected feature component. These time-domain statistical characteristics are common knowledge in the field of rotating machinery vibration analysis and fault diagnosis, and typically include, but are not limited to, root mean square value, peak factor, impulse factor, margin factor, kurtosis and waveform factor.

[0066] It should be noted that the time-domain statistical characteristics in this application refer to the time-domain judgment indicators commonly used in the field of rotating machinery vibration analysis and fault diagnosis, including but not limited to: root mean square value, peak factor, impulse factor, margin factor, kurtosis and waveform factor.

[0067] In practice, the extraction of multiple frequency domain features of the selected feature component after converting it to the frequency domain can be achieved in the following way: First, perform a fast Fourier transform on the time series of the selected feature component to convert it from the time domain to the frequency domain and obtain its amplitude spectrum. Then, extract a series of standard frequency domain features based on the amplitude spectrum. These frequency domain features are common knowledge in the field of rotating machinery vibration analysis and fault diagnosis, and usually include, but are not limited to: center of gravity frequency, frequency standard deviation, mean square frequency, spectral skewness, and spectral kurtosis.

[0068] It should be noted that the frequency domain characteristics in this application refer to the frequency domain judgment indicators commonly used in the field of rotating machinery vibration analysis and fault diagnosis, including but not limited to: center of gravity frequency, frequency standard deviation, mean square frequency, spectral skewness, and spectral kurtosis.

[0069] In practice, the fault feature vector representing the state of the elevator traction system can be constructed by sequentially concatenating all time-domain statistical features and all frequency-domain features of the selected feature component. This can be achieved as follows: After calculating the M time-domain statistical features and N frequency-domain statistical features of the feature component, they are combined into a one-dimensional array according to a predefined and fixed order (e.g., [time-domain feature 1, time-domain feature 2, ..., time-domain feature M, frequency-domain feature 1, frequency-domain feature 2, ..., frequency-domain feature N]). This array is the fault feature vector representing the state of the elevator traction system.

[0070] It should be noted that, in this application, the fault feature vector refers to a highly structured data vector used to transform the complex and non-stationary vibration state of the elevator traction system into a data vector that can be directly identified and classified by an artificial intelligence model.

[0071] Furthermore, it should be noted that step 104 in this application proposes a parameter refinement and feature selection mechanism based on dual-objective Pareto optimization, addressing the non-stationary characteristics of elevator vibration signals and the weakness of fault features. The core of this method is to treat the sufficiency of decomposition (measured by the proportion of residual energy entropy, requiring the extraction of as much useful information as possible) and the independence of components (measured by the average correlation between components, requiring the avoidance of information aliasing from different components) as a pair of mutually constraining optimization objectives. By traversing parameter combinations within the pre-converged search space and using the Pareto front to identify the optimal compromise point between these contradictory objectives, the optimal parameters are adaptively selected. The signal decomposition parameters are most suitable for the current elevator operating state (including specific component states and load conditions). Based on this, the impact characteristics (kurtosis) and energy concentration of elevator faults are further combined to lock the modal components carrying the core fault information from the optimal decomposition results. Finally, a complete vector set integrating time-frequency domain features is constructed. The advantage of doing this is that it systematically solves the problem of the failure of a single decomposition criterion due to the changing operating conditions in elevator scenarios. It ensures that the fault feature vector set input into the artificial intelligence model is not only complete in information but also high-quality data that can be classified, which fundamentally improves the accuracy and reliability of the subsequent intelligent diagnostic module.

[0072] In step 105, the fault feature vector set is input into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

[0073] In some embodiments, inputting the fault feature vector set into a trained artificial intelligence classification model for identification and classification, and then determining the fault type of the elevator traction system, can be achieved through the following steps: The fault feature vector set is used as input and imported into a pre-trained artificial intelligence classification model, and a vector containing the probability of each fault type of the elevator is output. Select the fault type with the highest probability value from the output vector as the diagnostic result of the elevator traction system fault type.

[0074] In specific implementation, the fault feature vector set is used as input and imported into a pre-trained artificial intelligence classification model. The output vector containing the probability of each fault type of the elevator can be implemented in the following way: the artificial intelligence classification model can directly select the short-sequence temporal convolutional neural network model trained in "Elevator Fault Diagnosis Method Based on Short-Sequence Temporal Convolutional Network". In other embodiments, other existing artificial intelligence models for fault identification based on elevator mechanical vibration signals can also be selected, which are not limited here.

