Escalator PHM system based on Internet of Things and intelligent algorithm

Abstract

The invention discloses an escalator PHM system based on the Internet of Things and an intelligent algorithm. The escalator PHM system comprises a signal collection and transmission module, a signal preprocessing module, a signal analysis module and an interface display module. The signal acquisition and transmission module comprises various sensors, an acquisition box, a gateway and a cable; the signal preprocessing module comprises signal noise processing, signal feature enhancement, signal time-frequency conversion processing and signal feature extraction; the signal analysis module obtains diagnosis and prediction results of the preprocessed signals based on vibration signal analysis and a deep learning algorithm; and the interface display module displays the running state of each escalator according to the result given by the signal analysis module, and displays the fault position, time and maintenance suggestion information of the faulted escalator at the same time. According to the method, the functions of escalator state detection, fault diagnosis, fault prediction, service life prediction and the like are achieved, the complete PHM of the escalator is achieved, and a solid step is made for the escalator monitoring industry.

Description

Technical Field

[0001] The present invention relates to the field of escalator PHM (fault prediction and health management) systems, and in particular to an escalator PHM system based on the Internet of Things and intelligent algorithms. Background Art

[0002] Most existing escalator PHM systems are single fault diagnosis systems, which cannot achieve comprehensive escalator status detection, fault diagnosis and maintenance suggestions, fault prediction and life prediction functions. Most existing escalator PHM systems monitor a single escalator and cannot be applied to scenarios where multiple escalators are running, such as line level. Most existing escalator PHM system algorithms are based on single vibration analysis or deep learning, and the lack of cross-validation leads to insufficient accuracy and interpretability. In the most critical rolling bearing and gear diagnosis, traditional escalator PHM systems directly use one-dimensional vibration signals as input, which cannot deeply mine abnormal information in the signal, resulting in insufficient accuracy.

[0003] The information disclosed in this background technology section is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art already known to those skilled in the art. Summary of the invention

[0004] In view of the problems existing in the prior art, the object of the present invention is to provide an escalator PHM system based on the Internet of Things and intelligent algorithms.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An escalator PHM system based on the Internet of Things and intelligent algorithms includes a signal acquisition and transmission module, a signal preprocessing module, a signal analysis module and an interface display module; wherein,

[0007] The signal acquisition and transmission module includes a variety of sensors, acquisition boxes, gateways and cables; after the data is collected, it is uniformly aggregated by the Ethernet port of each escalator to the switch set up in the escalator system of the weak current integrated equipment room, and the data is unified through the switch networking and then connected to the communication backbone network through the Ethernet interface and uploaded to the escalator central virtual server provided by the cloud platform;

[0008] The signal preprocessing module includes signal noise processing, signal feature enhancement, signal time-frequency conversion processing and signal feature extraction;

[0009] The signal analysis module obtains diagnosis and prediction results for the pre-processed signal based on vibration signal analysis and deep learning algorithm;

[0010] The interface display module displays the operating status of each escalator according to the result given by the signal analysis module, and displays the fault location, time and maintenance suggestion information of the faulty escalator.

[0011] Furthermore, the signal noise processing uses a filter to perform filtering and noise reduction, as follows:

[0012] Formula (1.1) shows the relationship between the effective signal and noise in the original signal; where y (t) is the original signal collected; η (t) is additive zero-mean Gaussian noise; x (t) is the target signal after denoising; the filter H is used to transform the original signal y containing Gaussian white noise into (t) Extract the center frequency as ω c The narrowband component signal x is used to separate the fault information and white noise in the vibration signal; the filter H transfer function is given by formula (1.2), where α is the factor affecting the filter bandwidth; the center frequency is continuously adjusted for iteration, and the original signal is divided into two parts, one part contains more fault information, and the other part contains more white noise; the Pearson correlation coefficient is used to extract a part of the component signal containing more fault information. The relationship between the Pearson correlation coefficient R between the original signal and the effective signal is given by formula (1.3), where and Represent the expectations of the effective signal and the original signal, R∈(0,1); finally, the weighted sum of each extracted signal is performed to realize signal reconstruction, so as to achieve the purpose of denoising and enhancing the features of the original signal;

[0013] y (t) =x (t) +η (t) (1.1)

[0014]

[0015] Furthermore, the specific process of the signal time-frequency conversion processing and signal feature extraction is as follows:

[0016] The time domain signal is converted into frequency domain and time-frequency domain signals, and the characteristics of the signal in various fields are extracted. The correlation calculation formula is used to calculate the correlation of each feature in the corresponding algorithm, and the feature with the largest correlation is selected as the algorithm input.

