A gate hoist online monitoring system and method

Abstract

The application provides a kind of gate hoist online monitoring system and method, monitoring system includes multi-source sensor array, edge computing gateway, cloud big data analysis platform and remote monitoring terminal, multi-source sensor array gathers mechanical, electrical, acoustic and environmental multidimensional data, edge computing gateway completes data preprocessing and local preliminary anomaly detection, cloud platform built-in deep learning prediction model, safety evaluation engine and dynamic threshold calculation module.The monitoring method is first processed by edge side data and feature extraction, then the deep learning prediction model is used to model the heterogeneous multivariate time series data, calculate the four-dimensional health safety index HSI, based on HSI and real-time working condition, the warning threshold is dynamically updated by adaptive algorithm, and finally the grading warning and maintenance suggestion are generated.The application realizes the precise monitoring of the whole state of the hoist, quantitative health assessment and intelligent warning, improves the monitoring sensitivity and operation science, and ensures the safe and stable operation of the equipment.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering equipment condition monitoring and intelligent operation and maintenance technology, and in particular to an online monitoring system and method for gate hoists. Background Technology

[0002] As a core and critical piece of equipment in water conservancy projects, gate hoists undertake important functions such as gate lifting and lowering control, water resource allocation, and flood control and drainage. The stability of their operation directly affects the safe operation of water conservancy projects and the safety of lives and property in the surrounding areas. With the development of water conservancy projects towards larger scale and greater intelligence, traditional gate hoist monitoring and maintenance models have gradually revealed many drawbacks:

[0003] Firstly, the monitoring dimensions are limited: traditional systems focus on a few electrical or mechanical parameters such as motor current and gate displacement, lacking monitoring of key indicators such as acoustic signals, lubricating oil status, structural torque, and ambient wind speed. This makes it difficult to detect potential equipment failures (such as wire rope fatigue, premature bearing wear, and abnormal gearbox meshing) in the early stages, which can easily lead to sudden failures.

[0004] Secondly, the data processing efficiency is low: existing systems mostly adopt the direct transmission mode of "sensor-cloud", without preprocessing the massive amount of raw data at the edge, resulting in large data transmission bandwidth consumption, concentrated cloud computing pressure, and inability to achieve rapid local anomaly response.

[0005] Third, the early warning mechanism is rigid: traditional monitoring systems generally use fixed thresholds to judge anomalies, without considering dynamic factors such as changes in the load of the gate hoist, environmental fluctuations and the decline in the health status of the equipment, resulting in a high false alarm rate and a high missed alarm rate, making it difficult to adapt to the monitoring needs under complex working conditions.

[0006] Fourth, health assessment lacks a systematic approach: existing technologies mostly rely on single parameters to judge equipment status, and have not established a multi-dimensional and quantitative health assessment system. This makes it impossible to provide accurate health status references for operation and maintenance decisions, resulting in blind operation and maintenance work and increasing operation and maintenance costs and equipment downtime risks.

[0007] Therefore, developing an online monitoring system and method for gate hoists with multi-source data fusion acquisition, edge-cloud collaborative processing, dynamic intelligent early warning, and quantitative health assessment capabilities has become an urgent need in the field of intelligent operation and maintenance of water conservancy projects. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an online monitoring system and method for gate hoists, which solves problems such as single monitoring dimensions, low data processing efficiency, rigid early warning mechanisms, and lack of systematic health assessment in existing technologies.

[0009] In a first aspect, the present invention proposes an online monitoring system for gate hoists, comprising:

[0010] A multi-source sensor array, which is set up in conjunction with the gate hoist body, is used to collect mechanical, electrical, acoustic and environmental data;

[0011] Edge computing gateways are used for sensor data preprocessing, feature extraction, and local lightweight anomaly detection;

[0012] The cloud-based big data analytics platform communicates with the edge computing gateway, receives feature data and some key raw data streams uploaded from the edge, and has built-in deep learning prediction models, security assessment engines and dynamic threshold calculation modules to achieve in-depth data analysis, status prediction, health assessment and intelligent threshold updates.

[0013] The remote monitoring terminal serves as the system's human-computer interaction interface, used for data visualization, early warning information display, and operation and maintenance decision support.

[0014] Preferably, the multi-source sensor array includes:

[0015] Mechanical condition sensors: vibration acceleration sensors, drive shaft torque sensors, wire rope absolute displacement sensors, lubricating oil particulate matter sensors;

[0016] Electrical condition sensors: motor three-phase current and voltage sensors, bearing and winding temperature sensors;

[0017] Acoustic state sensor: acoustic emission sensor;

[0018] Environmental condition sensors: ambient temperature and humidity sensors, wind speed sensors.

[0019] Furthermore, the edge computing gateway includes:

[0020] Data preprocessing module: performs synchronous alignment, cleaning, and standardization on raw data collected from multiple sources of sensors;

[0021] Feature extraction module: Extracts time-domain and frequency-domain features from the preprocessed data. The time-domain features include peak value, mean, variance, kurtosis, and skewness, while the frequency-domain features include spectral peak value, dominant frequency, and harmonic components, providing feature support for subsequent anomaly detection and fault diagnosis.

[0022] Anomaly detection module: Based on feature data extracted from the edge side, a simple threshold method is used to make preliminary anomaly judgment. For parameters that significantly exceed the safe range, a local audible and visual alarm is immediately triggered to achieve rapid response to emergency faults.

[0023] Furthermore, the cloud-based big data analytics platform includes:

[0024] Deep learning prediction model: It adopts a temporal fusion Transformer architecture and achieves high-precision prediction of key parameters and high-sensitivity detection of early anomalies through the collaborative design of gating mechanism, multi-head attention and time series-specific modules.

