Escalator health condition diagnosis system

Abstract

The present application relates to the technical field of escalator diagnosis, in particular to an escalator health condition diagnosis system, comprising: a data acquisition module, which acquires original data of escalator vibration signals obtained by a preset escalator side exploration module, forms escalator vibration signal data to be extracted after data preprocessing, and takes escalator vibration signal data to be extracted as data that meets the data specifications preset by a characteristic value extraction module; the characteristic value extraction module sequentially performs feature extraction and feature analysis on the escalator vibration signal data to be extracted, obtains characteristic values of vibration signals of each component of the escalator, integrates and inputs into a working condition recognition module; the working condition recognition module takes the characteristic values of the vibration signals of each component of the escalator as the input of the working condition recognition model that has been trained in advance, calibrates the running conditions and potential faults of each component of the escalator based on the output results, and completes the diagnosis of the health condition of the escalator.

Description

Technical Field

[0001] This invention relates to the field of escalator technology, and more specifically, to an escalator health status diagnosis system. Background Technology

[0002] With the continuous advancement of science and technology, the construction of escalators has also developed rapidly, and the number of escalator devices used in rail transit, commercial centers, and other areas has increased significantly. Escalators are complex special mechanical equipment. After long-term operation, the mechanical components of escalators usually experience aging and wear. Currently, escalator maintenance and inspection typically rely on experienced maintenance personnel to identify these issues. On the one hand, relying on the experience of maintenance personnel to identify escalator malfunctions carries a significant human element, increasing unreliability. On the other hand, some escalator components are enclosed within the assembly, making it difficult for maintenance personnel to visually inspect their wear and tear. Maintenance personnel can only detect these issues when the escalator exhibits vibrations or overheating, by which time the worn or faulty components have already been in operation for a considerable period. This not only affects subsequent escalator maintenance but also poses certain safety hazards. Summary of the Invention

[0003] The purpose of this invention is to provide an escalator health status diagnosis system to improve the low reliability of manual inspection.

[0004] The embodiments of the present invention are achieved through the following technical solutions:

[0005] Escalator health status diagnostic system, including:

[0006] The data acquisition module collects the raw data of escalator vibration signals obtained by the preset escalator side exploration module. After data preprocessing, it forms the escalator vibration signal data to be extracted, which is used to represent the data that meets the preset data specifications of the feature value extraction module.

[0007] The feature extraction module sequentially extracts and analyzes the features of the escalator vibration signal data to be extracted, so as to obtain the feature values ​​of the vibration signals of each component of the escalator, integrates them and inputs them into the working condition identification module.

[0008] The operating condition identification module takes the vibration signal characteristic values ​​of each component of the escalator as input to a pre-trained operating condition identification model. Based on the output results, it calibrates the operating conditions and potential faults of each component of the escalator to complete the diagnosis of the escalator's health status.

[0009] Optionally, the escalator-side exploration module consists of external sensors, an escalator mainboard, and a DTU device. The external sensors are connected to the DTU device via 485 communication, and the escalator mainboard is connected to the DTU device via 232 communication. The external sensors include: a 3-axis vibration sensor for the main unit, a 3-axis vibration sensor for the main drive chain, a 3-axis vibration sensor for the step chain tensioner, a handrail belt temperature sensor, a temperature sensor for the main unit and control cabinet, a power meter sensor, and a main unit noise sensor.

[0010] Optionally, the data acquisition module also acquires the time series and amplitude series of the corresponding time-domain signal each time it acquires the raw data of the escalator vibration signal.

[0011] Optionally, feature extraction is performed on the escalator vibration signal data to be extracted. The specific processing step for feature extraction is time-frequency transformation, and the specific calculation formula for time-frequency transformation is as follows:

[0012]

[0013]

[0014] Where FD_amp is the amplitude sequence of the frequency domain signal obtained after one-sided Fourier transform, N is the data length of the time domain signal, Y is the amplitude sequence of the frequency domain signal, Fs is the sampling frequency, and FD_f is the frequency sequence of the frequency domain signal obtained after one-sided Fourier transform.

