A step wheel state detection method and system

Abstract

The application discloses a kind of ladder wheel state detection method and system, the method includes: the vibration signal of ladder wheel is collected by high-frequency acceleration sensor, and double-scale feature is obtained by feature extraction;By pre-trained fault detection model, the initial weight corresponding to double-scale feature is dynamically adjusted to obtain target feature weight;Based on double-scale feature and target feature weight, prediction processing is carried out to obtain the corresponding prediction state type, and the prediction state type includes normal state and fault state;Confidence is calculated based on target feature weight to prediction state type, and confidence result is obtained, and the comprehensive state detection result of vibration signal is obtained based on confidence result and prediction state type.The application realizes real-time escalator ladder wheel state monitoring, can efficiently and accurately determine ladder wheel defect, improve ladder wheel maintenance efficiency, increase the safety and reliability of escalator.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal detection, in particular to a step wheel state detection method and system. BACKGROUND

[0002] The step wheel is a core component of the escalator system for power transmission and step support, and its running state directly affects the safety and stability of the entire escalator. In actual operation, this component is subjected to variable load impact and complex environmental effects for a long time, and is prone to typical faults such as tire rupture, bearing damage, and even tire falling off. If these faults are not identified in time, it will lead to an increase in the running resistance of the step wheel, causing abnormal phenomena such as abnormal noise, jamming or step inclination, and in severe cases, it may cause bearing jamming, step derailment, and even passenger falling, limb entrapment and other major safety accidents. At the same time, the continuous spread of faults will also accelerate the wear of related components such as step chain and guide rail, significantly increasing maintenance costs and equipment downtime.

[0003] At present, the detection of step wheels mainly relies on two methods. One is the traditional manual regular inspection, which is performed by maintenance personnel through subjective experience such as listening to sound and touching temperature, but due to the large number of step wheels and the hidden installation location, this method has the limitations of low efficiency, high missed detection rate and high risk of misjudgment. The second is a detection system based on multiple sensors, which requires the deployment of multiple sensors such as vibration, temperature and sound, which not only has complex site reconstruction and high cost, but also is easily disturbed by the environment in the multi-source signal fusion process, resulting in difficulty in meeting the actual engineering requirements in fault recognition accuracy.

[0004] However, the existing vibration detection technology relies on single domain features and is difficult to adapt to the non-stationary running conditions of step wheel bearings, resulting in insufficient generalization and anti-interference ability, high misjudgment rate, and inability to meet the high-precision safety monitoring requirements. Therefore, how to improve the reliability and accuracy of step wheel state monitoring is a problem to be solved. SUMMARY

[0005] In order to solve the above problems, the embodiments of the present application provide a step wheel state detection method and system, an electronic device, a computer readable storage medium and a computer program product.

[0006] In a first aspect, in order to solve the above technical problems, the present application provides a step wheel state detection method, comprising:

[0007] acquiring a vibration signal of an escalator step wheel through a high-frequency acceleration sensor;

[0008] extracting a double-scale feature from the vibration signal, wherein the double-scale feature includes micro transient features and macro trend features;

[0009] By using a pre-trained fault detection model, the corresponding initial weights are dynamically adjusted based on the dual-scale features to obtain the target feature weights corresponding to the vibration signal.

[0010] Based on the dual-scale features and the target feature weights, a prediction processing is performed to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state.

[0011] The confidence level of the predicted state type is calculated based on the target feature weights to obtain the confidence level result, and the comprehensive state detection result of the vibration signal is obtained based on the confidence level result and the predicted state type.

[0012] The beneficial effects are:

[0013] In the technical solution provided in the embodiments of this application, vibration signals of escalator step wheels are collected by a high-frequency accelerometer; dual-scale features are extracted from the vibration signals to obtain dual-scale features, which include microscopic instantaneous features and macroscopic trend features; a pre-trained fault detection model is used to dynamically adjust the corresponding initial weights based on the dual-scale features to obtain the target feature weights corresponding to the vibration signals; prediction processing is performed based on the dual-scale features and target feature weights to obtain the predicted state type corresponding to the vibration signals, which includes normal state and fault state; confidence is calculated based on the target feature weights to obtain the confidence result, and a comprehensive state detection result of the vibration signals is obtained based on the confidence result and the predicted state type. Thus, this application extracts the dual-scale features of the vibration signals as input to the fault detection model for multi-level judgment of the step wheel state, avoiding the situation where insufficient generalization and anti-interference ability, and excessively high false positive rate are caused by relying on single-domain features; and further, the pre-trained fault detection model uses dynamic weight diagnosis to specifically predict and judge the predicted state type corresponding to the vibration signals, further improving detection accuracy. Therefore, the provided step wheel condition detection method enables real-time monitoring of escalator step wheel conditions, which can efficiently and accurately identify step wheel defects, improve step wheel maintenance efficiency, reduce maintenance costs, and increase the safety and reliability of escalators.

[0014] Furthermore, the method also includes:

[0015] The current operating condition of the escalator is determined based on the vibration signal using a fundamental frequency identification algorithm.

[0016] Interference signals in the frequency band corresponding to the operating condition in the vibration signal are removed to obtain the processed signal, which is then used for dual-scale feature extraction.

[0017] Furthermore, the dual-scale feature extraction of the vibration signal to obtain dual-scale features includes:

[0018] The vibration signal is decomposed using wavelet packet transform to obtain multiple frequency bands, and the instantaneous impact energy entropy of the multiple frequency bands is calculated as a microscopic instantaneous feature.

[0019] The vibration signal is divided into windows based on a preset window length, and the RMS value in each window is linearly fitted to obtain the linear fitting slope, which is used as a macroscopic trend feature.

[0020] Obtain the correlation between the micro-instantaneous features and the macro-trend features, and construct a feature correlation matrix based on the correlation;

[0021] Dual-scale features are formed based on the micro-instantaneous features and macro-trend features in the feature correlation matrix.

