An elevator failure prediction method and system based on data processing

Abstract

The present application relates to elevator operation data processing technical field, especially to a kind of elevator fault prediction method and system based on data processing;When the present application is carried out denoising to elevator operation multi-source data, through the synergistic effect of adaptive shrinkage factor function, dynamic parameter steepness parameter, threshold offset parameter, signal and noise can be more accurately separated, compared with traditional soft / hard threshold or dynamic threshold method, the effect of the present application is more excellent in inhibiting random noise and periodic interference introduced by electromagnetic interference, mechanical friction, load fluctuation etc.

Description

Technical Field

[0001] This invention relates to the field of elevator operation data processing technology, and in particular to an elevator fault prediction method and system based on data processing. Background Technology

[0002] With the acceleration of urbanization, elevators have become an indispensable vertical transportation tool in modern buildings, and their safe operation is of paramount importance. Traditional elevator fault prediction methods mainly rely on periodic inspections, post-accident maintenance, or alarm systems based on simple thresholds. In recent years, with the development of Internet of Things (IoT) technology, elevator fault prediction methods based on sensor data have gradually increased. When using sensor data for elevator fault prediction, noise reduction processing is necessary because the data collected by sensors inevitably contains various noises, interferences, and invalid information.

[0003] Existing technologies include wavelet transform for denoising elevator operation data. In wavelet denoising, a threshold is typically used. To overcome the drawbacks of fixed thresholds, dynamic thresholding schemes have been developed. These schemes no longer use a globally uniform threshold but instead attempt to adjust the threshold based on local characteristics such as the amplitude and scale dependence of wavelet coefficients. For example, a common dynamic thresholding strategy is scale-based thresholding, where different thresholds are used for different decomposition levels (j). It is generally assumed that higher-level detail coefficients have more noise, so the threshold is set higher. Another approach is thresholding based on coefficient amplitude ranking, such as the SureShrink method, which estimates the optimal threshold based on the coefficients of each level.

[0004] The dynamism of existing dynamic thresholding schemes mainly lies in predefined, global rules, such as the threshold of layer j being twice that of layer 1. These rules are static; once determined, they apply the same to all signals and data segments throughout the denoising process, failing to truly consider the local signal-to-noise ratio environment of each specific wavelet coefficient. For a given coefficient, regardless of whether its amplitude is high or low relative to its local background noise, it will be processed by the same level of rules. This results in less than ideal denoising performance for non-stationary signals where noise levels change drastically with time and frequency. More importantly, existing techniques typically combine dynamic thresholding with static thresholding functions. That is, they simply replace the fixed threshold with a scale-varying threshold, but when processing individual coefficients, they still use discontinuous or biased soft / hard thresholding functions. For example, after calculating the optimal threshold for each layer, the SureShrink method still applies the standard soft thresholding function to all coefficients in that layer. This creates an inherent contradiction and inconsistency between dynamic thresholding and static processing methods, limiting the overall improvement in denoising performance. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a data processing-based elevator fault prediction method and system, which solves the problems existing in the prior art.

[0006] This invention provides a data processing-based elevator fault prediction method, the method comprising the following steps: S1: Install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; S2: Perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data; S3: Perform adaptive noise reduction processing based on improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data; Specifically, S3 consists of: S3.1: performing wavelet decomposition on the pre-processed multi-source data to obtain approximation coefficients and detail coefficients; S3.2: estimating the noise standard deviation of the detail coefficients; S3.3: performing adaptive denoising on the detail coefficients based on the noise standard deviation; S3.4: performing inverse wavelet transform on the adaptively denoised detail coefficients and the unprocessed approximation coefficients to obtain the denoised multi-source data. S4: Perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; S5: Input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction.