[0015] Optionally, after feature extraction, the escalator vibration signal data to be extracted forms the escalator vibration signal data to be analyzed. Based on various derivation models, feature analysis is performed on the escalator vibration signal data to be analyzed to obtain the feature values ​​of the vibration signals of each component of the escalator. Among them, the various derivation models include: peak value derivation model, peak-to-peak value derivation model, frequency value derivation model, acceleration set frequency derivation model, harmonic value derivation model, and steepness derivation model.

[0016] Optionally, the pre-trained working condition recognition model further includes a preset process before training is completed, the steps of which are as follows:

[0017] The mechanical simulation module acquires escalator vibration signal simulation data, and the feature value extraction module processes the escalator vibration signal simulation data to generate simulation feature values ​​of vibration signals of each escalator component. The simulation feature values ​​of vibration signals of escalator unit components are used as input data, and the output labels are calibrated according to the corresponding simulation working conditions to generate training data of unit components. The above steps are repeated to integrate and form a training dataset.

[0018] Input the training dataset into the Statistics and Machine Learning Toolbox module for data classification to obtain the preset working condition recognition model.

[0019] Optionally, after completing the above-mentioned preset process, the working condition identification model is trained by a machine learning algorithm to obtain the pre-trained working condition identification model. Specifically, the machine learning algorithm is one of linear regression algorithm, support vector machine algorithm, and K-nearest neighbor algorithm.

[0020] The technical solutions of the embodiments of the present invention have at least the following advantages and beneficial effects:

[0021] This embodiment analyzes the vibration acceleration spectrum of the main components of the escalator in real time, monitors the operating status and fault status of the escalator in real time, uses machine learning algorithms to study fault warning trends, and identifies potential operating conditions of the escalator, thereby changing "regular maintenance" to "needed maintenance", and ultimately achieving the goal of reducing the occurrence of accidents and reducing the maintenance costs throughout the entire life cycle of the escalator. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the escalator health status diagnosis system provided by the present invention;

[0023] Figure 2 This is a framework diagram of the escalator health status diagnosis system provided by the present invention;

[0024] Figure 3 A schematic diagram of the escalator-side exploration module provided by the present invention;

[0025] Figure 4 This is a schematic diagram of an example of using three important components of a motor as test components, provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0027] like Figure 1 , Figure 2 As shown, the present invention provides one embodiment: an escalator health status diagnosis system, comprising:

[0028] The data acquisition module collects the raw data of escalator vibration signals obtained by the preset escalator side exploration module. After data preprocessing, it forms the escalator vibration signal data to be extracted, which is used to represent the data that meets the preset data specifications of the feature value extraction module.

[0029] The feature extraction module sequentially extracts and analyzes the features of the escalator vibration signal data to be extracted, so as to obtain the feature values ​​of the vibration signals of each component of the escalator, integrates them and inputs them into the working condition identification module.

[0030] The operating condition identification module takes the vibration signal characteristic values ​​of each component of the escalator as input to a pre-trained operating condition identification model. Based on the output results, it calibrates the operating conditions and potential faults of each component of the escalator to complete the diagnosis of the escalator's health status.

[0031] In this embodiment, the completed working condition recognition model is specifically obtained by pre-training the working condition recognition model through machine learning algorithms. It is used to receive the vibration signal feature values ​​of each component of the escalator as input, and it is not necessary to retrain the model every time the vibration signal feature values ​​of each component of the escalator are input.

[0032] In one application of this embodiment, the function of the data acquisition module is to acquire the input raw data sequence into the system and realize the normalization of the sequence. For the acquisition of vibration signal data, each acquisition of each channel of data requires the time sequence and amplitude sequence of the time domain signal, and the data acquisition module will ensure the normalization of the corresponding data sequence. In this embodiment, the length of a single data sequence is 8192 points; the sampling frequency is 10KHz.