[0022] Furthermore, the step of obtaining the correlation between the microscopic instantaneous features and the macroscopic trend features, and constructing a feature correlation matrix based on the correlation, includes:

[0023] Based on the relationship between the energy entropy range corresponding to the preset frequency band and the instantaneous impact energy entropy, the energy proportion of the preset frequency band is determined;

[0024] Obtain a first relationship between the energy percentage and the corresponding first baseline, and a second relationship between the linear fitting slope and the corresponding second baseline;

[0025] By using preset association logic, the association relationship between the micro-instantaneous features and the macro-trend features is determined based on the first relationship and the second relationship;

[0026] Based on the aforementioned correlation, relevant target micro-instantaneous features and target macro-trend features are selected from the micro-instantaneous features and macro-trend features of the vibration signal;

[0027] A feature correlation matrix is ​​constructed based on the target's microscopic instantaneous features and the target's macroscopic trend features.

[0028] Furthermore, the method also includes:

[0029] The dual-scale features are normalized to obtain the processed features;

[0030] Outliers in the processed features are identified and removed based on the 3σ criterion to obtain target dual-scale features, which are then input into the pre-trained fault detection model to obtain comprehensive state detection results.

[0031] Furthermore, the step of dynamically adjusting the corresponding initial weights based on the dual-scale features using a pre-trained fault diagnosis model to obtain the target feature weights corresponding to the vibration signal includes:

[0032] Using a pre-trained fault diagnosis model, the distance information between the dual-scale features and the cluster centers of the pre-trained fault diagnosis model is obtained; the cluster centers correspond one-to-one with the normal state and the fault states, including tire rupture, bearing damage, and tire detachment.

[0033] Based on the distance information, the state type corresponding to the cluster center closest to the dual-scale feature is determined as the initial type corresponding to the vibration signal;

[0034] Obtain the initial weight corresponding to the initial type;

[0035] Obtain the feature deviation of each feature value in the dual-scale feature;

[0036] By dynamically adjusting the rules, the initial weights are dynamically adjusted based on the feature deviation to obtain the target feature weights corresponding to the vibration signal.

[0037] Furthermore, the prediction processing based on the dual-scale features and the target feature weights to obtain the predicted state type corresponding to the vibration signal includes:

[0038] The prediction results of each decision tree are obtained based on the dual-scale features through the random forest model in the pre-trained fault diagnosis model.

[0039] Based on the weight value corresponding to each decision tree in the target feature weights, the prediction results are weighted and fused to obtain the predicted state type corresponding to the vibration signal.

[0040] Furthermore, the step of calculating the confidence score of the predicted state type based on the target feature weights to obtain the confidence score result includes:

[0041] Obtain the prediction probability of each decision tree in the random forest model of the pre-trained fault diagnosis model and the predicted state type;

[0042] The confidence level is calculated based on the weight value of each decision tree in the target feature weight and the prediction probability to obtain the confidence level result of the predicted state type.

[0043] Furthermore, the comprehensive state detection result of the vibration signal obtained based on the confidence level result and the predicted state type includes:

[0044] Obtain the Euclidean distance between the dual-scale feature and the cluster center of the predicted state type;

[0045] When the Euclidean distance is less than the distance threshold and the confidence level of the confidence result is greater than or equal to the confidence threshold, a comprehensive state detection result is obtained, which characterizes the predicted state type as a valid detection result; otherwise, a suspected detection result is obtained, which characterizes the predicted state type as a false detection result.

[0046] Secondly, the present invention provides a system including a data acquisition unit for acquiring vibration signals of escalator step wheels via a high-frequency acceleration sensor;

[0047] The feature extraction unit is used to perform dual-scale feature extraction on the vibration signal to obtain dual-scale features, which include micro-instantaneous features and macro-trend features.

[0048] The dynamic weighting unit is used to dynamically adjust the corresponding initial weights based on the dual-scale features using a pre-trained fault detection model to obtain the target feature weights corresponding to the vibration signal.

[0049] A state determination unit is used to perform prediction processing based on the dual-scale features and the target feature weights to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state.

[0050] The result unit is used to calculate the confidence level of the predicted state type based on the target feature weights, obtain the confidence level result, and obtain the comprehensive state detection result of the vibration signal based on the confidence level result and the predicted state type.

[0051] Thirdly, this application also provides an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the ladder wheel state detection method as described above.

[0052] Fourthly, this application also provides a computer-readable storage medium storing computer-readable instructions thereon, which, when executed by a computer's processor, cause the computer to perform the ladder wheel state detection method as described above.

[0053] Fifthly, this application also provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the ladder wheel state detection method provided in the various alternative embodiments described above.

[0054] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0055] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0056] Figure 1 This is a schematic diagram of one implementation environment involved in this application;

[0057] Figure 2 This is a flowchart illustrating a ladder wheel state detection method in an exemplary embodiment of this application;

[0058] Figure 3 This is a schematic diagram of the installation of a high-frequency acceleration sensor in an escalator in an exemplary embodiment of this application;

[0059] Figure 4 This is a schematic diagram illustrating dual-scale feature extraction in an exemplary embodiment of this application;

[0060] Figure 5 This is a schematic diagram of time-domain signals of tire rupture, bearing damage, and tire detachment in a normal state and fault state, as described in an exemplary embodiment of this application.

[0061] Figure 6 This is a schematic diagram illustrating the output of predicted state type and confidence results by a pre-trained fault diagnosis model in an exemplary embodiment of this application;

[0062] Figure 7 This is an exemplary embodiment of the present application, showing the distribution of the dual-scale feature K-means clustering results for four states: normal state, tire rupture, bearing damage, and tire detachment.

[0063] Figure 8 In an exemplary embodiment of this application, the ladder wheel state classification confusion matrix has an overall accuracy of 95.8% on the test set;

[0064] Figure 9 This is a block diagram illustrating a ladder wheel state detection system, as shown in an exemplary embodiment of this application.

