Elevator system fault diagnosis method and device
By monitoring the three-axis vibration signals in the elevator and using the support vector machine for fault diagnosis, the problem of insufficient vertical signal analysis in the existing technology is solved, accurate identification and early warning of elevator faults is achieved, and the safety of the elevator system is improved and reliability.
Patent Information
- Application Number
- CN202411449244.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2024-11-29
AI Technical Summary
The existing elevator fault diagnosis methods mainly focus on the analysis of vibration signals in the vertical direction, which fails to fully reflect the impact of horizontal and vertical vibrations during elevator operation, resulting in inaccurate detection, inability to prevent potential faults in a timely manner, and safety hazards are posed.
Mobile phone equipment is used to monitor the vibration signals of the elevator in the x, y and z axes, with sampling frequency up to 500Hz, extract time and frequency domain features, use support vector machines to classify faults, and quantify the characteristic contribution degree through Sharpley values, and filter keys Features for troubleshooting.
It realizes a comprehensive capture of the operating status of the elevator, improves the accuracy and real-time nature of fault diagnosis, reduces calculation redundancy, and enhances the safety and reliability of the elevator system.
Smart Images

Figure CN119038346A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of elevators, and in particular to an elevator system fault diagnosis method and device. Background Art
[0002] With the acceleration of urbanization and the increase in the number of high-rise buildings, elevators have become an indispensable and important tool in modern buildings. Elevators operate frequently and involve the safety of a large number of people, so fault diagnosis is particularly important. The maintenance mode of traditional elevator systems mostly relies on regular manual inspections and repairs after sudden failures. This method is not only time-consuming and labor-intensive, but also fails to prevent and detect potential failures in a timely manner, which may lead to unnecessary safety hazards during operation. Therefore, sensor-based elevator fault diagnosis systems are receiving more and more attention, aiming to achieve early warning and prediction of faults through real-time data monitoring and analysis, and improve the safety and reliability of elevator systems.
[0003] Existing elevator fault diagnosis methods mainly focus on analyzing the operating status of the elevator by monitoring vibration signals through sensors. However, most existing technologies only monitor vibration signals in the vertical direction and fail to fully reflect the impact of lateral and longitudinal vibrations that may occur during the operation of the elevator. This single-axis signal analysis method has obvious limitations and is prone to inaccurate detection of certain faults, missing the early signs of certain faults.
[0004] To solve this problem, some studies have tried to introduce multi-axis sensor monitoring and combine it with machine learning algorithms for data analysis, hoping to more accurately identify potential faults in elevator systems. However, due to the large amount of data and complex features, existing methods still have room for improvement in feature selection and computational efficiency. Summary of the invention
[0005] The present invention aims to provide an elevator fault diagnosis system that fully utilizes the sensing capabilities of modern mobile devices to replace traditional IoT edge devices. The system uses mobile phones to obtain elevator vibration data, including vibration signals of the elevator in the x, y, and z axis directions, with a sampling frequency of up to 500Hz, ensuring high accuracy and comprehensiveness of the data.
[0006] The present invention provides an elevator system fault diagnosis method, comprising: in the elevator start-up, acceleration, uniform speed, deceleration and stop states, respectively monitoring the vibration signals of the x-axis, y-axis and z-axis through a mobile terminal fixed in the elevator; extracting the time domain features and frequency domain features in the vibration signal, wherein the time domain features include the variance, root mean square and kurtosis factor of the vibration signal, and the frequency domain features include the spectrum energy features obtained by wavelet packet decomposition and the elevator frequency features; using a support vector machine to classify the time domain features and the frequency domain features; using Shapley values to quantify the contribution of each feature in the time domain features and the frequency domain features; screening and obtaining key features for fault diagnosis according to the order of the contribution from large to small; and using the support vector machine to diagnose the elevator fault according to the key features.
[0007] Preferably, the spectrum energy characteristics obtained by wavelet packet decomposition include: Calculate the energy of each wavelet packet, where E 3j For data packet P j3 The energy of the discrete point, i is the subscript variable, j is the serial number of the data packet, N is the length of the data, x ij For data packet P j3 The amplitude of the discrete point; according to the formula Calculate the total energy E3 of the vibration signal; and according to the formula X f =E 3j / E3 performs normalization processing to obtain the spectrum energy characteristics of the vibration signal.
[0008] Preferably, before using the support vector machine to perform fault diagnosis based on the multi-domain features, the method further includes: Optimize the decision boundary, where w is the weight vector, is a nonlinear function, b is a bias term; using this nonlinear function Mapping the training features to a high-dimensional linear space; optimizing the hyperplane through the support vector machine according to the following formula to obtain the optimal values of the weight vector w and the bias term b; sty i (w T *x i +b)≥1-ξ i Among them, ξ i is a non-negative slack variable, i = 1, 2, ..., n, C is a penalty parameter indicating the degree of error tolerance; and according to the optimal value of the weight vector w and the bias term b, the decision function is obtained.