[0075] In some embodiments, after determining the fault type of the elevator traction system, corresponding elevator control commands can be generated based on the fault type. As a preferred embodiment, the generation of corresponding elevator control commands based on the fault type in this application can be implemented in the following way: corresponding elevator control commands can be pre-set for each type of elevator fault. Subsequently, after obtaining the fault type, the elevator control command corresponding to the fault type is queried by looking up a table. The pre-set corresponding elevator control commands for each type of elevator fault can be set according to experience and actual needs based on the instruction manual provided by the elevator manufacturer. For example, in this application, if the diagnosis is severe wear of the traction machine bearing (high risk), the command is generated as follows: "Decelerate to the nearest floor, open the door and stop the elevator, trigger the audible and visual alarm, and send an emergency maintenance notification to the monitoring center"; if the diagnosis is slight insufficient lubrication of the guide rail (medium risk), the command is generated as: "Run at a limited speed and lock the elevator after the end of this running cycle, and send a preventive maintenance reminder"; if the diagnosis is that the car vibration slightly exceeds the threshold (low risk), the command is generated as: "Record the status, include it in the health trend analysis, and prompt for inspection during the next maintenance". In other embodiments, it can also be set in other ways, which are not limited here.

[0076] Furthermore, in another aspect of this application, in some embodiments, this application provides an artificial intelligence-based elevator control system, which includes a fault detection unit, referencing... Figure 4 The figure is a schematic diagram of the structure of a fault detection unit according to some embodiments of this application. The fault detection unit 400 includes: a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire vibration signals of elevator mechanical components during elevator operation; Processing module 402 in this application is mainly used to determine the difference in vibration physical characteristics between various mechanical components during elevator operation based on the spectrum of the vibration signal, and to dynamically expand the difference according to the load change during elevator operation to obtain the search range of vibration signal decomposition layer. It should be noted that the processing module 402 in this application is also used to determine the candidate range of bandwidth constraints when each mechanical component of the elevator vibrates, with the goal of minimizing the degree of overlap of the vibration physical characteristics between each mechanical component. It should be noted that the processing module 402 in this application is also used to traverse different parameter combinations within the search range and the candidate range to perform mode decomposition, and to filter out feature components containing fault information of each mechanical component of the elevator based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components, and then construct a fault feature vector set characterizing the state of the elevator traction system through the time domain and frequency domain features of all feature components. The execution module 403 in this application is mainly used to input the fault feature vector set into the trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

[0077] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described artificial intelligence-based elevator fault detection method.

[0078] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing an AI-based elevator fault detection method according to some embodiments of this application. The AI-based elevator fault detection method in the above embodiments can... Figure 5 The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0079] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0080] The communication bus 502 can be used to transmit information between the aforementioned components.

[0081] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CDROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0082] The memory 503 stores program code for executing the solution of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. The artificial intelligence-based elevator fault detection method in the above embodiment can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0083] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0084] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0085] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0086] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based elevator fault detection method.

[0087] In summary, the elevator fault detection method and system based on artificial intelligence disclosed in this application firstly collects vibration signals of elevator mechanical components during elevator operation; based on the spectrum of the vibration signals, the degree of difference in vibration physical characteristics between various mechanical components during elevator operation is determined, and the degree of difference is dynamically expanded according to the load change situation during elevator operation to obtain the search range of vibration signal decomposition layers; with the goal of minimizing the degree of aliasing of vibration physical characteristics between various mechanical components, a candidate range of bandwidth constraints for vibration of each mechanical component of the elevator is determined; different parameter combinations are traversed within the search range and the candidate range to perform modal decomposition, and feature components containing fault information of each mechanical component of the elevator are screened based on the entropy ratio of residual energy in each decomposition result and the correlation between adjacent components; then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components; the fault feature vector set is input into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

[0088] Therefore, this application dynamically combines the differences in vibration physical characteristics between various mechanical components with load uncertainty, i.e., sudden load changes, to enable the number of decomposition layers to adapt to real-time changes in component vibration characteristics and load conditions during elevator operation, thus improving the targeting of feature extraction. Subsequently, bandwidth constraint optimization guided by aliasing degree reduces modal confusion and ensures effective separation of vibration characteristics of each component. A dual-objective screening mechanism of residual energy entropy ratio and component correlation is adopted to enhance component independence while ensuring sufficient decomposition, thereby extracting purer and more representative fault features. Finally, an artificial intelligence model is used to achieve accurate classification of fault types and generate control commands accordingly, realizing a closed-loop intelligent response from fault perception, diagnosis to control, significantly improving the fault early warning capability, operational safety, and maintenance intelligence level of the elevator system. In summary, the solution of this application can completely and independently extract the vibration characteristics of each mechanical component of the elevator under conditions of vibration coupling and sudden load changes.

