Escalator health assessment method and device based on deep learning and terminal equipment

Abstract

The application is suitable for the escalator monitoring technical field, and provides an escalator health assessment method, device and terminal equipment based on deep learning. The method comprises the following steps: collecting escalator operation data through a sensor, wherein the escalator operation data comprises an escalator vibration signal, an escalator temperature, an escalator operation noise and an escalator operation current; extracting local features for representing the health state of the escalator from the escalator operation data through a convolutional neural network, fusing the local features of various escalator operation data through an attention mechanism, and obtaining a global health state representation of the escalator; constructing an escalator field knowledge graph; when the global health state representation indicates that the escalator state is abnormal, reasoning the cause of the escalator state abnormality based on the relationship between entities in the escalator field knowledge graph, and assessing the health condition of the escalator according to the cause of the escalator state abnormality. The application can comprehensively analyze the escalator operation data, effectively assess the health state of the escalator, and improve the safety, reliability and operation efficiency of the escalator.

Description

Technical Field

[0001] This application relates to the field of escalator monitoring technology, specifically to an escalator health assessment method, device, and terminal equipment based on deep learning. Background Technology

[0002] Escalators are a common type of vertical transportation equipment, widely used in public places such as shopping malls, subways, and airports. The safe operation of escalators is directly related to the safety of public life and property; therefore, effective health monitoring and fault diagnosis of escalators are crucial. Traditional escalator maintenance typically relies on periodic inspections and manual patrols, which suffers from long maintenance cycles, high labor costs, and delayed fault detection.

[0003] In recent years, with the development of IoT, big data, and AI technologies, predictive maintenance has begun to be applied in the escalator field. By deploying various sensors on escalators to collect real-time operational data and using machine learning algorithms to analyze the data, it is possible to monitor the health status of escalators and provide early warnings of malfunctions. The inventors of this application have discovered that traditional escalator health monitoring methods are mainly based on single types of data, such as vibration or temperature data, which are insufficient to comprehensively reflect the health status of escalators; most methods employ traditional machine learning models, such as support vector machines or decision trees, which have limited ability to model complex escalator systems. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, device, and terminal equipment for escalator health assessment based on deep learning, which can comprehensively analyze escalator operation data and effectively assess the health status of escalators.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] In a first aspect, embodiments of this application provide a deep learning-based escalator health assessment method, including:

[0007] Escalator operation data is collected by sensors, including escalator vibration signals, escalator temperature, escalator operating noise, and escalator operating current. The sampling frequency and transmission cycle of the escalator operation data are adjusted based on the escalator operating status and the information entropy of the escalator operation data.

[0008] The local features representing the health status of the escalator are extracted from the escalator operation data by a convolutional neural network, and the local features of various escalator operation data are adaptively fused through an attention mechanism to obtain a global health status representation of the escalator.

[0009] A knowledge graph for the escalator domain is constructed, which includes entities and relationships between entities. Entities include escalator components, fault types, maintenance methods, and operating parameters. In constructing the knowledge graph, a low-dimensional structural embedding representation of each entity and the relationships between entities are learned based on a graph convolutional network.

[0010] When the global health status representation indicates that the escalator status is abnormal, the cause of the escalator status abnormality is inferred based on the relationships between entities in the escalator domain knowledge graph, and the health status of the escalator is assessed based on the cause of the escalator status abnormality.

[0011] Based on the first aspect, in some embodiments, when the global health status characterization indicates that the escalator is operating normally, the sensor collects escalator operation data at a first sampling frequency and transmits the escalator operation data at a first transmission cycle; when the global health status characterization indicates that the escalator is operating abnormally, the sensor collects escalator operation data at a second sampling frequency and transmits the escalator operation data at a second transmission cycle; wherein, the second sampling frequency is greater than the first sampling frequency, and the second transmission cycle is less than the first transmission cycle.

[0012] And / or, when the model performance index decreases, the sensor increases the sampling frequency and reduces the transmission cycle; wherein, the model performance index is the performance index of the model composed of the convolutional neural network and the attention mechanism, and the performance index of the model composed of the graph convolutional network and the escalator domain knowledge graph.

[0013] And / or, when the information entropy of the escalator operation data increases, the sensor increases the sampling frequency and reduces the transmission cycle.

[0014] Based on the first aspect, in some embodiments, if the escalator speed at time t is... Less than the escalator speed threshold v th And the model performance index p at time t (t) Less than the model performance metric threshold p th Then the sampling frequency at time t+1 is Let be the sampling frequency at time t, and α be the adjustment factor, where 0 < α < 1;

[0015] If the escalator speed at time t... Greater than or equal to the escalator speed threshold v th , or the model performance index p at time t (t) Greater than or equal to the model performance metric threshold p th Then the sampling frequency at time t+1 is β is an adjustment factor, and β > 1.

[0016] Based on the first aspect, in some embodiments, the information entropy of the escalator operation data is: X represents the data collected by the sensor, x i p(x) represents the i-th value of the data. i ) represents x i The probability of occurrence;

[0017] When the sensor transmits data, wavelet transform is used to compress the time-series data of the escalator operation, and principal component analysis is used to select the most important features for transmission.

[0018] Based on the first aspect, in some embodiments, the step of adaptively fusing local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator includes:

[0019] pass By integrating local features from various escalator operation data, a global health status representation of the escalator is obtained; where v (a) Let be the global health status representation vector of the escalator. This represents the feature vector of the data collected by the i-th sensor. Let be the attention weight for the i-th sensor, and N be the number of sensors. u is the hidden representation of the data collected by the i-th sensor. (a) For attention vectors, and These are the preset parameters for the linear transformation.

[0020] Based on the first aspect, in some embodiments, prior to the adaptive fusion of local features of various escalator operation data via an attention mechanism, the method further includes:

[0021] The local features of each sensor are adaptively weighted according to their importance by the gating unit, and then the step of adaptively fusing the local features of various escalator operation data through the attention mechanism is executed.

[0022] The process by which the gating unit adaptively weights the local features of each sensor according to their importance includes: through... Adaptive weighting is applied to the local features of the data collected by each sensor. For local features of the data collected by the i-th sensor after adaptive weighting, For local features of the data collected by the i-th sensor before adaptive weighting, This represents the gating signal corresponding to the i-th sensor in the gating unit, where ⊙ denotes element-wise multiplication. and σ represents the preset parameters of the gating unit, and σ is the sigmoid activation function.

[0023] Based on the first aspect, in some embodiments, the method further includes: fusing the attribute information of the entity with a low-dimensional structure embedding representation to obtain an enhanced entity; wherein, let the attribute features of the i-th entity be... Low-dimensional structure embedding representation is Then the enhanced entity e i for: and These are the transformation matrices for attribute information and low-dimensional structural embedding representation, respectively;

[0024] The reasoning for the abnormal escalator status based on the relationships between entities in the escalator domain knowledge graph includes: through... The reason for the abnormal status of the escalator. To deduce the cause of the escalator's abnormal condition, For target entity The enhancement indicates that Here is the parameter matrix for the attention mechanism, and | denotes vector concatenation. To target entity Related entities The information obtained after weighted aggregation For target entity With related entities Attention weights between them This is the parameter matrix for the attention mechanism.