[0065] Figure 10 This is a schematic diagram of the structure of a computer system suitable for implementing the electronic devices of the present application embodiments. Detailed Implementation

[0066] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0067] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0068] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0069] In this application, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0070] To address the problem that existing vibration detection technologies for ladder wheel bearings are ill-suited to the non-stable operating conditions of ladder wheel bearings, resulting in insufficient generalization and anti-interference capabilities, high misjudgment rates, and inability to meet the requirements of high-precision safety monitoring, embodiments of this application propose a ladder wheel state detection method and system, mainly involving ladder wheel state detection technology included in signal detection and processing technology. These embodiments will be described in detail below.

[0071] Please refer to the following first. Figure 1 , Figure 1 This is a schematic diagram of an implementation environment related to this application. The implementation environment includes a high-frequency accelerometer 10 and a server 20, which communicate with each other via a wired or wireless network.

[0072] Server 20 is used to collect vibration signals from escalator step wheels via high-frequency accelerometer 10; it performs dual-scale feature extraction on the vibration signals to obtain dual-scale features, including microscopic instantaneous features and macroscopic trend features; using a pre-trained fault detection model, it dynamically adjusts the corresponding initial weights based on the dual-scale features to obtain the target feature weights corresponding to the vibration signals; based on the dual-scale features and target feature weights, it performs prediction processing to obtain the predicted state type corresponding to the vibration signals, including normal state and fault state; based on the target feature weights, it calculates the confidence level of the predicted state type to obtain the confidence level result, and based on the confidence level result and the predicted state type, it obtains the comprehensive state detection result of the vibration signals. Compared with existing step wheel state detection schemes, the step wheel state detection method provided in this implementation environment can quickly and accurately monitor the state of escalator step wheels, thereby achieving efficient fault detection, improving step wheel maintenance efficiency, and reducing maintenance costs.

[0073] It should be noted that, Figure 1 The server 20 in the implementation environment shown can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. No restrictions are imposed here.

[0074] Figure 2 This is a flowchart illustrating a ladder wheel state detection method according to an exemplary embodiment of this application. This method can be applied to... Figure 1 The implementation environment shown, and by Figure 1 The method is specifically executed by server 20 in the illustrated embodiment environment. However, in other implementation environments, this method can be executed by devices in those environments, and this embodiment does not impose any limitations on this.

[0075] like Figure 2 As shown, in an exemplary embodiment, the ladder wheel state detection method may include steps S201 to S205, which are described in detail below:

[0076] Step S201: Vibration signals of the escalator step wheels are collected using a high-frequency acceleration sensor.

[0077] Step S202: Extract dual-scale features from the vibration signal to obtain dual-scale features, which include micro-instantaneous features and macro-trend features.

[0078] Step S203: Using a pre-trained fault detection model, the initial weights are dynamically adjusted based on dual-scale features to obtain the target feature weights corresponding to the vibration signal.

[0079] Step S204: Based on dual-scale features and target feature weights, prediction processing is performed to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state.

[0080] Step S205: Calculate the confidence level of the predicted state type based on the target feature weights to obtain the confidence level result, and obtain the comprehensive state detection result of the vibration signal based on the confidence level result and the predicted state type.

[0081] In this embodiment, vibration signals of the escalator step wheel bearings are collected using a high-frequency accelerometer, and accurate fault identification is achieved through dual-scale feature extraction and dynamic weighted diagnosis. In application, vibration signals of the escalator step wheels are collected using a high-frequency accelerometer according to a pre-configured signal sampling frequency and rules. For example, continuous sampling is performed at a signal sampling frequency of 10kHz, generating one data segment every 5 seconds to form the vibration signal required for feature extraction.

[0082] The vibration signal is then subjected to dual-scale feature extraction to obtain dual-scale features. These features are input into a pre-trained fault detection model. The model dynamically adjusts the initial weights based on the dual-scale features to obtain the target feature weights corresponding to the vibration signal. These target feature weights are then used for prediction processing to obtain the predicted state type of the vibration signal. The predicted state type includes normal state and fault state, with fault states including tire rupture, bearing damage, and tire detachment.

[0083] Meanwhile, the pre-trained fault detection model will also output the confidence result corresponding to the predicted state type based on the target feature weight, so that the server can obtain the comprehensive state detection result of the vibration signal based on the confidence result and the predicted state type, and use it to judge whether the detection result is effective.

[0084] As can be seen from the above, the method provided in this embodiment extracts dual-scale features of the vibration signal as input to the fault detection model for multi-level judgment of the step wheel state. This avoids the situation where relying on single-domain features leads to insufficient generalization and anti-interference ability, resulting in a high false positive rate. Furthermore, in the pre-trained fault detection model, dynamic weight diagnosis is used to specifically predict and judge the predicted state type corresponding to the vibration signal, further improving the detection accuracy. Thus, the provided step wheel state detection method realizes real-time escalator step wheel state monitoring, which can efficiently and accurately determine step wheel defects, improve step wheel maintenance efficiency, reduce maintenance costs, and increase the safety and reliability of escalators.

[0085] In an exemplary embodiment provided in this application, the high-frequency accelerometer can be a high-frequency piezoelectric accelerometer of model PCB352C33. Its measurement range is approximately ±50g (g is the acceleration due to gravity, 1g = 9.8m / s²), which can cover the vibration intensity of a ladder wheel fault; the sampling frequency is 10kHz, which can completely acquire the fault characteristic frequency band; the output signal is a 4-20mA current signal, with strong anti-electromagnetic interference capability, which can improve the accuracy of the acquired vibration signal.

[0086] In addition, the installation location of the high-frequency accelerometer also affects the accuracy and effectiveness of the vibration signal. Please refer to [link / reference]. Figure 3 , Figure 3 This is a schematic diagram illustrating the installation of a high-frequency acceleration sensor within an escalator in an exemplary embodiment of this application. Figure 3 As shown, the high-frequency acceleration sensor is magnetically fixed to the bottom of the horizontal section of the escalator track. One high-frequency acceleration sensor is installed on each side of the track, for a total of two sensors that can cover both sides of the escalator wheels. The signal cable is a shielded twisted pair cable with a length of ≤10m. One end is connected to the high-frequency acceleration sensor, and the other end is connected to the server's data processing terminal. The cable is laid along the track support and fixed to avoid running parallel to the escalator's power cable.

