Stereo garage predictive operation and maintenance method, device and equipment and storage medium

Abstract

The present application relates to the technical field of information processing, and particularly relates to a stereo garage predictive operation and maintenance method, device, equipment and storage medium, a health state baseline library is established by collecting multi-dimensional health state data of each target device of the stereo garage, environment data and current state data of each target device of the stereo garage are collected in real time by using a movable robot, the current state data is compared with the health state baseline library, the current state data and the environment data are input into a preset prediction model, so that the prediction model outputs a fault prediction result of each target device in a future preset time period, a preventive maintenance task is generated according to state deviation degree information and the fault prediction result, and the movable robot performs a maintenance or maintenance action according to the fault maintenance task or the preventive maintenance task, so that early prediction, automatic inspection and rapid disposal of stereo garage device faults are realized.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a predictive operation and maintenance method, apparatus, equipment, and storage medium for automated parking systems. Background Technology

[0002] As an important facility for alleviating urban parking difficulties, the operational stability and reliability of automated parking systems directly affect the efficiency, experience, and safety of users parking and retrieving their vehicles. Automated parking systems typically consist of barrier gates, license plate recognition cameras, billing terminals, surveillance cameras, lifting motors, traversing trolleys, vehicle carriers, parking space racks, a control system (PLC), and sensors (position switches, rangefinders). During long-term, high-frequency operation, these devices are prone to various malfunctions due to mechanical wear, electrical aging, environmental factors, and external interference.

[0003] Currently, the operation and maintenance of automated parking systems generally adopts a reactive management model of repairing after a failure. Specifically, when equipment malfunctions (such as failed vehicle retrieval, stuck vehicle platform, abnormal noise from the lift, or sensor failure), maintenance personnel are usually notified to go to the site for troubleshooting and handling after the user reports the abnormality or the management personnel discover it. After arriving at the site, the maintenance personnel manually check common fault points such as mechanical interference, limit switches, power connections, and sensor status to locate the problem and then perform operations such as restarting, mechanical reset, sensor cleaning, and lens cleaning. Statistics show that a considerable number of automated parking system equipment malfunctions are easily handled incidents, such as system crashes requiring restarts, photoelectric sensors being obscured by dust, slight limit switch misalignment requiring readjustment, motor overload protection triggering requiring resets, license plate recognition lenses being obscured by dust or foreign objects requiring cleaning, and traverse trolleys being stuck due to small foreign objects requiring removal.

[0004] However, this traditional operation and maintenance method has the following technical defects: (1) large response delay, maintenance personnel need a certain amount of time from receiving the notification to arriving at the site, resulting in congestion at the entrance and exit of the multi-level parking garage and a decline in user experience; (2) high labor cost, in order to ensure timely response of multiple dispersed multi-level parking garages, a large number of on-site or rotating maintenance personnel need to be configured; (3) lack of fault prediction capability, unable to intervene in maintenance in advance before the equipment is completely failed, often only passively dealing with the fault after it occurs, seriously affecting the normal operation of the multi-level parking garage, and may even cause safety accidents; (4) the fault types handled by maintenance personnel are highly repetitive and have low technical content, resulting in a waste of human resources. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a predictive operation and maintenance method, device, equipment and storage medium for automated parking garages that can realize early prediction of equipment failures, automated inspection and rapid handling.

[0006] The first aspect of this invention provides a predictive operation and maintenance method for an automated parking garage, comprising: collecting multi-dimensional health status data of each target device in the automated parking garage; establishing a health status baseline database based on the multi-dimensional health status data; scheduling at least one mobile robot to perform inspections within the automated parking garage, and using the mobile robot to collect environmental data of the automated parking garage and current status data of each target device in real time, comparing the current status data with the health status baseline database to obtain status deviation information; inputting the current status data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each target device within a preset future time period, generating preventive maintenance tasks based on the status deviation information and the fault prediction results; issuing fault repair tasks or preventive maintenance tasks to the mobile robot, so that the mobile robot performs repair or maintenance actions; generating operation and maintenance effect data, adding the operation and maintenance effect data as feedback samples to a training dataset, and periodically optimizing and updating the prediction model using the training dataset.

[0007] Optionally, in a first implementation of the first aspect of the present invention, the step of collecting multi-dimensional health status data of each target device in the automated parking garage and establishing a health status baseline library based on the multi-dimensional health status data includes: collecting vibration data, temperature data, current data, voltage data, and noise data of each target device in the automated parking garage during a fault-free operation cycle as multi-dimensional health status data; performing data cleaning, time alignment, and missing value imputation on the multi-dimensional health status data to obtain standardized time-series data; extracting time-domain features, frequency-domain features, and statistical features from the standardized time-series data and forming a health feature vector; clustering the health feature vector and establishing a health status baseline library for each target device, wherein the health status baseline library includes the normal value range, mean, and standard deviation of each feature dimension.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the step of scheduling at least one mobile robot to perform inspections within the automated parking garage, and using the mobile robot to collect environmental data of the automated parking garage and current status data of each target device in real time, comparing the current status data with the health status baseline database to obtain status deviation information, includes: planning the inspection path and inspection frequency of at least one mobile robot based on the spatial layout of the automated parking garage and the location distribution of the target devices; scheduling the mobile robot to perform inspections within the automated parking garage according to the inspection path and the inspection frequency; and using the cameras and sensors mounted on the mobile robot to collect environmental data and image data of the automated parking garage in real time. Based on the current status data of each target device, the environmental data includes ambient temperature and humidity information, and the current status data includes device vibration, device temperature, device current, device voltage, and device noise information; image processing technology is used to analyze the image data to obtain analysis results; based on the analysis results, it is determined whether there is an easily manageable accident problem at the location of the three-dimensional parking garage corresponding to the image; if so, an accident handling task is generated and sent to the mobile robot so that the mobile robot can immediately execute the handling action based on the accident handling task; the current status data is compared with the health status baseline database to obtain the degree of status deviation information.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the step of inputting the current state data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each of the target devices within a preset future time period, and generating preventive maintenance tasks based on the state deviation information and the fault prediction results, includes: inputting the current state data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each of the target devices within a preset future time period; analyzing the fault characteristics of the fault prediction results, and determining the maintenance type based on the fault characteristics; and matching the maintenance type with the corresponding preventive maintenance tasks in a preset maintenance knowledge base.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the step of issuing the fault repair task or the preventive maintenance task to the mobile robot so that the mobile robot can perform repair or maintenance actions includes: acquiring the operational status data of the automated parking garage and determining a low-interference period based on the operational status data; issuing the preventive maintenance task to the mobile robot so that the mobile robot can perform maintenance actions based on the preventive maintenance task during the low-interference period; when the status deviation information is greater than a preset deviation threshold, determining it as a device fault and generating a fault repair task; issuing the fault repair task to the mobile robot so that the mobile robot can immediately perform repair actions based on the fault repair task; after performing the repair or maintenance actions, using the mobile robot to re-collect the current status data to obtain new status data; re-inputting the new status data into the prediction model so that the prediction model outputs a new fault prediction result; when the fault probability in the new fault prediction result is greater than a preset fault threshold, generating a deep inspection task and pushing the deep inspection task to the management terminal.