[0009] Preferably, according to the optimized hyperplane, a kernel function K(x i , x j )=φ(x i )*φ(xj ), and obtain the optimal values of the weight vector w and the bias term b.
[0010] Preferably, the contribution of each feature in the time domain feature and the frequency domain feature is quantified using the Shapley value according to the following formula: φ i represents the contribution, n represents the total number of the time domain features and the frequency domain features, N represents the set of the time domain features and the frequency domain features, S represents the feature subset excluding the i-th feature, and the function f(·) represents the predicted value of the model of the feature subset.
[0011] Preferably, using the Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature further comprises: fitting a linear model; each weight determined by the linear model corresponds to the Shapley value; Ω(g)=0, where |z'| is the number of non-zero elements in z'.
[0012] Preferably, using Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature further comprises: using kernel SHAP to process a complex nonlinear model, wherein the objective function L(φ) of the nonlinear model is: φ i represents the Shapley value of the i-th feature, v(S i ) indicates that in this feature subset S i The predicted value of v(S i ∪{i}) means that in this feature subset S i The predicted value after adding the i-th feature is added to the basis, and the kernel SHAP generates a weighted data set by minimizing the objective function; a local linear model is fitted to optimize the contribution of each feature to obtain the Shapley value of each feature.
[0013] Preferably, after the support vector machine is used to perform fault diagnosis according to the key feature, the method further includes: returning to the step of performing fault classification on the time domain feature and the frequency domain feature using the support vector machine according to the result of the fault diagnosis and executing the step in a loop.
[0014] According to another aspect of the present invention, there is provided an elevator system fault diagnosis device, comprising: a vibration signal monitoring module, used to monitor the x-axis, y-axis and z-axis vibration signals respectively through a mobile terminal fixed in the elevator when the elevator is started, accelerated, at a constant speed, decelerated and stopped; a feature extraction module, used to extract time domain features and frequency domain features in the vibration signal, wherein the time domain features include the variance, root mean square and kurtosis factor of the vibration signal, and the frequency domain features include spectral energy features obtained by wavelet packet decomposition and elevator frequency features; a fault classification module, used to use a support vector machine to perform fault classification on the time domain features and frequency domain features; a contribution quantification module, used to use Shapley values to quantify the contribution of each feature in the time domain features and the frequency domain features; a feature screening module, used to screen and obtain key features for fault diagnosis according to the contribution from large to small; and a fault diagnosis module, used to use the support vector machine to perform fault diagnosis on the elevator according to the key features.
[0015] Preferably, the feature extraction unit further comprises: a wavelet packet energy calculation unit for calculating the energy of the wavelet packet according to the formula Calculate the energy of each wavelet packet, where E 3j For data packet P j3 The energy of the discrete point, i is the subscript variable, j is the serial number of the data packet, N is the length of the data, x ij For data packet P j3 The amplitude of the discrete point; the total energy calculation unit is used to calculate the total energy according to the formula Calculate the total energy E3 of the vibration signal; and a normalization processing unit for calculating the total energy E3 of the vibration signal according to the formula X f =E 3j / E3 performs normalization processing to obtain the spectrum energy characteristics of the vibration signal.
[0016] The present invention discloses an elevator system fault diagnosis method and system, which can comprehensively capture the fault characteristics of the elevator under the running state through the monitoring and analysis of three-axis vibration signals, and then combine the Shapley value method to effectively screen key characteristics, reduce calculation redundancy, and improve the accuracy and real-time performance of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:
[0018] Figure 1 is a flow chart of a fault diagnosis method for an elevator system according to an embodiment of the present invention;
[0019] Figure 2It is a diagram of the overall technical solution of elevator fault diagnosis according to an embodiment of the present invention;
[0020] Figure 3 is a z-axis vibration data diagram of a complete sample according to an embodiment of the present invention;
[0021] Figure 4 is a z-axis vibration data diagram of an incomplete sample according to an embodiment of the present invention;
[0022] Figure 5 is an elevator frequency band diagram for feature extraction according to an embodiment of the present invention; and
[0023] Figure 6 1 is a diagram showing the structure of an elevator system fault diagnosis device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0024] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0025] Reference will now be made in detail to various embodiments of the present invention, examples of which are shown in the accompanying drawings and described below. For ease of interpretation and precise definition in the appended claims, the terms "upper", "lower", "inner" and "outer" are used to describe features of the exemplary embodiments shown in the drawings with reference to their positions.
