Rail-mounted robot environmental monitoring system
By introducing an onboard edge computing module and a multi-robot collaborative scheduling module into the environmental monitoring system of the rail-mounted robot, the problems of false alarms and missed alarms caused by fixed weight discrimination, loss of local fusion analysis after network disconnection, difficulty in handling peak outliers by unidirectional filtering, parameter offset caused by labeling outliers, and repeated collection of data from multiple robot locations in the existing technology are solved, thus achieving efficient and reliable environmental monitoring and alarm output.
Patent Information
- Application Number
- CN202611632512.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing environmental monitoring systems for track-mounted robots suffer from several problems, including: false alarms and missed alarms due to fixed-weight single-parameter threshold discrimination; loss of local fusion and analysis capabilities after network outages; difficulty in handling peak outliers with unidirectional filtering; parameter shifts caused by labeling outliers; static manual configuration of risk weights; and repeated data collection at multiple locations during multi-robot collaboration.
The vehicle-mounted edge computing module is used to perform the variation coefficient weighted multi-parameter risk contribution inference, combined with bidirectional exponential moving median noise reduction processing logic, and a dynamic weight adaptive optimization module. It supports complete inspection and alarm in the event of network outage or offline, and realizes distributed collaborative monitoring of multiple robots through a multi-robot collaborative scheduling module and a global report merging module.
It improves the identification accuracy and anti-interference capability of environmental monitoring in prefabricated power modules, enhances operational reliability under complex working conditions, enables local risk assessment and alarm in unstable network scenarios, reduces false alarms, avoids resource waste, and improves inspection efficiency.
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent operation and maintenance technology for rail-mounted inspection robots for prefabricated power modules, specifically to an environmental monitoring system for rail-mounted robots. Background Technology
[0002] With the large-scale commissioning of prefabricated power modules, numerous high-voltage switchgear cabinets are installed inside, posing multiple safety risks such as sulfur hexafluoride leakage, oxygen deficiency, abnormal temperature and humidity, and excessive equipment operating noise. While existing track-mounted robots are equipped with environmental sensors, most employ fixed-weight single-parameter hard threshold discrimination logic. The weights of various environmental indicators are manually set and do not adaptively adjust based on the temporal fluctuations of the parameters themselves. Even small disturbances in a single parameter can easily generate numerous false alarms; when multiple parameters simultaneously experience slight degradation, none exceeding their respective alarm thresholds, traditional methods cannot identify complex environmental hazards. The fixed-weight model cannot reflect the actual sensitivity of different parameters to fluctuations, resulting in insufficient comprehensive judgment capabilities for multi-source environmental factors.
[0003] Existing monitoring solutions of this kind heavily rely on remote backend systems for data processing and risk assessment. All raw sensor observation data collected by the robot needs to be uploaded to the backend for processing via Ethernet. If the prefabricated cabin cable communication experiences jitter or link interruption, the robot can only complete the raw data collection. The on-board hardware does not have the ability to perform local multi-parameter risk contribution analysis and alarm output, resulting in the loss of real-time safety warnings on site. The continuous transmission of massive amounts of raw sampling data also brings a large network transmission load, making it difficult to adapt to the complex communication conditions of the prefabricated cabin.
[0004] Traditional environmental monitoring for track-mounted robots still has several practical shortcomings: residual mechanical vibration after robot parking and braking, and electromagnetic pulses in the field can easily introduce spike outliers; ordinary unidirectional sliding filters only utilize historical data and lack backward time-series information, resulting in limited spike noise suppression; a small number of manually labeled outlier samples exist in the training samples, and ordinary optimization methods are easily biased by outlier samples, causing parameter shifts; risk assessment weights are given entirely manually and statically, and cannot be iteratively optimized by combining parameter variation characteristics and historical risk samples; when multiple robots are deployed in the same field, there is a lack of point status interaction and avoidance logic, which can easily lead to repeated point collection, resulting in resource waste and limiting the overall efficiency of inspection operations. Summary of the Invention
[0005] To address the shortcomings of existing technologies, such as the susceptibility to false alarms and missed alarms due to fixed-weight single-parameter threshold discrimination, loss of local fusion analysis and alarm capabilities after network outages, difficulty in handling peak outliers with unidirectional filtering, parameter shifts caused by labeling outliers, static manual configuration of risk weights, and redundant data collection at multiple locations in multi-robot collaboration, this invention provides a rail-mounted robot environmental monitoring system. It utilizes an onboard edge computing module to perform multi-parameter risk contribution derivation based on the coefficient of variation, combined with bidirectional exponential moving median denoising logic, and a dynamic weight adaptive optimization module to achieve iterative parameter updates. This system supports complete offline inspection and alarm reporting even when the network is down. Furthermore, it leverages a multi-robot collaborative scheduling module and a global report merging module to achieve distributed collaborative monitoring of multiple robots, thereby improving the accuracy of environmental monitoring and identification in prefabricated power cabins, enhancing anti-interference capabilities, and improving operational reliability under complex conditions.