[0033] like Figure 3 As shown, in this embodiment, the escalator-side exploration module consists of external sensors, an escalator mainboard, and a DTU device. The external sensors are connected to the DTU device via 485 communication, and the escalator mainboard is connected to the DTU device via 232 communication. The external sensors include: a 3-axis vibration sensor for the main unit, a 3-axis vibration sensor for the main drive chain, a 3-axis vibration sensor for the step chain tensioner, a handrail temperature sensor, a temperature sensor for the main unit and control cabinet, a power meter sensor, and a main unit noise sensor.

[0034] In this embodiment, the data acquisition module also acquires the time sequence and amplitude sequence of the corresponding time-domain signal each time it acquires the raw data of the escalator vibration signal.

[0035] In this embodiment, feature extraction is performed on the escalator vibration signal data to be extracted. The specific processing step for feature extraction is time-frequency transformation, and the specific calculation formula for time-frequency transformation is as follows:

[0036]

[0037]

[0038] Where FD_amp is the amplitude sequence of the frequency domain signal obtained after one-sided Fourier transform, N is the data length of the time domain signal, Y is the amplitude sequence of the frequency domain signal, Fs is the sampling frequency, and FD_f is the frequency sequence of the frequency domain signal obtained after one-sided Fourier transform.

[0039] In one application of this embodiment, time-frequency transformation is a crucial step in feature extraction. Time-frequency transformation typically employs the Fourier transform algorithm, with the commonly used Fourier transform being the bilateral Fourier transform, which handles both positive and negative frequency values. This embodiment, focusing on the analysis of escalator vibration, only involves positive frequency values, employing a one-sided fast Fourier transform. The difference between the one-sided fast Fourier transform and the ordinary fast Fourier transform lies in the use of translation and truncation methods, mapping the transformation result only to the positive half-axis region. Based on this transformation, the derivation and calculation formula for the time-frequency transformation are as follows:

[0040] N = Length(X)

[0041]

[0042]

[0043]

[0044] P1(2:end-1)=2*P1(2:end-1)

[0045] FD_amp=P1

[0046]

[0047] In the above derived calculation formula, N is the data length of the time-series signal; Length is the length of the solution sequence; X is the amplitude sequence of the time-series sequence; Y is the amplitude sequence of the frequency domain signal; W N is the Fourier operator; e is the natural base; j is the unit sequence; k is the sequence number; P2 is the amplitude of the frequency domain signal after the bilateral Fourier transform; FD_amp(Frequency Domain_amplitude) is the amplitude sequence of the frequency domain signal obtained after the one-sided Fourier transform; FD_f(Frequency Domain_frequency) is the frequency sequence of the frequency domain signal obtained after the one-sided Fourier transform.

[0048] In addition, in this embodiment, time-domain feature values ​​and frequency-domain feature values ​​are applied simultaneously. After feature extraction, the escalator vibration signal data to be extracted forms the escalator vibration signal data to be analyzed. Based on various derivation models, feature analysis is performed on the escalator vibration signal data to be analyzed to obtain the feature values ​​of the vibration signals of each component of the escalator. Among them, the various derivation models include: peak value derivation model, peak-to-peak value derivation model, frequency conversion value derivation model, acceleration set frequency derivation model, harmonic value derivation model, and steepness derivation model.

[0049] In one application of this embodiment, the peak value derivation model refers to the maximum instantaneous value of the signal amplitude within the considered time interval. In this embodiment, the peak value considers both the maximum peak value (pvmax) and the minimum peak value (pvmin) of the signal. These two values ​​define the maximum and minimum amplitude of the signal, and their derivation model is as follows:

[0050] pv max =Max(TD_amp)

[0051] pv min =Min(TD_amp)

[0052] In the above derivation model, TD_amp (TimeDomain_amplitude) is the amplitude sequence of the time-domain signal (sampled signal); Max is the solver for the maximum value in the sequence; and Min is the solver for the minimum value in the sequence.