[0026] The embodiment of the present invention provides a method for diagnosing faults in an elevator system. Figure 1 As shown, the method includes: monitoring the vibration signals of the x-axis, y-axis and z-axis respectively through a mobile terminal fixed in the elevator in the starting, accelerating, constant speed, decelerating and stopping states of the elevator; extracting the time domain features and frequency domain features in the vibration signal, wherein the time domain features include the variance, root mean square and kurtosis factor of the vibration signal, and the frequency domain features include the spectrum energy features obtained by wavelet packet decomposition and the elevator frequency features; using a support vector machine to classify the time domain features and the frequency domain features; using Shapley values to quantify the contribution of each feature in the time domain features and the frequency domain features; screening and obtaining key features for fault diagnosis according to the order of the contribution from large to small; and using the support vector machine to diagnose the fault of the elevator according to the key features.
[0027] In the related art, the elevator system fault diagnosis method only analyzes the vibration signal in the vertical direction, so the diagnosis result is inaccurate. In the embodiment of the present invention, through the monitoring and analysis of the three-axis vibration signal, the fault characteristics of the elevator under the running state can be fully captured, and then combined with the Shapley value method, the key characteristics can be effectively screened, the calculation redundancy can be reduced, and the accuracy and real-time performance of the diagnosis can be improved.
[0028] According to an embodiment of the present invention, the spectrum energy features obtained by wavelet packet decomposition include: Calculate the energy of each wavelet packet, where E 3j For data packet P j3 The energy of the discrete point, i is the subscript variable, j is the serial number of the data packet, N is the length of the data, x ij For data packet P j3 The amplitude of the discrete point; according to the formula Calculate the total energy E3 of the vibration signal; and according to the formula X f =E 3j / E3 performs normalization processing to obtain the spectrum energy characteristics of the vibration signal.
[0029] In the embodiment of the present invention, wavelet packet decomposition can effectively capture the frequency domain characteristics in the elevator vibration signal, and the energy vector after normalization processing enables the contribution of the frequency characteristics in different frequency bands to be clearly presented, thereby improving the accuracy of fault diagnosis.
[0030] According to an embodiment of the present invention, the contribution of each feature in the time domain feature and the frequency domain feature is quantified using the Shapley value according to the following formula: φ i represents the contribution, n represents the total number of the time domain features and the frequency domain features, N represents the set of the time domain features and the frequency domain features, S represents the feature subset excluding the i-th feature, and the function f(·) represents the predicted value of the model of the feature subset.
[0031] In the embodiment of the present invention, the formula is used to quantify the contribution of each feature to the model prediction, effectively helping to screen out the most important features for fault diagnosis, reducing redundant features, and improving the accuracy and efficiency of the model.
[0032] According to an embodiment of the present invention, using the Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature further includes: fitting a linear model; each weight determined by the linear model corresponds to the Shapley value; Ω(g)=0, where |z'| is the number of non-zero elements in z'.
[0033] In an embodiment of the present invention, a weighted data set is generated and a linear model is fitted to optimize feature contribution, which is particularly suitable for processing complex nonlinear models (such as support vector machines or other kernel function models), thereby enhancing the efficiency of feature selection and the interpretability of the model.
[0034] According to an embodiment of the present invention, using Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature further includes: using kernel SHAP to process a complex nonlinear model, and the objective function L(φ) of the nonlinear model is: φ i represents the Shapley value of the i-th feature, v(S i ) represents the predicted value in the feature subset Si; v(S i ∪{i}) means that in this feature subset S i The predicted value after adding the i-th feature is added to the basis, and the kernel SHAP generates a weighted data set by minimizing the objective function; a local linear model is fitted to optimize the contribution of each feature to obtain the Shapley value of each feature.
[0035] In the embodiment of the present invention, the formula minimizes the loss L(φ) so that the Shapley value of each feature i can accurately reflect the feature in the specific subset S i The kernel SHAP method optimizes feature contribution by generating a weighted data set and fitting a local linear model. It is particularly suitable for processing complex nonlinear models (such as support vector machines or other kernel function models), thereby enhancing the efficiency of feature selection and the interpretability of the model. Through the definition of these parameters and the use of the kernel SHAP method, the importance of features in complex models can be analyzed more accurately, making machine learning models more interpretable and transparent in decision making.
[0036] According to an embodiment of the present invention, before the support vector machine is used to perform fault diagnosis according to the key feature, the method further includes: Optimize the decision boundary, where w is the weight vector, is a nonlinear function, b is a bias term; using this nonlinear function Mapping the training features to a high-dimensional linear space; optimizing the hyperplane through the support vector machine according to the following formula to obtain the optimal values of the weight vector w and the bias term b; sty i (w T *x i +b)≥1-ξ i , where ξ i is a non-negative slack variable, i = 1, 2, ..., n, C is a penalty parameter indicating the degree of error tolerance; and according to the optimal value of the weight vector w and the bias term b, the decision function is obtained.