[0025] Based on the first aspect, in some embodiments, the method further includes:

[0026] For multiple escalator operators, each escalator operator uses a federated learning framework to train a model locally using its own escalator operation data; wherein, the model is a model composed of the convolutional neural network and the attention mechanism, and a model composed of the graph convolutional network and the escalator domain knowledge graph;

[0027] After training, the various escalator operators exchange the trained model parameters through an encrypted communication protocol. In particular, differential privacy technology is introduced during the training of the model using the federated learning framework to add random noise to the gradient information shared between the various escalator operators.

[0028] Secondly, embodiments of this application provide a deep learning-based escalator health assessment device, comprising:

[0029] The data acquisition module is used to collect escalator operation data through sensors. The escalator operation data includes escalator vibration signals, escalator temperature, escalator operating noise, and escalator operating current. The acquisition frequency and transmission cycle of the escalator operation data are adjusted based on the escalator operating status and the information entropy of the escalator operation data.

[0030] The feature extraction and fusion module is used to extract local features representing the health status of the escalator from the escalator operation data through a convolutional neural network, and adaptively fuse the local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator.

[0031] The knowledge graph construction module is used to build a knowledge graph for the escalator domain. This knowledge graph contains entities and the relationships between entities. Entities include escalator components, fault types, maintenance methods, and operating parameters. In the process of building the knowledge graph for the escalator domain, a low-dimensional structural embedding representation of each entity and the relationships between entities are learned based on a graph convolutional network.

[0032] The health assessment module is used to infer the cause of the escalator's abnormal status based on the relationships between entities in the escalator domain knowledge graph when the global health status representation indicates that the escalator's status is abnormal, and to assess the health status of the escalator based on the cause of the abnormal status.

[0033] Thirdly, embodiments of this application provide a terminal device, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the deep learning-based escalator health assessment method as described in any of the first aspects.

[0034] The beneficial effects of the embodiments of this application compared with the prior art include:

[0035] In this embodiment, escalator operation data is collected by sensors, including escalator vibration signals, escalator temperature, escalator operating noise, and escalator operating current. A convolutional neural network extracts local features characterizing the escalator's health status from the operation data, and an attention mechanism adaptively fuses these local features to obtain a global health status representation of the escalator. An escalator domain knowledge graph is constructed. When the global health status representation indicates an abnormal escalator status, the cause of the abnormality is inferred based on the relationships between entities in the escalator domain knowledge graph. The escalator's health status is then assessed based on the cause of the abnormality. This embodiment comprehensively analyzes escalator operation data, effectively assesses escalator health status, promptly identifies potential faults, and provides maintenance decision recommendations, thereby improving the safety, reliability, and operational efficiency of the escalator.

[0036] Compared to traditional rule-based or single-model methods, the embodiments of this application can mine the health degradation patterns hidden in the data from multiple dimensions, exhibiting stronger predictive power and robustness. Furthermore, the embodiments of this application employ privacy protection measures such as data encryption, federated learning, and differential privacy during model training and deployment, ensuring model performance while maximizing the protection of data security for all parties. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the deep learning-based escalator health assessment method provided in the embodiments of this application;

[0039] Figure 2 This is a schematic diagram of a multi-view fusion model based on a convolutional neural network provided in an embodiment of this application;

[0040] Figure 3 This is a schematic diagram of a multi-view fusion model that integrates a fusion gating unit and an attention mechanism, as provided in an embodiment of this application.

[0041] Figure 4 This is a schematic diagram of the structure of the deep learning-based escalator health assessment device provided in the embodiments of this application;

[0042] Figure 5 This is a schematic diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0043] The present application will be described more clearly below with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the function of the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0045] See Figure 1 The deep learning-based escalator health assessment method provided in this application embodiment may include the following steps:

[0046] Step 101: Collect escalator operation data through sensors. The escalator operation data includes escalator vibration signal, escalator temperature, escalator operation noise and escalator operation current.

[0047] For example, various types of sensors, such as vibration sensors, temperature sensors, noise sensors, and current sensors, can be installed on key components of the escalator to collect escalator operation data in real time.

[0048] The sampling frequency and transmission cycle of the escalator operation data collected by the sensors are adjusted based on the escalator's operating status and the information entropy of the escalator's operation data. Specifically, an adaptive data sampling and transmission strategy can be adopted, dynamically adjusting the sensor's sampling frequency and data transmission cycle according to the escalator's operating status (e.g., speed, passenger capacity) and the performance indicators of the deep learning model (e.g., loss function, accuracy).

[0049] This embodiment involves three deep learning models: a multi-view fusion model based on convolutional neural networks (CNN), a knowledge graph embedding model based on graph convolutional networks (GCN), and a distributed training model based on federated learning. These three deep learning models will be introduced one by one in the relevant sections below.

[0050] In one scenario, when the global health status indicates that the escalator is operating normally, the sensor collects escalator operation data at a first sampling frequency and transmits the data over a first transmission cycle. When the global health status indicates that the escalator is operating abnormally, the sensor collects escalator operation data at a second sampling frequency and transmits the data over a second transmission cycle. The second sampling frequency is greater than the first sampling frequency, and the second transmission cycle is less than the first transmission cycle. That is, when the escalator is operating normally, the sensor reduces the sampling frequency and transmission cycle to decrease the amount of data and the transmission load.

[0051] For example, if the escalator speed v at time t s (t) Less than the escalator speed threshold v th And the model performance index p at time t (t) Less than the model performance metric threshold p th Then the sampling frequency at time t+1 is Let be the sampling frequency at time t, and α be the adjustment factor, where 0 < α < 1.

[0052] If the escalator speed at time t... Greater than or equal to the escalator speed threshold v th , or the model performance index p at time t (t) Greater than or equal to the model performance metric threshold p thThen the sampling frequency at time t+1 is β is an adjustment factor, and β > 1.

[0053] In another scenario, when model performance metrics decline, the sensor increases the sampling frequency and reduces the transmission cycle. Here, model performance metrics refer to the performance metrics of the model composed of the aforementioned convolutional neural network and attention mechanism, as well as the performance metrics of the model composed of the aforementioned graph convolutional network and the aforementioned escalator domain knowledge graph. These performance metrics can specifically be loss functions, accuracy, etc. Increasing the sampling frequency and transmission cycle when model performance metrics decline allows for the acquisition of data with higher accuracy and real-time performance.