[0017] Further, the process of the signal analysis module to obtain a diagnosis is as follows:

[0018] The escalator fault diagnosis is divided into three parts. The first part is direct threshold judgment, which takes the partial collected temperature signal and displacement signal as the diagnosis result to see whether they exceed the threshold specified by the technology.

[0019] The second part is vibration signal analysis. After processing the collected vibration signal, its envelope spectrum is analyzed. By detecting the fault characteristic frequency in the vibration envelope signal, the bearing fault can be diagnosed as a basis for diagnosis.

[0020] The third part is the deep learning algorithm analysis. The one-dimensional signal is converted into a two-dimensional grayscale image as input. First, the escalator database is called, and the grayscale recursive image after the one-dimensional data is converted is cut and thumbnailed, and then input into the convolutional neural network. After the convolution layer, pooling layer, and fully connected layer, the parameters are adjusted to obtain an algorithm model that meets the technical requirements. It is used for real escalator diagnosis as another result of the diagnosis. Finally, the comprehensive diagnosis result is obtained through weight analysis.

[0021] The specific method of converting a one-dimensional signal into a two-dimensional grayscale image is as follows:

[0022] For the vibration time series data y (t) , select the embedding parameter m and the lag time τ to reconstruct the sequence, and after reconstruction, we get formula (1.4). By calculating the distance between all phase points in the reconstructed phase space, we get a phase space distance matrix of size N×N. Compare the values ​​in the matrix with the given critical distance threshold ε to get a binary recursive matrix, as shown in (1.5), where ε represents the distance threshold, N represents the number of state vectors in the phase space, and “||·||” in the formula represents the distance measure between two vectors. We use the Euclidean distance to calculate, Θ represents the Heaviside step function, and each value in the matrix represents a pixel point, so as to perform grayscale conversion;

[0023] X (t) =(x i ,x i+τ , …x i+(m-1)τ ), (1.4)

[0024] R ij =Θ(ε-||X i -X j ||),i,j=1,2,...N (1.5).

[0025] Further, the process of the signal analysis module predicting the result is as follows:

[0026] Predict the data that is normal after fault diagnosis, including potential fault prediction and life prediction. For potential fault prediction, a recurrent neural network is used as the main algorithm to establish a part health regression model. The normal data of the parts in the escalator database are selected. After data processing and feature selection, the data is input into the recurrent neural network. Adam is used as the optimizer to adjust the parameters of the number of network layers, number of nodes, step size, batch size, learning rate, and number of training rounds. MAE, MSE, and MAPE are used as evaluation indicators. The model with the smallest evaluation indicator is selected and saved. The residual and residual standard deviation indicators are used for real escalator potential fault prediction.

[0027] The remaining life prediction uses a recurrent neural network to establish a full life model for parts. The full life cycle data of parts in the escalator database is selected. After data processing and feature selection, the corresponding life labels are added and input into the deep learning model. The remaining life prediction process of the bearing depends on the current and historical bearing vibration data. This process includes inputting the test data into the trained deep learning model to obtain a prediction of the remaining life of the bearing. Using Adam as the optimizer, adjust the parameters such as the number of network layers, the number of nodes, the batch size, the learning rate, and the number of training rounds. Using MAE, MSE, and MAPE as evaluation indicators, the accuracy of the remaining life prediction model is evaluated, and the model with the smallest evaluation index is selected and saved for actual escalator part life prediction.