[0025] Safety assessment engine: Based on multi-source fusion data, it constructs a four-dimensional health assessment system, quantitatively calculates the equipment health and safety index (HSI), and provides accurate basis for operation and maintenance decisions;

[0026] Dynamic threshold calculation module: It adopts an adaptive algorithm, combined with the real-time operating conditions and health status of the equipment, to dynamically adjust the warning thresholds of various monitoring parameters, thereby improving the accuracy and sensitivity of warnings.

[0027] Furthermore, the remote monitoring terminal includes:

[0028] Data visualization module: including web platform, mobile APP, local monitoring screen in data center and other terminals, to display raw data, preprocessed data, feature data, HSI index and dynamic threshold in real time in the form of dashboard, trend curve and heat map;

[0029] Early warning information display module: Based on cloud analysis results, it generates tiered early warnings, and pushes early warning information through sound and light alarms, SMS push, APP pop-ups, email notifications, etc.

[0030] Operation and maintenance decision support module: Based on HSI index, fault type and historical maintenance data, it automatically generates targeted maintenance suggestions and supports operation and maintenance work order creation, dispatch, tracking and closed-loop management.

[0031] Secondly, the present invention also proposes an online monitoring method for gate hoists, comprising the following steps:

[0032] S1. Simultaneously collect data from a multi-source sensor array and preprocess the data through an edge computing gateway to achieve data cleaning, alignment, and standardization;

[0033] S2. Extract key time-domain and frequency-domain feature data through the edge computing gateway, and compare the extracted feature data with a local temporary fixed threshold. If the feature data exceeds the temporary threshold, trigger a local audible and visual alarm and upload the anomaly to the cloud big data analysis platform. If the feature data does not exceed the temporary threshold, package the feature data and part of the original data stream and upload them to the cloud big data analysis platform.

[0034] S3, after receiving data uploaded from the edge side, the cloud-based big data analysis platform uses a time-series fusion architecture with a deep learning prediction model to predict key parameters and detect anomalies.

[0035] S4, the cloud-based big data analysis platform uses a security assessment engine to call multi-source fusion data to calculate the current device's Health and Safety Index (HSI);

[0036] S5, the cloud-based big data analysis platform calculates the dynamic thresholds of each parameter based on real-time operating conditions and HSI through the adaptive dynamic threshold algorithm of the dynamic threshold calculation module, thereby updating the early warning baseline;

[0037] The S6 cloud-based big data analytics platform combines early anomaly detection results, HSI index, and dynamic threshold comparison results to generate tiered early warnings. It then pushes the warning information, HSI index, possible causes of the fault, and maintenance recommendations to the remote monitoring terminal. Upon receiving the warning, maintenance personnel view detailed data and analysis reports through the terminal, formulate maintenance plans based on the recommendations, and create and dispatch maintenance work orders. After maintenance is completed, the processing results are entered into the terminal, and the system updates the equipment's historical maintenance records, providing data support for subsequent health assessments and threshold adjustments.

[0038] Furthermore, the data preprocessing in step S1 specifically includes:

[0039] Cleaning: Remove outliers caused by sensor malfunctions and electromagnetic interference, and use the 3σ criterion to remove data points that deviate from the mean by a certain multiple of the standard deviation;

[0040] Alignment: Synchronize and align the raw data collected by multiple sensors, and achieve multi-dimensional data time synchronization based on sensor timestamps;

[0041] Standardization: Converting data of different dimensions into standardized data within the interval [0, 1]. The formula is:

[0042] ,

[0043] Where, x std The data is standardized, and x is the original data. min x max These are the historical minimum and maximum values ​​of the parameter, respectively.

[0044] Furthermore, in step S3, the deep learning prediction model adopts a temporal fusion Transformer architecture, which, through the collaborative design of gating mechanisms, multi-head attention, and a dedicated time series module, specifically includes:

[0045] S3.1 Model Input: The input is a heterogeneous multivariate time series composed of data collected by a multi-source sensor array, specifically including two types of features: time series and static features. The time series features are the set of dynamic features in the time domain and frequency domain extracted by the edge computing gateway, and the static features are non-time-varying parameters such as the inherent parameters of the gate hoist and the characteristics of the installation position.

[0046] After all input features are standardized, a three-dimensional input tensor X∈ is constructed by aligning them according to time steps. Where batch_size is the batch size, step_steps is the step size, and feature_dim is the total dimension of the features;

[0047] S3.2 Introducing a gated selection unit: The importance weights w of each feature are calculated using the Sigmoid activation function. i ∈[0,1], dynamically filter effective features and suppress noise interference, the formula is:

[0048] ,

[0049] in, Let be the value of the i-th temporal feature at time t, s be the static feature embedding vector, W be the weight matrix, b be the bias term, and σ be the Sigmoid function;

[0050] S3.3 Multi-scale temporal convolutional layer: A multi-scale convolutional module is constructed using 1D convolutional kernels to capture local temporal features at different time granularities, output multi-scale feature maps and stitch them together;

[0051] S3.4, Temporal Multi-Head Attention Layer: By using multiple attention heads in parallel, it simultaneously captures both global long-term dependencies and local critical dependencies in time series data.

[0052] Each attention head computes the mapping of query Q, key K, and value V, and calculates the feature association weights using the scaled dot product attention formula:

[0053] ,

[0054] Where, d k The dimensions of Q and K are set to avoid gradient vanishing due to excessively high dimensionality;

[0055] S3.5, Gated Recurrent Unit (GRU) Layer:

[0056] Received from the attention layer output, the model dynamically memorizes key temporal information and forgets redundant noise through the reset and update gates of GRU, further enhancing the model's ability to capture temporal trends;

[0057] S3.6 Output layer: A fully connected layer with a linear activation function is used to output the predicted values ​​of key parameters in the next few seconds, with one predicted value for each time step.