[0037] In the embodiment of the present invention, by using support vector machines, nonlinear classification problems can be effectively processed in high-dimensional space. By mapping key features into a high-dimensional linear space, the support vector machine can find the optimal hyperplane that maximizes the classification interval. At the same time, the optimization process introduces a penalty parameter C and a non-negative slack variable ξ i , thus ensuring that the model has good generalization ability while considering the tolerance of classification error. This method improves the accuracy and robustness of fault diagnosis, ensures the fault identification performance of the elevator system under complex operating conditions, and reduces the possibility of false alarms and missed alarms.
[0038] According to an embodiment of the present invention, based on the optimized hyperplane, a kernel function K(x i , x j )=φ(x i )*φ(x j ), and obtain the optimal values of the weight vector w and the bias term b.
[0039] In the embodiment of the present invention, the input features are mapped to a high-dimensional space by using a kernel function, thereby finding the optimal hyperplane in this space, which can effectively handle nonlinear classification problems. This method improves the classification ability of the support vector machine, so that the accuracy and robustness of elevator system fault diagnosis are significantly improved.
[0040] According to an embodiment of the present invention, after the support vector machine is used to perform fault diagnosis based on the key feature, it also includes: based on the result of the fault diagnosis, returning to the step of using the support vector machine to classify the time domain features and frequency domain features and executing it in a loop.
[0041] In the embodiment of the present invention, by introducing the step of loop execution, the fault diagnosis model can be adaptively adjusted according to the previous diagnosis result. This loop strategy helps to improve the learning ability of the model, enabling it to more accurately classify new fault features, and improve the dynamic adaptability and accuracy of elevator system fault diagnosis.
[0042] Another embodiment of the present invention provides an elevator system fault diagnosis device, such as Figure 6As shown, it includes: a vibration signal monitoring module, which is used to monitor the vibration signals of the x-axis, y-axis and z-axis respectively through a mobile terminal fixed in the elevator when the elevator is started, accelerated, uniform, decelerated and stopped; a feature extraction module, which is used to extract the time domain features and frequency domain features in the vibration signal, wherein the time domain features include the variance, root mean square and kurtosis factor of the vibration signal, and the frequency domain features include the spectrum energy features and elevator frequency features obtained by wavelet packet decomposition; a fault classification module, which is used to classify the time domain features and frequency domain features by using a support vector machine; a contribution quantification module, which is used to quantify the contribution of each feature in the time domain features and the frequency domain features by using Shapley values; a feature screening module, which is used to screen and obtain key features for fault diagnosis according to the order of the contribution from large to small; and a fault diagnosis module, which is used to diagnose the fault of the elevator by using the support vector machine according to the key features.
[0043] In the related art, the elevator system fault diagnosis method only analyzes the vibration signal in the vertical direction, so the diagnosis result is inaccurate. In the embodiment of the present invention, through the monitoring and analysis of the three-axis vibration signal, the fault characteristics of the elevator under the running state can be fully captured, and then combined with the Shapley value method, the key characteristics can be effectively screened, the calculation redundancy can be reduced, and the accuracy and real-time performance of the diagnosis can be improved.
[0044] According to an embodiment of the present invention, the feature extraction unit further includes: an energy calculation unit for calculating the energy according to the formula Calculate the energy of each wavelet packet, where E 3j For data packet P j3 The energy of the discrete point, i is the subscript variable, j is the serial number of the data packet, N is the length of the data, x ij For data packet P j3 The amplitude of the discrete point; the total energy calculation unit is used to calculate the total energy according to the formula Calculate the total energy E3 of the vibration signal; and a normalization processing unit for calculating the total energy E3 of the vibration signal according to the formula X f =E 3j / E3 performs normalization processing to obtain the spectrum energy characteristics of the vibration signal.
[0045] In the embodiment of the present invention, wavelet packet decomposition can effectively capture the frequency domain characteristics in the elevator vibration signal, and the energy vector after normalization processing enables the contribution of the frequency characteristics in different frequency bands to be clearly presented, thereby improving the accuracy of fault diagnosis.
[0046] The key contents of the present invention are:
[0047] This paper proposes an elevator fault diagnosis system that makes full use of the sensing capabilities of modern mobile devices to replace traditional IoT edge devices. The system uses mobile phones to obtain elevator vibration data, including vibration signals of the elevator in the x, y, and z axes, with a sampling frequency of up to 500Hz, ensuring high accuracy and comprehensiveness of the data.