[0054] In another scenario, when the information entropy of the escalator operation data increases, the sensor increases the sampling frequency and reduces the transmission cycle. When adjusting the sensor sampling frequency and transmission cycle, not only the escalator's operating status and model performance indicators can be considered, but also the information entropy of the data. When the information entropy of the data is high, it indicates that the data contains more useful information; in this case, the sampling frequency can be appropriately increased and the transmission cycle reduced. Conversely, when the information entropy of the data is low, it indicates high data redundancy; in this case, the sampling frequency can be reduced and the transmission cycle increased to reduce the amount of data.

[0055] Specifically, the information entropy of escalator operation data can be... X represents the data collected by the sensor, x i p(x) represents the i-th value of the data. i ) represents x i The probability of occurrence.

[0056] In addition, when the sensor transmits data, wavelet transform can be used to compress the time-series data of escalator operation, and principal component analysis can be used to select the most important features for transmission.

[0057] During data transmission, this embodiment introduces data compression and feature selection techniques to reduce the amount of data transmitted. Specifically, wavelet transform and other methods can be used to compress the time-series data of escalator operation, reducing the data length. Then, principal component analysis (PCA) and other methods can be used to select the most important features from the compressed data for transmission, reducing the data dimensionality. Assume the original data is X∈R. n×m Where n is the number of sampling points and m is the feature dimension. Wavelet transform is applied to X to obtain the compressed data. Where n c <n. For X c PCA was performed to obtain the dimensionality-reduced data. Where m r <m. The final transmitted data X r While retaining critical information, it significantly reduced the amount of data, thereby lowering transmission load and storage overhead.

[0058] In another scenario, when the global health status indicates that the escalator is in an abnormal operating state and the model performance indicators are declining, the sampling frequency of the sensors can be increased and the transmission cycle reduced, thereby obtaining more accurate and real-time data to monitor the escalator's operating status.

[0059] In some embodiments, a distributed big data platform (such as Hadoop, Spark, etc.) can be used to store and manage the massive amounts of data collected by sensors, supporting efficient data read, write, and query operations. By designing a reasonable data storage format and indexing mechanism, it is possible to meet both real-time data writing and rapid data retrieval and analysis.

[0060] Step 102: Extract local features representing the health status of the escalator from the escalator operation data using a convolutional neural network, and adaptively fuse the local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator.

[0061] Specifically, after obtaining the raw data collected by the sensors, the raw data undergoes preprocessing operations such as cleaning, denoising, and normalization to improve data quality. Then, using various feature extraction methods such as time domain, frequency domain, and time-frequency domain, key features reflecting the health status of the escalator are extracted from multiple sensor data such as vibration, temperature, noise, and current. These features include the root mean square value, peak factor, and spectral entropy of the vibration signal, as well as the trend and abrupt change points of the temperature signal.

[0062] The convolutional neural network and attention mechanism in this step constitute the first type of deep learning model mentioned above: a multi-view fusion model based on convolutional neural networks (CNN), used to fuse features extracted from different sensors to learn a unified representation of the escalator's health status. For example... Figure 2 As shown, this deep learning model designs an independent CNN sub-network for each type of sensor data, extracts local features, and then adaptively fuses the features extracted by different sub-networks through an attention mechanism to obtain a global health status representation of the escalator.

[0063] In some embodiments, step 102, "adaptively fusing local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator," may include: through... By integrating local features from various escalator operation data, a global health status representation of the escalator is obtained; where v (a) Let be the global health status representation vector of the escalator. This represents the feature vector of the data collected by the i-th sensor. Let be the attention weight for the i-th sensor, and N be the number of sensors. u is the hidden representation of the data collected by the i-th sensor. (a) For attention vectors, and These are the preset parameters for the linear transformation.

[0064] Through an attention mechanism, the system can adaptively adjust the importance of different sensor features. For example, in some cases, vibration features may receive higher attention weights, indicating that the current health status of the escalator is primarily determined by vibration data. This dynamic adjustment capability allows the system to assess the overall health status of the escalator more flexibly and accurately.

[0065] In one example, 40% weighting can be assigned to vibration information, 20% to temperature information, 10% to noise information, and 30% to current information. This allocation indicates that, under the current conditions, the system considers vibration and current information to be the most important, while noise information is relatively less critical. The system will then integrate all this information to form a global health status representation. This global health status representation can include information from multiple aspects, such as the overall state of the mechanical system, the smoothness of operation, energy efficiency, and potential failure risks.

[0066] In this way, the system can flexibly adjust the emphasis on various information based on different situations, thereby arriving at a comprehensive and accurate assessment of the escalator's health status. The advantage of this method is its ability to automatically adapt to different situations without requiring fixed, manually set rules, thus better handling complex and changing circumstances.

[0067] As one possible implementation, before adaptively fusing the local features of various escalator operation data through an attention mechanism, the above-mentioned deep learning-based escalator health assessment method may further include: adaptively weighting the local features of each sensor according to their importance through a gating unit, and then performing the step of adaptively fusing the local features of various escalator operation data through an attention mechanism.

[0068] Specifically, the process by which the gating unit adaptively weights the local features of each sensor according to their importance can include: through... Adaptive weighting is applied to the local features of the data collected by each sensor. For local features of the data collected by the i-th sensor after adaptive weighting, For local features of the data collected by the i-th sensor before adaptive weighting, This represents the gating signal corresponding to the i-th sensor in the gating unit, where ⊙ denotes element-wise multiplication.

[0069] and σ represents the preset parameters of the gating unit, and σ is the sigmoid activation function.

[0070] For example, such as Figure 3 As shown, a corresponding CNN subnetwork is set up for each sensor to extract local features from the data collected by each sensor. After the local features are extracted by each CNN subnetwork, the local features of the data collected by different sensors are adaptively weighted by a gating unit, and then fused across views through an attention mechanism. The gating unit automatically adjusts the weights according to the importance of the local features of the data collected by each sensor, highlighting the role of key sensors. The attention mechanism calculates the fusion weights based on the correlation between sensors, utilizing the complementary information of different sensors.

[0071] In cross-view attention fusion, the correlation between views (i.e., sensors) is first calculated, and then the fusion weights are calculated based on the correlation. The formula for calculating the correlation between the i-th sensor and the j-th sensor is:

[0072]

[0073] in, The parameter matrix for correlation calculation. Let represent the correlation between the i-th sensor and the j-th sensor.

[0074] The formula for calculating the fusion weight is:

[0075]

[0076]

[0077] in, v is the fusion weight of the i-th sensor. (c) This is the final representation of health status.

[0078] Through the synergistic effect of gating units and attention mechanisms, the importance of different sensor data can be adaptively adjusted, making full use of complementary information from different perspectives, thereby obtaining more comprehensive and accurate escalator health status assessment results.

[0079] Step 103: Construct a knowledge graph for the escalator domain. This knowledge graph includes entities and relationships between entities. Entities include escalator components, fault types, maintenance methods, and operating parameters.