[0087] In an exemplary embodiment of this application, after the vibration signal is acquired, a working condition adaptive anti-interference preprocessing step is further provided, which may specifically include:

[0088] The current operating condition of the escalator is determined based on the vibration signal using a fundamental frequency identification algorithm.

[0089] Interference signals in the frequency band corresponding to the operating conditions in the vibration signal are removed to obtain the processed signal, which is then used for dual-scale feature extraction.

[0090] In this embodiment, the corresponding interference signals are identified in a targeted and accurate manner through the working condition adaptive anti-interference preprocessing step, and a pure vibration signal is obtained, which further improves the accuracy of the dual-scale features extracted based on the vibration signal.

[0091] In an exemplary embodiment of this application, the specific steps for extracting dual-scale features from a vibration signal to obtain dual-scale features may include:

[0092] The vibration signal was decomposed using wavelet packet transform to obtain multiple frequency bands, and the instantaneous impact energy entropy of multiple frequency bands was calculated as microscopic instantaneous features.

[0093] The vibration signal is divided into windows based on a preset window length, and the RMS value in each window is linearly fitted to obtain the linear fitting slope, which is used as a macro trend feature.

[0094] Obtain the correlation between micro-level instantaneous features and macro-level trend features, and construct a feature correlation matrix based on the correlation;

[0095] Dual-scale features are formed based on the micro-instantaneous features and macro-trend features in the feature correlation matrix.

[0096] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating dual-scale feature extraction in an exemplary embodiment of this application. For example... Figure 4 As shown, the dual-scale feature extraction in this embodiment includes micro-instantaneous feature extraction by identifying the instantaneous manifestations of the fault, macro-trend feature extraction by identifying the development patterns of the fault, and feature correlation matrix construction. Specifically, micro-instantaneous feature extraction: wavelet packet transform is used to decompose the vibration signal and calculate the instantaneous impact energy entropy of each frequency band; macro-trend feature extraction: a sliding window trend coefficient is introduced. After the window is divided, the slope of the linear fitting of the root mean square of the vibration signal in each window (sliding window trend coefficient) is calculated based on the RMS value calculated based on the vibration signal, thereby realizing the extraction of the macro-trend coefficient and reflecting the long-term development trend of the bearing fault; fault feature correlation matrix construction: a mapping relationship between micro and macro features is established to eliminate misjudgment based on a single feature.

[0097] In an exemplary embodiment, when extracting microscopic instantaneous features, the db4 wavelet can be selected, with a wavelet decomposition level of 14. The calculation steps are as follows:

[0098] Step 1: Perform 14-layer wavelet packet decomposition on the vibration signal to obtain 16384 frequency bands.

[0099] Step 2: Calculate the energy E for each frequency band. i ;

[0100]

[0101] Where, N i X is the number of wavelet coefficients in the i-th frequency band. i，kRepresents the k-th data point (wavelet coefficient) within the i-th frequency band;

[0102] And normalize to obtain the normalized energy E i,nom ;

[0103]

[0104] Where M represents the total number of frequency bands, which is 16384 in this embodiment, and E j This represents the total energy of the j-th frequency band.

[0105] Step 3: Calculate the energy entropy to obtain the microscopic instantaneous impact energy entropy. As a microscopic instantaneous feature.

[0106] In one exemplary embodiment, when extracting macroscopic trend features, trend features are specifically extracted from time-domain signals, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of time-domain signals for normal and fault states, including tire rupture, bearing damage, and tire detachment, in an exemplary embodiment of this application. Figure 5 Figure (a) shows the time-domain signal under normal conditions, Figure (b) shows the time-domain signal under fault conditions including tire rupture, Figure (c) shows the time-domain signal under fault conditions including bearing damage, and Figure (d) shows the time-domain signal under fault conditions including tire detachment.

[0107] The calculation steps for macroeconomic trend characteristics can be as follows:

[0108] Step 1: Calculate the RMS (Root Mean Square) of the signal within each 10-second window. The formula for RMS is as follows:

[0109]

[0110] Among them, X i Let N be the amplitude of the i-th data point in the time-domain signal, and N represent the total number of sampling points within the 10-second time window.

[0111] Step 2: Perform linear fitting on the RMS values ​​of 5 consecutive windows totaling 25 seconds. The slope is the macroscopic RMS linear fitting slope. If the RMS value fluctuation within the window exceeds the normal baseline by 3... The window is marked as potentially abnormal.

[0112] In another exemplary embodiment, the specific steps for obtaining the correlation between microscopic instantaneous features and macroscopic trend features, and constructing a feature correlation matrix based on the correlation, may include:

[0113] Based on the relationship between the energy entropy range corresponding to the preset frequency band and the instantaneous impact energy entropy, the energy proportion of the preset frequency band is determined.

[0114] Obtain the first relationship between the energy percentage and the corresponding first baseline, and the second relationship between the linear fitting slope and the corresponding second baseline;

[0115] By pre-setting association logic, the association relationship between micro-instantaneous features and macro-trend features is determined based on the first and second relationships;

[0116] Based on the correlation, relevant target micro-instantaneous features and target macro-trend features are selected from the micro-instantaneous features and macro-trend features of the vibration signal;

[0117] A feature correlation matrix is ​​constructed based on the target's micro-instantaneous features and macro-trend features.

[0118] In this embodiment, the preset frequency band and the corresponding energy entropy range can be low frequency (0-500Hz), mid frequency (500-1500Hz), and high frequency (2000-3000Hz). A first relationship between the energy percentage and the corresponding first baseline, and a second relationship between the linear fitting slope and the corresponding second baseline are obtained. Preset association logic is used to mark the micro-instantaneous features and macro-trend features corresponding to a high-frequency energy percentage exceeding the normal baseline by 50% and an RMS slope exceeding the normal baseline by 50% as associated and suspected of tire detachment; the micro-instantaneous features and macro-trend features corresponding to a mid-frequency energy percentage exceeding the normal baseline by 30% and an RMS slope exceeding the normal baseline by 30% are marked as associated and suspected of bearing damage. This excludes anomalies in single features (unassociated features), such as high high-frequency energy with a normal slope, avoiding misjudgments.