[0006] The rail-mounted robot environmental monitoring system includes a rail-mounted patrol robot body, a multi-source environmental sensing and acquisition kit, an on-board edge computing module, and a remote control backend at the site. The rail-mounted patrol robot body is equipped with a Siling Diana7 robotic arm, enabling it to walk and stop along a galvanized rack and pinion track within a prefabricated cabin. The multi-source environmental sensing and acquisition kit is deployed on the robot body, simultaneously collecting raw environmental observation data on temperature, humidity, sulfur hexafluoride, oxygen, and noise. The on-board edge computing module is deployed on the robot's onboard industrial control hardware, performing raw observation data cleaning, multi-parameter coupled environmental risk simulation, and local audible and visual alarm signal output. The remote control backend at the site uses a wired Ethernet cable and the on-board edge computing module to complete data interaction, enabling the issuance of inspection tasks, reception of monitoring and alarm data, and storage of historical data. The robot traverses preset points along the track to complete environmental data collection. The on-board edge computing module can independently conduct monitoring and analysis without the remote control backend at the site, building an on-board... A two-tiered, collaborative environmental monitoring system for prefabricated power cabins.
[0007] Preferably, the vehicle-mounted edge computing module is configured with a multi-parameter risk contribution inference logic based on the coefficient of variation, and risk assessment is carried out according to the following processing steps: S1. Perform statistical analysis on the time-series sampling sequences of each environmental parameter, calculate the time-series mean and standard deviation of each parameter, and calculate the dynamic weight of each parameter based on the coefficient of variation; the more severe the time-series fluctuation of the parameter, the higher the dynamic weight is assigned. S2. Perform normalized risk mapping on each effective environmental parameter, and output 0. Risk characterization value in interval 1; Read the sensor hardware self-test status code, identify the faulty measurement point, and reset the dynamic weight of the measurement point to zero, so that it does not participate in subsequent risk calculation; S3. Combine the dynamic weights of each parameter with the risk characterization value to perform weighted aggregation and obtain the comprehensive environmental risk assessment result; S4. Compare the comprehensive environmental risk assessment results with the preset risk thresholds. If the assessment results exceed the thresholds, output a comprehensive environmental hazard risk warning signal.
[0008] Preferably, the multi-source environmental perception and acquisition kit includes an NDIR-based sulfur hexafluoride sensor, an oxygen detection module, a digital temperature and humidity sensor, and an A-weighted noise sensor; the output messages of each sensor carry a unified timestamp and a hardware self-test status code; the vehicle-mounted edge computing module reads the self-test status code to identify sensor offline and zero-point drift faults, performs shielding marking on faulty measurement points, and dynamically resets the weights of shielded measurement points to zero, so that they no longer participate in the multi-parameter risk contribution inference and processing flow.
[0009] Preferably, the vehicle-mounted edge computing module is configured with bidirectional exponential moving median noise reduction processing logic to perform the following noise reduction processing steps: S1. Take the original sampled data sequence within the bidirectional sliding window before and after the current sampling time; calculate the median of the forward window and backward window sequences respectively; S2. Perform exponential smoothing on the forward and backward median results respectively to obtain the forward smoothed output and the backward smoothed output; S3. Average the forward smoothing output and the backward smoothing output to output the final noise-reduced observation data. This processing logic simultaneously utilizes bidirectional timing information to suppress residual vibrations during robot docking and spike outlier interference from instantaneous electromagnetic pulses at the site, outputting high-quality observation sequences for use in the risk simulation phase.