[0007] Preferably, in step S3.2, for the detail coefficients of the j-th layer, the noise standard deviation is estimated based on the median absolute deviation. The specific formula is as follows: ; In the formula, Here is the noise standard deviation estimate for the detail coefficients of the j-th layer. Let be the median of the absolute values ​​of the coefficient sequence of the detail coefficients at level j. Let be the detail coefficient of the j-th layer.

[0008] Preferably, in S3.3, for each coefficient in the detail coefficients of the j-th layer... , The length of the detail coefficients in this layer, and the detail coefficients for noise reduction. Given by the following formula: ; In the formula, This represents the result of denoising the k-th wavelet coefficient of the j-th layer. For the k-th detail coefficient of the j-th layer, The base threshold for the detail coefficients of the j-th layer is calculated using the following formula: , Let be the noise standard deviation estimate of the detail coefficients of the j-th layer, α be the steepness parameter, β be the threshold offset parameter, ψ() be the adaptive shrinkage factor function, and be the normalized coefficient magnitude. The function is given by the formula: ; In the formula, Let k be the value of the adaptive shrinkage factor function at the kth detail coefficient. For the normalized detail coefficient magnitude, Let be the steepness parameter corresponding to the k-th detail coefficient. >0; This is the threshold offset parameter corresponding to the k-th detail coefficient. >0.

[0009] Preferably, the size of the local window is defined as N, where N is an odd number between 5 and 11, and the normalized magnitude of the N detail coefficients within the local window is taken. The ratio of the mean of the normalized amplitudes of the N detail coefficients within the local window to the standard deviation of the noise is used as the local signal-to-noise ratio of the k-th detail coefficient. The specific formula is as follows: ; In the formula, Let be the local signal-to-noise ratio of the k-th detail coefficient. The mean of the normalized magnitudes of the N detail coefficients within the local window; The local signal-to-noise ratio gradient of the k-th detail coefficient The center difference of the local signal-to-noise ratio for the N detail coefficients within the local window; The specific formula is as follows: ; In the formula, and These represent the local signal-to-noise ratios of the adjacent windows on the right and left sides of the center, respectively. The steepness parameter corresponding to the kth detail coefficient The calculation formula is: ; In the formula, and These are the minimum steepness parameter and the maximum steepness parameter, respectively, and γ is the attenuation coefficient; The threshold offset parameter corresponding to the kth detail coefficient The calculation formula is: ; In the formula, and These are the minimum and maximum offset parameters, respectively, and η is the growth coefficient.

[0010] Preferably, in step S3.1, the db4 wavelet basis function is selected to perform wavelet decomposition on the pre-processed multi-source data; the number of decomposition layers is set to 5.

[0011] Preferably, in step S1, the various types of sensors include vibration sensors, temperature sensors, current sensors, voltage sensors, and speed sensors.

[0012] Preferably, in step S2, the preliminary data preprocessing operation includes three steps: data cleaning, data synchronization, and data formatting.

[0013] Preferably, the data cleaning is used to identify and remove or correct obvious outliers and erroneous data in the data; the data synchronization includes timestamp unification and sampling frequency alignment; the timestamp alignment is to assign a unified millisecond timestamp to each collected data point, and the sampling frequency alignment is to interpolate and generate high-frequency data points as needed between two sampling time points for low-frequency sensor data; The data formatting refers to organizing the cleaned and synchronized data into a standardized format suitable for computer algorithm processing, specifically including data structuring and metadata annotation.

[0014] Preferably, the data is structured into a two-dimensional matrix D, where the rows of the two-dimensional matrix D are discrete sampling times, the columns of the two-dimensional matrix D are different sensor channels, and the elements of the two-dimensional matrix D are the measurement values ​​of the j-th sensor channel at the i-th sampling time. The metadata annotation adds detailed metadata information to each data matrix, including: the elevator's unique identifier, the start and end times of data acquisition, the name, unit, range, and sampling frequency of each sensor channel, and the elevator's current operating condition label.