[0058] Furthermore, the security assessment engine in step S4 constructs a four-dimensional health assessment system based on multi-source fusion data, quantifies and calculates the device health and safety index (HSI), and provides accurate basis for operation and maintenance decisions. The specific operation steps are as follows:

[0059] S4.1 Calculate the four-dimensional health sub-scores:

[0060] Mechanical structure health score S m Based on data from vibration sensors, torque sensors, wire rope displacement sensors, and lubricating oil particulate matter sensors, the weights of each indicator are determined using the analytic hierarchy process (AHP). The weighted summation is then used to reflect the health status of the mechanical structure, with values ​​ranging from [0, 1], where 1 represents complete health and 0 represents complete failure.

[0061] Electrical drive health score S e Based on data from the three-phase current, voltage, and temperature sensors of the motor, and combined with electrical fault diagnosis rules, the fuzzy comprehensive evaluation method is used to calculate and reflect the electrical health status of the motor and power supply circuit. The value range is [0, 1], where 1 represents complete health and 0 represents complete fault.

[0062] Operational performance health score p Based on the operating efficiency and response speed parameters of the gate hoist, combined with the comparison and calculation of historical best operating data, it reflects the operating efficiency of the equipment. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure.

[0063] Historical maintenance of health points S h Based on the equipment's historical maintenance records, the score is calculated using a decay function. The more timely the maintenance and the fewer the failures, the higher the score. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure.

[0064] S4.2 Determining the overall weight using the combined weighting method:

[0065] A combined weighting method, employing both subjective and objective weighting, was used to determine the overall weight W for each health sub-score. m W e W p W h The subjective weighting method is the Analytic Hierarchy Process (AHP): by constructing a judgment matrix and combining expert experience, the subjective importance weight of each sub-score is determined; the objective weighting method is the entropy weight method: based on the information entropy of multi-source data, the objective weight is calculated, and the greater the data dispersion, the higher the weight.

[0066] The formula for calculating the overall weight is:

[0067] ,

[0068] Where α is the weighting balance coefficient. Subjective weighting, For objective weights, satisfying W m +W e +W p +W h =1;

[0069] S4.3 Calculate the Health and Safety Index (HSI):

[0070] The health and safety index is calculated using the geometric weighted average method, and the formula is as follows:

[0071] ,

[0072] HSI is a scalar value between 0 and 1, where 0 represents complete failure and inoperability, and 1 represents absolute health and optimal performance. The specific grading criteria are as follows: HSI∈[0.9, 1.0] is "excellent", [0.7, 0.9) is "good", [0.5, 0.7) is "medium", [0.3, 0.5) is "poor", and [0, 0.3) is "failed".

[0073] Furthermore, in step S5, the dynamic threshold calculation module employs an adaptive algorithm, combining the real-time operating conditions and health status of the equipment to dynamically adjust the warning thresholds of each monitoring parameter, thereby improving the accuracy and sensitivity of the warnings. The specific formula is as follows:

[0074] ,

[0075] The meanings of each character are as follows:

[0076] T current The dynamic early warning threshold at the current moment;

[0077] T base The basic threshold is determined by the rated parameters of the gate hoist, industry standards, and historical normal operation data.

[0078] f(Load): Load correction function, calculated based on the real-time load Load measured by the drive shaft torque sensor. The calculation formula is as follows:

[0079] ,

[0080] Load max k1 is the rated maximum load of the equipment, and k1 is the load correction factor.

[0081] g(Env): Environmental correction function, calculated based on ambient temperature, humidity, and wind speed data. The calculation formula is as follows:

[0082] ,

[0083] Where T ref For standard ambient temperature, T max V represents the extreme ambient temperature threshold. max k2 and k3 are environmental correction factors, representing the maximum permissible wind speed.

[0084] β: Attenuation coefficient, used to adjust the degree of influence of health status on the threshold;

[0085] HSI t : The current health and safety index.

[0086] In this invention, the core advantage of the dynamic threshold calculation module algorithm lies in: when the device health status declines (HSI)... t When [1-β*(1-HSI) decreases], t The term decreases, causing T to... current Automatic tightening improves the system's sensitivity to faults; when the equipment is under heavy load, harsh environment or other special operating conditions, the threshold is corrected by f (Load) and g (Env) to avoid false alarms.

[0087] Compared with the prior art, the present invention has the following beneficial effects:

[0088] 1. The multi-source sensor array of the present invention can simultaneously collect mechanical, electrical, acoustic and environmental data, thereby greatly expanding the data collection range. In addition to traditional electrical or mechanical parameters, it can also monitor key indicators such as acoustic signals, lubricating oil status, structural torque, and ambient wind speed, thereby detecting potential abnormalities such as wire rope fatigue, early bearing wear, and abnormal gearbox meshing. It can realize early fault monitoring and complete the elimination and maintenance before the fault occurs or in its early stages, thereby avoiding larger faults and major losses.

[0089] 2. This invention adopts a three-step transmission mode of multi-source sensor array - edge computing gateway - cloud big data analysis platform. The edge computing gateway preprocesses the collected data and performs preliminary threshold comparison, thereby detecting some obvious and direct faults in advance. This not only improves the response and monitoring speed of some major faults, but also reduces the amount of data that needs to be transmitted to the server, saving transmission bandwidth and reducing the pressure on cloud computing.

[0090] 3. The cloud-based big data analysis platform of this invention has a built-in deep learning prediction model, a security assessment engine, and a dynamic threshold calculation module, which realizes in-depth data analysis, status prediction, health assessment, and intelligent threshold update. It uses information such as changes in shutdown load, environmental fluctuations, and equipment health status decay in actual working conditions to adjust the dynamic threshold, which greatly improves the accuracy of early warning and enhances the detection capability under complex working conditions.

[0091] 4. This invention establishes a multi-dimensional and quantitative health assessment system, which can realize accurate monitoring of the entire status of gate opening and closing mechanism, quantitative health assessment and intelligent early warning, improve monitoring sensitivity and scientific operation and maintenance, and ensure the safe and stable operation of equipment. Attached Figure Description

[0092] Figure 1This is a diagram illustrating the architecture of the online monitoring system for gate opening and closing mechanisms in an embodiment of the present invention.