[0010] Preferably, the remote control backend of the site is configured with an environmental lifecycle archive module; the environmental lifecycle archive module persistently stores the time-series historical datasets of each inspection point, supports multi-dimensional conditional retrieval by site number, time range, and monitoring parameters, and supports the export of inspection statistical reports and alarm event reports; it receives alarm events reported by the vehicle edge computing module and realizes two-level alarm push: WEB pop-up and in-site message.
[0011] Preferably, the system has an offline autonomous operation mode; when the wired Ethernet communication link of the tow cable is interrupted, the rail-mounted patrol robot continues to execute the local pre-set inspection plan, and the multi-source environmental perception and acquisition kit continues to carry out environmental sampling; the on-board edge computing module independently completes bidirectional exponential moving median noise reduction, coefficient of variation weighted multi-parameter risk contribution estimation, and local audible and visual alarms; all monitoring alarm original records are cached in the on-board local storage medium; after the communication link is restored, the cached data is automatically batch-transmitted to the remote control backend of the site.
[0012] Preferably, the repeatability of the rack and pinion servo drive for the rail-mounted patrol robot is ≤ ±2mm; after the robot completes the point braking, an editable static delay parameter is configured, and the sensor sampling is only started after the static delay timer ends, thereby avoiding the sampling distortion problem caused by the residual mechanical vibration of the vehicle body.
[0013] Preferably, the remote management backend of the site is configured with a dynamic weight adaptive optimization module, which performs the following parameter iterative optimization steps: S1. Retrieve the historical labeled sample set with manual risk labels stored in the background, and read the dynamic weights of each parameter in the current version; S2. Input historical samples into the risk extrapolation process to obtain the sample prediction risk results. Compare the prediction results with the human real labels to calculate the overall sample error using robust loss. Add regularization constraints to the weight parameters to suppress parameter overfitting and parameter shift caused by outlier labeled samples. S3. Update the dynamic weights of each environmental parameter according to the gradient descent direction; S4. Determine if the iteration has reached the convergence termination condition; if it has not converged, return to S2 to continue the iteration; after the iteration converges, send the updated set of dynamic weights to the vehicle edge computing module for subsequent location risk assessment.
[0014] Preferably, the remote control backend of the station is configured with a multi-machine collaborative scheduling module; the multi-machine collaborative scheduling module divides the inspection point set for multiple rail-mounted patrol robots, generates non-overlapping sub-tasks and sends them to each robot; the on-board edge computing module of each robot interacts with the point occupancy status based on the station's local area network to avoid multiple robots sampling the same monitoring point repeatedly.
[0015] Preferably, the remote control backend of the site is configured with a full-domain report merging module; the full-domain report merging module receives monitoring datasets transmitted from multiple robots, completes the fusion and merging of multi-source monitoring data, and automatically outputs a complete prefabricated cabin full-domain environmental inspection report, which supports online preview and local file export.
[0016] The derivation logic for the risk contribution of the coefficient of variation weighted multi-parameter model can be realized through the following mathematical expression: In the formula, m represents the number of environmental parameters effectively participating in the assessment; The coefficient of variation of the k-th parameter; This is the time series observation standard deviation of this parameter; This is the time-series observation mean of this parameter; The dynamic weights are obtained by normalizing the coefficient of variation. The normalized risk mapping function for the k-th parameter has an output range of [0,1]; the fault measurement points will correspond to dynamic weights. Set directly to zero; do not participate in the risk value summation calculation.
[0017] Mathematical expression of the bidirectional exponential moving median noise reduction processing logic: In the formula, The original sampled value at time i; W is the half-width of the forward and backward sliding window; (0,1) is the exponential smoothing coefficient; ( This is the median operation for a sequence; Output the forward exponentially shifted median; Output the backward exponentially shifted median; The output observations are used to reduce noise.