[0011] Optionally, in the fifth implementation of the first aspect of the present invention, the step of generating operation and maintenance effect data, adding the operation and maintenance effect data as a feedback sample to the training dataset, and periodically optimizing and updating the prediction model using the training dataset includes: after each preventive maintenance or fault repair, recording the change in the failure probability before and after the corresponding equipment maintenance to generate operation and maintenance effect data; adding the operation and maintenance effect data as a feedback sample to the training dataset; and periodically using the training dataset to incrementally train the prediction model, wherein the prediction model includes a fusion network based on an attention mechanism and a temporal prediction network, and the temporal prediction network is a long short-term memory network or a Transformer network.

[0012] Optionally, in the sixth implementation of the first aspect of the present invention, after generating the operation and maintenance effect data, adding the operation and maintenance effect data as a feedback sample to the training dataset, and periodically optimizing and updating the prediction model using the training dataset, the method further includes: re-predicting the failure probability of all the target devices in the automated parking garage using the updated prediction model; and dynamically adjusting the inspection path and inspection frequency of the mobile robot according to the failure probability.

[0013] A second aspect of the present invention provides a predictive operation and maintenance device for an automated parking garage, comprising: a data acquisition and establishment module, used to acquire multi-dimensional health status data of each target device in the automated parking garage, and establish a health status baseline database based on the multi-dimensional health status data; a scheduling, acquisition, and comparison module, used to schedule at least one mobile robot to perform inspections within the automated parking garage, and to use the mobile robot to acquire environmental data of the automated parking garage and current status data of each target device in real time, and to compare the current status data with the health status baseline database to obtain status deviation information; an input generation module, used to input the current status data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each target device within a preset future time period, and generates preventive maintenance tasks based on the status deviation information and the fault prediction results; a distribution module, used to distribute fault repair tasks or preventive maintenance tasks to the mobile robot, so that the mobile robot performs repair or maintenance actions; and a generation, addition, and update module, used to generate operation and maintenance effect data, add the operation and maintenance effect data as feedback samples to a training dataset, and periodically optimize and update the prediction model using the training dataset.

[0014] Optionally, in the first implementation of the second aspect of the present invention, the acquisition and establishment module includes: a first acquisition unit, used to acquire vibration data, temperature data, current data, voltage data, and noise data of each target device in the three-dimensional parking garage during a fault-free operation cycle as multi-dimensional health status data; a cleaning, alignment, and imputation unit, used to perform data cleaning, time alignment, and missing value imputation on the multi-dimensional health status data to obtain standardized time-series data; an extraction and composition unit, used to extract time-domain features, frequency-domain features, and statistical features from the standardized time-series data and form a health feature vector; and a clustering and establishment unit, used to cluster the health feature vector and establish a health status baseline library for each target device, wherein the health status baseline library includes the normal value range, mean, and standard deviation of each feature dimension.

[0015] Optionally, in a second implementation of the second aspect of the present invention, the scheduling, acquisition, and comparison module includes: a planning unit, configured to plan the inspection path and inspection frequency of at least one mobile robot based on the spatial layout of the automated parking garage and the location distribution of the target equipment; a scheduling unit, configured to schedule the mobile robot to perform inspections within the automated parking garage according to the inspection path and the inspection frequency; and a second acquisition unit, configured to use the camera and sensors mounted on the mobile robot to collect environmental data, image data, and current status data of each target equipment in the automated parking garage in real time, wherein the environmental data includes environmental temperature information and environmental humidity information, and the current status data includes... The system includes equipment vibration information, equipment temperature information, equipment current information, equipment voltage information, and equipment noise information; an analysis unit for analyzing the image data using image processing technology to obtain analysis results; a determination unit for determining whether there is an easily manageable accident problem at the location of the three-dimensional parking garage corresponding to the image based on the analysis results; a generation and distribution unit for generating an accident handling task if so, and distributing the accident handling task to the mobile robot so that the mobile robot can immediately perform a handling action based on the accident handling task; and a comparison unit for comparing the current state data with the health state baseline database to obtain information on the degree of state deviation.

[0016] Optionally, in a third implementation of the second aspect of the present invention, the input generation module includes: a first input unit, configured to input the current state data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each of the target devices within a preset future time period; an analysis and determination unit, configured to analyze the fault characteristics of the fault prediction results and determine the maintenance type based on the fault characteristics; and a matching unit, configured to match the maintenance type with the corresponding preventive maintenance tasks in a preset maintenance knowledge base.

[0017] Optionally, in a fourth implementation of the second aspect of the present invention, the distribution module includes: an acquisition and determination unit, configured to acquire operational status data of the automated parking garage and determine low-interference periods based on the operational status data; a first distribution unit, configured to distribute the preventive maintenance task to the mobile robot, so that the mobile robot performs maintenance actions based on the preventive maintenance task during the low-interference period; a determination and generation unit, configured to determine a device malfunction when the status deviation information is greater than a preset deviation threshold, and generate a fault repair task; a second distribution unit, configured to distribute the fault repair task to the mobile robot, so that the mobile robot immediately performs maintenance actions based on the fault repair task; a third acquisition unit, configured to re-acquire current status data using the mobile robot after performing maintenance or repair actions to obtain new status data; a second input unit, configured to re-input the new status data into the prediction model, so that the prediction model outputs a new fault prediction result; and a generation and push unit, configured to generate a deep inspection task and push the deep inspection task to the management terminal when the fault probability in the new fault prediction result is greater than a preset fault threshold.

[0018] Optionally, in the fifth implementation of the second aspect of the present invention, the generation, addition, and update module includes: a record generation unit, used to record the change in the failure probability of the corresponding equipment before and after each preventive maintenance or fault repair, and generate operation and maintenance effect data; an addition unit, used to add the operation and maintenance effect data as a feedback sample to the training dataset; and a training unit, used to periodically use the training dataset to incrementally train the prediction model, wherein the prediction model includes a fusion network based on an attention mechanism and a temporal prediction network, and the temporal prediction network is a long short-term memory network or a Transformer network.

[0019] Optionally, in a sixth implementation of the second aspect of the present invention, the method further includes: a prediction module, configured to re-predict the failure probability of all target devices in the automated parking garage using the updated prediction model; and an adjustment module, configured to dynamically adjust the inspection path and inspection frequency of the mobile robot according to the failure probability.

[0020] A third aspect of the present invention provides a predictive maintenance device for automated parking garages, the device comprising: a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the device to perform the steps of the predictive maintenance method for automated parking garages described in any of the preceding claims.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the predictive operation and maintenance method for the automated parking garage described in any of the preceding claims.

[0022] In the technical solution of this invention, a health status baseline database is established by collecting multi-dimensional health status data of each target device in the automated parking garage. A mobile robot is used to collect environmental data of the automated parking garage and the current status data of each target device in real time. The current status data is compared with the health status baseline database. The current status data and environmental data are input into a preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset time period. Based on the status deviation information and the fault prediction results, preventive maintenance tasks are generated. The mobile robot performs maintenance or repair actions according to the fault repair task or preventive maintenance task, realizing early prediction of equipment faults in the automated parking garage, automated inspection and rapid handling. The operation and maintenance effect data is added to the training dataset as feedback samples. The prediction model is periodically optimized and updated using the training dataset to improve the accuracy of fault prediction. Attached Figure Description

[0023] Figure 1 This is a first flowchart of the predictive operation and maintenance method for automated parking garages provided in an embodiment of the present invention; Figure 2 This is a second flowchart of the predictive operation and maintenance method for automated parking garages provided in an embodiment of the present invention; Figure 3 This is a third flowchart of the predictive operation and maintenance method for automated parking garages provided in this embodiment of the invention; Figure 4 This is a fourth flowchart of the predictive operation and maintenance method for automated parking garages provided in this embodiment of the invention; Figure 5 A schematic diagram of a predictive operation and maintenance device for an automated parking garage provided in an embodiment of the present invention; Figure 6 This is another structural schematic diagram of the predictive operation and maintenance device for automated parking garages provided in an embodiment of the present invention; Figure 7 A schematic diagram of the structure of the predictive maintenance equipment for a three-dimensional parking garage provided in an embodiment of the present invention. Detailed Implementation

[0024] This invention provides a predictive operation and maintenance method, device, equipment, and storage medium for automated parking garages, which can realize early prediction of equipment failures, automated inspection, and rapid handling of automated parking garage issues.