[0080] The escalator domain knowledge graph is a structured knowledge base containing various entities related to escalators and the relationships between them. The specific content of the escalator domain knowledge graph can include entity types, relationship types, attribute information, rules and constraints, and historical data.

[0081] 1. Entity types can include escalator components, fault types, maintenance methods, and operating parameters.

[0082] Escalator components include: drive system (including motor, reducer, drive chain, etc.), step system (including steps, step chain, etc.), handrail system (including handrail belt, handrail drive device, etc.), control system (including control cabinet, sensors, safety switches, etc.), truss structure (including truss, cover plate, etc.), and braking system (including working brake, auxiliary brake, etc.).

[0083] Fault types include: mechanical faults (such as stuck steps, broken handrails, etc.), electrical faults (such as motor overheating, control system malfunction, etc.), safety device faults (such as safety switch failure, etc.), abnormal noise, abnormal vibration, and abnormal speed.

[0084] Maintenance methods include: daily inspections, regular maintenance, fault repairs, component replacements, and system upgrades.

[0085] Operating parameters include: speed, load, running time, energy consumption, etc.

[0086] 2. Relationship types can include composition relationships, fault-symptom relationships, fault-cause relationships, fault-solution relationships, component-parameter relationships, and maintenance-cycle relationships, etc.

[0087] Compositional relationship: For example, "drive system - components - motor", the drive system is composed of motors.

[0088] Fault-symptom relationship: For example, "step jamming - leading to - abnormal vibration", step jamming leads to abnormal vibration.

[0089] Fault-cause relationship: For example, "insufficient lubrication - causes - abnormal noise", meaning insufficient lubrication causes abnormal noise.

[0090] Fault-Solution Relationship: For example, "Handrail belt broken - Need - Replace handrail belt", the solution to the broken handrail belt is to replace the handrail belt.

[0091] Component-parameter relationship: For example, "motor-influence-energy consumption", where energy consumption is a parameter of the motor.

[0092] Maintenance-cycle relationship: For example, "Lubrication - Recommended - Monthly", the lubrication maintenance cycle is monthly.

[0093] 3. Attribute information can include component attributes, fault attributes, and maintenance attributes, etc.

[0094] Component attributes: such as manufacturer, model, and service life. Failure attributes: such as severity and frequency of occurrence. Maintenance attributes: such as required tools, estimated time, and skill requirements.

[0095] 4. Rules and constraints may include security specifications, maintenance rules, etc.

[0096] Safety regulations: For example, "The load must not exceed the rated value." Maintenance rules: For example, "A safety test must be performed after replacing a part."

[0097] 5. Historical data may include: past fault records, maintenance history, component replacement records, etc.

[0098] By constructing such a comprehensive escalator domain knowledge graph, the system can achieve the following:

[0099] 1. More accurate fault diagnosis: For example, by analyzing symptoms and possible causes, problems can be quickly located.

[0100] 2. Provide targeted maintenance recommendations: For example, provide appropriate maintenance methods based on the type and severity of the fault.

[0101] 3. Predict potential problems: For example, predict possible failures based on component lifespan and historical data.

[0102] 4. Optimize maintenance plans: For example, develop reasonable maintenance plans based on the status and maintenance cycle of each component.

[0103] 5. Support decision-making: For example, provide managers with comprehensive information support to assist in decision-making.

[0104] The aforementioned knowledge graph in the escalator field not only contains static structured information, but can also be continuously updated and expanded through machine learning algorithms to incorporate dynamic information such as new failure modes and maintenance experience, thereby continuously improving the system's intelligence and adaptability.

[0105] In constructing the escalator domain knowledge graph, a low-dimensional structural embedding representation of each entity and the relationships between entities are learned based on a graph convolutional network. In this step, the escalator domain knowledge graph and the graph convolutional network constitute the second deep learning model mentioned above: a knowledge graph embedding model based on graph convolutional networks (GCN).

[0106] This deep learning model first constructs a knowledge graph for the escalator domain, containing entities and their relationships related to escalator component structures, fault types, maintenance methods, etc. Then, it uses a Geographic Knowledge Network (GCN) to learn a low-dimensional embedding representation for each entity, while also considering semantic relationships between entities. When an abnormal state of the escalator is identified, the model uses the fault propagation relationships in the knowledge graph to infer possible causes, assisting in fault diagnosis and prediction. The forward propagation process of the GCN is as follows:

[0107]

[0108] Among them, H (l) Let the matrix represent the nodes of the l-th layer. To add a self-loop adjacency matrix, for The degree matrix, W (l) Let σ be the weight matrix of the l-th layer, and σ be the activation function.

[0109] Low-dimensional structural embedding maps entities from their original high-dimensional space (which may contain a large number of attributes and relationships) to a low-dimensional vector space. For example, the entity "motor" may have many attributes (model, power, usage time, etc.). Through low-dimensional structural embedding, a motor can be represented as a vector such as [0.5, 0.3, -0.2, 0.8]. This vector can capture the key features of the motor and its relationships with other entities.

[0110] The relationships between entities can be specifically defined as semantic associations between entities, which refer to various relationships between entities. In escalator data, these relationships can include: "composition" relationships, such as the motor being part of the drive system; "cause" relationships, such as bearing wear potentially causing abnormal vibration; "affect" relationships, such as increased load potentially affecting energy consumption; and "need" relationships, such as abnormal noise potentially requiring lubrication and maintenance.

[0111] By using low-dimensional structural embedding, complex escalator states can be represented by concise vectors that contain information about component states, operating parameters, and potential faults. Graph convolutional networks can learn implicit relationships between entities; for example, abnormal combinations of parameters may predict specific faults. Increased vibration and energy consumption may both point to bearing problems. By learning semantic associations between entities, knowledge of known faults can be transferred to new situations. If it learns that "bearing wear causes vibration," it may infer that wear on other similar components may also cause vibration. Low-dimensional structural embedding allows the system to simultaneously consider sensor data, historical fault records, maintenance information, etc. This fusion can provide more comprehensive and accurate anomaly identification. By analyzing the relationships between entities, the system can provide more meaningful explanations; for example, it can not only point out the existence of an anomaly but also explain its possible causes and effects.

[0112] By establishing a knowledge graph for the escalator domain, the system can more intelligently understand the overall status of the escalator, not only identifying current anomalies but also predicting potential problems and providing targeted maintenance suggestions. It combines the advantages of data-driven and knowledge-driven approaches, making the identification of abnormal states more accurate, comprehensive, and explainable.

[0113] In some embodiments, the deep learning-based escalator health assessment method described above may further include: fusing the attribute information of an entity with a low-dimensional structural embedding representation to obtain an enhanced entity; wherein, let the attribute features of the i-th entity be... Low-dimensional structure embedding representation is Then the enhanced entity e i for: and These are the transformation matrices for attribute information and low-dimensional structure embedding representation, respectively.