[0028] Furthermore, the interface display module includes a homepage, station management, escalator management, fault warning analysis and a large screen; the homepage interface mainly displays the basic information of the entire line-level escalator, the number and trend of faults and warnings, and is accompanied by a line-level subway map, in which escalator stations with faults are displayed in red, and escalator stations with warnings are displayed in yellow; the station management interface displays the basic information of the station-level escalator, including the station name, the number of escalators in the station, and the number of overall faults and warnings in the station; at the same time, the operating time and operating speed of each escalator, the operating status of each escalator, the number and time of faults and warnings are displayed respectively; the escalator management interface focuses on a single escalator, displays various basic information of a certain escalator, and displays a three-dimensional stereogram of the escalator. The figure shows the various components of the escalator. If a fault or warning occurs, the components will be displayed in red or yellow respectively. You can click to view the specific information of the fault and the recommended maintenance method. The fault warning analysis interface shows that you can choose to view all the fault and warning information within the storage time of the line level, station level, and single escalator, including the number of faults and warnings, the location of the fault and warning, the fault and repair time, and a diagnostic analysis result diagram. The diagram shows the time domain diagram, frequency domain diagram, envelope diagram, machine learning result diagram, and intensity and kurtosis feature diagrams. The large-screen interface is based on the theme of technology and intelligence, aiming to improve the user's comfort when using the system. It contains basic information of escalators at all levels and click channels leading to the above sub-interfaces, giving users a simple, efficient and beautiful user experience.

[0029] By adopting the above technical solution, the present invention has the following beneficial effects:

[0030] 1. The administrator can intuitively see the operating status, fault time and maintenance suggestions of each escalator through the display interface, which helps the administrator to discover and quickly locate and solve the fault in advance, reducing business interruptions caused by the fault.

[0031] 2. The system realizes the functions of escalator status detection, fault diagnosis, fault prediction, life prediction, etc., realizes the complete PHM of escalators, and takes a solid step for the escalator monitoring industry.

[0032] 3. By combining vibration signals and deep learning algorithms, the diagnostic accuracy is greatly improved.

[0033] 4. The system realizes PHM of escalator groups at line level, station level, etc., and realizes unified monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 It is the overall block diagram of the PHM system of the present invention;

[0036] Figure 2 This is the general block diagram of the fault diagnosis of the present invention;

[0037] Figure 3 This is a fault diagnosis flow chart based on deep learning of the present invention;

[0038] Figure 4 This is a fault prediction flow chart of the present invention;

[0039] Figure 5 This is a life prediction flow chart of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0042] Combination Figure 1-5 As shown, the present invention requires the cooperation of multiple modules to realize the complete escalator PHM function, including a signal acquisition and transmission module, a signal processing module, a signal analysis module and an interface display module.

[0043] (1) The signal acquisition module includes a variety of sensors (vibration, temperature vibration, displacement, temperature, etc.), acquisition boxes, gateways and cables, etc.

[0044] (2) The signal processing module includes signal noise processing, signal feature enhancement, signal time-frequency conversion processing and signal feature extraction.

[0045] (3) The signal analysis module obtains diagnostic and prediction results for the preprocessed signals based on vibration signal analysis and deep learning algorithms.

[0046] (4) The interface display module displays the operating status of each escalator based on the results given by the signal analysis module, and also displays information such as the fault location, time, and maintenance suggestions of the faulty escalator.

[0047] Signal preprocessing: Signal noise processing uses filter filtering or EMD (empirical mode decomposition) and other methods to remove noise from the original signal, retaining the signal rich in abnormal information to achieve the purpose of feature enhancement; signal time-frequency domain conversion processing includes the conversion between signal time domain, frequency domain, and time-frequency domain, which facilitates the subsequent extraction of features in the corresponding domain. Signal feature extraction selects the features with the strongest correlation with diagnosis or prediction through correlation algorithms as the input of the subsequent PHM algorithm.

[0048] Fault diagnosis: In order to solve the defects of low fault diagnosis accuracy, high false alarm rate and insensitivity to early faults in traditional escalators, the present invention uses a fault diagnosis algorithm based on vibration signal analysis and deep learning. The data signals collected from the same part are diagnosed using two algorithms respectively, and then a comprehensive diagnosis result is obtained based on weight analysis. The accuracy of diagnosis is greatly enhanced.

[0049] (1) Vibration signal analysis: The preprocessed vibration signal is analyzed in the frequency domain, time domain, and time-frequency domain. The features of each domain are calculated and extracted, and the escalator parts are judged to be faulty based on the feature values.