[0093] Figure 2 This is a flowchart of the online monitoring method for gate hoisting mechanism in an embodiment of the present invention.

[0094] Figure 3 This is a flowchart of the temporal fusion algorithm for the deep learning prediction model in an embodiment of the present invention;

[0095] Figure 4 This is a flowchart of the security assessment engine calculation process in an embodiment of the present invention; Detailed Implementation

[0096] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0097] like Figure 1 As shown in the figure, an embodiment of the present invention proposes an online monitoring system for gate hoists, comprising:

[0098] A multi-source sensor array, which is set up in conjunction with the gate hoist body, is used to collect mechanical, electrical, acoustic and environmental data;

[0099] Edge computing gateways are used for sensor data preprocessing, feature extraction, and local lightweight anomaly detection;

[0100] The cloud-based big data analytics platform communicates with the edge computing gateway, receives feature data and some key raw data streams uploaded from the edge, and has built-in deep learning prediction models, security assessment engines and dynamic threshold calculation modules to achieve in-depth data analysis, status prediction, health assessment and intelligent threshold updates.

[0101] The remote monitoring terminal serves as the system's human-computer interaction interface, used for data visualization, early warning information display, and operation and maintenance decision support.

[0102] Specifically, the multi-source sensor array includes:

[0103] Mechanical condition sensors:

[0104] Vibration acceleration sensors are attached to key vibration source locations such as motor housing, gearbox, and drive shaft bearing housing to collect vibration acceleration signals and reflect structural loosening and component wear faults. The selected model is PCB352C65, with a measurement range of 0-50g, frequency response of 1-10000Hz, accuracy of ±0.01g, and output of 4-20mA analog signal.

[0105] Torque sensor for drive shaft (installed in the middle section of the drive shaft of the hoist to measure torque value and monitor the load distribution and fatigue state of the transmission system): HBMT40B type, range 0-1000N・m, accuracy class 0.1.

[0106] Absolute displacement sensor for wire rope (deployed on the side of the wire rope drum to accurately measure the expansion and contraction displacement of the wire rope and assess the risk of wire rope slack, wear and breakage): KEYENCEIL-700 laser displacement sensor is selected, with a measurement range of 0-1000mm and an accuracy of ±0.02mm.

[0107] Lubricating oil particulate matter sensor (installed in the gearbox lubricating oil circulation loop to detect the concentration and size of metal particles in the lubricating oil, reflecting the wear degree of gears and bearings): Parker SCM-100 model is selected, which can detect metal particles ≥4μm, with a measurement range of 0-10000 particles / mL.

[0108] Electrical status sensors:

[0109] Three-phase current and voltage sensor for motor (connected in series in the power supply circuit of the hoist motor to collect three-phase current Ia / Ib / Ic and voltage Ua / Ub / Uc signals, and monitor electrical faults such as motor overload, phase loss, and insulation aging): LEMCASR600-S type is selected, with current range 0-600A, voltage range 0-1000V, and accuracy class 0.2.

[0110] Bearing and winding temperature sensors (embedded in the outer ring of the motor bearing and inside the stator winding to monitor temperature changes in real time and prevent overheating damage): PT100 platinum resistance thermometers are used, with a measurement range of -50℃ to 200℃ and an accuracy of ±0.1℃.

[0111] Acoustic state sensor:

[0112] Acoustic emission sensor (attached to the surface of gearbox and motor housing to collect acoustic emission signals during equipment operation and capture early fault characteristics of crack propagation and friction impact in parts): PACR15 type, resonant frequency 150kHz, sensitivity ≥75dB, suitable for acoustic emission signal acquisition of mechanical faults.

[0113] Environmental condition sensors:

[0114] Ambient temperature and humidity sensor: SHT35+RS485 type temperature and humidity sensor is selected.

[0115] Wind speed sensor (deployed in the hoist room and surrounding area of ​​the gate to collect ambient temperature Tenv, humidity Henv, and wind speed Vwind data to provide basic data for operating condition correction and threshold adjustment): WindSensor05106 wind speed sensor.

[0116] All sensors support the industry-standard RS485 communication protocol, and their measurement accuracy meets the requirements for water conservancy project monitoring.

[0117] Edge computing gateways include:

[0118] Data preprocessing module: performs synchronous alignment, cleaning, and standardization on raw data collected from multiple sources of sensors;

[0119] Feature extraction module: Extracts time-domain and frequency-domain features from the preprocessed data. The time-domain features include peak value, mean, variance, kurtosis, and skewness, while the frequency-domain features include spectral peak value, dominant frequency, and harmonic components, providing feature support for subsequent anomaly detection and fault diagnosis.

[0120] Anomaly detection module: Based on feature data extracted from the edge side, a simple threshold method is used to make preliminary anomaly judgment. For parameters that significantly exceed the safe range, a local audible and visual alarm is immediately triggered to achieve rapid response to emergency faults.

[0121] Specifically, the edge computing gateway selected is the Huawei AR502H edge gateway, which supports RS485, CAN bus and 5G+LoRa dual-mode communication, has a local cache capacity of 32GB, and can run lightweight Python algorithm scripts to realize data preprocessing and edge anomaly detection.

[0122] Cloud-based big data analytics platforms include:

[0123] Deep learning prediction model: It adopts a temporal fusion Transformer architecture and achieves high-precision prediction of key parameters and high-sensitivity detection of early anomalies through the collaborative design of gating mechanism, multi-head attention and time series-specific modules.

[0124] Safety assessment engine: Based on multi-source fusion data, it constructs a four-dimensional health assessment system, quantitatively calculates the equipment health and safety index (HSI), and provides accurate basis for operation and maintenance decisions;

[0125] Dynamic threshold calculation module: It adopts an adaptive algorithm, combined with the real-time operating conditions and health status of the equipment, to dynamically adjust the warning thresholds of various monitoring parameters, thereby improving the accuracy and sensitivity of warnings.