[0025] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the predictive operation and maintenance method for automated parking garages in this invention includes: 101. Collect multi-dimensional health status data of each target device in the automated parking garage, and establish a health status baseline database based on the multi-dimensional health status data; In this embodiment, the system first identifies key target equipment within the automated parking garage, such as barrier gates, license plate recognition cameras, billing terminals, surveillance cameras, lifting motors, traverse trolleys, vehicle carriers, parking space racks, PLC controllers, and position sensors. Through IoT sensors deployed on the equipment (such as vibration sensors, temperature sensors, current / voltage transformers, and sound sensors) or via the equipment's own API interface, the system continuously collects multi-dimensional health status data, including vibration data, temperature data, current data, voltage data, and noise data, during the equipment's fault-free operation cycle (e.g., the first 30 days after a new device is put into operation). The system performs data cleaning on the collected raw data, removing noise data caused by network fluctuations or sensor anomalies, and fills in missing values ​​using an interpolation algorithm. In addition, data from different sampling frequencies are time-aligned to form a standardized time-series dataset. Then, time-domain features (such as peak and mean values ​​of vibration amplitude), frequency-domain features (such as the dominant frequency of the spectrum obtained through fast Fourier transform), and statistical features (such as the standard deviation of temperature data and the skewness of current data) are extracted from the standardized data. These features are combined into a high-dimensional health feature vector. Finally, density-based clustering algorithms (such as DBSCAN) are used to perform cluster analysis on the health feature vector. After removing outliers, a dedicated health status baseline library is established for each target device. This baseline library specifically includes the normal value range of each feature dimension in the health state (such as the upper and lower quartiles of vibration frequency), mean (μ), and standard deviation (σ), which are used for subsequent real-time comparison.

[0027] 102. Schedule at least one mobile robot to perform inspections in the automated parking garage, and use the mobile robot to collect environmental data and current status data of each target device in real time. Compare the current status data with the health status baseline database to obtain information on the degree of status deviation. In this embodiment, based on a high-precision map of the automated parking garage, the system uses path planning algorithms (such as A* or Dijkstra's algorithm) combined with the physical locations of all target devices to plan the optimal inspection path and frequency for each mobile robot (e.g., once every hour during peak hours and once every two hours during off-peak hours). The system schedules the mobile robots via a wireless network to automatically execute inspection tasks according to the planned paths. During inspection, the mobile robots utilize their onboard high-definition cameras, infrared thermal imagers, vibration sensors, microphone arrays, and other equipment to collect the current status data of each target device at close range. The data collection includes: acquiring the surface temperature of the equipment through infrared thermal imaging; acquiring the vibration waveform of the equipment through contact or non-contact vibration sensors; collecting the operating noise of the equipment through a microphone array; and simultaneously, the robot also collects surrounding environmental data (such as ambient temperature and humidity). After the data collection is completed, the system compares the various feature values ​​of the current state data (such as current temperature, vibration frequency, and noise decibel value) with the normal range, mean, and standard deviation of the corresponding features in the health state baseline library. The deviation of the current state from the health baseline is quantified by calculating Mahalanobis distance or Z-score, and finally, the state deviation information is generated.

[0028] 103. Input the current status data and environmental data into the preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset time period, and generates preventive maintenance tasks based on the status deviation information and fault prediction results. In this embodiment, the system uses real-time data collected by the mobile robot, such as current vibration, temperature, current, voltage, and noise, and environmental data (ambient temperature and humidity), to form a multidimensional input tensor. This tensor is then input into a pre-trained prediction model. The core of this model is a time-series prediction network (such as Transformer or LSTM) incorporating an attention mechanism. Based on the time-series characteristics of the input data, the model outputs the failure probability, estimated remaining lifespan, and possible failure types (such as gate motor overheating, power module voltage instability, etc.) for each target device within the next 24 or 48 hours. The system then comprehensively analyzes the degree of state deviation and fault prediction results. For example, when the state deviation of a certain device is small but the predicted probability of future failure is high, or when the state deviation is large, the system will generate specific preventive maintenance tasks. The tasks include: analyzing the fault characteristics of the fault prediction results and determining the maintenance type based on the fault characteristics (e.g., abnormal vibration spectrum corresponds to mechanical lubrication, and continuous temperature rise corresponds to heat dissipation cleaning or motor bearing inspection); matching the maintenance type with the corresponding preventive maintenance tasks in the preset maintenance knowledge base (e.g., matching the specific operating steps and required tools for adding lubricating oil to the gate motor bearing or lifting motor bearing).

[0029] 104. Assign fault repair or preventive maintenance tasks to mobile robots so that they can perform repair or maintenance actions; In this embodiment, the system first acquires the operational status data of the automated parking garage (such as the flow of vehicles entering and exiting), and dynamically determines the optimal low-interference time period for performing maintenance tasks based on this data (such as 2:00-4:00 AM). For preventative maintenance tasks, the system queues them and issues them to the mobile robot during the low-interference time period so that the mobile robot can perform maintenance actions. Simultaneously, when the status deviation exceeds a preset emergency deviation threshold, the system determines that the equipment has experienced a fault affecting operation and immediately generates a fault repair task (such as restarting the gate controller or resetting the power switch). This task is preferentially issued to the nearest mobile robot. The system instructs the robot to immediately proceed to the site to perform maintenance actions, such as physically pressing the restart button or sending a reset command via an infrared transmitter. After the maintenance action is completed, the system commands the mobile robot to immediately re-collect the current status data of the same device and re-input it into the prediction model to obtain new fault prediction results for effect verification. If the fault probability in the new fault prediction results is still greater than the preset fault threshold, it indicates that simple automated maintenance has failed to solve the fundamental problem. In this case, the system will generate a deep inspection task that requires manual intervention and push this task, along with the device's historical data, prediction report, and records of executed actions, to the administrator interface on the management terminal.

[0030] 105. Generate operation and maintenance effect data, add the operation and maintenance effect data as feedback samples to the training dataset, and use the training dataset to regularly optimize and update the prediction model; In this embodiment, after each preventive maintenance or fault repair task is completed, the system automatically records the specific change in the failure probability output by the prediction model before and after the maintenance action (for example, the failure probability of a certain device is 85% before maintenance and decreases to 20% after maintenance), thereby generating quantitative operation and maintenance effect data. The system adds the status data fragment, the executed maintenance action, and the effect data as a high-quality feedback sample to the training dataset of the prediction model. The system sets a regular update cycle (for example, every Sunday at 2:00 AM), at which time the expanded training dataset is used to incrementally train the prediction model. During the incremental training process, the model will focus on learning those effective maintenance patterns that successfully reduce the failure probability, thereby continuously adjusting its own weights, improving the accuracy of future failure prediction, and generating more effective preventive maintenance tasks.