[0119] In an exemplary embodiment of this application, after extracting the dual-scale features of the vibration signal, a data preprocessing step is further provided, which may specifically include:

[0120] The dual-scale features are normalized to obtain the processed features;

[0121] Outliers in the processed features are identified and removed based on the 3σ criterion to obtain the target dual-scale features, which are then input into a pre-trained fault detection model to obtain the comprehensive state detection results.

[0122] In this embodiment, Z-score normalization can be used to obtain the processed feature X. norm To eliminate dimensional differences, the formula is expressed as:

[0123]

[0124] Among them, X rawThis is a two-scale feature, where μ is the arithmetic mean and σ is the standard deviation.

[0125] Then, outliers that significantly deviate from the normal range are identified and removed using the 3σ criterion, such as removing data with "absolute value of normalized eigenvalue > 3", to ensure the reliability of the target dual-scale features.

[0126] In an exemplary embodiment of this application, the specific steps for obtaining the target feature weights corresponding to the vibration signal by dynamically adjusting the corresponding initial weights based on dual-scale features using a pre-trained fault diagnosis model may include:

[0127] By using a pre-trained fault diagnosis model, the distance information between the dual-scale features and the cluster centers of the pre-trained fault diagnosis model is obtained; the cluster centers correspond one-to-one with the normal state and the fault state, including tire rupture, bearing damage and tire detachment.

[0128] Based on distance information, the state type corresponding to the cluster center closest to the dual-scale feature is determined as the initial type of the vibration signal;

[0129] Obtain the initial weight corresponding to the initial type;

[0130] Obtain the feature deviation of each feature value in the dual-scale feature;

[0131] By dynamically adjusting the rules, the initial weights are dynamically adjusted based on the feature deviation to obtain the target feature weights corresponding to the vibration signal.

[0132] In this embodiment, the distance information between the dual-scale features and the cluster centers of the pre-trained fault diagnosis model is obtained through the pre-trained fault diagnosis model. Based on the distance information, the state type corresponding to the cluster center closest to the dual-scale features is determined as the initial type of the vibration signal. For example, if the distance information indicates that the state type corresponding to the closest cluster center is bearing damage, then the initial type is the initial type.

[0133] After determining the initial type, obtain the initial weights corresponding to the initial type, and at the same time calculate the feature deviation of each feature value in the dual-scale feature. The formula for calculating the feature deviation D can be expressed as:

[0134]

[0135] Among them, F current μ is the eigenvalue in the dual-scale feature set. baseline σ is the mean of all eigenvalues ​​in the dual-scale feature. baseline This represents the standard deviation corresponding to the characteristic deviation.

[0136] Then, by dynamically adjusting the rules, the initial weights are dynamically adjusted based on the feature deviation to obtain the target feature weights corresponding to the vibration signal:

[0137] If the deviation of a certain feature is greater than 50%, the weight is increased by 20% (from 0.5 to 0.6).

[0138] If the deviation of a certain feature is less than 20%, the weight is reduced by 10% (from 0.2 to 0.18).

[0139] The sum of the adjusted weights remains at 1.

[0140] In an exemplary embodiment of this application, the specific steps for performing prediction processing based on dual-scale features and target feature weights to obtain the predicted state type corresponding to the vibration signal may include:

[0141] The prediction results of each decision tree are obtained by using the random forest model in the pre-trained fault diagnosis model based on dual-scale features;

[0142] Based on the weight value of each decision tree in the target feature weight, the prediction results are weighted and fused to obtain the predicted state type of the vibration signal.

[0143] In this embodiment, the target feature weights include the weight values ​​corresponding to each decision tree. For example, the random forest model in the pre-trained fault diagnosis model has 5 decision trees, and the weights of the 5 decision trees are [0.22, 0.18, 0.20, 0.25, 0.15] (the sum of the weights is 1). After determining the target feature weights, the dual-scale features are input into all decision trees in the random forest model to obtain the prediction results output by each decision tree. Then, based on the weight values ​​corresponding to each decision tree in the target feature weights, the prediction results are weighted and fused to obtain the predicted state type corresponding to the vibration signal.

[0144] In an exemplary embodiment of this application, the specific steps for calculating the confidence level of the predicted state type based on the target feature weights to obtain the confidence result may include:

[0145] Obtain the prediction probability of each decision tree and the predicted state type in the pre-trained fault diagnosis model's random forest model;

[0146] The confidence score is calculated based on the weight value and prediction probability of each decision tree in the target feature weights to obtain the confidence score result of the predicted state type.

[0147] In this embodiment, the confidence score can be calculated using the formula: Confidence Score = Σ (Prediction probability of each decision tree × Weight value), where the weight value is the weight value corresponding to each decision tree in the target feature weights. For example, the weights of the 5 decision trees in the random forest model are [0.22, 0.18, 0.20, 0.25, 0.15] (the sum of the weights is 1), and the corresponding calculation process is: Confidence Score = (0.8*0.22) + (0.1*0.18) + (0.5*0.20) + (1.0*0.25) + (0.3*0.15) = 0.176 + 0.018 + 0.100 + 0.250 + 0.045 = 0.589.

[0148] Furthermore, the confidence level results can be classified according to preset grading rules. For example, a confidence level ≥ 0.9 is considered high confidence, 0.7 ≤ confidence level < 0.9 is considered medium confidence, and a confidence level < 0.7 is considered low confidence.

[0149] In another exemplary embodiment, the process by which a pre-trained fault diagnosis model obtains the predicted state type corresponding to a vibration signal based on the input dual-scale features includes determining the target feature weights, outputting the predicted state type, and calculating the confidence score. Please refer to [link to relevant documentation]. Figure 6 , Figure 6 This is a schematic diagram illustrating the output of predicted state type and confidence results by a pre-trained fault diagnosis model in an exemplary embodiment of this application.