[0050] (2) Deep learning analysis: An innovation of the present invention is that the input signal is not a traditional one-dimensional vibration signal, but is converted into a two-dimensional grayscale recursive graph, which is then input into the convolutional neural network after image cropping, thumbnail processing, etc. The advantage of this is that in the two-dimensional grayscale graph, the data anomalies will be presented in a more intuitive and easier to detect form such as texture, color, and shape, which greatly improves the accuracy of both early fault detection and overall diagnosis compared to traditional one-dimensional input. The diagnostic results are obtained by using the two-dimensional grayscale graph as the input of the neural network, which serves as another basis for escalator component fault diagnosis and prediction.

[0051] Failure prediction:

[0052] (1) Potential fault prediction: By analyzing historical operating status parameter data and combining specific theories and analysis methods, the possible faults of the equipment in the future are predicted, analyzed and judged. A large amount of normal data is used to establish a healthy component regression model for the escalator diagnosis part through a deep learning algorithm. After adjusting the model parameters, the parameters whose predicted values ​​are closest to the actual values ​​are found and the model is saved. Then, the fault warning threshold is set through the deviation analysis between the predicted value and the actual value to achieve fault warning for the escalator.

[0053] (2) Remaining life prediction: Remaining life prediction (RUL) can determine the remaining life of parts, and then arrange production tasks and formulate reasonable maintenance plans in a more targeted manner, maximize production efficiency, and avoid the occurrence of serious accidents. The processed full life cycle data of the escalator diagnosis part is used as the input of the deep learning life prediction model, and the data is labeled accordingly. The remaining life is output through the neural network. After artificial model parameter adjustment, the parameter with the predicted value closest to the actual value is found through the evaluation index, and the model is saved at this time. It is used to predict the life of each component during the actual operation of the escalator.

[0054] The interface displays:

[0055] The interface display module includes several sub-interface modules such as homepage, station management, escalator management, fault warning analysis and large screen. The homepage interface mainly displays the basic information of the entire line-level escalator, the number and trend of faults and warnings, and is attached with a line-level subway map. In the map, escalator stations with faults will be displayed in red, and escalator stations with warnings will be displayed in yellow; the station management interface displays the basic information of the station-level escalator, including the station name, the number of escalators in the station, the overall number of faults and warnings in the station; at the same time, it displays the operating time and operating speed of each escalator, the operating status of each escalator, the number and time of faults and warnings, etc.; the escalator management interface focuses on a single escalator, displays various basic information of a certain escalator, and displays a three-dimensional stereogram of the escalator, marking each component of the escalator in the map. If a fault or warning occurs, , the components are displayed in red and yellow respectively, and you can click to view the specific information of the fault and the recommended maintenance method; the fault warning analysis interface shows that you can choose to view all the fault and warning information within the storage time of the line level, station level, and single escalator, including the number of faults and warnings, fault and warning locations, fault and repair time, etc., and is accompanied by a diagnostic analysis result graph, which shows the time domain graph, frequency domain graph, envelope graph and machine learning result graph as well as characteristic graphs such as intensity and kurtosis; the large-screen interface is based on the theme of technology and intelligence, aiming to improve the user's comfort when using the system. It contains basic information of escalators at all levels and click channels leading to the above sub-interfaces, giving users a simple, efficient and beautiful user experience.

[0056] The workflow of the present invention is as follows:

[0057] Step 1, collection and transmission: After the data is collected by the sensor, it is uniformly collected by the Ethernet port of each escalator and aggregated to the switch set up in the escalator system in the weak current integrated equipment room. After the data is unified through the switch networking, it is connected to the communication backbone network through the Ethernet interface and uploaded to the cloud platform to provide the escalator central virtual server. The Ethernet interface uses TCP / IP or modbus TCP / IP protocol.

[0058] Step 2: Noise reduction processing: The directly collected signal contains a lot of noise, which makes the analysis accuracy of the original signal low. Therefore, it is necessary to perform noise reduction processing on the signal, using filters for filtering and noise reduction.