[0015] According to another aspect of the present invention, a data processing-based elevator fault prediction system is provided, the system employing the above-described data processing-based elevator fault prediction method, the system comprising: The data acquisition module is used to install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; The preliminary preprocessing module is used to perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data. The noise reduction module is used to perform adaptive noise reduction processing based on an improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data. The feature extraction module is used to perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; The fault prediction module is used to input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction.

[0016] The embodiments of the present invention have the following technical effects: When performing noise reduction processing on multi-source elevator operation data, this invention achieves more accurate separation of signal and noise through the synergistic effect of an adaptive shrinkage factor function and dynamic parameters such as steepness and threshold offset. Compared with traditional soft / hard thresholding or dynamic thresholding methods, this invention is more effective in suppressing random noise and periodic interference introduced by electromagnetic interference, mechanical friction, load fluctuations, etc.

[0017] Early elevator faults often manifest as transient impact signals with extremely small amplitude and short duration. Traditional denoising methods often smooth out these weak features due to excessive smoothing, leading to missed detections. This invention significantly improves the ability to retain weak signals through flexible threshold transition and local signal-to-noise ratio sensing. By using LSG-driven local adaptive parameters, the denoising process can sense and respond to dynamic changes in the local noise environment, significantly improving the robustness of the algorithm under non-stationary conditions.

[0018] This invention introduces local signal-to-noise ratio (SNR) as a parameter driving factor for steepness and local SNR gradient as a parameter driving factor for threshold offset for the first time. It not only considers the magnitude of local SNR but also captures its changing trend, thereby identifying the region where fault features are emerging and dynamically adjusting the contraction behavior accordingly. This allows for a more accurate distinction between elevator operating noise and weak fault features. In elevator operating signals, weak features such as the minor impact of early bearing failures and the modulation sidebands of gear meshing can be preserved more completely, while background noise is more effectively suppressed. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of an elevator fault prediction method based on data processing provided in an embodiment of the present invention; Figure 2 This is a flowchart of adaptive noise reduction processing based on improved wavelet threshold for multi-source data after preliminary preprocessing, provided by an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0022] Example 1, Figure 1 A flowchart of a data processing-based elevator fault prediction method is shown, such as... Figure 1 As shown, an elevator fault prediction method based on data processing includes the following steps: S1: Install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; The various types of sensors include vibration sensors, temperature sensors, current sensors, voltage sensors, and speed sensors.

[0023] The vibration sensor is preferably a piezoelectric vibration sensor or an IEPE accelerometer, as it has high sensitivity, wide frequency response range and good dynamic characteristics, making it suitable for capturing the impact and modulation components in mechanical vibration signals. The vibration sensor is installed above the non-drive end bearing seat of the traction machine housing, on the traction machine base, at the center or bottom of the car top, or near the guide rail joint. The sampling frequency is set to 1kHz to 10kHz to meet the requirements for capturing high-frequency vibration fault characteristics.

[0024] The temperature sensor is used to monitor the stator winding of the traction motor, the bearings of the traction motor, the oil temperature of the gearbox, the brake coil and friction pads, and the ambient temperature of the machine room. The temperature sensor adopts a high-precision NTC thermistor or PT100 platinum resistance temperature sensor, etc., and is installed inside the motor stator winding or through a dedicated temperature measuring hole for contact installation, in the upper space of the gearbox oil sump, on the return oil pipeline, or near the brake coil. The sampling frequency is set to 1 time / minute to 1 time / 5 minutes to monitor the slow temperature change trend and detect overheating in a timely manner.

[0025] The current and voltage sensors are used to monitor the three-phase operating current and voltage of the traction motor, as well as the current of key circuits such as the brake contactor and safety circuit. The current and voltage sensors are Hall effect current sensors, voltage divider / isolation acquisition modules, etc., to achieve non-contact measurement. The current and voltage sensors are installed at the main power input terminal of the traction motor or at relevant circuit nodes in the control cabinet. The sampling frequency is set to 1kHz to 5kHz to analyze the distortion and imbalance of the current waveform, as well as the current surge during the start-up / braking process.