[0031] In this embodiment of the invention, a health status baseline database is established by collecting multi-dimensional health status data of each target device in the automated parking garage. A mobile robot collects environmental data and current device status data in real time, compares the current status data with the health status baseline database, and inputs the current status data and environmental data into a preset prediction model to output fault prediction results for a preset future time period. Preventive maintenance tasks are generated based on the degree of status deviation and the fault prediction results. These tasks are then distributed to the mobile robot for maintenance or repair. Simultaneously, operation and maintenance effect data is generated and added to the training dataset as feedback samples. The prediction model is periodically optimized and updated. This achieves early prediction, automated inspection, and rapid handling of equipment faults in the automated parking garage. Furthermore, a self-optimization mechanism continuously improves prediction accuracy, reduces equipment downtime and manual intervention costs, and enhances the intelligence level and operational efficiency of the automated parking garage's operation and maintenance.

[0032] Please see Figure 2 In the second embodiment of the predictive operation and maintenance method for automated parking garages in this invention, steps 101 and 102 include: 201. Collect vibration data, temperature data, current data, voltage data, and noise data of each target device in the automated parking garage during a fault-free operation cycle as multi-dimensional health status data; In this embodiment, the system continuously collects vibration data (via patch-type vibration sensor), temperature data (via contact-type temperature sensor or the device's own IC temperature), current and voltage data (via closed-loop Hall current / voltage sensor connected in series / parallel to the device's power line), and noise data (via MEMS microphone deployed near the device) of each target device (such as a barrier gate, camera, or lifting motor) for 30 days at a sampling frequency of 1Hz through a pre-deployed IoT gateway.

[0033] 202. Perform data cleaning, time alignment, and missing value imputation on multi-dimensional health status data to obtain standardized time-series data; In this embodiment, the system uses a median filtering algorithm to remove impulse noise from vibration and noise data. Since the sampling frequencies of different sensors may be different (e.g., vibration is 512Hz and temperature is 1Hz), the system uses a linear interpolation method to resample all numerical data to a uniform frequency (e.g., 1Hz) and performs strict time alignment based on high-precision timestamps provided by GPS or NTP. For continuous data gaps caused by network interruptions, the system uses spline interpolation based on historical data from the same period to fill in the gaps.

[0034] 203. Extract time-domain features, frequency-domain features, and statistical features from standardized time-series data, and construct a health feature vector; In this embodiment, the system uses a sliding window (window size 10 minutes, step size 5 minutes) to perform feature engineering on standardized time-series data. The extracted time-domain features include: root mean square value and peak value of vibration waveform; frequency-domain features include: main frequency energy ratio and 1 / 3 octave spectrum calculated by fast Fourier transform; statistical features include: skewness and kurtosis of temperature data, and coefficient of variation of current data. Finally, all numerical features and visual features are fused within the window to form a high-dimensional health feature vector.

[0035] 204. Cluster the health feature vectors and establish a health status baseline library for each target device. The health status baseline library includes the normal range, mean and standard deviation of each feature dimension. In this embodiment, the system uses the health feature vectors of all windows as a sample set, performs unsupervised clustering using a Gaussian mixture model (GMM), automatically identifies sample clusters representing health status, removes outliers outside such clusters, calculates the mean and standard deviation of each feature dimension (including visual feature dimensions, such as the proportion of unobstructed area and edge integrity score) within the health cluster, and defines the normal value range accordingly (e.g., mean ± 3 times standard deviation), ultimately forming a structured health status baseline library for the device.

[0036] 205. Based on the spatial layout of the automated parking garage and the location distribution of the target equipment, plan the inspection path and inspection frequency of at least one mobile robot; In this embodiment, the system loads a high-precision map of the automated parking garage (containing semantic information such as lane lines, parking space levels, lateral movement channels, parking spaces, equipment installation points, and control cabinet locations), marks the locations of all target equipment as points of interest, and uses the Multiple Traveling Salesman Problem (MTSP) algorithm, combined with the number of mobile robots (e.g., 2), to calculate the globally optimal inspection path with the goal of minimizing the total inspection path and balancing tasks. At the same time, the system assigns differentiated inspection frequencies to equipment based on its importance (e.g., entrance / exit gates are more important than parking space indicator lights) and historical failure rate. For example, core equipment is inspected every 30 minutes, and ordinary equipment is inspected every 2 hours. For equipment that is prone to obstruction or structural damage (e.g., parking space sensors, limit switches, entrance / exit gates), the system can dynamically increase its inspection frequency and image acquisition accuracy requirements (e.g., by increasing the shooting angle and resolution).

[0037] 206. Dispatch mobile robots to perform inspections within the automated parking garage based on the inspection route and frequency; In this embodiment, the system's task scheduler generates a timestamped inspection task queue based on the planned path and frequency. The scheduler sends the tasks to the designated mobile robots via the MQTT protocol. The robot's built-in real-time operating system starts the chassis navigation module to travel along the predetermined route according to the task instructions, and hovers at each path point to perform data collection. When the robot's battery level is below 20%, the task scheduler will dynamically insert a high-priority task to return to the charging station for charging. After charging is completed, the unfinished inspection tasks will be automatically resumed. The scheduler will also monitor the robot's operating status, and when a robot malfunctions, its tasks will be automatically reassigned to other idle robots.

[0038] 207. Use the cameras and sensors mounted on the mobile robot to collect environmental data, image data and current status data of each target device in real time in the automated parking garage. The environmental data includes environmental temperature information and environmental humidity information, and the current status data includes device vibration information, device temperature information, device current information, device voltage information and device noise information. In this embodiment, when the mobile robot arrives at the designated inspection point, it extends its multi-degree-of-freedom robotic arm and aligns its onboard sensor probes with key parts of the target equipment (such as the heat dissipation holes of the barrier gate motor and the camera lens). It then uses a 2D / 3D vision camera to capture images of the equipment's appearance, collecting image data. This data is used to determine, through image processing technology, whether the equipment has structural damage (such as a broken barrier gate arm, deformed vehicle platform, broken steel cable, or misaligned limit switch bracket) or obstructions (such as spider webs blocking the lens or dirt covering the recognition area) or hinders normal operation (such as stones stuck in the barrier gate track or foreign objects stuck in the transmission chain). A thermal imaging sensor captures the surface temperature distribution of the equipment, a contact vibration sensor probe briefly contacts the equipment casing to collect vibration data, and a current clamp connects to the equipment cable to collect current waveforms. The microphone array focuses on the sound source of the device to collect noise, while the ambient temperature and humidity data are acquired in real time by environmental sensors on the robot itself. Ambient temperature and humidity information, as key environmental features, are used together with the current status data of the device (such as vibration, temperature, current, noise, etc.) to input into the prediction model. This enables the model to distinguish between normal fluctuations caused by environmental factors and the precursors of device failure, thereby significantly improving the accuracy of fault prediction and anti-interference ability. On the other hand, ambient temperature and humidity information help to establish a health status baseline library for different scenarios, support environmental compensation comparison to avoid misjudgment, help determine the appropriate timing for preventive maintenance tasks (such as avoiding high humidity or extreme temperature periods), help identify environmentally related easily handled accidents such as lens fogging, and add environmental tags to the operation and maintenance effect data for closed-loop feedback optimization of the model.