[0059] The filtering and noise reduction steps are as follows: Equation (1.1) shows the relationship between the effective signal and noise in the original signal. (t) is the original signal collected; η (t) is additive zero-mean Gaussian noise; x (t) is the target signal after denoising. A filter H is used to convert the original signal y containing Gaussian white noise into (t) Extract the center frequency as ω cThe narrowband component signal x can be used to separate the fault information from the white noise in the vibration signal. The transfer function of the filter H is given by formula (1.2), where α is the factor affecting the filter bandwidth. The center frequency is continuously adjusted for iteration to divide the original signal into two parts, one containing more fault information and the other containing more white noise. The Pearson correlation coefficient is used to extract a component signal containing more fault information. The relationship between the Pearson correlation coefficient R between the original signal and the effective signal is given by formula (1.3), where and Represent the expectation of the effective signal and the original signal, R∈(0,1). Finally, the weighted sum of each extracted signal is performed to achieve signal reconstruction, so as to achieve the purpose of denoising and enhancing the features of the original signal.

[0060] y (t) =x (t) +η (t) (1.1)

[0061]

[0062] Step 3: Signal conversion and feature extraction: Convert the time domain signal into frequency domain and time-frequency domain signals, and extract the features of the signal in various fields. Apply the correlation calculation formula to calculate the correlation of each feature in the corresponding algorithm, and select the feature with the largest correlation as the algorithm input.

[0063] The extracted features are shown in the table:

[0064]

[0065] Step 4, fault diagnosis: escalator fault diagnosis can be divided into three parts. The first part is direct threshold judgment, which mainly targets some collected temperature signals, displacement signals, etc., and uses whether they exceed the threshold specified by the technology as the diagnosis result; the second part is vibration signal analysis. After processing the collected vibration signal, its envelope spectrum is analyzed. By detecting the fault characteristic frequency in the vibration envelope signal, the bearing fault can be diagnosed as a basis for diagnosis; the third part is deep learning algorithm analysis. In terms of deep learning algorithm, the present invention abandons the traditional mode of using one-dimensional signal as input, but converts the one-dimensional signal into a two-dimensional grayscale image as input. First, the escalator database is called, and the grayscale recursive image converted from the one-dimensional data is processed by image cutting, thumbnail, etc., and then input into the convolutional neural network. After the convolution layer, pooling layer, and fully connected layer, the parameters are adjusted to obtain an algorithm model that meets the technical requirements for real elevator diagnosis, which is used as another result of the diagnosis. Finally, the comprehensive result of the diagnosis is obtained through weight analysis.

[0066] The specific method of converting a one-dimensional signal into a two-dimensional grayscale image is as follows:

[0067] For the vibration time series data y (t) , select the embedding parameter m and the lag time τ to reconstruct the sequence, and after reconstruction, we get formula (1.4). By calculating the distance between all phase points in the reconstructed phase space, we can get a phase space distance matrix of size N×N. Compare the values ​​in the matrix with the given critical distance threshold ε to get a binary recursive matrix, as shown in (1.5), where ε represents the distance threshold value, N represents the number of state vectors in the phase space, and “||·||” in the formula represents the distance measure between two vectors. Euclidean distance is used for calculation, Θ represents the Heaviside step function, and each value in the matrix represents a pixel point, so as to perform grayscale conversion.

[0068] X (t) =(x i ,x i+τ , …x i+(m-1)τ ) (1.4)

[0069] R ihj Θ(ε-||X i -X j ||),i,j=1,2,...N (1.5).

[0070] Step 5, fault prediction: predict the data with normal results after fault diagnosis, including potential fault prediction and life prediction. Among them, the potential fault prediction uses recurrent neural network as the main algorithm to establish a part health regression model, select the normal data of the parts in the escalator database, and input them into the recurrent neural network after data processing and feature selection. Adam is used as the optimizer to adjust the parameters such as the number of network layers, number of nodes, step size, batch size, learning rate, and number of training rounds; according to MAE (mean absolute error), MSE (mean square error) and MAPE (mean absolute percentage error) as evaluation indicators, select the model with the smallest evaluation index and save it, and use residuals, residual standard deviation and other indicators to predict potential faults of the escalator.