[0114] Correspondingly, the above-mentioned reasoning for the abnormal escalator status based on the relationships between entities in the escalator domain knowledge graph may include: through... The reason for the abnormal status of the escalator. To deduce the cause of the escalator's abnormal condition, For target entity The enhancement indicates that Here is the parameter matrix for the attention mechanism, and | denotes vector concatenation. To target entity Related entities The information obtained after weighted aggregation For target entity With related entities Attention weights between them This is the parameter matrix for the attention mechanism.

[0115] Step 104: When the global health status representation indicates that the escalator status is abnormal, the cause of the escalator status abnormality is inferred based on the relationship between entities in the escalator domain knowledge graph, and the health status of the escalator is evaluated based on the cause of the escalator status abnormality.

[0116] For example, after identifying an abnormal escalator status, using an escalator domain knowledge graph to assist in fault diagnosis, prediction, and reasoning about the cause of the escalator status abnormality can be divided into the following steps:

[0117] 1. Abnormal state mapping:

[0118] First, the identified abnormal states are mapped to relevant entities in the escalator domain knowledge graph. For example, if abnormal vibration is detected, the entity "abnormal vibration" is located in the escalator domain knowledge graph.

[0119] 2. Related entity retrieval:

[0120] Retrieve entities directly related to the abnormal state, including possible causes, affected components, and relevant parameters. For example, for "abnormal vibration," potentially associated entities include: possible causes such as bearing wear, loose steps, and motor imbalance; affected components such as the drive system and step system; and relevant parameters such as vibration frequency and vibration amplitude.

[0121] 3. Path Analysis:

[0122] Analyze the relationships between entities in the escalator domain knowledge graph to find paths from abnormal states to possible causes. These paths can include: direct relationships, such as "abnormal vibration" being directly connected to "bearing wear"; and indirect relationships, such as "abnormal vibration" being indirectly connected to "motor imbalance" through "drive system".

[0123] 4. Probabilistic Reasoning:

[0124] Based on historical data and expert knowledge, a probability is assigned to each possible failure path. For example, the probability that "abnormal vibration" is caused by "bearing wear" is 60%; the probability that "abnormal vibration" is caused by "loose steps" is 30%; and the probability that "abnormal vibration" is caused by "motor imbalance" is 10%.

[0125] 5. Evidence integration:

[0126] The system integrates data from other sensors and operating parameters as additional evidence. For example, if an increase in temperature is detected simultaneously, it can increase the probability of "bearing wear"; if the noise level also increases, it can further support the hypothesis of "step loosening".

[0127] 6. Multi-source information fusion:

[0128] By combining the reasoning results from the escalator domain knowledge graph with the output of the deep learning model, the system performs multi-source information fusion. This can be achieved through the following methods: using Bayesian networks to integrate probabilistic information from different sources; employing fuzzy logic methods to handle uncertain information; and utilizing methods such as decision trees or random forests to integrate multiple factors.

[0129] 7. Diagnosis results generation:

[0130] Based on the above analysis, the system generates a sorted list of fault diagnoses, which may include: possible causes of the fault and their probability, severity assessment of each fault, and suggested further inspections or tests.

[0131] 8. Predictive analytics:

[0132] By leveraging causal relationships and historical data in the knowledge graph, the system can also perform predictive analysis: assess the potential chain reactions caused by the current failure, predict the possible impact on other components, and estimate the possible development trend of the failure if maintenance is not carried out.

[0133] 9. Maintenance suggestion generation:

[0134] Finally, based on the diagnostic and prediction results, and combined with maintenance information from the knowledge graph, the system generates maintenance recommendations, including: providing specific repair steps for the most likely causes of failure, suggesting preventative maintenance measures to avoid the potential problems from worsening, and providing a priority list of maintenance tasks.

[0135] 10. Continuous learning and updating:

[0136] The system records the results of each diagnosis and the actual maintenance situation, which is used to update the probabilities and relationships in the escalator domain knowledge graph, optimize fault diagnosis and prediction models, and accumulate new fault modes and solutions.

[0137] Through this process, the escalator domain knowledge graph not only provides structured domain knowledge but also, when combined with data-driven anomaly detection methods, enables a more accurate, interpretable, and predictable fault diagnosis system. This approach is particularly suitable for handling complex systems such as escalators because it can capture the complex interactions between various components and potential fault propagation paths.

[0138] In some embodiments, the above-mentioned deep learning-based escalator health assessment method may further include: for multiple escalator operating units, each escalator operating unit uses a federated learning framework to train a model locally using its own escalator operation data; wherein the model is a model composed of the convolutional neural network and the attention mechanism, and a model composed of the graph convolutional network and the escalator domain knowledge graph; after training, each escalator operating unit exchanges the trained model parameters through an encrypted communication protocol; wherein, differential privacy technology is introduced during the model training process using the federated learning framework to add random noise to the gradient information shared among the escalator operating units.

[0139] Here, the gradient is the partial derivative of the loss function with respect to the model parameters. During model training, gradient information indicates how to adjust the parameters to reduce the loss and is used to update the model parameters. Individual escalator operators do not directly share raw data or complete model parameters; instead, they share gradient information or parameter updates. This shared information is used to aggregate and update the global model on a central server.

[0140] This embodiment relates to the third type of deep learning model mentioned above: a distributed training model based on federated learning. This deep learning model employs a federated learning framework across multiple escalator operating units to achieve collaborative model training. Each operating unit trains CNN and GCN models locally using its own data, and then exchanges model parameters through an encrypted communication protocol (such as isomorphic encryption), optimizing the global model without sharing the original data. To protect data privacy, differential privacy technology is introduced during the federated learning process, adding random noise to the shared gradient information.

[0141] Some embodiments also involve model interpretation and decision support, utilizing techniques such as SHAP (SHapley Additive Explanations) to analyze the prediction results of deep learning models, calculate the contribution of each input feature to the prediction results, and provide interpretability reports. Simultaneously, combining domain expert knowledge and historical maintenance data, this module provides a comprehensive score of the escalator's health status, as well as maintenance priority and decision recommendations for potential faults.

[0142] In some embodiments, the above-mentioned deep learning-based escalator health assessment method further includes: displaying the escalator's real-time operating status, health metrics, prediction results, and other information through web and mobile applications to facilitate managers' understanding of the escalator's operating status; and issuing alarm notifications to relevant personnel when an abnormality is detected in the escalator's status or when an impending malfunction is predicted, so as to facilitate rapid response and handling.

[0143] In some embodiments, the aforementioned deep learning-based escalator health assessment method further includes: utilizing an active learning strategy to select sample points that have the greatest impact on the judgment result based on the uncertainty of the model prediction, actively requesting expert annotations or conducting experimental verification, thereby expanding the training data in a targeted manner. Furthermore, newly collected data can be periodically used to incrementally train and fine-tune the deep learning model, enabling it to adapt to changes in escalator operating conditions.