[0126] Specifically, the cloud-based big data analytics platform uses Alibaba Cloud ECS g7 instances as cloud servers, runs on the CentOS 7 operating system, develops time-series fusion models based on the Python PyTorch framework, stores historical data using a MySQL database, develops a web-based monitoring platform using the Spring Boot + Vue framework, and develops a mobile app based on uni-app.

[0127] The remote monitoring terminal includes:

[0128] Data visualization module: including web platform, mobile APP, local monitoring screen in data center and other terminals, to display raw data, preprocessed data, feature data, HSI index and dynamic threshold in real time in the form of dashboard, trend curve and heat map;

[0129] Early warning information display module: Based on cloud analysis results, it generates tiered early warnings, and pushes early warning information through sound and light alarms, SMS push, APP pop-ups, email notifications, etc.

[0130] Operation and maintenance decision support module: Based on HSI index, fault type and historical maintenance data, it automatically generates targeted maintenance suggestions and supports operation and maintenance work order creation, dispatch, tracking and closed-loop management.

[0131] Specifically, the local monitoring screen in the computer room uses a 15-inch industrial touch screen with a resolution of 1920×1080, supporting real-time data display and local operation; the mobile APP supports Android 9.0 and iOS 13.0 and above, providing functions such as alert push, data query, and work order management; the web platform supports access from mainstream browsers such as Chrome and Firefox, providing a full-featured operation interface.

[0132] like Figure 2 As shown in the figure, this invention also proposes an online monitoring method for gate hoists, comprising the following steps:

[0133] (1) Data from a multi-source sensor array is collected synchronously and preprocessed through an edge computing gateway to achieve data cleaning, alignment, and standardization. The specific steps are as follows:

[0134] Cleaning (removing outliers caused by sensor malfunctions and electromagnetic interference, and using the 3σ criterion to remove data points that deviate from the mean by 3 times the standard deviation): removing outliers caused by sensor malfunctions and electromagnetic interference, and using the 3σ criterion to remove data points that deviate from the mean by a certain multiple of the standard deviation.

[0135] Alignment (multi-dimensional data time synchronization based on sensor timestamps): Synchronize and align raw data collected from multiple sources of sensors, and achieve multi-dimensional data time synchronization based on sensor timestamps;

[0136] Standardization (converting data of different dimensions into standardized data in the interval [0, 1]) is calculated using the following formula:

[0137] ,

[0138] Where, x std The data is standardized, and x is the original data. min x max These are the historical minimum and maximum values ​​of the parameter, respectively.

[0139] (2) Extract key time-domain and frequency-domain feature data through the edge computing gateway. The time-domain features include peak value, mean, variance, kurtosis and skewness, and the frequency-domain features include spectral peak value, main frequency and harmonic components, providing feature support for subsequent anomaly detection and fault diagnosis.

[0140] Then, local lightweight anomaly detection is performed: the extracted feature data is compared with a local temporary fixed threshold (determined by the device's rated parameters and industry standards) to make a preliminary anomaly judgment. If the feature data exceeds the temporary threshold, a local audible and visual alarm is triggered to achieve rapid emergency fault response, and the anomaly is uploaded to the cloud big data analysis platform at the same time; if it does not exceed the temporary threshold, the feature data and part of the original data stream are packaged and uploaded to the cloud big data analysis platform.

[0141] Meanwhile, the edge computing gateway has a local data caching function, which can temporarily store data when the network is interrupted and automatically retransmit it after the network is restored to ensure data integrity; it supports 5G+LoRa dual-mode communication, taking into account both long-distance high-speed transmission and stable communication needs in complex environments.

[0142] (3) After receiving the data uploaded from the edge, the cloud-based big data analysis platform uses a time-series fusion Transformer architecture with a deep learning prediction model. Through the collaborative design of gating mechanisms, multi-head attention, and time-series dedicated modules, such as... Figure 3 As shown, the deep learning prediction model is as follows:

[0143] (3.1) Model input: The input is a heterogeneous multivariate time series composed of data collected by a multi-source sensor array, specifically including two types of features: time series and static features. The time window length is set to 60s. The time series features are the set of dynamic features in the time domain and frequency domain extracted by the edge computing gateway, and the static features are non-time-varying parameters such as the inherent parameters of the gate hoist and the installation position characteristics.

[0144] After all input features are standardized, a three-dimensional input tensor X∈ is constructed by aligning them according to time steps. Where batch_size is the batch size, step_steps is the step size, and feature_dim is the total dimension of the features;

[0145] The model adopts a modular layered design, with input embedding and gating feature selection layers. Temporal features are injected with time information through position encoding, and static features are mapped to embedding vectors of the same dimension as the temporal features through fully connected layers.

[0146] (3.2) Introducing a gated selection unit: The importance weight w of each feature is calculated using the Sigmoid activation function. i ∈[0,1], dynamically filter effective features and suppress noise interference, the formula is:

[0147] ,

[0148] in, Let be the value of the i-th temporal feature at time t, s be the static feature embedding vector, W be the weight matrix, b be the bias term, and σ be the Sigmoid function;

[0149] (3.3) Multi-scale temporal convolutional layer: A multi-scale convolutional module is constructed using 1D convolutional kernels (kernel sizes of 3, 5, and 7 respectively) to capture local temporal features at different time granularities, output multi-scale feature maps and splice and fuse them;

[0150] (3.4) Temporal Multi-Head Attention Layer: By using multiple attention heads in parallel, it simultaneously captures the global long-term dependencies and local critical dependencies of the time series.

[0151] Each attention head computes the mapping of query Q, key K, and value V, and calculates the feature association weights using the scaled dot product attention formula:

[0152] ,

[0153] Where, d k The dimensions of Q and K are set to avoid gradient vanishing due to excessively high dimensionality;

[0154] (3.5) Gated Recurrent Unit (GRU) layer:

[0155] Receive the output of the attention layer, and use the reset gate and update gate of GRU to dynamically memorize key temporal information and forget redundant noise, further enhancing the model's ability to capture temporal trends. The number of neurons in the hidden layer is set to 256.