[0071] Similar to potential fault prediction, the remaining life prediction uses a recurrent neural network to establish a full life model for parts. The full life cycle data of parts in the escalator database is selected. After data processing and feature selection, the corresponding life labels are marked and input into the deep learning model. The remaining life prediction process of the bearing depends on the current and historical bearing vibration data. This process includes inputting the test data into the trained deep learning model to obtain a prediction of the remaining life of the bearing. Using Adam as the optimizer, adjust the parameters such as the number of network layers, the number of nodes, the batch size, the learning rate, and the number of training rounds; use MAE (mean absolute error), MSE (mean square error), and MAPE (mean absolute percentage error) as evaluation indicators to evaluate the accuracy of the remaining life prediction model, and select the model with the smallest evaluation index to save for the actual escalator part life prediction.

[0072] Troubleshooting list:

[0073]

[0074]

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An escalator PHM system based on the Internet of Things and intelligent algorithms, characterized in that: It includes a signal acquisition and transmission module, a signal preprocessing module, a signal analysis module and an interface display module; wherein, The signal acquisition and transmission module includes a variety of sensors, acquisition boxes, gateways and cables; after the data is collected, it is uniformly aggregated by the Ethernet port of each escalator to the switch set up in the escalator system of the weak current integrated equipment room, and the data is unified through the switch networking and then connected to the communication backbone network through the Ethernet interface and uploaded to the escalator central virtual server provided by the cloud platform; The signal preprocessing module includes signal noise processing, signal feature enhancement, signal time-frequency conversion processing and signal feature extraction; The signal analysis module obtains diagnosis and prediction results for the pre-processed signal based on vibration signal analysis and deep learning algorithm; The interface display module displays the operating status of each escalator according to the result given by the signal analysis module, and displays the fault location, time and maintenance suggestion information of the faulty escalator.

2. The escalator PHM system based on the Internet of Things and intelligent algorithms according to claim 1 is characterized in that: The signal noise processing uses a filter to perform filtering and noise reduction, as follows: Formula (1.1) shows the relationship between the effective signal and noise in the original signal; where y (t) is the original signal collected; η (t) is additive zero-mean Gaussian noise; x (t) is the target signal after denoising; the filter H is used to transform the original signal y containing Gaussian white noise into (t) Extract the center frequency as ω c The narrowband component signal x is used to separate the fault information and white noise in the vibration signal; the filter H transfer function is given by formula (1.2), where α is the factor affecting the filter bandwidth; the center frequency is continuously adjusted for iteration, and the original signal is divided into two parts, one part contains more fault information, and the other part contains more white noise; the Pearson correlation coefficient is used to extract a part of the component signal containing more fault information. The relationship between the Pearson correlation coefficient R between the original signal and the effective signal is given by formula (1.3), where and Represent the expectations of the effective signal and the original signal respectively, R∈(0,1); finally, the weighted sum of each extracted signal is performed to realize signal reconstruction, so as to achieve the purpose of denoising and enhancing the features of the original signal; y (t) =x (t) +n (t) (1.1) 3. The escalator PHM system based on the Internet of Things and intelligent algorithms according to claim 1 is characterized in that: The specific process of the signal time-frequency conversion processing and signal feature extraction is as follows: The time domain signal is converted into frequency domain and time-frequency domain signals, and the characteristics of the signal in various fields are extracted. The correlation calculation formula is used to calculate the correlation of each feature in the corresponding algorithm, and the feature with the largest correlation is selected as the algorithm input.