[0039] 208. Use image processing techniques to analyze image data and obtain analysis results; In this embodiment, the system performs a series of image processing operations on the acquired image data. First, it uses a target detection algorithm (such as YOLOv8) to identify the main body of the equipment in the image (such as the barrier gate arm and camera lens area). Then, it uses an image segmentation algorithm (such as U-Net) to extract the key component areas of the equipment. Next, it uses edge detection (Canny operator) and contour analysis to determine whether the equipment has structural breaks or missing parts. It uses texture analysis and anomaly detection (such as isolated forest) to identify surface cracks, wear and other defects. It uses background subtraction and connected component analysis to detect obstructions in front of the lens (such as spider webs, dust, foreign objects) and calculate their area ratio. It uses color histogram analysis to determine the status of the indicator lights (such as a constantly lit red light indicating a fault). Finally, the system outputs structured analysis results, including: equipment ID, detection time, whether there is structural damage (yes / no and damage type), whether there is obstruction (yes / no and obstruction type and area ratio), and whether the indicator light status is normal.

[0040] 209. Based on the analysis results, determine whether there are any easily manageable accident issues at the location of the automated parking garage corresponding to the image; In this embodiment, the system matches the image analysis results with a predefined rule base for easily handled incidents. Easily handled incidents include, but are not limited to: accidental shutdown of the device power switch (e.g., determined by the power indicator light not being lit in the image), the license plate recognition lens being obstructed by dust or foreign objects (e.g., determined by the image detecting that the obstruction area in front of the lens exceeds 50%), and small stones stuck in the barrier gate track (e.g., determined by the image detecting foreign objects in the track). The system determines whether the current problem belongs to an easily handled incident based on the rule base. For example, if a spider web is detected covering more than 30% of the area in front of the camera lens, it is determined to be an easily handled incident (it can be restored by cleaning); if a broken barrier gate arm is detected, it is determined to be a non-easily handled incident (the part needs to be replaced). This determination result will determine the urgency and handling method of subsequent tasks.

[0041] 210. If so, generate an accident handling task and send the accident handling task to the mobile robot so that the mobile robot can immediately perform the handling action based on the accident handling task. In this embodiment, if step 209 determines that there is an easily manageable accident problem, the system immediately generates a high-priority accident handling task. The task content includes: accident type (such as lens obstruction, foreign object in the barrier gate track), target equipment location, and specific handling action instructions (such as using a miniature air nozzle carried by a robotic arm to blow away the lens, or using a robotic arm to pick up stones in the track). The system sends the task to the mobile robot closest to the accident point through a low-latency communication channel (such as 5G URLLC). The robot pauses its current inspection task, navigates to the accident point, and after confirming the accident location a second time through machine vision, executes the corresponding handling action (for example, extending a robotic arm with a sponge brush to wipe the lens, or using a vacuum cleaner to remove dust). After the handling is completed, the robot re-collects image data and uploads it to the system for effect verification to ensure that the accident problem has been resolved.

[0042] 211. Compare the current status data with the health status baseline database to obtain information on the degree of deviation from the status; In this embodiment, the system performs the same preprocessing and feature extraction procedures as steps 202 and 203 on the collected current state data to generate a current feature vector. Then, it calculates the Mahalanobis distance between this current feature vector and the center of the health feature vector cluster stored in the health state baseline database. This distance value represents the degree of state deviation. The formula for calculating the Mahalanobis distance is: ; in, The feature vector of the current state (such as temperature, vibration, current, etc.). This is the mean vector of the feature vectors in the health baseline database. The covariance matrix of the eigenvectors in the healthy baseline database. This represents the Mahalanobis distance; the larger the value, the more severe the deviation.

[0043] In this embodiment of the invention, multi-dimensional health status data of each target device in the automated parking garage during a fault-free operation cycle are collected, cleaned, aligned, filled, and feature-extracted to form a health feature vector. Cluster analysis is then performed to establish a health status baseline library containing the normal value range, mean, and standard deviation of each feature dimension, providing a reliable benchmark for subsequent status comparison. Simultaneously, based on the spatial layout of the automated parking garage and the distribution of equipment locations, the inspection path and frequency of mobile robots are planned. The robots are scheduled to perform inspections within the automated parking garage, using cameras and sensors mounted on the robots to collect environmental data, image data, and current equipment status data in real time. Image processing technology is used to analyze the image data to determine whether there are any easily manageable accidents. If so, an accident handling task is immediately generated and issued, enabling the robot to immediately execute the handling action. This achieves immediate response and rapid handling of common easily manageable accidents, significantly reducing the workload of manual inspection and maintenance, and improving the automation level and response speed of automated parking garage operation and maintenance.

[0044] Please see Figure 3 In the third embodiment of the predictive operation and maintenance method for automated parking garages in this invention, steps 103 and 104 include: 301. Input the current status data and environmental data into the preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset time period; In this embodiment, the system concatenates the current feature vector generated in step 211 with environmental data (temperature, humidity) to form the input vector of the model. The prediction model is an LSTM network based on an attention mechanism. The network first processes the input time series data (data from the past hour) through an encoder to capture the long-term dependencies between data. Then, it outputs the time series prediction for the next 24 hours through a decoder. The prediction result passes through a Softmax layer and is finally mapped to the probability distribution of various fault states.

[0045] 302. Analyze the fault characteristics of the fault prediction results and determine the maintenance type based on the fault characteristics; In this embodiment, the system analyzes the intermediate layer attention weights of the prediction model output and locates the input features that contribute the most to fault prediction (for example, the harmonic distortion rate of motor current and the high-frequency components in the vibration spectrum are the core features that cause bearing wear faults). The system maps these significant features to specific fault modes (such as bearing oil shortage, bearing inner and outer ring wear) through a predefined feature-fault type mapping table, and then determines the corresponding maintenance type (such as adding lubricating oil or replacing the bearing).

[0046] 303. Match the maintenance type with the corresponding preventive maintenance task in the preset maintenance knowledge base; In this embodiment, the system uses the determined maintenance type as a keyword to search in a structured maintenance knowledge base. This knowledge base stores detailed operation guides for each maintenance type, including: required tools (such as Allen wrenches and lubricating guns), operation steps, safety precautions, and estimated time. The system instantiates the operation guide and generates a task sheet containing device ID, execution time, and detailed operation instructions.

[0047] 304. Obtain the operational status data of the automated parking garage and determine low-interference periods based on the operational status data; In this embodiment, the system obtains real-time traffic flow statistics (such as the number of vehicles entering and exiting per hour) by connecting to the gate management system and billing system of the automated parking garage. By statistically analyzing historical data from the past week, the system identifies the time window with the lowest traffic flow (for example, 2:30-4:00 AM from Tuesday to Thursday). The system marks this time window as a low-interference period and uses it as the primary constraint for the execution time of preventive maintenance tasks. For emergency fault repair tasks, this restriction does not apply.

[0048] 305. Assign preventive maintenance tasks to mobile robots so that they can perform maintenance actions during low-interference periods based on preventive maintenance tasks; In this embodiment, the system stores the generated preventive maintenance tasks (including navigation location and operation instructions) in the task database and sets their execution timestamp to the next low-interference period. When the preset time is reached, the scheduler wakes up the mobile robot in standby or charging state through the message queue. The robot navigates to the target device, first confirms the device status through machine vision (comparing it with the image template in the task sheet), and then calls the corresponding robotic arm end effector (such as an electric screwdriver bit for tightening screws, an air knife for blowing dust, an oil nozzle, and a sponge brush for cleaning) to perform maintenance actions strictly according to the operation steps in the task sheet. The entire process is recorded by the robot's built-in camera and transmitted back to the management terminal in real time.