[0053] In one application of this embodiment, the peak-to-peak value derivation model refers to the difference between the highest and lowest values ​​of the signal within one period, describing the magnitude of the signal value variation range. In this embodiment, the peak-to-peak value is the difference between the maximum and minimum amplitude of the vibration signal, and is a positive number, describing the magnitude of the amplitude span of the vibration signal. Its derivation model is as follows:

[0054] ppv = pv max -pv min

[0055] In the above derivation model, pv max It is the maximum peak value; PV min It is the minimum peak value.

[0056] In one application of this embodiment, the passband value derivation model refers to the peak-to-peak value of the signal's frequency domain amplitude, describing the breadth of the frequency domain amplitude range. In this embodiment, it is used to characterize the degree of difference in frequency domain energy distribution. The larger the passband value, the greater the difference in the signal's energy distribution; the smaller the passband value, the smaller the difference in the signal's energy distribution. Its derivation model is as follows:

[0057] tfv = Max(FD_amp) - Min(FD_amp)

[0058] In the above derivation model, FD_amp(Frequency Domain_amplitude) is the amplitude sequence of the frequency domain signal obtained after one-sided Fourier transform.

[0059] In one application of this embodiment, the acceleration set-frequency derivation model refers to calculating the maximum value of the frequency domain amplitude within each frequency band, used to characterize the energy proportion of the signal in each frequency band. If the high-frequency acceleration value is much greater than the low-mid-frequency value, it indicates that most of the signal energy is distributed in the high-frequency signal; conversely, if the low-frequency acceleration value is much greater than the mid-high-frequency value, it indicates that most of the signal energy is distributed in the low-frequency band. In this embodiment, the acceleration set-frequency refers to the high, medium, and low frequencies of acceleration, and its division adopts an equal division method, that is, each frequency band has an equal length. The specific derivation model is as follows:

[0060] pvifds_h = Max(FD_amp(high))

[0061] pvifds_m = Max(FD_amp(mid))

[0062] pvifds_l = Max(FD_amp(low))

[0063] In the above derivation model, FD_amp(high) is the frequency domain amplitude sequence of the high-frequency band; FD_amp(mid) is the frequency domain amplitude sequence of the mid-frequency band; and FD_amp(low) is the frequency domain amplitude sequence of the low-frequency band.

[0064] In one application of this embodiment, the harmonic value derivation model refers to the harmonic amplitude of a specific frequency. In this embodiment, the specified specific frequency value is the fundamental frequency value, and the harmonics are taken as the 1st, 2nd, and 3rd harmonics. The physical meaning of this characteristic value is to directly monitor the fundamental frequency vibration, attempting to find the differences in the fundamental frequency vibration of components corresponding to different types of signals. The model derivation is as follows:

[0065] index=Find(FD_f(:)==N*bf)

[0066] Nbf_amp = FD_amp(index)

[0067] In the above derivation model, index is the serial number of the frequency value in the frequency sequence of the frequency domain signal that is equal to the harmonic value; FD_f is the frequency sequence of the frequency domain signal; N is the multiplier of the harmonic value; bf is the fundamental frequency value of the component; Find is the serial number of a specific value in the sequence; Nbf_amp sequence is the amplitude of the harmonic value from 1 to N.

[0068] In one application of this embodiment, the kurtosis derivation model refers to the peak state of the signal sequence distribution, which is used to characterize the tightness of the signal value distribution. The higher the kurtosis of the acquired vibration signal, the larger its variance value and the more extreme the signal distribution (i.e., the vibration signal distribution is closer to the left and right ends of the horizontal axis). In this embodiment, the model derivation is as follows:

[0069] n = Length(TD_amp)

[0070]

[0071]

[0072]

[0073] In the above derivation model, n is the data length of the amplitude sequence of the time-domain signal; μ is the data mean of the amplitude sequence of the time-domain signal; σ is the data variance of the amplitude sequence of the time-domain signal; and kv is the kurtosis.