[0018] The dynamic weight adaptive optimization module uses the Huber robust loss function with L2 regularization constraints: In the formula, N is the total number of historical labeled samples; The comprehensive environmental risk value calculated for the t-th sample; For unique heat risk labels of samples; This is the Huber loss cutoff threshold; The L2 regularization coefficient; The L2 regularization term is the parameter weight matrix.
[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a two-tier system consisting of an onboard edge computing module and a remote control backend at the site. When the network is normal, it completes data backhaul, archive statistics, and report management. After communication is interrupted, the onboard unit, without relying on the backend, fully performs inspection sampling, bidirectional exponential moving median noise reduction, coefficient of variation weighted multi-parameter risk contribution estimation, local audible and visual alarms, local data caching, and batch retransmission upon communication recovery, solving the business pain point of no on-site early warning in prefabricated cabin communication unstable scenarios.
[0020] 2. This invention abandons the traditional manual fixed-weight discrimination mode. It first calculates the time series mean and standard deviation of each parameter, and uses the coefficient of variation to characterize the severity of parameter fluctuations. The greater the parameter fluctuation, the higher the dynamic weight is assigned. The comprehensive environmental risk value is obtained by integrating the normalized risk mapping of each parameter, which effectively identifies the compound hidden dangers caused by the simultaneous slight deterioration of multiple parameters. Fault measurement points are directly set to zero weight for exclusion, avoiding the interference of abnormal sensor data with the judgment results. The entire processing flow has a small amount of computation and is suitable for real-time execution by vehicle-mounted embedded edge hardware.
[0021] 3. Unlike traditional unidirectional sliding filters that only use historical sampling data, this invention simultaneously takes time series data from both the preceding and following windows, calculates the median for each, and then performs exponential smoothing. The average of the preceding and following results is used to output the noise-reduced observation value. This fully leverages the median's advantage in resisting outliers, effectively suppressing residual vehicle vibration and spike abrupt sampling outliers introduced by on-site electromagnetic pulses, and outputting a high-quality time series observation sequence for use in risk projection models.
[0022] 4. This invention utilizes manually labeled risk samples accumulated at the site to compare the predicted risk results with the actual labels to calculate the overall error, and adds regularization constraints to suppress overfitting and parameter shifts caused by outlier samples; it breaks away from the mode of relying entirely on manual experience to configure weights, and as field samples continue to accumulate, the ability to identify complex hazards can be continuously iterated and improved. Detailed Implementation
[0023] Example 1 Environmental conditions: Wired communication at the site is normal, and the single-unit rail-mounted robot is conducting routine inspection and monitoring.
[0024] This embodiment is deployed in a 35kV prefabricated substation. The substation is equipped with a 9000mm long galvanized rack and pinion track and a rail-mounted patrol robot with a Siling Diana7 robotic arm. The multi-source environmental perception and acquisition kit integrates an NDIR sulfur hexafluoride sensor, an oxygen detection module, a digital temperature and humidity sensor, and an A-weighted noise sensor. The robot is mounted on a vehicle-mounted edge computing module. The remote control backend of the substation is deployed on the substation server, and the towed Ethernet communication link is stable.
[0025] Pre-enter all inspection points corresponding to switchgear; configure bidirectional sliding window size and exponential smoothing parameters; set risk assessment thresholds and configure convergence conditions related to weight iteration; control the robot's rack and pinion servo drive to stop and repeat positioning accuracy ≤ ±2mm, and configure a 0.9s static delay parameter after the point braking is completed.
[0026] The track-mounted patrol robot receives inspection tasks from the remote control backend of the station and traverses all preset monitoring points along the galvanized rack and pinion track. Upon reaching each target point, it waits for the static delay timer to end, and the multi-source environmental perception acquisition kit outputs raw observation data carrying a unified timestamp and hardware self-test status code. The onboard edge computing module reads the sensor self-test status code, identifies faulty measurement points, and resets the corresponding dynamic weights to zero shielding. It performs bidirectional exponential moving median noise reduction processing to suppress sampling spikes and outliers. It then extrapolates S1 based on the coefficient of variation weighted multi-parameter risk contribution. Step S4 completes the risk assessment; if the comprehensive assessment results reach the threshold conditions, an on-site audible and visual alarm is triggered.