[0049] 306. When the deviation information of the status exceeds the preset deviation threshold, it is determined to be a device malfunction, and a fault repair task is generated; In this embodiment, the system sets a dynamic deviation threshold (e.g., Mahalanobis distance > 15). Once the state deviation information calculated in step 211 exceeds this threshold, the system immediately determines that the equipment is in a serious abnormal state (a failure that is about to occur or has already occurred that affects operation). The trigger is activated, and the system bypasses the complex calculation of the prediction model and directly retrieves the corresponding fault repair task from the rapid response knowledge base based on the characteristic pattern of the current state deviation. For example, the solution corresponding to abnormally high current and no vibration is to reset the motor controller.

[0050] 307. Distribute the fault repair task to the mobile robot so that the mobile robot can immediately perform repair actions based on the fault repair task; In this embodiment, the system prioritizes this urgent fault repair task to the highest level and immediately sends it through a high-priority communication channel (such as a URLLC slice of a 5G network). Regardless of what task the robot is currently performing (unless it is a safety emergency task, such as avoiding pedestrians), the robot will immediately pause the current task, save the current state, navigate to the fault point, and perform repair actions such as physically pressing the restart button of the gate controller, using the robotic arm to replug the loose network cable, and pressing the power switch to reset. After the repair is completed, the robot sends a task completion signal.

[0051] 308. After performing repair or maintenance actions, use a mobile robot to re-collect the current status data to obtain new status data; In this embodiment, after performing the repair or maintenance action, the robot does not leave immediately, but waits in place for 10-30 seconds until the equipment status stabilizes. Then, following the standard data acquisition process, it re-acquires a complete set of status data for the target equipment to obtain new status data.

[0052] 309. Re-input the new state data into the prediction model so that the prediction model outputs new fault prediction results; In this embodiment, the system takes the new status data reported by the robot and uses the prediction model described in step 301 to generate a new fault prediction result to verify the effect of this maintenance operation. The system compares the changes in fault probability before and after maintenance.

[0053] 310. When the probability of a new fault prediction is greater than the preset fault threshold, a deep maintenance task is generated and pushed to the management terminal. In this embodiment, when the probability of a new fault prediction is greater than a preset fault threshold, it is determined that the automated maintenance has failed to solve the fundamental problem. The system will automatically generate an in-depth maintenance task report, which includes the equipment ID, fault history, maintenance actions performed, status data before and after maintenance, and a comparison of prediction results. This report will be pushed to the management terminal of the automated parking garage via an App or Web terminal, prompting that human experts are needed for in-depth analysis or on-site repair.

[0054] In this embodiment of the invention, by inputting current state data and environmental data into a preset prediction model, and using an attention-based LSTM or Transformer network to output the fault prediction results of each target device within a preset future time period, the fault characteristics are analyzed and the maintenance type is determined. The maintenance type is then matched with preventive maintenance tasks in a preset maintenance knowledge base, achieving forward-looking prediction of equipment faults and accurate generation of maintenance tasks. Simultaneously, based on the operational status data of the automated parking garage, low-interference periods are determined, and preventive maintenance tasks are issued to mobile robots for execution during these periods to avoid interfering with normal parking operations. When the degree of state deviation exceeds a preset deviation threshold, it is immediately determined as an equipment fault, and a fault repair task is generated and issued to the robot for immediate execution of the repair action. After the repair or maintenance is performed, the state data is re-collected and re-input into the prediction model for effect verification. If the fault probability in the new fault prediction result is still greater than the preset fault threshold, a deep inspection task is generated and pushed to the management end, forming a closed-loop processing mechanism of prediction-maintenance-verification-upgrade, effectively ensuring the continuous and stable operation of the automated parking garage equipment.

[0055] Please see Figure 4 In the fourth embodiment of the predictive operation and maintenance method for automated parking garages in this invention, steps 105 and 105 thereafter include: 401. After each preventive maintenance or fault repair, record the change in the failure probability of the corresponding equipment before and after maintenance, and generate operation and maintenance effect data. In this embodiment, the system creates a unique record entry for each maintenance task. The entry includes: task type (preventive / fault-related / accident handling), execution time, maintenance action code, failure probability given by the prediction model before maintenance, new failure probability given by the prediction model after maintenance, and changes in image analysis indicators (such as the occlusion area decreasing from 30% to 0%). The operation and maintenance effect data is quantified into failure risk reduction value and maintenance effectiveness label.

[0056] 402. Add the operation and maintenance performance data as feedback samples to the training dataset; In this embodiment, the system associates and packages the complete record entry generated in step 401 with the historical state data fragments that triggered this maintenance (including numerical data and image features within a certain period before maintenance) and the specific maintenance actions performed. This state-action-effect triplet is timestamped and written asynchronously into the training dataset as a labeled strong feedback sample. For samples marked as invalid, the system will make special annotations for subsequent manual review or model training to increase weight.

[0057] 403. Regularly use the training dataset to incrementally train the prediction model. The prediction model includes an attention-based fusion network and a temporal prediction network. The temporal prediction network is a long short-term memory network or a Transformer network. In this embodiment, the system sets a scheduled task (e.g., every Monday at 3 AM) to trigger the online learning process of the prediction model. This process reads all new feedback samples from the database and merges them with the historical training set. Then, the system uses a gradient descent algorithm (such as the Adam optimizer) to incrementally train the currently deployed prediction model, fine-tuning the network weights. To prevent catastrophic forgetting, the system adopts the Elastic Weight Consolidation (EWC) technique to constrain the important weights of old tasks. The goal of incremental training is to enable the model to learn under what feature trends which maintenance actions can most effectively reduce the risk of failure, thereby improving the prediction accuracy of the model and the rationality of maintenance recommendations.

[0058] 404. Re-predict the failure probability of all target equipment in the automated parking garage using the updated prediction model; In this embodiment, after the model optimization and update are completed, the system will trigger a global assessment. The system will load the historical status data of all target devices stored in the previous week and use the new version model to perform backtracking prediction to generate a list of failure probabilities of all devices in the future at the current moment.

[0059] 405. Dynamically adjust the inspection path and frequency of the mobile robot according to the failure probability; In this embodiment, the system's task scheduler obtains an updated global fault probability list. For devices with a fault probability exceeding the high warning threshold (e.g., greater than 70%), the scheduler dynamically increases their inspection frequency (e.g., from once every 2 hours to once every 30 minutes) and marks these high-probability fault devices as key focus devices. Then, the scheduler re-runs the Multiple Traveling Salesman Problem (MTSP) algorithm, sets these key focus devices as high-priority points of interest in path planning, reduces the path weight of low-probability devices, and re-optimizes and generates new inspection paths. At the same time, the scheduler also adjusts the robot's inspection strategy, such as requiring the robot to stay longer and collect images from more angles when passing high-probability fault devices. The updated path and frequency take effect immediately for the next round of inspection tasks.

[0060] In this embodiment of the invention, after each preventive maintenance or fault repair, the change in fault probability before and after equipment maintenance is recorded, and operation and maintenance effect data is generated. The operation and maintenance effect data is added to the training dataset as feedback samples. The prediction model is incrementally trained periodically using the training dataset. The prediction model includes a fusion network based on an attention mechanism and a temporal prediction network. The temporal prediction network is a long short-term memory network or a Transformer network, which realizes continuous optimization and self-evolution of the prediction model, and continuously improves the accuracy of fault prediction and the rationality of maintenance suggestions. At the same time, the updated prediction model is used to re-predict the fault probability of all target equipment in the automated parking garage, and the inspection path and inspection frequency of the mobile robot are dynamically adjusted according to the fault probability, so that inspection resources are tilted towards equipment with high fault risk, improving the targeting and efficiency of inspection, forming an intelligent operation and maintenance closed loop of prediction-maintenance-feedback-optimization-re-prediction, which significantly improves the adaptive capability and long-term operation and maintenance effect of the automated parking garage predictive operation and maintenance system.