[0048] During the data collection process, the mobile device is fixed inside the elevator to monitor the vibration of the elevator in real time. The collected vibration signals provide the basis for subsequent data processing. The system eliminates incomplete samples through a series of data cleaning and preprocessing processes to ensure that each sample fully covers all stages of elevator operation, including static, acceleration, constant speed, deceleration and stop states.
[0049] In order to improve the effectiveness of data analysis, the present invention has conducted in-depth research on feature extraction. First, through time domain and frequency domain analysis, various statistical features such as variance, peak-to-peak value, root mean square, kurtosis factor, skewness factor, crest factor, shape factor, margin factor, pulse factor, etc. are extracted from the vibration signal. These features can reflect the time domain vibration characteristics of the elevator in different operating stages.
[0050] In addition, considering the particularity of the elevator structure, the present invention introduces the structural frequency characteristics of the elevator. By analyzing the relevant frequencies and their multiples of the traction wheel, traction motor, guide wheel, anti-rope wheel and wire rope, the system can identify potential fault characteristics related to these components.
[0051] In frequency domain analysis, wavelet packet decomposition technology is applied to signal processing to decompose the original vibration signal into multiple wavelet packets. By calculating the energy of each wavelet packet, the system obtains rich frequency domain features.
[0052] The present invention uses support vector machines to classify faults. The system can find the optimal hyperplane in high-dimensional space and accurately distinguish between normal operation and fault states. SVM solves complex classification problems by mapping nonlinear data to high-dimensional space.
[0053] In order to improve the accuracy and computational efficiency of the model, the present invention also uses the SHapley AdditiveExplanations (SHAP) method for feature selection. SHAP is based on the Shapley value theory of cooperative game theory and provides a highly interpretable tool for feature selection. By calculating the marginal contribution of each feature, the system can identify the features that have the greatest impact on the results, thereby optimizing the feature set. SHAP effectively reduces redundant features, reduces computational costs, and enhances the generalization ability of the model by quantifying the contribution of each feature to the model output.
[0054] In general, the present invention provides an elevator fault detection and diagnosis system based on mobile devices, which has the advantages of high accuracy, high efficiency and low cost. By comprehensively utilizing the time domain and frequency domain characteristics of vibration signals, combined with advanced machine learning models and feature selection methods, the system provides strong technical support for the safe operation of elevators.
[0055] The present invention is further described in detail below in conjunction with specific implementation modes and accompanying drawings.
[0056] Figure 2 The overall technical solution flow chart of the present invention is as follows:
[0057] Data collection: The mobile phone is fixed on the floor inside the elevator car to monitor the vibration in real time and collect vibration signals in the x, y, and z axes. The sampling frequency is as high as 500Hz to ensure high accuracy and comprehensiveness of the data.
[0058] Data processing: Perform data cleaning and preprocessing to remove incomplete samples and ensure that each sample covers all operation stages of the elevator, including stationary, acceleration, constant speed, deceleration and stop.
[0059] Feature extraction: Extract various statistical features such as variance, root mean square, etc. through time domain analysis, introduce the frequency characteristics of the elevator structure, and analyze the relevant frequency and frequency doubling characteristics of key components.
[0060] Wavelet packet decomposition: Apply wavelet packet decomposition technology to decompose the original vibration signal into multiple wavelet packets, calculate the energy of each wavelet packet, and obtain frequency domain characteristics.
[0061] Fault diagnosis model: Use support vector machine to classify faults and find the optimal hyperplane to distinguish between normal and faulty states.
[0062] SHAP feature selection: Apply the SHAP method for feature selection to quantify the contribution of each feature to the model output, reduce redundant features, reduce computational costs, and enhance the generalization ability of the model.
[0063] (1) Data Collection
[0064] Mobile phones are used as data collection tools specifically for monitoring and analyzing elevator vibration signals. This method significantly improves the accuracy and response speed of elevator fault detection with its high efficiency, convenience and low cost. Mobile phones have powerful sensing capabilities and can simultaneously record the vibration data of the elevator in the x-axis, y-axis and z-axis directions. The sampling frequency of up to 500Hz ensures high resolution and precision of the data, so that even small vibration changes can be accurately captured. This high-frequency sampling provides a solid foundation for subsequent data analysis.
[0065] During the data collection process, the mobile phone was fixed in the center of the elevator car to capture the vibration of the elevator in real time. This setup can fully cover all stages of the elevator operation cycle, including starting, acceleration, constant speed, deceleration and stop.
[0066] (2) Data processing
[0067] Before feature extraction, the present invention needs to perform a series of steps to ensure the quality and availability of data.