[0156] (3.6) Output layer: A fully connected layer with a linear activation function is used to output the predicted values ​​of key parameters in the next 30 seconds. Each time step corresponds to a predicted value, and the final output sequence length is 30.

[0157] (4) The cloud-based big data analysis platform, through a security assessment engine based on multi-source fusion data, constructs a four-dimensional health assessment system, quantifies and calculates the equipment health and safety index (HSI), and provides accurate basis for operation and maintenance decisions, such as... Figure 4 As shown, the specific operation steps are as follows:

[0158] (4.1) Calculate the four-dimensional health sub-scores:

[0159] Mechanical structure health score S mBased on data from vibration sensors, torque sensors, wire rope displacement sensors, and lubricating oil particulate matter sensors, the weights of each indicator are determined using the analytic hierarchy process (AHP). The weighted summation is then used to reflect the health status of the mechanical structure, with values ​​ranging from [0, 1], where 1 represents complete health and 0 represents complete failure.

[0160] Electrical drive health score S e Based on data from the three-phase current, voltage, and temperature sensors of the motor, and combined with electrical fault diagnosis rules (such as current imbalance ≤10%, winding temperature ≤120℃), the fuzzy comprehensive evaluation method is used to calculate and reflect the electrical health status of the motor and power supply circuit. The value range is [0, 1], where 1 represents complete health and 0 represents complete fault.

[0161] Operational performance health score p Based on the parameters of hoist operating efficiency (actual output power / rated input power) and response speed (actual gate lifting time / rated time), combined with historical best operating data comparison calculation, it reflects the equipment operating efficiency. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure.

[0162] Historical maintenance of health points S h Based on the equipment's historical maintenance records (maintenance frequency, fault repair quality, spare parts replacement cycle), the score is calculated using a decay function. The more timely the maintenance and the fewer the faults, the higher the score. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure.

[0163] (4.2) Determining the overall weight using the combined weighting method:

[0164] A combined weighting method, employing both subjective and objective weighting, was used to determine the overall weight W for each health sub-score. m W e W p W h The subjective weighting method is the Analytic Hierarchy Process (AHP): by constructing a judgment matrix and combining expert experience, the subjective importance weight of each sub-score is determined; the objective weighting method is the entropy weight method: based on the information entropy of multi-source data, the objective weight is calculated, and the greater the data dispersion, the higher the weight.

[0165] The formula for calculating the overall weight is:

[0166] ,

[0167] Where α is the weighting balance coefficient. Subjective weighting, For objective weights, satisfying W m +W e +W p +W h =1.

[0168] (4.3) Calculate the Health and Safety Index (HSI):

[0169] The health and safety index is calculated using the geometric weighted average method, and the formula is as follows:

[0170] ,

[0171] HSI is a scalar value between 0 and 1, where 0 represents complete failure and inoperability, and 1 represents absolute health and optimal performance. The specific grading criteria are as follows: HSI∈[0.9, 1.0] is “excellent”, [0.7, 0.9) is “good”, [0.5, 0.7) is “medium”, [0.3, 0.5) is “poor”, and [0, 0.3) is “failure”.

[0172] (5) The cloud-based big data analysis platform, based on real-time operating conditions and HSI, uses the adaptive dynamic threshold algorithm of the dynamic threshold calculation module, combined with the real-time operating conditions and health status of the equipment, to dynamically adjust the warning thresholds of each monitoring parameter, thereby improving the accuracy and sensitivity of the warnings. The specific formula is as follows:

[0173] ,

[0174] The meanings of each character are as follows:

[0175] T current The dynamic early warning threshold at the current moment;

[0176] T base The basic threshold is determined by the rated parameters of the gate hoist, industry standards (such as GB / T30038-2013 "General Technical Conditions for Gate Hoists in Water Conservancy Projects"), and historical normal operation data statistics.

[0177] f(Load): Load correction function, calculated based on the real-time load Load measured by the drive shaft torque sensor. The calculation formula is as follows:

[0178] ,

[0179] Load max The rated maximum load of the equipment is given by k1=0.4, which is the load correction factor.

[0180] g(Env): Environmental correction function, calculated based on ambient temperature, humidity, and wind speed data. The calculation formula is as follows:

[0181] ,

[0182] Where T ref =25℃ is the standard ambient temperature, T max =60℃ is the extreme ambient temperature threshold, Vmax =15m / s is the maximum allowable wind speed, and k2=0.3 and k3=0.2 are environmental correction factors;

[0183] β: Attenuation coefficient, with a value of 0.6, used to adjust the degree of influence of health status on the threshold;

[0184] HSI t : The current health and safety index.

[0185] (6) The cloud-based big data analysis platform combines early anomaly detection results, HSI index and dynamic threshold comparison results to generate graded early warnings and pushes the early warning information, HSI index, possible causes of failure and maintenance suggestions to the remote monitoring terminal. After receiving the early warning, the maintenance personnel can view detailed data and analysis reports through the terminal, formulate maintenance plans according to the maintenance suggestions (such as "HSI=0.45, the mechanical structure health score is low, it is recommended to check the wear status of the wire rope and the lubrication of the bearing"), create and dispatch maintenance work orders. After the maintenance is completed, the processing results are entered into the terminal, the system updates the equipment's historical maintenance records, and provides data support for subsequent health assessment and threshold adjustment.

[0186] The warning information is displayed as follows: Based on cloud analysis results, a tiered warning system is generated (three levels: "Attention," "Abnormal," and "Alarm"). Specifically: "Attention" indicates HSI ∈ [0.5, 0.7) and the residual does not exceed the limit; "Abnormal" indicates HSI ∈ [0.3, 0.5) or the residual exceeds the limit; and "Alarm" indicates HSI ∈ [0, 0.3) or the key parameter exceeds the dynamic threshold. See Table 1 for details.