[0026] The speed sensor is used to monitor the real-time operating speed of the car. An encoder is used to collect the real-time operating speed of the car. The encoder is installed at the end of the traction machine shaft and the car speed is indirectly calculated by measuring the rotational speed of the traction sheave. The sampling frequency of the speed sensor is set to 1kHz to 2kHz to obtain an accurate speed curve.

[0027] During the data acquisition process, the raw signals output by the sensors cannot usually be directly processed by the digital system and need to be conditioned and digitized. Therefore, a signal conditioning module and a data acquisition card are set up to achieve signal conditioning and digitization of the sensor sampling signals.

[0028] The signal conditioning module is used to amplify weak signals with programmable gain to improve the signal-to-noise ratio; to perform anti-aliasing filtering in the analog domain to filter out noise above the Nyquist frequency and prevent spectral aliasing during subsequent sampling; to provide photoelectric or magnetic isolation for strong electrical signals to protect subsequent acquisition equipment from high-voltage impacts and ground loop interference; and to linearize the nonlinear outputs of sensors such as temperature sensors.

[0029] The data acquisition card converts the conditioned analog signal into a digital signal and performs timestamp marking and multi-channel synchronous acquisition.

[0030] S2: Perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data; The raw data inevitably contains various noises, interferences, and invalid information, and must undergo preliminary data preprocessing before the next step of deep denoising. In this step, the preliminary data preprocessing operation includes three stages: data cleaning, data synchronization, and data formatting.

[0031] The data cleaning is used to identify and remove or correct obvious outliers and erroneous data. Specifically, for isolated, short-lived spikes, the data point is directly removed; for continuous outliers, the mean, median, or linear interpolation of the nearest normal data points is used to replace them to ensure the continuity of the data.

[0032] The data synchronization includes timestamp unification and sampling frequency alignment. Timestamp alignment involves assigning a unified timestamp accurate to milliseconds to each acquired data point, typically done by the clock module of the data acquisition card, and synchronizing with the master clock via Network Time Protocol (NTP) or GPS timing to ensure that the time base of all data acquisition nodes is consistent. Sampling frequency alignment involves interpolating high-frequency data points between two sampling time points for low-frequency sensor data as needed. In this embodiment, linear interpolation is used to achieve sampling frequency alignment.

[0033] The data formatting involves organizing the cleaned and synchronized data into a standardized format suitable for computer algorithm processing, specifically including data structuring and metadata annotation. Data structuring involves organizing the data into a two-dimensional matrix D, where rows of matrix D represent discrete sampling times, columns represent different sensor channels, and elements represent the measurement values ​​of the j-th sensor channel at the i-th sampling time. Metadata annotation adds detailed metadata information to each data matrix, including: a unique elevator identifier, data acquisition start and end times, the name, unit, range, and sampling frequency of each sensor channel, and the elevator's current operating condition label. In this embodiment, the elevator's current operating condition label includes normal operation, full load on the uphill direction, no load on the downhill direction, and maintenance mode, which can be obtained by reading the elevator control system status word.

[0034] S3: Perform adaptive noise reduction processing based on improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data; In wavelet denoising, a threshold is typically used to achieve noise reduction. To overcome the drawbacks of fixed thresholds, dynamic thresholding schemes have been developed. These schemes no longer use a globally uniform threshold but instead attempt to adjust the threshold based on local characteristics such as the amplitude and scale dependence of wavelet coefficients. For example, a common dynamic thresholding strategy is scale-based thresholding, where different thresholds are used for different decomposition levels j, with higher-level detail coefficients typically considered to have greater noise and thus set higher thresholds; or thresholding based on coefficient amplitude ranking, such as the SureShrink method, which estimates the optimal threshold based on the coefficients of each level.