[0027] All monitoring and alarm data are uploaded to the remote management and control backend of the site, and the environmental lifecycle archive module completes the time-series dataset storage, supports multi-dimensional retrieval and report export, and triggers two-level alarm pushes: WEB pop-up and in-site message.
[0028] In this embodiment, a simulated working condition is constructed: the parameters of sulfur hexafluoride and temperature and humidity deviate slightly from the normal values, and neither of them reaches their respective independent alarm thresholds; the time-series fluctuation of the sulfur hexafluoride parameter is larger, and the simulation process automatically assigns a higher dynamic weight. The system identifies the compound hidden danger and outputs an early warning; this risk sample is stored in the background labeled sample library and supplied to the dynamic weight adaptive optimization module for subsequent parameter iteration and optimization.
[0029] Example 2 Environmental conditions: The Ethernet cable at the station is disconnected, triggering an offline autonomous inspection alarm for the standalone rail-mounted robot. The hardware equipment and all algorithm parameters are completely consistent with those in Example 1; Environmental conditions: the wired Ethernet cable between the robot and the remote control backend of the site is disconnected, the local area network inside the prefabricated cabin site is intact, and the robot enters the offline autonomous operation mode.
[0030] The robot reads the locally fixed inspection plan on the vehicle and executes the entire inspection operation. The robot moves along the track, brakes and stops, and samples the sensor. The on-board edge computing module independently completes sensor fault identification, exponential weighted noise reduction, and multi-parameter risk contribution calculation based on the coefficient of variation. It identifies environmental risks and directly triggers on-site audible and visual alarms. All sampling and alarm data are cached in the robot's on-board local storage and are not transmitted to the backend.
[0031] A minor SF6 leak occurred in the simulated prefabricated cabin, and multiple environmental parameters deteriorated simultaneously. The vehicle-mounted terminal independently identified the potential hazards in the complex environment, and the audible and visual alarms were output normally. Later, the tow cable Ethernet communication was restored, and the cached data was batch-transmitted to the remote control backend of the site. The backend completed the data entry and inspection report generation.
[0032] Example 3 Environmental conditions: The hardware, prefabricated cabin environment, and number of monitoring points are exactly the same as in Example 1; independent hard threshold alarms for each parameter are used, along with manually fixed weight logic; ordinary one-way fixed window weighted filtering is used; all weights are manually statically configured, and there is no dynamic weight adaptive optimization module based on robust loss; the vehicle-mounted terminal does not perform multi-parameter risk extrapolation calculations, and all raw sampled data is uploaded to the backend for analysis; after communication is lost, the robot only collects raw data and there is no local alarm output; there is no multi-machine collaborative scheduling module and a global report merging module, and duplicate data collection at multiple points may occur under multi-machine deployment.
[0033] Comparative Example 1: Environmental Monitoring Solution for Traditional Rail-Mounted Robots Environmental conditions: The hardware, prefabricated cabin environment, and number of monitoring points are exactly the same as in Example 1; independent hard threshold alarms for each parameter are used, along with manually fixed weight logic; ordinary one-way fixed window weighted filtering is used; all weights are manually statically configured, and there is no dynamic weight adaptive optimization module based on robust loss; the vehicle-mounted terminal does not perform multi-parameter risk extrapolation calculations, and all raw sampled data is uploaded to the backend for analysis; after communication is lost, the robot only collects raw data and there is no local alarm output; there is no multi-machine collaborative scheduling module and a global report merging module, and duplicate data collection at multiple points may occur under multi-machine deployment.
[0034] Comparative Example 2: Monitoring Solution for Vehicle-Mounted Simple Fixed Weighted Fusion Rail-Mounted Robot Environmental conditions: The hardware equipment and site environment are exactly the same as in Example 1; the vehicle-mounted system can achieve simple fixed-weighted fusion and can output alarms even when the network is down; all weights are manually pre-defined and cannot be adaptively allocated according to the time-series variation characteristics of parameters, and the ability to identify fluctuating parameters is insufficient; there is no bidirectional exponential moving median noise reduction processing logic; there is no dynamic weight adaptive optimization module; there is a lack of multi-machine collaborative scheduling module and full-domain report merging module, and there is duplicate data collection at multiple locations in multi-machine deployment scenarios, making it impossible to automatically generate a full-domain fusion inspection report.