[0150] like Figure 6 As shown, after obtaining the dual-scale features and determining the state type corresponding to the cluster center closest to the dual-scale features as the initial type of the vibration signal, the initial weights corresponding to the initial types are obtained. The initial weights are dynamically adjusted based on the feature deviation to obtain the target feature weights.

[0151] The process of determining the target feature weights is as follows:

[0152] D = |x - μ| / σ;

[0153] D_avg=(D1+D2+ … +Dn) / n;

[0154] W'=W o (1+k*D_avg);

[0155] Where D is the single feature deviation, D_avg is the association deviation, W' is the adjusted weight (target feature weight), and W o The initial weights are denoted by k, which is an adjustment coefficient determined by the correlation deviation. Specifically, this may include the following steps:

[0156] Step 1: Calculate the deviation of a single feature.

[0157] Calculate the single-feature deviation of the micro-instantaneous features and macro-trend features in the dual-scale features respectively:

[0158]

[0159] in, For the current eigenvalue, The characteristic mean, The characteristic standard deviation is denoted as .

[0160] Step 2: Calculate the correlation deviation:

[0161]

[0162] Step 3: Determine the adjustment coefficient k.

[0163] like If the weight needs to be increased by 20%, then... Solving for .

[0164] like If the weight needs to be reduced by 10%, then... Solving for .

[0165] like If k=0 (weights remain unchanged).

[0166] Step 4: Calculate and normalize the target weights.

[0167] Let the initial feature weights be... , (and Adjusted weights: , ,because The weights need to be normalized: ,at this time That is, the final target feature weight.

[0168] After determining the target feature weights, the prediction results of each decision tree are obtained based on dual-scale features using the random forest model in the pre-trained fault diagnosis model. The prediction results are then weighted and fused based on the weight values ​​of each decision tree in the target feature weights to obtain the predicted state type of the vibration signal. Next, the prediction probabilities corresponding to each decision tree and the predicted state type in the pre-trained random forest model are obtained. Confidence is calculated based on the weight values ​​and prediction probabilities of each decision tree in the target feature weights to obtain the confidence result of the predicted state type, which is used to calculate the overall state detection results.

[0169] In an exemplary embodiment provided in this application, the training process of the pre-trained fault diagnosis model is as follows:

[0170] We acquire dual-scale feature data corresponding to vibration signals in four states: normal state, tire rupture, bearing damage, and tire detachment, and set the number of K-means clusters to 4.

[0171] During clustering, the distance between each feature vector and the four cluster centers is calculated using Euclidean distance, and the feature vector is assigned to the cluster with the closest distance. The algorithm iterates a total of 100 times, and the execution steps are as follows:

[0172] Step 1: Prepare training data. There are 50 sets of data for each category: normal, tire rupture, bearing damage, and tire detachment, for a total of 200 sets of feature data.

[0173] Step 2: Select the feature values ​​of 4 typical samples as the initial centers;

[0174] Step 3: Calculate the Euclidean distance between each feature vector and the four centers, and assign it to the nearest cluster; update the cluster centers, and repeat until the center change is <0.01 or 100 iterations;

[0175] Step 4: Clustering result verification. Cluster purity must be ≥95%, such as... Figure 7 As shown, Figure 7 This is an exemplary embodiment of the present application, showing the distribution of K-means clustering results for four states: normal state, tire rupture, bearing damage, and tire detachment, based on dual-scale features.

[0176] 200 clustered labeled data sets (50 sets per class) were divided into 140 training sets and 60 test sets in a 7:3 ratio. There were 100 decision trees with a maximum depth of 10. The initial weights of feature importance were: normal state [0.2, 0.2, 0.2, 0.2, 0.2], tire rupture [0.4, 0.1, 0.1, 0.3, 0.1], bearing damage [0.1, 0.4, 0.1, 0.3, 0.1], and tire detachment [0.2, 0.2, 0.5, 0.1, 0.0].

[0177] Train a random forest model using the training set, calculating the feature importance of each decision tree; then adjust the initial weights; finally, validate the model using the test set, and deploy it only if the test set accuracy is ≥95%. Figure 8 As shown, Figure 8 This is an exemplary embodiment of the present application, showing a ladder wheel state classification confusion matrix with an overall accuracy of 95.8% on the test set.

[0178] In an exemplary embodiment of this application, the specific steps for obtaining the comprehensive state detection result of the vibration signal based on the confidence level result and the predicted state type may include:

[0179] Obtain the Euclidean distance between the dual-scale features and the cluster centers of the predicted state type;

[0180] When the Euclidean distance is less than the distance threshold and the confidence level of the confidence result is greater than or equal to the confidence threshold, a comprehensive state detection result representing the predicted state type as a valid detection result is obtained; otherwise, a suspected detection result representing the predicted state type is obtained.

[0181] This embodiment determines the validity of the predicted state type by setting distance thresholds and confidence thresholds. For example, using the ladder wheel state detection method provided in this application, if the data points of the collected vibration signal are clustered into three fault clusters—tire rupture, bearing damage, and tire detachment—and the Euclidean distance between the data points and the corresponding cluster center is less than 0.5, and the pre-trained fault diagnosis model outputs a specific predicted state type with a confidence score of not less than 0.7, the system is determined to be a valid fault. If only one of the above clustering or model determination conditions is met, the system is determined to be a suspected fault, and further continuous observation is required to confirm the state.

[0182] In another exemplary embodiment, this application also includes an early warning mechanism. Specifically, it implements tiered early warning and maintenance strategies for predicted state types representing different frequency characteristics of the escalator wheel, such as normal condition, tire rupture, bearing damage, and tire detachment. No early warning information is pushed during normal conditions; a yellow early warning information is pushed for tire rupture or bearing damage, prompting inspection and replacement; a red early warning information is pushed for tire detachment, and the escalator is stopped for inspection when no passengers are passing through.

[0183] Figure 9 This is a block diagram illustrating a ladder wheel state detection system 900, as shown in an exemplary embodiment of this application. Figure 9 As shown, the system includes:

[0184] The acquisition unit 901 is used to acquire vibration signals of the escalator step wheels through a high-frequency acceleration sensor;

[0185] The feature extraction unit 902 is used to extract dual-scale features from the vibration signal to obtain dual-scale features, which include micro-instantaneous features and macro-trend features.