[0187]

[0188] Table 1

[0189] Early warning information is pushed out through various means such as audible and visual alarms (local in the data center), SMS push, APP pop-ups, and email notifications to ensure that maintenance personnel receive it in a timely manner.

[0190] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An online monitoring system for gate hoists, characterized in that, include: A multi-source sensor array, which is set up in conjunction with the gate hoist body, is used to collect mechanical, electrical, acoustic and environmental data; Edge computing gateways are used for sensor data preprocessing, feature extraction, and local lightweight anomaly detection; The cloud-based big data analytics platform communicates with the edge computing gateway, receives feature data and some key raw data streams uploaded from the edge, and has built-in deep learning prediction models, security assessment engines and dynamic threshold calculation modules to achieve in-depth data analysis, status prediction, health assessment and intelligent threshold updates. The remote monitoring terminal serves as the system's human-computer interaction interface, used for data visualization, early warning information display, and operation and maintenance decision support.

2. The online monitoring system for gate hoists as described in claim 1, characterized in that, The multi-source sensor array includes: Mechanical condition sensors: vibration acceleration sensors, drive shaft torque sensors, wire rope absolute displacement sensors, lubricating oil particulate matter sensors; Electrical condition sensors: motor three-phase current and voltage sensors, bearing and winding temperature sensors; Acoustic state sensor: acoustic emission sensor; Environmental condition sensors: ambient temperature and humidity sensors, wind speed sensors.

3. The online monitoring system for gate hoists as described in claim 1, characterized in that, The edge computing gateway includes: Data preprocessing module: performs synchronous alignment, cleaning, and standardization on raw data collected from multiple sources of sensors; Feature extraction module: Extracts time-domain and frequency-domain features from the preprocessed data. The time-domain features include peak value, mean, variance, kurtosis, and skewness, while the frequency-domain features include spectral peak value, dominant frequency, and harmonic components, providing feature support for subsequent anomaly detection and fault diagnosis. Anomaly detection module: Based on feature data extracted from the edge side, a simple threshold method is used to make preliminary anomaly judgment. For parameters that significantly exceed the safe range, a local audible and visual alarm is immediately triggered to achieve rapid response to emergency faults.

4. The online monitoring system for gate hoists as described in claim 1, characterized in that, The cloud-based big data analytics platform includes: Deep learning prediction model: It adopts a temporal fusion Transformer architecture and achieves high-precision prediction of key parameters and high-sensitivity detection of early anomalies through the collaborative design of gating mechanism, multi-head attention and time series-specific modules. Safety assessment engine: Based on multi-source fusion data, it constructs a four-dimensional health assessment system, quantitatively calculates the equipment health and safety index (HSI), and provides accurate basis for operation and maintenance decisions; Dynamic threshold calculation module: It adopts an adaptive algorithm, combined with the real-time operating conditions and health status of the equipment, to dynamically adjust the warning thresholds of various monitoring parameters, thereby improving the accuracy and sensitivity of warnings.

5. The online monitoring system for gate hoists as described in claim 1, characterized in that, The remote monitoring terminal includes: Data visualization module: including web platform, mobile APP, local monitoring screen in data center and other terminals, to display raw data, preprocessed data, feature data, HSI index and dynamic threshold in real time in the form of dashboard, trend curve and heat map; Early warning information display module: Based on cloud analysis results, it generates tiered early warnings, and pushes early warning information through sound and light alarms, SMS push, APP pop-ups, email notifications, etc. Operation and maintenance decision support module: Based on HSI index, fault type and historical maintenance data, it automatically generates targeted maintenance suggestions and supports operation and maintenance work order creation, dispatch, tracking and closed-loop management.

6. A method for online monitoring of a gate hoist based on the system described in claim 1, characterized in that, Includes the following steps: S1. Simultaneously collect data from a multi-source sensor array and preprocess the data through an edge computing gateway to achieve data cleaning, alignment, and standardization; S2. Extract key time-domain and frequency-domain feature data through the edge computing gateway, and compare the extracted feature data with local temporary fixed thresholds; If the feature data exceeds the temporary threshold, a local audible and visual alarm is triggered, and the anomaly is uploaded to the cloud-based big data analysis platform; if it does not exceed the temporary threshold, the feature data and part of the original data stream are packaged and uploaded to the cloud-based big data analysis platform. S3, after receiving data uploaded from the edge side, the cloud-based big data analysis platform uses a time-series fusion architecture with a deep learning prediction model to predict key parameters and detect anomalies. S4, the cloud-based big data analysis platform uses a security assessment engine to call multi-source fusion data to calculate the current device's Health and Safety Index (HSI); S5, the cloud-based big data analysis platform calculates the dynamic thresholds of each parameter based on real-time operating conditions and HSI through the adaptive dynamic threshold algorithm of the dynamic threshold calculation module, thereby updating the early warning baseline; S6, the cloud-based big data analysis platform combines early anomaly detection results, HSI index and dynamic threshold comparison results to generate graded early warnings, and pushes the early warning information, HSI index, possible causes of failure and maintenance suggestions to the remote monitoring terminal; After receiving the alert, the maintenance personnel can view detailed data and analysis reports through the terminal, formulate maintenance plans based on maintenance suggestions, and create and dispatch maintenance work orders. After maintenance is completed, the processing results are entered into the terminal, and the system updates the device's historical maintenance records to provide data support for subsequent health assessments and threshold adjustments.

7. The online monitoring method for a gate hoist as described in claim 6, characterized in that, The data preprocessing in step S1 specifically includes: Cleaning: Remove outliers caused by sensor malfunctions and electromagnetic interference, and use the 3σ criterion to remove data points that deviate from the mean by a certain multiple of the standard deviation; Alignment: Synchronize and align the raw data collected by multiple sensors, and achieve multi-dimensional data time synchronization based on sensor timestamps; Standardization: Converting data of different dimensions into standardized data within the interval [0, 1]. The formula is: ， Where, x std The data is standardized, and x is the original data. min x max These are the historical minimum and maximum values ​​of the parameter, respectively.