[0068] First, data cleaning is crucial. To ensure the integrity of the training data, all incomplete samples, such as Figure 4 As shown in the figure, they will be deleted. Figure 3 As shown in Figure 1, a complete sample must cover the entire operation cycle of the elevator, including stationary, accelerating, constant speed, decelerating, and stopping states. These states provide comprehensive operation information, ensuring that the model can learn effective features from various operating conditions.
[0069] Next, it is also necessary to segment the elevator's operating phases. The vibration characteristics of the elevator in different operating phases vary significantly, especially in the acceleration and deceleration phases, where the vibration mode may show different characteristics compared to the uniform speed operation phase.
[0070] Therefore, the present invention divides the entire elevator operation cycle into three stages based on acceleration: acceleration, uniform speed and deceleration. Each stage has unique vibration characteristics, so it is necessary to perform feature extraction separately.
[0071] During the acceleration phase, the elevator accelerates from a stationary state. The vibration data during this process can reveal the mechanical characteristics of the elevator when it starts. During the uniform speed phase, the elevator maintains a constant speed and the vibration characteristics are relatively stable. The data in this phase is particularly important for detecting long-term wear or structural changes. The deceleration phase involves the elevator slowing down until it stops. Its vibration data can reflect the performance and potential problems of the braking system.
[0072] By analyzing the vibration data in stages, the features of each stage can be accurately extracted and analyzed, ensuring that the fault detection model can identify abnormal conditions in operation. This meticulous segmentation method enhances the model's diagnostic capabilities and improves the accuracy of fault prediction.
[0073] (3) Feature extraction
[0074] 1. Time domain characteristics
[0075] The vibration signal is x = {x1, x2, ..., x n The time domain characteristics of the x-axis, y-axis, and z-axis directions during the elevator operation phase are as follows:
[0076] Variance: Variance measures the deviation of the signal value from its mean, that is:
[0077] where x i is the amplitude of the discrete points of the signal, and n is the length of the vibration signal.
[0078] Peak-to-peak value: The peak-to-peak value is the amplitude range of the signal, that is: Peak_peak(x)=max(x)-min(x).
[0079] Root Mean Square (RMS): RMS reflects the overall energy of the signal, that is:
[0080] Kurtosis Factor: Kurtosis factor is a measure of the sharpness or peakedness of the vibration signal, namely:
[0081] Skewness factor: The skewness factor represents the asymmetry of the vibration signal distribution, that is:
[0082] Crest Factor: The crest factor provides insight into the magnitude of the instantaneous peaks within the signal, i.e.:
[0083] Shape Factor: The shape factor is the ratio of the RMS value to the mean value. It provides information about the flatness of the signal, i.e.:
[0084] Margin factor: The margin factor is defined as the ratio of the signal peak value to the square root of the amplitude, that is:
[0085] Pulse Factor: The pulse factor compares the height of the signal's peak to the signal's average level, i.e.
[0086] 2. Elevator frequency characteristics
[0087] Considering the structure and operation characteristics of the elevator, the present invention introduces the frequency characteristics of the elevator. The components that the present invention focuses on include the traction wheel, traction motor, guide wheel, anti-rope wheel and wire rope. The relevant frequencies are:
[0088] Traction wheel rotation frequency (f s ):
[0089] Where V is the speed, I is the suspension ratio, and D s is the diameter of the elevator traction sheave.
[0090] Traction motor pulse frequency (f m ):
[0091] Where P is the number of motor poles.
[0092] Guide wheel rotation frequency (f bp ):
[0093] Where D bp is the guide wheel diameter.
[0094] The rotation frequency of the anti-ropes (f cp ):
[0095] Where D cp is the diameter of the car return pulley.
[0096] Wire rope plying frequency (f rp ):
[0097] where pr is the distance between the two strands in the rope.
[0098] In the analysis of elevator fault vibration, it is found that the multiples of these frequencies are also part of the fault characteristics. Therefore, the present invention also focuses on the frequencies of 2 and 3 times the above elevator frequency values, such as Figure 5 shown.
[0099] 3. Wavelet Packet Energy
[0100] Wavelet packet decomposition decomposes the original vibration signal into multiple wavelet packets. For each packet, its energy is used as a feature of the signal. After decomposition, 8 data packets of the third layer are obtained, denoted as P j3 , where j = 0, 1, 2, 3, 4, 5, 6, 7. Calculate the energy of each packet:
[0101] Where i is the subscript variable, j is the number of the data packet, N is the length of the data, and x ij For data packet P j3 The amplitude of the discrete point. The total energy E3 of the signal is equal to the cumulative energy sum, that is:
[0102] After normalizing the energy, the frequency domain feature vector X of the original vibration signal is obtained. f as follows:
[0103] During the feature extraction process, a total of 165 features were obtained. Among them, there were 81 time domain features of the three-axis vibration data in the acceleration, deceleration and uniform speed stages. At the same time, the elevator frequency and its multiples of the elevator model extracted a total of 12 elevator features in the frequency. A total of 72 features were extracted from the wavelet packet energy values on each axis in each operating stage. Normalizing the extracted features helps to accelerate the gradient descent, so that the training network converges faster. The normalized data was then used as the input of the support vector machine classifier in the fault diagnosis process.