[0074] In this embodiment, the function of the feature value extraction module is to extract feature values ​​from signal data. The most important part of the feature value extraction module is feature value analysis. The input to the feature value analysis system is the data from the data acquisition module; the output is a set of feature values. The feature value information that can be extracted in this embodiment is shown in the table below:

[0075]

[0076] In this embodiment, the preset process of the working condition identification model includes the following steps:

[0077] The mechanical simulation module acquires escalator vibration signal simulation data, and the feature value extraction module processes the escalator vibration signal simulation data to generate simulation feature values ​​of vibration signals of each escalator component. The simulation feature values ​​of vibration signals of escalator unit components are used as input data, and the output labels are calibrated according to the corresponding simulation working conditions to generate training data of unit components. The above steps are repeated to integrate and form a training dataset.

[0078] Input the training dataset into the Statistics and Machine Learning Toolbox module for data classification to obtain the preset working condition recognition model.

[0079] In one application of this embodiment, the MATLAB platform is selected for model training. The specific operation is as follows:

[0080] 1) Constructing the training dataset. Collect raw vibration signal data (from the mechanical simulation section), and obtain the signal feature values ​​using a feature value extraction algorithm. Use the collected feature values ​​as a set of input data, and assign output labels according to the corresponding simulation conditions to generate a set of data. Repeat the above process to construct the dataset required for training.

[0081] 2) Training Environment Preparation. Import the dataset into the Statistics and Machine Learning Toolbox. Select the data type to complete the training environment configuration.

[0082] 3) Select the desired machine learning algorithm as the kernel, train it, and obtain the trained model. View the model's detailed information and perform analysis. Export and save the model for subsequent improvement and testing.

[0083] In this embodiment, the machine learning algorithm specifically selected is one of the following: linear regression algorithm, support vector machine algorithm, and K-nearest neighbor algorithm.

[0084] In one application of this embodiment, the linear regression algorithm is divided into two types: simple linear regression, with only one independent variable; and multivariate regression, with at least two or more independent variables. The modeling process of the multivariate regression algorithm used in this embodiment involves using data points to find the best-fit line. Its model is as follows:

[0085] f(x) = β1x1 + β2x2 + ... + β n x n +β0

[0086] In one application of this embodiment, the Support Vector Machine (SVM) algorithm is a classification algorithm. The SVM model represents instances as points in space and uses a hyperplane to separate data points. The distance of a point from the hyperplane reflects the degree of difference in classification. The hyperplane model is as follows:

[0087] ωx+b=0

[0088] |ωx+b|=0 can relatively represent the distance of point x from the hyperplane. Whether the sign of ωx+b is consistent with the sign of the class label y can indicate whether the classification is correct. The quantity y(ωx+b) can be used to represent the correctness and confidence of the classification.

[0089] When this model is placed on a hyperplane, and some constraints are applied to w, it can be transformed into a geometric margin problem, thus making it applicable to multivariate problems. The mathematical model of the geometric margin is as follows:

[0090]

[0091] In one application of this embodiment, the K-nearest neighbors algorithm uses the nearest neighbor (k) to predict unknown data points. The value of k is a key factor in prediction accuracy. Whether for classification or regression, measuring the weight of neighbors is very useful, as closer neighbors have greater weights than distant neighbors.

[0092] The K-Nearest Neighbors algorithm calculates the distance between samples to determine which training samples are closer to the test samples. In practical applications, the distance calculation method is often selected based on the application scenario and the characteristics of the data itself. When existing distance methods cannot meet the needs of practical applications, it is necessary to propose a distance metric suitable for the specific problem. Therefore, this embodiment uses L... p Distance, its mathematical model is as follows:

[0093]

[0094] Where, x i x j ∈X, where X is the eigenspace of n-dimensional real eigenvectors, and when p=1, L p L is the Manhattan distance; when p=2, p For Euclidean distance; when p = ∞, L p This represents the maximum distance between each coordinate.