[0144] In some embodiments, before extracting local features characterizing the health status of the escalator from the escalator operation data using a convolutional neural network, the aforementioned deep learning-based escalator health assessment method may further include: performing spectral analysis on the escalator operation data using a fast Fourier transform to calculate the spectral energy and spectral entropy of the escalator operation data; comparing the spectral energy with a preset spectral energy range and comparing the spectral entropy with a preset spectral entropy range; if the spectral energy exceeds the preset spectral energy range and / or the spectral entropy exceeds the preset spectral entropy range, then inputting the escalator operation data into a convolutional neural network to extract local features characterizing the health status of the escalator from the escalator operation data using the convolutional neural network.

[0145] By calculating the spectral energy and spectral entropy of escalator operation data using Fast Fourier Transform, and pre-determining potentially abnormal escalator operation data based on the spectral energy and spectral entropy, and then inputting the potentially abnormal escalator operation data into a convolutional neural network to extract local features used to characterize the health status of the escalator, the amount of data processing required by the convolutional neural network can be reduced, and the processing efficiency can be improved.

[0146] Specifically, it can be done through Calculate the spectral energy E of the escalator operation data, through Calculate the spectral entropy H of the escalator operation data. x(t) represents the escalator operation data.

[0147] The above-mentioned deep learning-based escalator health assessment method has the following beneficial effects:

[0148] 1. Comprehensively utilize multi-source heterogeneous data to fully assess the health status of escalators. This invention employs multi-modal sensors to collect escalator operation data, covering multiple dimensions such as vibration, temperature, noise, and current. Through deep learning models, feature extraction and fusion of different modal data are achieved, constructing a more comprehensive and accurate escalator health assessment model, effectively improving the accuracy of fault detection and early warning.

[0149] 2. Adaptive adjustment of data sampling and transmission strategies improves data processing efficiency. This invention adaptively adjusts the sensor sampling frequency and data transmission cycle based on the escalator's operating status and model performance. This reduces data redundancy and transmission overhead while ensuring data quality, thereby improving the system's real-time performance and efficiency.

[0150] 3. Introducing domain knowledge to enhance the interpretability of fault diagnosis and prediction. This invention constructs a domain knowledge graph for escalators, embeds the knowledge graph into a deep learning model, and uses graph convolutional networks to learn the representation of entities and relationships. This enables the model to use domain knowledge for reasoning and decision-making, improving the interpretability and credibility of fault diagnosis and prediction results.

[0151] 4. Employing a federated learning framework to achieve secure data sharing and collaborative modeling among multiple escalator operators. This invention uses a federated learning framework among multiple escalator operators, where each operator trains its model locally using its own data, sharing only the model parameters and not the original data. This protects data privacy while enabling collaborative training and optimization of the model, improving its generalization performance.

[0152] 5. Active learning and incremental learning are introduced to achieve continuous model optimization and adaptive updates. This invention employs an active learning strategy, selecting the most valuable samples for annotation and learning, improving annotation efficiency and model performance. Simultaneously, an incremental learning method is used to periodically fine-tune and update the model using newly collected data, enabling the model to adapt to changes in escalator operating conditions and maintain high prediction accuracy.

[0153] 6. Provides visual and interactive decision support, facilitating monitoring and maintenance by management personnel. This invention has developed web and mobile applications to display the real-time operating status, health indicators, and early warning information of escalators in an intuitive and interactive manner, and provides maintenance decision suggestions based on diagnostic results, providing management personnel with an efficient and convenient decision support tool.

[0154] The aforementioned deep learning-based escalator health assessment method, through key technologies such as adaptive data sampling, multi-view feature fusion, knowledge graph embedding, federated learning, and active learning, can comprehensively analyze escalator operation data, effectively assess the escalator's health status, promptly identify potential faults, and provide maintenance decision suggestions, thereby improving the escalator's safety, reliability, and operational efficiency. Compared with traditional rule-based or single-model-based methods, the embodiments of this application can mine the health degradation patterns contained in the data from multiple dimensions, exhibiting stronger predictive capabilities and robustness. Furthermore, the embodiments of this application employ privacy protection measures such as data encryption, federated learning, and differential privacy during model training and deployment, ensuring model performance while maximizing the protection of data security for all parties.

[0155] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0156] Corresponding to the deep learning-based escalator health assessment method described in the above embodiments, Figure 4 The diagram shows a structural block diagram of the escalator health assessment device based on deep learning provided in the embodiments of this application. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0157] See Figure 4 This application provides a deep learning-based escalator health assessment device, including a data acquisition module 201, a feature extraction and fusion module 202, a knowledge graph construction module 203, and a health assessment module 204.

[0158] The data acquisition module 201 is used to collect escalator operation data through sensors. The escalator operation data includes escalator vibration signals, escalator temperature, escalator operating noise, and escalator operating current. The acquisition frequency and transmission cycle of the escalator operation data are adjusted based on the escalator operating status and the information entropy of the escalator operation data.

[0159] The feature extraction and fusion module 202 is used to extract local features representing the health status of the escalator from the escalator operation data through a convolutional neural network, and adaptively fuse the local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator.

[0160] The knowledge graph construction module 203 is used to construct a knowledge graph for the escalator domain. This knowledge graph includes entities and relationships between entities. Entities include escalator components, fault types, maintenance methods, and operating parameters. Specifically, during the construction of the escalator domain knowledge graph, a low-dimensional structural embedding representation of each entity and the relationships between entities are learned based on a graph convolutional network.

[0161] The health assessment module 204 is used to infer the cause of the escalator status abnormality based on the relationship between entities in the escalator domain knowledge graph when the global health status representation indicates that the escalator status is abnormal, and to assess the health status of the escalator based on the cause of the escalator status abnormality.

[0162] Optionally, when the global health status indicates that the escalator is operating normally, the sensor collects escalator operation data at a first sampling frequency and transmits the escalator operation data at a first transmission cycle; when the global health status indicates that the escalator is operating abnormally, the sensor collects escalator operation data at a second sampling frequency and transmits the escalator operation data at a second transmission cycle; wherein the second sampling frequency is greater than the first sampling frequency, and the second transmission cycle is less than the first transmission cycle.

[0163] And / or, when the model performance index decreases, the sensor increases the sampling frequency and reduces the transmission cycle; wherein, the model performance index is the performance index of the model composed of the convolutional neural network and the attention mechanism, and the performance index of the model composed of the graph convolutional network and the escalator domain knowledge graph.

[0164] And / or, when the information entropy of the escalator operation data increases, the sensor increases the sampling frequency and reduces the transmission cycle.

[0165] Optionally, if the escalator speed at time t... Less than the escalator speed threshold v th And the model performance index p at time t (t) Less than the model performance metric threshold p thThen the sampling frequency at time t+1 is Let be the sampling frequency at time t, and α be the adjustment factor, where 0 < α < 1;

[0166] If the escalator speed at time t... Greater than or equal to the escalator speed threshold v th , or the model performance index p at time t (t) Greater than or equal to the model performance metric threshold p th Then the sampling frequency at time t+1 is β is an adjustment factor, and β > 1.