[0104] (4) Fault diagnosis model
[0105] 1. Fault diagnosis using support vector machines
[0106] The support vector machine constructs a classification model by converting the initial nonlinear data space into a high-dimensional space and using the support vector to calculate the optimal hyperplane. Mapping the training set x into a high-dimensional linear space enables linear modeling to estimate the decision function. The classifier plane is represented as The decision function is given by Given this, this approach allows linear methods to solve nonlinear problems efficiently.
[0107] In the nonlinear case, the optimization problem of the classification hyperplane is expressed as follows: sty i (w T *x i +b)≥1-ξ i ,
[0108] In the formula, ξ i ≥0, I=1, 2, …, n, C is the penalty parameter indicating the error tolerance, ξ i is a non-negative slack variable. By introducing the Lagrangian function, the problem is transformed into:
[0109] where α i is the support vector value, y i and j is the classification label. If 0≤a i ≤C, then x i is the support vector. The kernel function is defined as K(x i , x j )=φ(x i )*φ(x j ).
[0110] 2. SHapleyAdditive exPlanations selection features
[0111] SHapleyAdditive exPlanations (SHAP) is used to select features, reduce redundant features, lower computational costs, and ensure high accuracy and generalization ability of the model.
[0112] SHAP is a common method for explaining the relationship between input and output, which helps to gain insight into how each feature contributes to the model prediction for a particular instance. It is based on the Shapley value.
[0113] a. Shapley Value
[0114] The Shapley value, based on cooperative game theory principles, quantifies the contribution of each feature to the outcome. It is calculated as the average marginal contribution of all feature subsets. The Shapley value of the i-th feature It is expressed as:
[0115] Where n represents the total number of features, N represents the set of all features, and S represents the feature subset excluding the i-th feature. The function f(·) represents the predicted value of the model given the feature subset.
[0116] b. Core SHAP
[0117] Kernel SHAP is an extension of the SHAP framework designed to explain complex machine learning models. Kernel SHAP extends SHAP values to handle models that contain kernels, such as SVM and XGBoost. In the case of complex models, it is not feasible to calculate Shapley values exactly. Instead, Kernel SHAP uses an approximate method similar to Local Interpretable model-agnostic Explanations (LIME). The method involves generating a weighted dataset of perturbed instances and fitting an interpretable model (e.g., linear regression) to this dataset.
[0118] The specific value (π x ', L, Ω) are assigned to a predefined form. This process consists in fitting a linear model, where each weight determined by the final linear model corresponds to a Shapley value. Ω(g)=0,
[0119] where |z'| is the number of non-zero elements in z'.
[0120] The SHAP feature selection method first comprehensively analyzes the contribution of each feature in the model decision-making. The SHAP value is used as a quantitative indicator to sort the features according to their contribution to the model prediction accuracy. Then, from the sorted feature list, the feature subset that most significantly improves the model performance is selected from top to bottom (i.e., in descending order of importance). This process ensures that even when the number of features is limited, the information that is critical to the prediction results can be retained to the maximum extent, thereby effectively avoiding overfitting and improving the generalization ability and interpretability of the model.
[0121] The main advantages of the present invention are summarized as follows:
[0122] (1) Compared with traditional vertical signal feature extraction, the features based on triaxial signals retain more spatiotemporal information and are more suitable for practical applications.
[0123] (2) The frequency characteristics of the elevator model can improve the accuracy of fault diagnosis. These features change with the elevator parameters, which is conducive to migrating feature extraction and fault diagnosis algorithms to different types of elevators.
[0124] (3) Due to computing load limitations, we use SHAP feature selection to remove redundant features, so that the diagnostic method can maintain high accuracy at a lower computational cost. It effectively reduces the chance of overfitting.
[0125] The above embodiments are only examples for clearly explaining the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications may be made based on the following description, and these various changes, modifications, substitutions and deformations derived from the principles and spirit of the present invention still fall within the scope of protection of the present invention.