[0061] The predictive operation and maintenance method for automated parking garages in this invention has been described above. The predictive operation and maintenance device for automated parking garages in this invention is described below. Please refer to [link / reference]. Figure 5 One embodiment of the predictive maintenance device for automated parking garages in this invention includes: The data acquisition module 501 is used to collect multi-dimensional health status data of each target device in the automated parking system and to establish a health status baseline library based on the multi-dimensional health status data. The scheduling, acquisition, and comparison module 502 is used to schedule at least one mobile robot to perform inspections in the automated parking garage, and to use the mobile robot to collect environmental data and current status data of each target device in the automated parking garage in real time. The current status data is then compared with the health status baseline database to obtain information on the degree of deviation from the status. The input generation module 503 is used to input current status data and environmental data into a preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset period of time, and generates preventive maintenance tasks based on the status deviation information and fault prediction results. The dispatch module 504 is used to dispatch fault repair tasks or preventive maintenance tasks to the mobile robot so that the mobile robot can perform repair or maintenance actions. The generation and addition update module 505 is used to generate operation and maintenance effect data, add the operation and maintenance effect data as feedback samples to the training dataset, and use the training dataset to periodically optimize and update the prediction model.

[0062] In this embodiment, a health status baseline database is established by collecting multi-dimensional health status data of each target device in the automated parking garage. A mobile robot is used to collect environmental data of the automated parking garage and the current status data of each target device in real time. The current status data is compared with the health status baseline database. The current status data and environmental data are input into a preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset time period. Based on the status deviation information and fault prediction results, preventive maintenance tasks are generated. The mobile robot performs maintenance or repair actions according to the fault repair task or preventive maintenance task, realizing early prediction of equipment faults in the automated parking garage, automated inspection and rapid handling. The operation and maintenance effect data is added to the training dataset as feedback samples. The prediction model is periodically optimized and updated using the training dataset to improve the accuracy of fault prediction.

[0063] Please see Figure 6 Another embodiment of the predictive maintenance device for automated parking garages in this invention includes: The data acquisition module 501 is used to collect multi-dimensional health status data of each target device in the automated parking system and to establish a health status baseline library based on the multi-dimensional health status data. The scheduling, acquisition, and comparison module 502 is used to schedule at least one mobile robot to perform inspections in the automated parking garage, and to use the mobile robot to collect environmental data and current status data of each target device in the automated parking garage in real time. The current status data is then compared with the health status baseline database to obtain information on the degree of deviation from the status. The input generation module 503 is used to input current status data and environmental data into a preset prediction model so that the prediction model outputs the fault prediction results of each target device in the future preset period of time, and generates preventive maintenance tasks based on the status deviation information and fault prediction results. The dispatch module 504 is used to dispatch fault repair tasks or preventive maintenance tasks to the mobile robot so that the mobile robot can perform repair or maintenance actions. The generation and update module 505 is used to generate operation and maintenance effect data, add the operation and maintenance effect data as feedback samples to the training dataset, and use the training dataset to periodically optimize and update the prediction model. In this embodiment, the data acquisition and establishment module 501 includes: a first acquisition unit 5011, used to acquire vibration data, temperature data, current data, voltage data, and noise data of each target device in the automated parking garage during a fault-free operation cycle as multi-dimensional health status data; a cleaning, alignment, and filling unit 5012, used to clean, align, and fill missing values ​​in the multi-dimensional health status data to obtain standardized time-series data; an extraction and composition unit 5013, used to extract time-domain features, frequency-domain features, and statistical features from the standardized time-series data and form a health feature vector; and a clustering and establishment unit 5014, used to cluster the health feature vector and establish a health status baseline library for each target device, the health status baseline library including the normal value range, mean, and standard deviation of each feature dimension.

[0064] In this embodiment, the scheduling, acquisition, and comparison module 502 includes: a planning unit 5021, used to plan the inspection path and frequency of at least one mobile robot based on the spatial layout of the automated parking garage and the location distribution of the target equipment; a scheduling unit 5022, used to schedule the mobile robot to perform inspections within the automated parking garage according to the inspection path and frequency; and a second acquisition unit 5023, used to collect environmental data, image data, and current status data of each target equipment in the automated parking garage in real time using cameras and sensors mounted on the mobile robot. The environmental data includes environmental temperature information and environmental humidity information, and the current status data includes equipment vibration information, etc. The system includes equipment temperature information, equipment current information, equipment voltage information, and equipment noise information; an analysis unit 5024, used to analyze image data using image processing technology to obtain analysis results; a determination unit 5025, used to determine whether there is an easily manageable accident problem at the location of the three-dimensional parking garage corresponding to the image based on the analysis results; a generation and distribution unit 5026, used to generate an accident handling task if so, and distribute the accident handling task to the mobile robot so that the mobile robot can immediately execute the handling action based on the accident handling task; and a comparison unit 5027, used to compare the current state data with the health state baseline library to obtain information on the degree of state deviation.

[0065] In this embodiment, the input generation module 503 includes: a first input unit 5031, used to input current status data and environmental data into a preset prediction model so that the prediction model outputs the fault prediction results of each target device in a future preset time period; an analysis and determination unit 5032, used to analyze the fault characteristics of the fault prediction results and determine the maintenance type based on the fault characteristics; and a matching unit 5033, used to match the maintenance type with the corresponding preventive maintenance tasks in the preset maintenance knowledge base.

[0066] In this embodiment, the distribution module 504 includes: an acquisition and determination unit 5041, used to acquire the operational status data of the automated parking garage and determine low-interference periods based on the operational status data; a first distribution unit 5042, used to distribute preventive maintenance tasks to the mobile robot, so that the mobile robot can perform maintenance actions based on the preventive maintenance tasks during low-interference periods; a judgment and generation unit 5043, used to determine that the equipment is faulty when the status deviation information is greater than a preset deviation threshold, and generate a fault repair task; a second distribution unit 5044, used to distribute the fault repair task to the mobile robot, so that the mobile robot can immediately perform maintenance actions based on the fault repair task; a third acquisition unit 5045, used to re-acquire the current status data using the mobile robot after performing maintenance or repair actions, and obtain new status data; a second input unit 5046, used to re-input the new status data into the prediction model, so that the prediction model outputs a new fault prediction result; and a generation and push unit 5047, used to generate a deep inspection task and push the deep inspection task to the management terminal when the fault probability in the new fault prediction result is greater than a preset fault threshold.

[0067] In this embodiment, the generation and addition update module 505 includes: a record generation unit 5051, used to record the change in the failure probability of the corresponding equipment before and after each preventive maintenance or fault repair, and generate operation and maintenance effect data; an addition unit 5052, used to add the operation and maintenance effect data as feedback samples to the training dataset; and a training unit 5053, used to periodically use the training dataset to incrementally train the prediction model, the prediction model including a fusion network based on an attention mechanism and a temporal prediction network, the temporal prediction network being a long short-term memory network or a Transformer network.