[0186] The dynamic weighting unit 903 is used to dynamically adjust the corresponding initial weights based on dual-scale features through a pre-trained fault detection model to obtain the target feature weights corresponding to the vibration signal.

[0187] The state determination unit 904 is used to perform prediction processing based on dual-scale features and target feature weights to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state.

[0188] The result unit 905 is used to calculate the confidence level of the predicted state type based on the target feature weight, obtain the confidence level result, and obtain the comprehensive state detection result of the vibration signal based on the confidence level result and the predicted state type.

[0189] This system applies the step wheel state detection method provided in this application. By extracting dual-scale features of vibration signals as input to a fault detection model, it assesses the step wheel state from multiple perspectives. This avoids the problems of insufficient generalization and anti-interference ability, and excessively high false positive rates, caused by relying on single-domain features. Furthermore, through dynamic weight diagnosis in the pre-trained fault detection model, it specifically predicts and judges the predicted state type corresponding to the vibration signal, further improving detection accuracy. Therefore, the provided step wheel state detection method achieves real-time monitoring of escalator step wheel states, enabling efficient and accurate identification of step wheel defects, improving step wheel maintenance efficiency, reducing maintenance costs, and increasing the safety and reliability of escalators.

[0190] In another exemplary embodiment, the system further includes:

[0191] The adaptive anti-interference preprocessing unit is used to determine the current operating condition of the escalator based on the vibration signal using a fundamental frequency identification algorithm; it removes interference signals in the vibration signal corresponding to the operating condition frequency band to obtain the processed signal, which is then used for dual-scale feature extraction.

[0192] In another exemplary embodiment, the feature extraction unit 902 is further configured to decompose the vibration signal using wavelet packet transform to obtain multiple frequency bands, and calculate the instantaneous impact energy entropy of the multiple frequency bands as micro-instantaneous features; divide the vibration signal into windows based on a preset window length, and perform linear fitting on the RMS value in each window to obtain the linear fitting slope as macro-trend features; obtain the correlation between micro-instantaneous features and macro-trend features, construct a feature correlation matrix based on the correlation relationship; and form dual-scale features based on the micro-instantaneous features and macro-trend features in the feature correlation matrix.

[0193] In another exemplary embodiment, the feature extraction unit 902 is further configured to determine the energy proportion of the preset frequency band based on the relationship between the energy entropy range corresponding to the preset frequency band and the instantaneous impact energy entropy; obtain a first relationship between the energy proportion and the corresponding first baseline, and a second relationship between the linear fitting slope and the corresponding second baseline; determine the correlation between micro-instantaneous features and macro-trend features based on the first and second relationships through preset correlation logic; select associated target micro-instantaneous features and target macro-trend features from the micro-instantaneous features and macro-trend features of the vibration signal based on the correlation relationship; and construct a feature correlation matrix based on the target micro-instantaneous features and target macro-trend features.

[0194] In another exemplary embodiment, the system further includes:

[0195] The feature data preprocessing unit is used to normalize the dual-scale features to obtain the processed features; based on the 3σ criterion, outliers in the processed features are identified and removed to obtain the target dual-scale features, which are then input into the pre-trained fault detection model to obtain the comprehensive state detection results.

[0196] In another exemplary embodiment, the dynamic weighting unit 903 is further configured to obtain distance information between the dual-scale features and the cluster centers of the pre-trained fault diagnosis model through the pre-trained fault diagnosis model; the cluster centers correspond one-to-one with normal states and fault states including tire rupture, bearing damage, and tire detachment; based on the distance information, determine the state type corresponding to the cluster center closest to the dual-scale features as the initial type corresponding to the vibration signal; obtain the initial weight corresponding to the initial type; obtain the feature deviation of each feature value in the dual-scale features; and dynamically adjust the initial weights based on the feature deviation using dynamic adjustment rules to obtain the target feature weights corresponding to the vibration signal.

[0197] In another exemplary embodiment, the state determination unit 904 is further configured to obtain the prediction results output by each decision tree based on dual-scale features through the random forest model in the pre-trained fault diagnosis model; and to perform weighted fusion of the prediction results based on the weight values ​​corresponding to each decision tree in the target feature weights to obtain the predicted state type corresponding to the vibration signal.

[0198] In another exemplary embodiment, the result unit 905 is further configured to obtain the prediction probability of each decision tree of the random forest model in the pre-trained fault diagnosis model and the prediction state type; and to calculate the confidence level based on the weight value and prediction probability of each decision tree in the target feature weight to obtain the confidence level result of the prediction state type.

[0199] In another exemplary embodiment, the result unit 905 is further configured to obtain the Euclidean distance between the dual-scale features and the cluster centers of the predicted state type; when the Euclidean distance is less than the distance threshold and the confidence of the confidence result is greater than or equal to the confidence threshold, a comprehensive state detection result representing the predicted state type as a valid detection result is obtained; otherwise, a suspected detection result representing the predicted state type is obtained.

[0200] It should be noted that the ladder wheel state detection system and the ladder wheel state detection method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the ladder wheel state detection system provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0201] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the ladder wheel state detection method provided in the above embodiments.

[0202] Figure 10 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown. It should be noted that... Figure 10 The computer system 1000 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0203] like Figure 10 As shown, the computer system 1000 includes a Central Processing Unit (CPU) 1001, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on programs stored in Read-Only Memory (ROM) 1002 or programs loaded from storage portion 1008 into Random Access Memory (RAM) 1003. The RAM 1003 also stores various programs and data required for system operation. The CPU 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An Input / Output (I / O) interface 1005 is also connected to the bus 1004.

[0204] The following components are connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. Removable media 1011, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 1010 as needed so that computer programs read from them can be installed into storage section 1008 as needed.

[0205] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 1009, and / or installed from removable medium 1011. When the computer program is executed by central processing unit (CPU) 1001, it performs various functions defined in the system of this application.