[0089] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An artificial intelligence-based elevator fault detection method, used by an artificial intelligence-based elevator control system to detect elevator faults, characterized in that, The method includes the following steps: Vibration signals of elevator mechanical components are collected during elevator operation; The difference in vibration physical characteristics between various mechanical components during elevator operation is determined based on the spectrum of the vibration signal, and the difference is dynamically expanded according to the load change during elevator operation to obtain the search range of vibration signal decomposition layers. The candidate range of bandwidth constraints for vibration of each mechanical component of the elevator is determined with the goal of minimizing the degree of overlap of vibration physical characteristics among the mechanical components. Modal decomposition is performed by traversing different parameter combinations within the search range and the candidate range. Based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components, feature components containing fault information of each mechanical component of the elevator are selected. Then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components. The fault feature vector set is input into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

2. The method as described in claim 1, characterized in that, Determining the difference in vibration physical characteristics between various mechanical components during elevator operation based on the spectrum of the vibration signal specifically includes: Based on the vibration signals, determine the vibration characteristic diagrams of each mechanical component during elevator operation; The resonance peaks in the vibration characteristic diagram are extracted to obtain the degree of difference in the vibration physical characteristics between the various mechanical components during elevator operation.

3. The method as described in claim 1, characterized in that, The difference is dynamically expanded based on the sudden changes in load during elevator operation, and the search range for the number of vibration signal decomposition layers specifically includes: The uncertainty of load change during elevator operation is determined based on the vibration signal. Based on the uncertainty, the difference is dynamically expanded to obtain the search range for the number of vibration signal decomposition layers.

4. The method as described in claim 1, characterized in that, The candidate range for determining the bandwidth constraint of each mechanical component of the elevator during vibration, with the goal of minimizing the aliasing of vibration physical characteristics among the components, specifically includes: The initial range of bandwidth constraint is preset for each mechanical component of the elevator during vibration; The initial range is divided into three equal parts to obtain two internal dividing points; The vibration signal is decomposed at each internal segmentation point to obtain the degree of aliasing between each component in each decomposition result. Based on all degrees of aliasing, the candidate range for bandwidth constraints during vibration of each mechanical component of the elevator is determined.

5. The method as described in claim 1, characterized in that, The modality decomposition process, which involves traversing different parameter combinations within the search range and the candidate range, specifically includes: Filter out multiple integer parameter combinations within the search range and the candidate range; The vibration signal is modally decomposed using different parameter combinations to obtain multiple modal components and residual signals corresponding to each parameter combination.

6. The method as described in claim 1, characterized in that, Based on the entropy ratio of residual energy in each decomposition result and the correlation between adjacent components, feature components containing fault information of various mechanical parts of the elevator are selected, specifically including: Select one decomposition result as the selected decomposition result, and obtain all modal components and residual signals corresponding to the selected decomposition result; Determine the entropy ratio of the residual energy in the vibration signal; Determine the correlation between every two modal components in all modal components; The average relevance is determined based on all relevance scores. Further determine the entropy percentage and average correlation of the residual energy in the remaining decomposition results; Feature components containing fault information of various mechanical components of the elevator were selected by filtering out the entropy ratio of all residual energy and all average correlations.

7. The method as described in claim 6, characterized in that, By filtering the entropy percentage of all residual energy and all average correlations, feature components containing fault information of various mechanical components of the elevator are selected. Specifically, this includes: The entropy percentage of all residual energy and all average correlations constitute the solution set; Based on the set of solutions, a frontier curve of the non-dominated solution is constructed on a two-dimensional plane with the entropy ratio of residual energy and the average correlation as coordinate axes; The optimal point for adapting to the vibration and load changes of various mechanical components of the elevator is identified on the leading edge curve. All modal components corresponding to the optimal point are obtained, and multiple feature components containing fault information of various mechanical components of the elevator are selected from them based on the vibration and impact during elevator failure.

8. An elevator control system based on artificial intelligence, comprising a fault detection unit, characterized in that, The fault detection unit includes: The acquisition module is used to collect vibration signals of the elevator's mechanical components during elevator operation; The processing module is used to determine the degree of difference in the vibration physical characteristics between the mechanical components during elevator operation based on the spectrum of the vibration signal, and to dynamically expand the degree of difference according to the load change during elevator operation to obtain the search range of the vibration signal decomposition layer. The processing module is also used to determine the candidate range of bandwidth constraints for vibration of each mechanical component of the elevator with the goal of minimizing the degree of overlap of vibration physical characteristics between each mechanical component. The processing module is also used to perform mode decomposition by traversing different parameter combinations within the search range and the candidate range, and to filter out feature components containing fault information of each mechanical component of the elevator based on the entropy ratio of the residual energy of each decomposition result and the correlation between adjacent components. Then, a fault feature vector set characterizing the state of the elevator traction system is constructed through the time domain and frequency domain features of all feature components. The execution module is used to input the fault feature vector set into a trained artificial intelligence classification model for identification and classification, thereby determining the fault type of the elevator traction system.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the AI-based elevator fault detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the artificial intelligence-based elevator fault detection method as described in any one of claims 1 to 7.

Images