[0068] In this embodiment, it also includes: a prediction module 506, used to re-predict the failure probability of all target equipment in the automated parking garage using the updated prediction model; and an adjustment module 507, used to dynamically adjust the inspection path and inspection frequency of the mobile robot according to the failure probability.

[0069] above Figure 5 and Figure 6 The predictive operation and maintenance device for the automated parking garage in this embodiment of the invention is described in detail from the perspective of modular functional entities. The predictive operation and maintenance device for the automated parking garage in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0070] Figure 7This is a schematic diagram of the structure of a predictive maintenance device for an automated parking garage according to an embodiment of the present invention. The predictive maintenance device 600 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the predictive maintenance device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the predictive maintenance device 600 to implement the steps of the predictive maintenance method for automated parking garages provided in the above-described method embodiments.

[0071] The predictive maintenance equipment 600 for automated parking systems may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The illustrated structure of the predictive maintenance equipment for automated parking garages does not constitute a limitation on predictive maintenance equipment for automated parking garages. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0072] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a predictive maintenance method for an automated parking system.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0074] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A predictive operation and maintenance method for a multi-level parking garage, characterized in that, include: Collect multi-dimensional health status data of each target device in the automated parking system, and establish a health status baseline database based on the multi-dimensional health status data. At least one mobile robot is dispatched to perform inspections in the automated parking garage, and the mobile robot is used to collect environmental data of the automated parking garage and the current status data of each target device in real time. The current status data is compared with the health status baseline database to obtain information on the degree of status deviation. The current state data and the environmental data are input into a preset prediction model so that the prediction model outputs the fault prediction results of each of the target devices in a future preset period of time, and preventive maintenance tasks are generated based on the state deviation information and the fault prediction results. The fault repair task or the preventive maintenance task is assigned to the mobile robot so that the mobile robot can perform the repair or maintenance action; Generate operation and maintenance performance data, add the operation and maintenance performance data as feedback samples to the training dataset, and use the training dataset to periodically optimize and update the prediction model.

2. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The process involves collecting multi-dimensional health status data from each target device in the automated parking garage, and establishing a health status baseline database based on this data, including: Vibration data, temperature data, current data, voltage data, and noise data of each target device in the automated parking garage are collected during the fault-free operation cycle as multi-dimensional health status data; The multi-dimensional health status data is cleaned, time-aligned, and missing values ​​are imputed to obtain standardized time-series data. Time-domain features, frequency-domain features, and statistical features are extracted from the standardized time-series data and used to form a health feature vector; The health feature vectors are clustered and a health status baseline library for each target device is established. The health status baseline library includes the normal range, mean, and standard deviation of each feature dimension.

3. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The system schedules at least one mobile robot to perform inspections within the automated parking garage. The mobile robot collects real-time environmental data and current status data of each target device within the garage. This current status data is then compared with a health status baseline database to obtain information on the degree of deviation from the target status, including: Based on the spatial layout of the automated parking garage and the location distribution of the target equipment, plan the inspection path and frequency of at least one mobile robot. The mobile robot is scheduled to perform inspections within the automated parking garage according to the inspection path and the inspection frequency. The mobile robot uses cameras and sensors to collect environmental data, image data, and current status data of each target device in the automated parking garage in real time. The environmental data includes ambient temperature and humidity information, and the current status data includes device vibration information, device temperature information, device current information, device voltage information, and device noise information. The image data is analyzed using image processing techniques to obtain analysis results; Based on the analysis results, determine whether there are any easily manageable accident issues at the location of the automated parking garage corresponding to the image; If so, an accident handling task is generated and sent to the mobile robot so that the mobile robot can immediately perform a handling action based on the accident handling task. The current status data is compared with the health status baseline database to obtain information on the degree of status deviation.

4. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The step involves inputting the current state data and the environmental data into a preset prediction model, so that the prediction model outputs fault prediction results for each of the target devices within a preset future time period. Based on the state deviation information and the fault prediction results, preventative maintenance tasks are generated, including: The current state data and the environmental data are input into a preset prediction model so that the prediction model outputs the fault prediction results of each of the target devices in a future preset time period; Analyze the fault characteristics of the fault prediction results, and determine the maintenance type based on the fault characteristics; The maintenance type is matched with the corresponding preventive maintenance task in the preset maintenance knowledge base.

5. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The step of assigning the fault repair task or the preventive maintenance task to the mobile robot, so that the mobile robot can perform repair or maintenance actions, includes: Acquire operational status data of the automated parking garage and determine low-interference periods based on the operational status data; The preventive maintenance task is assigned to the mobile robot so that the mobile robot performs maintenance actions based on the preventive maintenance task during the low-interference period. When the deviation information of the state exceeds a preset deviation threshold, it is determined to be a device malfunction, and a fault repair task is generated; The fault repair task is sent to the mobile robot so that the mobile robot can immediately perform repair actions based on the fault repair task; After performing repair or maintenance actions, the mobile robot is used to re-collect the current state data to obtain new state data; The new state data is re-input into the prediction model so that the prediction model outputs a new fault prediction result. When the probability of a fault in the new fault prediction result is greater than the preset fault threshold, a deep maintenance task is generated and pushed to the management terminal.

6. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The process of generating operation and maintenance performance data, using this data as feedback samples in the training dataset, and periodically optimizing and updating the prediction model using the training dataset includes: After each preventative maintenance or fault repair, record the change in the failure probability of the corresponding equipment before and after maintenance, and generate operation and maintenance effect data; The aforementioned operation and maintenance performance data is added to the training dataset as feedback samples. The prediction model is incrementally trained periodically using the training dataset. The prediction model includes an attention-based fusion network and a temporal prediction network, wherein the temporal prediction network is a long short-term memory network or a Transformer network.

7. The predictive operation and maintenance method for automated parking garages according to claim 1, characterized in that, The process of generating operation and maintenance performance data, adding this data as feedback samples to the training dataset, and periodically optimizing and updating the prediction model using the training dataset further includes: The updated prediction model is used to re-predict the failure probability of all target equipment in the automated parking system. The inspection path and frequency of the mobile robot are dynamically adjusted based on the failure probability.

8. A predictive operation and maintenance device for a three-dimensional parking garage, characterized in that, include: The data acquisition and establishment module is used to collect multi-dimensional health status data of each target device in the automated parking system and establish a health status baseline library based on the multi-dimensional health status data. The scheduling, acquisition, and comparison module is used to schedule at least one mobile robot to perform inspections in the automated parking garage, and to use the mobile robot to collect environmental data of the automated parking garage and the current status data of each target device in real time. The current status data is then compared with the health status baseline database to obtain information on the degree of status deviation. The input generation module is used to input the current state data and the environmental data into a preset prediction model so that the prediction model outputs the fault prediction results of each of the target devices in a future preset period of time, and generates preventive maintenance tasks based on the state deviation information and the fault prediction results. The dispatch module is used to dispatch fault repair tasks or preventive maintenance tasks to the mobile robot, so that the mobile robot can perform repair or maintenance actions; A generation and update module is used to generate operation and maintenance effect data, which is then added to the training dataset as feedback samples. The prediction model is periodically optimized and updated using the training dataset.

9. A predictive maintenance device for a three-dimensional parking garage, characterized in that, The predictive maintenance equipment for the automated parking garage includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the automated parking garage predictive maintenance device to perform the steps of the automated parking garage predictive maintenance method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the predictive operation and maintenance method for automated parking garages as described in any one of claims 1-7.

Images