[0206] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0207] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0208] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0209] Another aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned ladder wheel state detection method. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0210] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the ladder wheel state detection method provided in the various embodiments described above.

[0211] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for detecting the state of a ladder wheel, characterized in that, The method includes: Vibration signals of escalator step wheels are collected using a high-frequency accelerometer. The vibration signal is subjected to dual-scale feature extraction to obtain dual-scale features, which include micro-instantaneous features and macro-trend features. By using a pre-trained fault detection model, the corresponding initial weights are dynamically adjusted based on the dual-scale features to obtain the target feature weights corresponding to the vibration signal. Based on the dual-scale features and the target feature weights, a prediction processing is performed to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state. The confidence level of the predicted state type is calculated based on the target feature weights to obtain the confidence level result, and the comprehensive state detection result of the vibration signal is obtained based on the confidence level result and the predicted state type.

2. The method according to claim 1, characterized in that, The method further includes: The current operating condition of the escalator is determined based on the vibration signal using a fundamental frequency identification algorithm. Interference signals in the frequency band corresponding to the operating condition in the vibration signal are removed to obtain the processed signal, which is then used for dual-scale feature extraction.

3. The method according to claim 1, characterized in that, The process of extracting dual-scale features from the vibration signal to obtain dual-scale features includes: The vibration signal is decomposed using wavelet packet transform to obtain multiple frequency bands, and the instantaneous impact energy entropy of the multiple frequency bands is calculated as a microscopic instantaneous feature. The vibration signal is divided into windows based on a preset window length, and the RMS value in each window is linearly fitted to obtain the linear fitting slope, which is used as a macroscopic trend feature. Obtain the correlation between the micro-instantaneous features and the macro-trend features, and construct a feature correlation matrix based on the correlation; Dual-scale features are formed based on the micro-instantaneous features and macro-trend features in the feature correlation matrix.

4. The method according to claim 3, characterized in that, The step of obtaining the correlation between the microscopic instantaneous features and the macroscopic trend features, and constructing a feature correlation matrix based on the correlation, includes: Based on the relationship between the energy entropy range corresponding to the preset frequency band and the instantaneous impact energy entropy, the energy proportion of the preset frequency band is determined; Obtain a first relationship between the energy percentage and the corresponding first baseline, and a second relationship between the linear fitting slope and the corresponding second baseline; By using preset association logic, the association relationship between the micro-instantaneous features and the macro-trend features is determined based on the first relationship and the second relationship; Based on the aforementioned correlation, relevant target micro-instantaneous features and target macro-trend features are selected from the micro-instantaneous features and macro-trend features of the vibration signal; A feature correlation matrix is ​​constructed based on the target's microscopic instantaneous features and the target's macroscopic trend features.

5. The method according to claim 1, characterized in that, The method further includes: The dual-scale features are normalized to obtain the processed features; Outliers in the processed features are identified and removed based on the 3σ criterion to obtain target dual-scale features, which are then input into the pre-trained fault detection model to obtain comprehensive state detection results.

6. The method according to claim 1, characterized in that, The step of dynamically adjusting the corresponding initial weights based on the dual-scale features using a pre-trained fault diagnosis model to obtain the target feature weights corresponding to the vibration signal includes: Using a pre-trained fault diagnosis model, the distance information between the dual-scale features and the cluster centers of the pre-trained fault diagnosis model is obtained; the cluster centers correspond one-to-one with the normal state and the fault states, including tire rupture, bearing damage, and tire detachment. Based on the distance information, the state type corresponding to the cluster center closest to the dual-scale feature is determined as the initial type corresponding to the vibration signal; Obtain the initial weight corresponding to the initial type; Obtain the feature deviation of each feature value in the dual-scale feature; By dynamically adjusting the rules, the initial weights are dynamically adjusted based on the feature deviation to obtain the target feature weights corresponding to the vibration signal.

7. The method according to claim 1, characterized in that, The prediction processing based on the dual-scale features and the target feature weights to obtain the predicted state type corresponding to the vibration signal includes: The prediction results of each decision tree are obtained based on the dual-scale features through the random forest model in the pre-trained fault diagnosis model. Based on the weight value corresponding to each decision tree in the target feature weights, the prediction results are weighted and fused to obtain the predicted state type corresponding to the vibration signal.

8. The method according to claim 1, characterized in that, The step of calculating the confidence score of the predicted state type based on the target feature weights to obtain the confidence score result includes: Obtain the prediction probability of each decision tree in the random forest model of the pre-trained fault diagnosis model and the predicted state type; The confidence level is calculated based on the weight value of each decision tree in the target feature weight and the prediction probability to obtain the confidence level result of the predicted state type.

9. The method according to claim 1, characterized in that, The comprehensive state detection result of the vibration signal obtained based on the confidence level and the predicted state type includes: Obtain the Euclidean distance between the dual-scale feature and the cluster center of the predicted state type; When the Euclidean distance is less than the distance threshold and the confidence level of the confidence result is greater than or equal to the confidence threshold, a comprehensive state detection result is obtained, which characterizes the predicted state type as a valid detection result; otherwise, a suspected detection result is obtained, which characterizes the predicted state type as a false detection result.

10. A ladder wheel condition detection system, characterized in that, include: The acquisition unit is used to acquire vibration signals of the escalator step wheels through a high-frequency acceleration sensor; The feature extraction unit is used to perform dual-scale feature extraction on the vibration signal to obtain dual-scale features, which include micro-instantaneous features and macro-trend features. The dynamic weighting unit is used to dynamically adjust the corresponding initial weights based on the dual-scale features using a pre-trained fault detection model to obtain the target feature weights corresponding to the vibration signal. A state determination unit is used to perform prediction processing based on the dual-scale features and the target feature weights to obtain the predicted state type corresponding to the vibration signal. The predicted state type includes normal state and fault state. The result unit is used to calculate the confidence level of the predicted state type based on the target feature weights, obtain the confidence level result, and obtain the comprehensive state detection result of the vibration signal based on the confidence level result and the predicted state type.

Images