[0035] The dynamism of existing dynamic thresholding schemes mainly lies in predefined, global rules, such as the threshold of layer j being twice that of layer 1. These rules are static; once determined, they apply the same to all signals and data segments throughout the denoising process, failing to truly consider the local signal-to-noise ratio environment of each specific wavelet coefficient. For a given coefficient, regardless of whether its amplitude is high or low relative to its local background noise, it will be processed by the same level of rules. This results in less than ideal denoising performance for non-stationary signals where noise levels change drastically with time and frequency. More importantly, existing techniques typically combine dynamic thresholding with static thresholding functions. That is, they simply replace the fixed threshold with a scale-varying threshold, but when processing individual coefficients, they still use discontinuous or biased soft / hard thresholding functions. For example, after calculating the optimal threshold for each layer, the SureShrink method still applies the standard soft thresholding function to all coefficients in that layer. This creates an inherent contradiction and inconsistency between dynamic thresholding and static processing methods, limiting the overall improvement in denoising performance.

[0036] Based on this, this embodiment provides a wavelet thresholding denoising method based on a hyperbolic tangent threshold function with local signal-to-noise ratio adaptation. Specifically, as... Figure 2 As shown, S3 specifically refers to: S3.1: Perform wavelet decomposition on the pre-processed multi-source data to obtain approximation coefficients and detail coefficients; In this step, each type of data from the multi-source data is decomposed into sub-bands of different scales and frequencies to allow for separate processing of different frequency components. Noise is mainly concentrated in the high-frequency detail components, while the main energy of the useful signal is concentrated in the low-frequency approximation components and some high-frequency detail components.

[0037] Based on the characteristics of elevator operation signals, orthogonal or bioorthogonal wavelet bases with good regularity and tight support properties are selected. In this step, the db4 wavelet basis function is selected to perform wavelet decomposition on the pre-processed multi-source data.

[0038] The number of decomposition layers determines the fineness of frequency division. If there are too few layers, high-frequency noise may be confused with the useful signal; if there are too many layers, boundary effects and computational redundancy may be introduced. In this step, the number of decomposition layers is set to 5.

[0039] Wavelet decomposition is performed on the pre-processed multi-source data to obtain one set of approximation coefficients and five sets of detail coefficients. The approximation coefficients represent the lowest frequency trend components of the signal, and the detail coefficients represent the high frequency detail components of the signal at different decomposition scales.

[0040] S3.2: Estimate the noise standard deviation of the aforementioned detail coefficients; For the detail coefficients of the j-th layer, the noise standard deviation is estimated based on the median absolute deviation. The specific formula is as follows: ; In the formula, Here is the noise standard deviation estimate for the detail coefficients of the j-th layer. Let be the median of the absolute values ​​of the coefficient sequence of the detail coefficients at level j. Let be the detail coefficient of the j-th layer.

[0041] S3.3: Perform adaptive noise reduction on the detail coefficients based on the noise standard deviation of the detail coefficients; For each coefficient in the detail coefficients of the j-th level , The length of the detail coefficients in this layer, and the detail coefficients for noise reduction. Given by the following formula: ; In the formula, This represents the result of denoising the k-th wavelet coefficient of the j-th layer. For the k-th detail coefficient of the j-th layer, The base threshold for the detail coefficients of the j-th layer is calculated using the following formula: , Let be the noise standard deviation estimate of the detail coefficients of the j-th layer, α be the steepness parameter, β be the threshold offset parameter, ψ() be the adaptive shrinkage factor function, and be the normalized coefficient magnitude. The function is given by the formula: ; In the formula, Let k be the value of the adaptive shrinkage factor function at the kth detail coefficient. For the normalized detail coefficient magnitude, Let be the steepness parameter corresponding to the k-th detail coefficient. >0, controls the steepness of the transition of the control function ψ() from 0 to 1. The larger the value, the narrower the transition region, and the more decisive the distinction between high and low signal-to-noise ratio coefficients. The smaller the value, the smoother the transition zone and the gentler the processing; This is the threshold offset parameter corresponding to the k-th detail coefficient. >0 indicates the starting point at which the adaptive shrinkage factor function begins to increase significantly, equivalent to a dynamic, soft threshold position related to the coefficient magnitude. This makes the threshold no longer a fixed value, but rather dynamically changes with r.