Claims
1. A method for diagnosing faults in an elevator system, characterized in that: include: When the elevator is started, accelerated, kept at a constant speed, decelerated, and stopped, the vibration signals of the x-axis, y-axis, and z-axis are monitored respectively by a mobile terminal fixed in the elevator; Extracting time domain features and frequency domain features from the vibration signal, wherein the time domain features include the variance, root mean square, and kurtosis factor of the vibration signal, and the frequency domain features include spectrum energy features obtained by wavelet packet decomposition and elevator frequency features; Using a support vector machine to perform fault classification on the time domain features and frequency domain features; The Shapley value is used to quantify the contribution of each feature in the time domain feature and the frequency domain feature; According to the order of the contribution from large to small, key features for fault diagnosis are screened and obtained; as well as According to the key features, the support vector machine is used to perform fault diagnosis on the elevator.
2. The method according to claim 1, characterized in that: The spectrum energy features obtained by wavelet packet decomposition include: According to the formula Calculate the energy of each wavelet packet, where E 3j For data packet P j3 The energy of the discrete point, i is the subscript variable, j is the serial number of the data packet, N is the length of the data, x ij For data packet P j3 The amplitude of the discrete points; According to the formula Calculating the total energy E3 of the vibration signal; and According to the formula X f =E 3j / E3 performs normalization processing to obtain the spectrum energy characteristics of the vibration signal.
3. The method according to claim 2, characterized in that According to the following formula, the Shapley value is used to quantify the contribution of each feature in the time domain feature and the frequency domain feature: φ i represents the contribution, n represents the total number of the time domain features and the frequency domain features, N represents a set of the time domain features and the frequency domain features, S represents the feature subset that does not include the i-th feature, The function f(·) represents the predicted value of the model for the feature subset.
4. The method according to claim 3, characterized in that Using the Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature also includes: Fit a linear model; Each weight determined by the linear model corresponds to the Shapley value; Ω(g)=0, where |z'| is the number of non-zero elements in z'.
5. The method according to claim 4, characterized in that Using the Shapley value to quantify the contribution of each feature in the time domain feature and the frequency domain feature also includes: Kernel SHAP is used to process complex nonlinear models, where the objective function L(φ) of the nonlinear model is: φ i represents the Shapley value of the ith feature, v(S i ) indicates that in the feature subset S i The predicted value of v(S i ∪{i}) represents the feature subset S i The predicted value after adding the i-th feature is By minimizing the objective function, the kernel SHAP generates a weighted data set; A local linear model is fitted to optimize the contribution of each feature to obtain the Shapley value of each feature.
6. The method according to any one of claims 1 to 5, characterized in that Before using the support vector machine to perform fault diagnosis according to the key features, the method further includes: pass Optimize the decision boundary, where w is the weight vector, is a nonlinear function, b is a bias term; Using the nonlinear function Mapping training features into a high-dimensional linear space; According to the following formula, the optimal values of the weight vector w and the bias term b are obtained by optimizing the hyperplane through the support vector machine; s.t.y i (w T *x i +b)≥1-ξ i . Among them, ξ i is a non-negative slack variable, i=1, 2, ..., n, C is a penalty parameter indicating the error tolerance; and According to the optimal value of the weight vector w and the bias term b, the decision function is obtained 7. The method according to claim 6, characterized in that According to the optimized hyperplane, the kernel function K(x i , x j )=φ(x i )*φ(x j ), and obtain the optimal values of the weight vector w and the bias term b.
8. The method according to any one of claims 1 to 5, characterized in that After the support vector machine is used to perform fault diagnosis according to the key features, the method further includes: returning to the step of using the support vector machine to perform fault classification on the time domain features and the frequency domain features according to the result of the fault diagnosis and executing the step in a loop.
9. An elevator system fault diagnosis device, characterized in that: include: A vibration signal monitoring module is used to monitor the vibration signals of the x-axis, y-axis and z-axis respectively through a mobile terminal fixed in the elevator when the elevator is started, accelerated, kept at a constant speed, decelerated and stopped; A feature extraction module, used to extract time domain features and frequency domain features from the vibration signal, wherein the time domain features include the variance, root mean square, and kurtosis factor of the vibration signal, and the frequency domain features include spectrum energy features obtained by wavelet packet decomposition and elevator frequency features; A fault classification module, used for performing fault classification on the time domain features and frequency domain features using a support vector machine; A contribution quantification module, used for quantifying the contribution of each feature in the time domain feature and the frequency domain feature by using Shapley value; A feature screening module, used to screen key features for fault diagnosis according to the order of contribution from large to small; as well as A fault diagnosis module is used to perform fault diagnosis on the elevator using the support vector machine according to the key features.
Citation Information
Cited By
Elevator fault diagnosis system based on AI intelligent data analysis
CN119503572A
Elevator fault diagnosis system based on AI intelligent data analysis
CN119503572B
Elevator brake fault lossless online diagnosis and fatigue prediction method and system
CN119953996A
Method and system for non-destructive online diagnosis of elevator brake faults and fatigue prediction
CN119953996B
Fault signal detection method and device
CN120970955A