4. The escalator PHM system based on the Internet of Things and intelligent algorithms according to claim 1 is characterized in that: The process of the signal analysis module to obtain a diagnosis is as follows: The escalator fault diagnosis is divided into three parts. The first part is direct threshold judgment, which takes the partial collected temperature signal and displacement signal as the diagnosis result to see whether they exceed the threshold specified by the technology. The second part is vibration signal analysis. After processing the collected vibration signal, its envelope spectrum is analyzed. By detecting the fault characteristic frequency in the vibration envelope signal, the bearing fault can be diagnosed as a basis for diagnosis. The third part is the deep learning algorithm analysis. The one-dimensional signal is converted into a two-dimensional grayscale image as input. First, the escalator database is called, and the grayscale recursive image after the one-dimensional data is converted is cut and thumbnailed, and then input into the convolutional neural network. After the convolution layer, pooling layer, and fully connected layer, the parameters are adjusted to obtain an algorithm model that meets the technical requirements. It is used for real escalator diagnosis as another result of the diagnosis. Finally, the comprehensive diagnosis result is obtained through weight analysis. The specific method of converting a one-dimensional signal into a two-dimensional grayscale image is as follows: For the vibration time series data y (t) , select the embedding parameter m and the lag time τ to reconstruct the sequence, and after reconstruction, we get formula (1.4). By calculating the distance between all phase points in the reconstructed phase space, we get a phase space distance matrix of size N×N. Compare the values ​​in the matrix with the given critical distance threshold ε to get a binary recursive matrix, as shown in (1.5), where ε represents the distance threshold, N represents the number of state vectors in the phase space, and "||·||" in the formula represents the distance measure between two vectors. We use the Euclidean distance to calculate, Θ represents the Heaviside step function, and each value in the matrix represents a pixel point, so as to perform grayscale conversion; X (t) =(x i ,x i+τ ,...x i+(m-1)τ ), (1.4) R ij =Θ(ε-||X i -X j ||),i,j=1,2,...N (1.5)。 5. The escalator PHM system based on the Internet of Things and intelligent algorithms according to claim 1 is characterized in that: The process of predicting the results of the signal analysis module is as follows: Predict the data that is normal after fault diagnosis, including potential fault prediction and life prediction. For potential fault prediction, a recurrent neural network is used as the main algorithm to establish a part health regression model. The normal data of the parts in the escalator database are selected. After data processing and feature selection, the data is input into the recurrent neural network. Adam is used as the optimizer to adjust the parameters of the number of network layers, number of nodes, step size, batch size, learning rate, and number of training rounds. MAE, MSE, and MAPE are used as evaluation indicators. The model with the smallest evaluation indicator is selected and saved. The residual and residual standard deviation indicators are used for real escalator potential fault prediction. Remaining life prediction uses a recurrent neural network to build a full life model for parts. The full life cycle data of parts in the escalator database is selected. After data processing and feature selection, the corresponding life labels are marked and input into the deep learning model. The remaining life prediction process of the bearing depends on the current and historical bearing vibration data. This process includes inputting the test data into the trained deep learning model to obtain a prediction of the remaining life of the bearing. Using Adam as the optimizer, adjust the parameters such as the number of network layers, number of nodes, batch size, learning rate, and number of training rounds; using MAE, MSE, and MAPE as evaluation indicators to evaluate the accuracy of the remaining life prediction model, select the model with the smallest evaluation index and save it for actual ladder parts life prediction.

6. The escalator PHM system based on the Internet of Things and intelligent algorithms according to claim 1 is characterized in that: The interface display module includes a homepage, station management, escalator management, fault warning analysis and a large screen; the homepage interface mainly displays the basic information of the entire line-level escalator, the number and trend of faults and warnings, and is accompanied by a line-level subway map. In the map, escalator stations with faults will be displayed in red, and escalator stations with warnings will be displayed in yellow; the station management interface displays the basic information of station-level escalators, including the station name, the number of escalators in the station, and the number of overall faults and warnings in the station; at the same time, the operating time and operating speed of each escalator, the operating status of each escalator, and the number and time information of faults and warnings are displayed separately; the escalator management interface focuses on a single escalator, displays various basic information of a certain escalator, and displays a three-dimensional stereogram of the escalator, marking each component of the escalator in the map. If a fault or warning occurs, the components will be displayed in red and yellow respectively, and you can click to view the specific information of the fault and the recommended maintenance method; The fault warning analysis interface displays all fault and warning information within the storage time at the line level, station level, and individual escalator, including the number of faults and warnings, fault and warning locations, fault and repair time, and is accompanied by a diagnostic analysis result graph, which displays time domain graphs, frequency domain graphs, envelope graphs, machine learning result graphs, and intensity and kurtosis feature graphs. The large-screen interface is themed on technology and intelligence, aiming to improve the user's comfort when using the system. It contains basic information on escalators at all levels and click channels leading to the above sub-interfaces, giving users a simple, efficient and beautiful user experience.

Images