[0042] Furthermore, from the above formula, it can be seen that the steepness parameter and threshold offset parameter This method, which is the core of calculating the adaptive shrinkage factor function, aims to address the problems of static parameters and lack of local awareness in existing dynamic thresholding methods. Compared with existing technologies, this embodiment proposes an adaptive calculation method based on the local signal-to-noise ratio gradient, which makes the steepness parameter... and threshold offset parameter It can dynamically adjust according to the local environment of the wavelet coefficients of each data point, thereby more accurately matching the non-stationary characteristics of elevator vibration signals.

[0043] Specifically, the size of the local window is defined as N, where N is an odd number between 5 and 11, and the normalized magnitude of the N detail coefficients within the local window is taken. The ratio of the mean of the normalized amplitudes of the N detail coefficients within the local window to the standard deviation of the noise is used as the local signal-to-noise ratio of the k-th detail coefficient. The specific formula is as follows: ; In the formula, Let be the local signal-to-noise ratio of the k-th detail coefficient. It is the mean of the normalized magnitudes of the N detail coefficients within the local window.

[0044] The local signal-to-noise ratio gradient of the k-th detail coefficient The center difference of the local signal-to-noise ratio for the N detail coefficients within the local window; The specific formula is as follows: ; In the formula, and These represent the local signal-to-noise ratios of the adjacent windows on the right and left sides of the center, respectively.

[0045] The steepness parameter corresponding to the kth detail coefficient The calculation formula is: ; In the formula, and These are the minimum steepness parameter and the maximum steepness parameter, respectively, and γ is the attenuation coefficient, which is set to 3 in this embodiment.

[0046] The threshold offset parameter corresponding to the kth detail coefficient The calculation formula is: ; In the formula, and These are the minimum and maximum offset parameters, respectively, where η is the growth coefficient, and β is controlled by... In this embodiment, the rate of ascent is set to 2.

[0047] This embodiment introduces local signal-to-noise ratio (SNR) as a parameter driving factor for steepness and local SNR gradient as a parameter driving factor for threshold offset for the first time. It not only considers the magnitude of local SNR but also captures its changing trend, thereby identifying the region where fault features are emerging and dynamically adjusting the contraction behavior accordingly. This allows for a more accurate distinction between elevator operating noise and weak fault features. In the elevator operating signal, weak features such as the small impact of early bearing failure and the modulation sideband of gear meshing can be preserved more completely, while background noise is more effectively suppressed.

[0048] S3.4: Perform inverse wavelet transform on the detail coefficients after adaptive denoising and the unprocessed approximation coefficients to obtain the denoised multi-source data; This embodiment achieves more precise signal-to-noise separation through the synergistic effect of an adaptive shrinkage factor function, dynamic steepness parameters, and threshold offset parameters. Compared to traditional soft / hard thresholding or dynamic thresholding methods, this embodiment is superior in suppressing random noise and periodic interference introduced by electromagnetic interference, mechanical friction, load fluctuations, etc.

[0049] Early elevator faults often manifest as transient impact signals with extremely small amplitude and short duration. Traditional denoising methods often smooth out these weak features due to excessive smoothing, leading to missed detections. This embodiment significantly improves the ability to retain weak signals by using flexible threshold transition and local signal-to-noise ratio sensing. Through LSG-driven local adaptive parameters, the denoising process can sense and respond to dynamic changes in the local noise environment, significantly improving the robustness of the algorithm under non-stationary conditions.