[0035] The experimental procedure and results are as follows: A 35kV prefabricated cabin semi-physical simulation + physical robot joint testing environment was constructed; a test dataset was built, which included samples of normal environment, peak outliers caused by vehicle vibration, composite hidden danger samples of multiple parameters with simultaneous minor degradation, and sensor zero-point drift fault samples; at the same time, offline working condition tests and dual-robot distributed collaborative monitoring experiments were carried out. Examples 1, 2, and 3, Comparative Example 1 and Comparative Example 2 were subjected to multiple rounds of repeated tests under the condition of completely identical hardware and test sample sets.
[0036] This experiment focuses on examining: the ability to identify hidden dangers in complex environments, the adaptability of risk judgment under parameter fluctuation conditions, the false alarm situation caused by vibration pulse outliers, the effectiveness of local alarms under network outage conditions, the change in identification ability after parameter iterative optimization, the phenomenon of repeated data collection at multiple inspection points, the data volume of network transmission in a single round of inspection, and the ability to automatically generate inspection reports across the entire region.
[0037] Under normal network connectivity conditions, Example 1, relying on the coefficient of variation weighted multi-parameter risk contribution deduction logic, can adaptively allocate dynamic weights based on the temporal fluctuation characteristics of each parameter, and can identify complex hidden dangers caused by small shifts in multiple parameters. Its ability to identify complex hidden dangers is significantly better than that of Comparative Example 1 and Comparative Example 2. The weights of the two sets of comparative examples are manually fixed values and cannot be adjusted with the fluctuation characteristics of the parameters, resulting in the omission of some composite deterioration samples. The bidirectional exponential moving median noise reduction logic effectively suppresses the sampling outliers caused by vehicle vibration and electromagnetic pulses. The false alarms generated by Example 1 are significantly less than those of the two sets of comparative examples. The dynamic weight adaptive optimization module iteratively updates the parameter weights based on historical labeled samples. It has small parameter drift when facing a small number of outlier labeled samples, and its identification ability continues to improve with the accumulation of samples. The weight parameters of Comparative Example 1 and Comparative Example 2 are completely manually set and do not have self-optimization capabilities. Comparative Example 1 requires uploading all original sampling data, resulting in high network transmission overhead for a single round of inspection. The vehicle-mounted edge computing module of Example 1 completes the main calculations and only transmits back the judgment results and alarms, resulting in lower network load.
[0038] A simulation experiment of Ethernet cable disconnection was conducted. Example 2 was in offline autonomous mode, and it completed sampling, bidirectional time-series noise reduction, risk contribution deduction by coefficient of variation weighting, and local audible and visual alarms, realizing a complete local service closed loop. Comparative Example 1 completely lost its risk assessment and alarm output capabilities after the network was disconnected. Although Comparative Example 2 could output alarms after the network was disconnected, it used manual fixed weights, and its judgment adaptability was poor when the parameters fluctuated.
[0039] A dual-robot distributed collaborative monitoring experiment was conducted. In Example 3, the multi-robot collaborative scheduling module was used to split the monitoring points, and the vehicle-mounted edge computing module interacted with the point occupancy status to eliminate duplicate point collection. The global report merging module automatically completed the fusion of multi-source data and output a global inspection report. Comparative Examples 1 and 2 did not have point avoidance logic. During the test, the same point was repeatedly collected by the two robots, resulting in invalid collection overhead, which lengthened the overall inspection time. In addition, multiple inspection data sets needed to be manually integrated, and a global report could not be automatically output.
[0040] Based on the results of multiple rounds of comparative experiments, the system of this invention can adapt to various working conditions such as normal network, offline network disconnection, and multi-machine collaboration; it has the ability to suppress vehicle vibration pulse outliers, and can adaptively assign weights based on parameter temporal fluctuations to identify complex environmental hazards formed by the superposition of multiple parameters. The risk assessment parameters support sample-driven adaptive optimization, which is more in line with the high-reliability environmental monitoring engineering requirements of the electric prefabricated cabin rail patrol robot.