[0167] Optionally, the information entropy of the escalator operation data is: X represents the data collected by the sensor, x i p(x) represents the i-th value of the data. i ) represents x i The probability of occurrence;

[0168] When the sensor transmits data, wavelet transform is used to compress the time-series data of the escalator operation, and principal component analysis is used to select the most important features for transmission.

[0169] Optionally, the feature extraction and fusion module 202 is specifically used for: through By integrating local features from various escalator operation data, a global health status representation of the escalator is obtained; where v (a) Let be the global health status representation vector of the escalator. This represents the feature vector of the data collected by the i-th sensor. Let be the attention weight for the i-th sensor, and N be the number of sensors. u is the hidden representation of the data collected by the i-th sensor. (a) For attention vectors, and These are the preset parameters for the linear transformation.

[0170] Optionally, the aforementioned deep learning-based escalator health assessment device may further include a gating weighting module. The gating weighting module is used to adaptively weight the local features of each sensor according to their importance through a gating unit. Then, the feature extraction and fusion module 202 performs the process of adaptively fusing the local features of various escalator operation data through an attention mechanism.

[0171] The process by which the gating unit adaptively weights the local features of each sensor according to their importance includes: through... Adaptive weighting is applied to the local features of the data collected by each sensor. For local features of the data collected by the i-th sensor after adaptive weighting, For local features of the data collected by the i-th sensor before adaptive weighting, This represents the gating signal corresponding to the i-th sensor in the gating unit, where ⊙ denotes element-wise multiplication. and σ represents the preset parameters of the gating unit, and σ is the sigmoid activation function.

[0172] Optionally, the aforementioned deep learning-based escalator health assessment device may further include an information fusion module. This module fuses the entity's attribute information with a low-dimensional structure embedding representation to obtain an enhanced entity. Here, let the attribute features of the i-th entity be... Low-dimensional structure embedding representation is The enhanced entity for: and These are the transformation matrices for attribute information and low-dimensional structure embedding representation, respectively.

[0173] Correspondingly, health assessment module 204 is specifically used for: through The reason for the abnormal status of the escalator. To deduce the cause of the escalator's abnormal condition, For target entity The enhancement indicates that Here is the parameter matrix for the attention mechanism, and | denotes vector concatenation. To target entity Related entities The information obtained after weighted aggregation For target entity With related entities Attention weights between them This is the parameter matrix for the attention mechanism.

[0174] Optionally, the aforementioned deep learning-based escalator health assessment device may further include a federated learning module. Specifically, the federated learning module is used for: for multiple escalator operating units, each escalator operating unit trains a model locally using its own escalator operation data using a federated learning framework; wherein the model is a model composed of the convolutional neural network and the attention mechanism, and a model composed of the graph convolutional network and the escalator domain knowledge graph; after training, each escalator operating unit exchanges the trained model parameters through an encrypted communication protocol; wherein, differential privacy technology is introduced during the model training process using the federated learning framework to add random noise to the gradient information shared among the escalator operating units.

[0175] Figure 5 This is a schematic diagram of a terminal device provided in an embodiment of the present invention. Figure 5 As shown, the terminal device 300 in this embodiment includes a processor 310 and a memory 320. The memory 320 stores a computer program that can run on the processor 310, such as a deep learning-based escalator health assessment program. When the processor 310 executes the computer program, it implements the steps in the above-described deep learning-based escalator health assessment method embodiment, for example... Figure 1 Steps 101 to 104 are shown. Alternatively, when the processor 310 executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 4 The functions of modules 201 to 204 are shown.

[0176] For example, the computer program can be divided into one or more modules / units, which are stored in the memory 320 and executed by the processor 310 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device 300. For example, the computer program can be divided into a data acquisition module, a feature extraction and fusion module, a knowledge graph construction module, and a health assessment module.

[0177] The terminal device 300 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The terminal device may include, but is not limited to, a processor 310 and a memory 320. Those skilled in the art will understand that... Figure 5 This is merely an example of terminal device 300 and does not constitute a limitation on terminal device 300. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0178] The processor 310 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0179] The memory 320 can be an internal storage unit of the terminal device 300, such as a hard disk or memory of the terminal device 300. The memory 320 can also be an external storage device of the terminal device 300, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal device 300. Furthermore, the memory 320 can include both internal and external storage units of the terminal device 300. The memory 320 is used to store the computer program and other programs and data required by the terminal device. The memory 320 can also be used to temporarily store data that has been output or will be output.

[0180] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A deep learning-based escalator health assessment method, characterized in that, The application relates to a health status evaluation method for an escalator, and belongs to the technical field of health status evaluation. Escalator operation data is collected through sensors, vibration sensors, temperature sensors, noise sensors and current sensors are installed on the escalator, real-time escalator operation data is collected, the escalator operation data includes escalator vibration signals, escalator temperature, escalator operation noise and escalator operation current, the collection frequency of the sensor for collecting the escalator operation data and the transmission period of the escalator operation data are adjusted based on the escalator operation state and the information entropy of the escalator operation data; Local features for representing the health state of the escalator are extracted from the escalator operation data through a convolutional neural network, and the local features of various escalator operation data are adaptively fused through an attention mechanism to obtain a global health state representation of the escalator; An escalator field knowledge graph is constructed, the escalator field knowledge graph includes entities and the relationships between the entities, the entities include escalator components, fault types, maintenance methods and operation parameters, and when the global health state representation indicates that the escalator state is abnormal, the reason for the escalator state abnormality is inferred based on the relationships between the entities in the escalator field knowledge graph, and the health state of the escalator is evaluated according to the reason for the escalator state abnormality; Wherein, after identifying that the escalator state is abnormal, the escalator field knowledge graph is used to assist in fault diagnosis and prediction and to infer the reason for the escalator state abnormality, including: mapping the identified abnormal state to related entities in the escalator field knowledge graph; retrieving entities directly related to the abnormal state, including possible causes, affected components, and related parameters; analyzing the relationships between the entities in the escalator field knowledge graph to find a path from the abnormal state to the possible cause; based on historical data and expert knowledge, assigning a probability to each possible fault path; the system integrates other sensor data and operation parameters as additional evidence; combining the inference results of the escalator field knowledge graph and the output of the deep learning model, the system performs multi-source information fusion; the system generates a ranked fault diagnosis list, including: possible fault causes and their probabilities, and severity assessment of each fault; When the global health state representation indicates that the escalator operation state is normal, the sensor collects the escalator operation data at a first sampling frequency and transmits the escalator operation data at a first transmission period; when the global health state representation indicates that the escalator operation state is abnormal, the sensor collects the escalator operation data at a second sampling frequency and transmits the escalator operation data at a second transmission period; wherein the second sampling frequency is greater than the first sampling frequency, and the second transmission period is less than the first transmission period; The local features of various escalator operation data are adaptively fused through the attention mechanism to obtain a global health state representation of the escalator, including: through fusing the local features of various escalator operation data to obtain a global health state representation of the escalator; wherein v (a) is a global health state representation vector of the escalator, is a feature vector of data collected by the i th sensor, is an attention weight of the i th sensor, N is the number of sensors, and is a hidden representation of data collected by the i th sensor, u (a) is an attention vector, and is a preset parameter of linear transformation.