[0050] S4: Perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; In step 3, multi-level and multi-dimensional features are extracted from the denoised sensor channel data to form a comprehensive feature vector F. The comprehensive feature vector includes time-domain features, frequency-domain features, time-frequency-domain features, and multi-sensor data correlation features.

[0051] The time-domain features include the mean, variance, standard deviation, maximum value, minimum value, peak value, skewness, and kurtosis of each type of signal within a preset time window; the frequency-domain features include the spectral center, root mean square frequency, and frequency variance of each type of signal within a preset time window; the time-frequency-domain features include wavelet packet node energy entropy and Hilbert marginal spectral energy moment; the multi-sensor data correlation features include the correlation coefficients of different sensors, phase synchronization features, and linkage anomaly indicators; the phase synchronization features are used to analyze the phase difference stability or synchronization index of different sensor signals in a specific frequency band; the linkage anomaly indicators are a measure of the linkage between door opening / closing time and abnormal car vibration.

[0052] S5: Input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction; In this step, the elevator fault prediction model is an artificial neural network model.

[0053] The comprehensive feature vector F obtained through step S4 and its corresponding labels are divided into a training set, a validation set, and a test set. The artificial neural network model is trained using a training set. During the training process, the hyperparameters of the artificial neural network model are adjusted using a validation set to optimize the performance of the artificial neural network model on the validation set. The comprehensive feature vector F is input into the trained artificial neural network model to obtain the elevator fault prediction result.

[0054] Example 2: The present invention also provides an elevator fault prediction system based on data processing. The system employs an elevator fault prediction method based on data processing as described in Example 1. The system includes: The data acquisition module is used to install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; The preliminary preprocessing module is used to perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data. The noise reduction module is used to perform adaptive noise reduction processing based on an improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data. The feature extraction module is used to perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; The fault prediction module is used to input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction.

[0055] Example 3: The present invention also provides an electronic device, including one or more processors and a memory.

[0056] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0057] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and a processor may execute the program instructions to implement a data processing-based elevator fault prediction method and / or other desired functions as described in any embodiment of this application. Various contents such as initial extrinsic parameters and thresholds may also be stored in the computer-readable storage medium.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A data processing-based elevator fault prediction method, characterized in that, The method includes the following steps: S1: Install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; S2: Perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data; S3: Perform adaptive noise reduction processing based on improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data; Specifically, S3 consists of: S3.1: performing wavelet decomposition on the pre-processed multi-source data to obtain approximation coefficients and detail coefficients; S3.2: estimating the noise standard deviation of the detail coefficients; S3.3: performing adaptive denoising on the detail coefficients based on the noise standard deviation; S3.4: performing inverse wavelet transform on the adaptively denoised detail coefficients and the unprocessed approximation coefficients to obtain the denoised multi-source data. In S3.3, for each coefficient in the detail coefficients of the j-th layer... , The length of the detail coefficients in this layer, and the detail coefficients for noise reduction. Given by the following formula: ； In the formula, This represents the result of denoising the k-th wavelet coefficient of the j-th layer. For the k-th detail coefficient of the j-th layer, The base threshold for the detail coefficients of the j-th layer is calculated using the following formula: , Let be the noise standard deviation estimate of the detail coefficients at layer j, α be the steepness parameter, and β be the threshold offset parameter. Here, is the adaptive shrinkage factor function, and is the normalized coefficient magnitude. The function is given by the formula: ； In the formula, Let k be the value of the adaptive shrinkage factor function at the kth detail coefficient. For the normalized detail coefficient magnitude, Let be the steepness parameter corresponding to the k-th detail coefficient. >0; This is the threshold offset parameter corresponding to the k-th detail coefficient. >0; Define the local window size as N, where N is an odd number between 5 and 11, and take the normalized magnitude of the N detail coefficients within the local window. The ratio of the mean of the normalized amplitudes of the N detail coefficients within the local window to the standard deviation of the noise is used as the local signal-to-noise ratio of the k-th detail coefficient. The specific formula is as follows: ； In the formula, Let be the local signal-to-noise ratio of the k-th detail coefficient. The mean of the normalized magnitudes of the N detail coefficients within the local window; The local signal-to-noise ratio gradient of the k-th detail coefficient The center difference of the local signal-to-noise ratio for the N detail coefficients within the local window; The specific formula is as follows: ； In the formula, and These represent the local signal-to-noise ratios of the adjacent windows on the right and left sides of the center, respectively. The steepness parameter corresponding to the kth detail coefficient The calculation formula is: ； In the formula, and These are the minimum steepness parameter and the maximum steepness parameter, respectively, and γ is the attenuation coefficient; The threshold offset parameter corresponding to the kth detail coefficient The calculation formula is: ； In the formula, and These are the minimum and maximum offset parameters, respectively, and η is the growth coefficient. S4: Perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; S5: Input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction.