[0095] like Figure 4 As shown, this embodiment also provides an example using three important components of the motor as test components: the main motor, the main gearbox, and the anchor bolts. Sensors are installed on these three components to collect vibration data. The collected vibration data are used to extract feature values, which are then used as input to the working condition identification model. The output of the working condition identification model represents the potential working conditions of each component. Based on the identified working condition results, subsequent safety analysis and risk mitigation measures can be specified.

[0096] This embodiment also provides a mapping table between the dataset and labels used for model training. Operating condition numbers 1-3 correspond to the operating conditions of the motor components, which corresponds to operating condition recognition model 1; operating condition numbers 4-8 correspond to the operating conditions of the gearbox, which corresponds to operating condition recognition model 2; and operating condition numbers 9-11 correspond to the operating conditions of the anchor bolts, which corresponds to operating condition recognition model 3. The mapping table is shown below:

[0097]

[0098] This embodiment converts the output of the working condition identification model into predicted working condition results, and compares the predicted working condition results with the actual working condition results. The accuracy rates of the models corresponding to each algorithm are shown in the table below:

[0099] motor Model 1 99.3％ 100％ 99.3％ Gearbox Model 2 98.8％ 99.6％ 98.4％ Anchor bolts Model 3 58％ 58％ 62.7％

[0100] In summary, this embodiment analyzes the vibration acceleration spectrum of the main components of the escalator in real time, monitors the operating and fault states of the escalator in real time, uses machine learning algorithms to study fault warning trends, and identifies potential operating conditions of the escalator, thereby changing "periodic maintenance" to "needed maintenance", ultimately achieving the goal of reducing accidents and reducing maintenance costs throughout the escalator's life cycle.