2. The deep learning-based escalator health assessment method of claim 1, wherein, And / or, when the model performance indicator decreases, the sensor increases the sampling frequency and reduces the transmission period; wherein the model performance indicator is the performance indicator of the model composed of the convolutional neural network and the attention mechanism, and the performance indicator of the model composed of the graph convolutional network and the escalator field knowledge graph; And / or, when the information entropy of the escalator operation data increases, the sensor increases the sampling frequency and reduces the transmission period. ​ 3. The deep learning-based escalator health assessment method of claim 2, wherein, If the escalator speed at the tth moment is less than the escalator speed threshold value v th , and the model performance index p (t) at the tth moment is less than the model performance index threshold value p th , then the sampling frequency at the t+1th moment is the sampling frequency at the tth moment, and a is an adjustment factor, and 0 If the escalator speed at the tth moment is greater than or equal to the escalator speed threshold value v th , or the model performance index p at the tth moment (t) is greater than or equal to the model performance index threshold value p th , then the sampling frequency at the t+1th moment β is an adjustment factor, and β>

1.

4. The deep learning-based escalator health assessment method of claim 2, wherein, The information entropy of the escalator operation data is X represents the data collected by the sensor, x i represents the i-th value of the data, p(x i ) represents the probability of occurrence of x i ; When the sensor transmits data, wavelet transform is used to compress the time series data of the escalator operation, and principal component analysis is used to select the most important features for transmission.

5. The deep learning-based escalator health assessment method of claim 1, wherein, Before adaptively fusing the local features of various escalator operation data through the attention mechanism, the method further comprises: Adaptively weighting the local features of each sensor according to importance through a gating unit, and then performing the step of adaptively fusing the local features of various escalator operation data through the attention mechanism; wherein the process of adaptively weighting the local features of each sensor according to the importance by the gating unit comprises: adaptively weighting the local features of the data collected by each sensor, the local feature of the data collected by the i-th sensor after adaptive weighting, the local feature of the data collected by the i-th sensor before adaptive weighting, the gating signal corresponding to the i-th sensor in the gating unit, and represents element-wise multiplication, and is a preset parameter of the gating unit, and σ is a sigmoid activation function.

6. The deep learning-based escalator health assessment method of claim 1, wherein, The method further comprises: fusing attribute information of the entity with the low-dimensional structure embedding representation to obtain an enhanced entity; wherein attribute features of an i th entity are denoted as The low-dimensional structure embedding representation is The enhanced entity is is: and are respectively a transformation matrix of the attribute information and the low-dimensional structure embedding representation. The escalator state abnormality reason is inferred based on the relationship between the entities in the escalator domain knowledge graph, including: inference of escalator state abnormality reason, for the inferred escalator state abnormality reason, for the target entity enhanced representation, for the parameter matrix of the attention mechanism, | represents vector splicing, for the information obtained after weighting aggregation of the related entities related to the target entity , for the attention weight between the target entity and the related entity , for the parameter matrix of the attention mechanism.

7. The deep learning-based escalator health assessment method of claim 1, wherein, The method further comprises: For multiple escalator operating units, each escalator operating unit trains a model locally using its own escalator operation data using a federated learning framework; wherein the model is a model composed of the convolutional neural network and the attention mechanism, and a model composed of the graph convolutional network and the escalator domain knowledge graph; After training, each escalator operating unit exchanges the trained model parameters through an encrypted communication protocol; wherein differential privacy technology is introduced in the process of training the model through the federated learning framework, and random noise is added to the gradient information shared between each escalator operating unit.

8. An escalator health assessment device based on deep learning, characterized by, Comprise: A data acquisition module for acquiring escalator operation data through sensors, installing vibration sensors, temperature sensors, noise sensors and current sensors on the escalator, and acquiring real-time escalator operation data, the escalator operation data including escalator vibration signals, escalator temperature, escalator operating noise and escalator operating current; wherein the acquisition frequency of the sensor for acquiring escalator operation data and the transmission period of the transmission of escalator operation data are adjusted based on the escalator operating state and the information entropy of the escalator operation data; A feature extraction and fusion module for extracting local features for representing the health status of the escalator from the escalator operation data through a convolutional neural network, and adaptively fusing the local features of various escalator operation data through an attention mechanism to obtain a global health status representation of the escalator; A knowledge graph construction module for constructing an escalator domain knowledge graph, which contains entities and relationships between entities, and entities include escalator components, fault types, maintenance methods and operating parameters; wherein when constructing the escalator domain knowledge graph, the low-dimensional structural embedding representation of each entity and the relationship between entities are learned based on the graph convolutional network; A health assessment module for reasoning the cause of the escalator state anomaly based on the relationship between entities in the escalator domain knowledge graph when the global health status representation indicates that the escalator state is abnormal, and assessing the health status of the escalator according to the cause of the escalator state anomaly. Wherein, after identifying that the escalator state is abnormal, the escalator field knowledge graph is used to assist in fault diagnosis and prediction and to reason the cause of the escalator state anomaly, including: mapping the identified abnormal state to related entities in the escalator field knowledge graph; searching for entities directly related to the abnormal state, including possible causes, affected components, and related parameters; analyzing the relationship between entities in the escalator field knowledge graph to find a path from the abnormal state to the possible cause; based on historical data and expert knowledge, assigning a probability to each possible fault path; the system integrates other sensor data and operating parameters as additional evidence; combining the reasoning results of the escalator field knowledge graph and the output of the deep learning model, the system performs multi-source information fusion; the system generates a ranked fault diagnosis list, including: possible fault causes and their probabilities, and severity assessment of each fault; The feature extraction and fusion module is specifically configured to: obtain the local feature of the escalator running data through fuse the local features of various escalator running data to obtain a global health state representation of the escalator; wherein, v (a) is a global health state representation vector of the escalator, is a feature vector of data collected by the i th sensor, is an attention weight of the i th sensor, N is the number of sensors, and is a hidden representation of data collected by the i th sensor, u (a) is an attention vector, and is a preset parameter of linear transformation.

9. A terminal device comprising a memory and a processor, said memory having stored a computer program executable on said processor, characterized in that, The processor implements the steps of the escalator health assessment method based on deep learning according to any one of claims 1-7 when executing the computer program.

Images