2. The elevator fault prediction method based on data processing according to claim 1, characterized in that, In S3.2, for the detail coefficients of the j-th layer, the noise standard deviation is estimated based on the median absolute deviation. The specific formula is as follows: ； In the formula, This is an estimate of the noise standard deviation of the detail coefficients at layer j. Let be the median of the absolute values ​​of the coefficient sequence of the detail coefficients at level j. Let be the detail coefficient of the j-th layer.

3. The elevator fault prediction method based on data processing according to claim 2, characterized in that, In step S3.1, the db4 wavelet basis function is selected to perform wavelet decomposition on the pre-processed multi-source data; the number of decomposition layers is set to 5.

4. The elevator fault prediction method based on data processing according to claim 1, characterized in that, In S1, the various types of sensors include vibration sensors, temperature sensors, current sensors, voltage sensors, and speed sensors.

5. The elevator fault prediction method based on data processing according to claim 1, characterized in that, In step S2, the preliminary data preprocessing operation includes three steps: data cleaning, data synchronization, and data formatting.

6. The elevator fault prediction method based on data processing according to claim 5, characterized in that, The data cleaning is used to identify and remove or correct obvious outliers and erroneous data; the data synchronization includes timestamp unification and sampling frequency alignment. The timestamp alignment is to assign a uniform millisecond timestamp to each collected data point, and the sampling frequency alignment is to interpolate and generate high-frequency data points as needed between two sampling time points for low-frequency sensor data. The data formatting refers to organizing the cleaned and synchronized data into a standardized format suitable for computer algorithm processing, specifically including data structuring and metadata annotation.

7. The elevator fault prediction method based on data processing according to claim 6, characterized in that, The data is structured into a two-dimensional matrix D, where the rows of the two-dimensional matrix D are discrete sampling times, the columns of the two-dimensional matrix D are different sensor channels, and the elements of the two-dimensional matrix D are the measurement values ​​of the j-th sensor channel at the i-th sampling time. The metadata annotation adds detailed metadata information to each data matrix, including: elevator unique identifier, data acquisition start and end time, name, unit, range, and sampling frequency of each sensor channel, and elevator current operating condition label.

8. A data processing-based elevator fault prediction system, characterized in that, The data-processing-based elevator fault prediction system employs the data-processing-based elevator fault prediction method according to any one of claims 1-7, and the data-processing-based elevator fault prediction system includes: The data acquisition module is used to install various types of sensors in key parts of the elevator to collect multi-source data on elevator operation; The preliminary preprocessing module is used to perform preliminary data preprocessing on the multi-source data of the elevator operation to obtain the pre-processed multi-source data. The noise reduction module is used to perform adaptive noise reduction processing based on an improved wavelet threshold on the pre-processed multi-source data to obtain the noise-reduced multi-source data. The feature extraction module is used to perform multi-scale, multi-feature fusion feature extraction on the denoised multi-source data to obtain a comprehensive feature vector F; The fault prediction module is used to input the comprehensive feature vector F into the elevator fault prediction model to realize elevator fault prediction.

Images