8. The online monitoring method for a gate hoist as described in claim 6, characterized in that, In step S3, the deep learning prediction model adopts a temporal fusion Transformer architecture, which, through the collaborative design of gating mechanisms, multi-head attention, and time series-specific modules, specifically includes: S3.1 Model Input: The input is a heterogeneous multivariate time series composed of data collected by a multi-source sensor array, specifically including two types of features: time series and static features. The time series features are the set of dynamic features in the time domain and frequency domain extracted by the edge computing gateway, and the static features are non-time-varying parameters such as the inherent parameters of the gate hoist and the characteristics of the installation position. After all input features are standardized, a three-dimensional input tensor X∈ is constructed by aligning them according to time steps. Where batch_size is the batch size, step_steps is the step size, and feature_dim is the total dimension of the features; S3.2 Introducing a gated selection unit: The importance weights w of each feature are calculated using the Sigmoid activation function. i ∈[0,1], dynamically filter effective features and suppress noise interference, the formula is: ， in, Let be the value of the i-th temporal feature at time t, s be the static feature embedding vector, W be the weight matrix, b be the bias term, and σ be the Sigmoid function; S3.3 Multi-scale temporal convolutional layer: A multi-scale convolutional module is constructed using 1D convolutional kernels to capture local temporal features at different time granularities, output multi-scale feature maps and stitch them together; S3.4, Temporal Multi-Head Attention Layer: By using multiple attention heads in parallel, it simultaneously captures both global long-term dependencies and local critical dependencies in time series data. Each attention head computes the mapping of query Q, key K, and value V, and calculates the feature association weights using the scaled dot product attention formula: ， Where, d k The dimensions of Q and K are set to avoid gradient vanishing due to excessively high dimensionality; S3.5, Gated Recurrent Unit (GRU) Layer: Received from the attention layer output, the model dynamically memorizes key temporal information and forgets redundant noise through the reset and update gates of GRU, further enhancing the model's ability to capture temporal trends; S3.6 Output layer: A fully connected layer with a linear activation function is used to output the predicted values ​​of key parameters in the next few seconds, with one predicted value for each time step.

9. The online monitoring method for a gate hoist as described in claim 6, characterized in that, The security assessment engine in step S4 is based on multi-source fusion data to construct a four-dimensional health assessment system, quantitatively calculate the device health and safety index (HSI), and provide accurate basis for operation and maintenance decisions. The specific operation steps are as follows: S4.1 Calculate the four-dimensional health sub-scores: Mechanical structure health score S m Based on data from vibration sensors, torque sensors, wire rope displacement sensors, and lubricating oil particulate matter sensors, the weights of each index are determined using the analytic hierarchy process (AHP). The weighted summation is then used to reflect the health status of the mechanical structure, with values ​​ranging from [0, 1], where 1 represents complete health and 0 represents complete failure. Electrical drive health score S e Based on data from the three-phase current, voltage, and temperature sensors of the motor, and combined with electrical fault diagnosis rules, the fuzzy comprehensive evaluation method is used to calculate and reflect the electrical health status of the motor and power supply circuit. The value range is [0, 1], where 1 represents complete health and 0 represents complete fault. Operational performance health score p Based on the operating efficiency and response speed parameters of the gate hoist, combined with the comparison and calculation of historical best operating data, it reflects the operating efficiency of the equipment. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure. Historical maintenance of health points S h Based on the equipment's historical maintenance records, the score is calculated using a decay function. The more timely the maintenance and the fewer the failures, the higher the score. The value range is [0, 1], where 1 represents complete health and 0 represents complete failure. S4.2 Determining the overall weight using the combined weighting method: A combined weighting method, employing both subjective and objective weighting, was used to determine the overall weight W for each health sub-score. m W e W p W h The subjective weighting method is the Analytic Hierarchy Process (AHP): by constructing a judgment matrix and combining expert experience, the subjective importance weights of each sub-score are determined. The objective weighting method is the entropy weighting method: it calculates objective weights based on the information entropy of multi-source data, and the greater the data dispersion, the higher the weight. The formula for calculating the overall weight is: ， Where α is the weighting balance coefficient. Subjective weighting, For objective weights, satisfying W m +W e +W p +W h =1; S4.3 Calculate the Health and Safety Index (HSI): The health and safety index is calculated using the geometric weighted average method, and the formula is as follows: ， HSI is a scalar value between 0 and 1, where 0 represents complete failure and inoperability, and 1 represents absolute health and optimal performance. The specific grading criteria are as follows: HSI∈[0.9, 1.0] is "excellent", [0.7, 0.9) is "good", [0.5, 0.7) is "medium", [0.3, 0.5) is "poor", and [0, 0.3) is "failed".

10. The online monitoring method for a gate hoist as described in claim 6, characterized in that, In step S5, the dynamic threshold calculation module employs an adaptive algorithm, combining the real-time operating conditions and health status of the equipment to dynamically adjust the warning thresholds of various monitoring parameters, thereby improving the accuracy and sensitivity of the warnings. The specific formula is as follows: ， The meanings of each character are as follows: T current The dynamic early warning threshold at the current moment; T base The basic threshold is determined by the rated parameters of the gate hoist, industry standards, and historical normal operation data. f(Load): Load correction function, calculated based on the real-time load Load measured by the drive shaft torque sensor. The calculation formula is as follows: ， Load max k1 is the rated maximum load of the equipment, and k1 is the load correction factor. g(Env): Environmental correction function, calculated based on ambient temperature, humidity, and wind speed data. The calculation formula is as follows: ， Where T ref For standard ambient temperature, T max V represents the extreme ambient temperature threshold. max k2 and k3 are environmental correction factors, representing the maximum permissible wind speed. β: Attenuation coefficient, used to adjust the degree of influence of health status on the threshold; HSI t : The current health and safety index.

Images