[0101] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An escalator health status diagnosis system, characterized in that, include: The data acquisition module collects the raw data of escalator vibration signals obtained by the preset escalator side exploration module. After data preprocessing, it forms the escalator vibration signal data to be extracted, which is used to represent the data that meets the preset data specifications of the feature value extraction module. The feature extraction module sequentially extracts and analyzes the features of the escalator vibration signal data to be extracted, so as to obtain the feature values ​​of the vibration signals of each component of the escalator, integrates them and inputs them into the working condition identification module. Feature extraction is performed on the escalator vibration signal data. The specific processing step for feature extraction is time-frequency transformation, and the specific calculation formula for time-frequency transformation is as follows: in, This is the amplitude sequence of the frequency domain signal obtained after a one-sided fast Fourier transform. The data length of the time-domain signal. It is the amplitude sequence of the frequency domain signal. Sampling frequency, This is the frequency sequence of the frequency domain signal obtained after a one-sided fast Fourier transform; The formula for calculating the time-frequency transform of a one-sided fast Fourier transform is as follows: In the above calculation formula, It is the data length of the time-domain signal; It is to find the length of the sequence; It is the amplitude sequence of a time series; It is a sequence of amplitude values ​​of a frequency domain signal; It is a Fourier operator; It is the natural base; It is a unit sequence; It is a serial number; It is the amplitude of the frequency domain signal obtained by the bilateral Fourier transform; It is the amplitude sequence of the frequency domain signal obtained after a one-sided fast Fourier transform; It is the frequency sequence of the frequency domain signal obtained after a one-sided fast Fourier transform; After feature extraction, the escalator vibration signal data to be extracted is formed into escalator vibration signal data to be analyzed. Based on various derivation models, feature analysis is performed on the escalator vibration signal data to be analyzed to obtain the feature values ​​of the vibration signals of each component of the escalator. Among them, the various derivation models include: peak value derivation model, peak-to-peak value derivation model, frequency value derivation model, acceleration set frequency derivation model, harmonic value derivation model, and steepness derivation model. The acceleration setpoint frequency derivation model refers to calculating the maximum value of the frequency domain amplitude within each frequency band, used to characterize the energy ratio of the signal in each frequency band. The acceleration setpoint frequency refers to the high, medium, and low frequencies of acceleration, and its division adopts an equal division method, that is, each frequency band has an equal length. The model derivation is as follows: in, (high) is a frequency domain amplitude sequence in the high-frequency band; (mid) is the frequency domain amplitude sequence of the mid-frequency band; (low) is the frequency domain amplitude sequence of the low-frequency band; The derivation model for harmonic values ​​refers to the harmonic amplitude at a specific frequency, where the harmonics are 1, 2, or 3. The model derivation is as follows: in, It is the series number of the frequency sequence of the frequency domain signal in which the frequency value is equal to the harmonic value; It is the frequency sequence of the frequency domain signal; N' is the multiplier of the harmonic value; It is the fundamental frequency value of the component; It is to find the sequence number of a specific value in the sequence; The sequence consists of the amplitudes of 1 to N' times the harmonic frequency. The kurtosis derivation model refers to the peak state of a signal sequence distribution, used to characterize the density of the signal value distribution. Its model derivation is as follows: in, TD_amp is the data length of the amplitude sequence of the time-domain signal; TD_amp is the amplitude sequence of the time-domain signal. It is the data mean of the amplitude sequence of the time-domain signal; It is the data variance of the amplitude sequence of the time-domain signal. It is steepness; The operating condition identification module takes the vibration signal characteristic values ​​of each component of the escalator as input to the pre-trained operating condition identification model, and calibrates the operating conditions and potential faults of each component of the escalator based on the output results, so as to complete the diagnosis of the health status of the escalator. The pre-trained working condition recognition model includes a preset process before training, which includes the following steps: acquiring escalator vibration signal simulation data based on the mechanical simulation module, processing the escalator vibration signal simulation data through the feature value extraction module to generate simulation feature values ​​of vibration signals of each escalator component, using the simulation feature values ​​of vibration signals of escalator unit components as input data, and calibrating the output labels according to the corresponding simulation working conditions to generate training data for unit components, repeating the above steps to integrate and form a training dataset; inputting the training dataset into the Statistics and MachineLearning Toolbox module for data classification to obtain the preset working condition recognition model; After completing the above-mentioned preset process, the working condition identification model is trained by a machine learning algorithm to obtain the pre-trained working condition identification model. Specifically, the machine learning algorithm is one of linear regression algorithm, support vector machine algorithm, and K-nearest neighbor algorithm. The Support Vector Machine (SVM) algorithm represents instances as points in space, using a hyperplane to separate the data points. The distance of a point from the hyperplane reflects the degree of difference in classification. The hyperplane model is as follows: It can relatively represent the distance of point x from the hyperplane, while Symbols and class tags Whether the symbols are consistent can indicate whether the classification is correct, and can be measured. To indicate the correctness and confidence level of the classification; When this model is placed on a hyperplane, for By imposing certain constraints, the problem can be transformed into a geometric margin, which can then be used for multivariate problems. The mathematical model for the geometric margin is as follows: 。 2. The escalator health status diagnosis system according to claim 1, characterized in that, The escalator side exploration module consists of external sensors, an escalator mainboard, and a DTU device. The external sensors are connected to the DTU device via 485 communication, and the escalator mainboard is connected to the DTU device via 232 communication. The external sensors include: a three-axis vibration sensor for the main unit, a three-axis vibration sensor for the main drive chain, a three-axis vibration sensor for the step chain tensioner, a handrail belt temperature sensor, a temperature sensor for the main unit and control cabinet, a power meter sensor, and a main unit noise sensor.

3. The escalator health status diagnosis system according to claim 1, characterized in that, When acquiring raw data of escalator vibration signals each time, the data acquisition module also acquires the time series and amplitude series of the corresponding time-domain signals.

Images