Claims
1. An environmental monitoring system for a track-mounted robot, characterized in that: It includes the main body of the rail-mounted patrol robot, the multi-source environmental perception and acquisition kit, the vehicle-mounted edge computing module, and the remote control backend of the site; the main body of the rail-mounted patrol robot is equipped with the Siling Diana7 robotic arm, which can walk and stop along the galvanized rack and pinion track in the prefabricated cabin; the multi-source environmental perception and acquisition kit is deployed on the robot body to simultaneously collect raw environmental observation data of multiple dimensions such as temperature, humidity, sulfur hexafluoride, oxygen, and noise. The vehicle-mounted edge computing module is deployed on the robot's onboard industrial control hardware, completing the cleaning of raw observation data, multi-parameter coupled environmental risk simulation, and local audible and visual alarm signal output. The remote control backend at the site uses a wired Ethernet cable to complete data interaction with the vehicle-mounted edge computing module, enabling the issuance of inspection tasks, reception of monitoring and alarm data, and storage of historical archives. The robot traverses preset points along the track to complete environmental data collection. The vehicle-mounted edge computing module can independently conduct monitoring and analysis without the remote control backend at the site, building a vehicle-mounted... A two-tiered, collaborative environmental monitoring system for prefabricated power cabins.
2. The environmental monitoring system for a track-mounted robot according to claim 1, characterized in that, The risk contribution derivation logic of the weighted multi-parameter risk factor based on the coefficient of variation of the vehicle edge computing module is carried out according to the following processing steps: S1. Perform statistical analysis on the time-series sampling sequences of each environmental parameter, calculate the time-series mean and standard deviation of each parameter, and calculate the dynamic weight of each parameter based on the coefficient of variation. The more drastic the temporal fluctuations of the parameters, the higher the dynamic weights assigned. S2. Perform normalized risk mapping on each effective environmental parameter, and output 0. Risk characterization value in interval 1; Read the sensor hardware self-test status code, identify the faulty measurement point, and reset the dynamic weight of the measurement point to zero, so that it does not participate in subsequent risk calculation; S3. Combine the dynamic weights of each parameter with the risk characterization value to perform weighted aggregation and obtain the comprehensive environmental risk assessment result; S4. Compare the comprehensive environmental risk assessment results with the preset risk thresholds. If the assessment results exceed the thresholds, output a comprehensive environmental hazard risk warning signal.
3. The environmental monitoring system for the track-mounted robot according to claim 1, characterized in that, The multi-source environmental sensing and acquisition kit includes an NDIR-based sulfur hexafluoride sensor, an oxygen detection module, a digital temperature and humidity sensor, and an A-weighted noise sensor; the output messages of each sensor carry a unified timestamp and a hardware self-test status code. The vehicle-mounted edge computing module reads the self-test status code to identify sensor offline and zero-point drift faults, performs masking marking on the faulty measurement points, and resets the dynamic weight of the masked measurement points to zero, so that they no longer participate in the multi-parameter risk contribution inference and processing process.
4. The environmental monitoring system for the track-mounted robot according to claim 1, characterized in that, The vehicle-mounted edge computing module is configured with bidirectional exponentially moving median noise reduction processing logic, and performs the following noise reduction processing steps: S1. Take the original sampled data sequence within the bidirectional sliding window before and after the current sampling time; calculate the median of the forward window and backward window sequences respectively; S2. Perform exponential smoothing on the forward and backward median results respectively to obtain the forward smoothed output and the backward smoothed output; S3. Average the forward smoothing output and the backward smoothing output to output the final noise-reduced observation data. This processing logic simultaneously utilizes bidirectional timing information to suppress residual vibrations during robot docking and spike outlier interference from instantaneous electromagnetic pulses at the site, outputting high-quality observation sequences for use in the risk simulation phase.
5. The environmental monitoring system for a track-mounted robot according to claim 1, characterized in that, The remote management and control backend of the site configures the environmental lifecycle archive module; the environmental lifecycle archive module persistently stores the time-series historical dataset of each inspection point, supports multi-dimensional conditional retrieval by site number, time range, and monitoring parameters, and supports the export of inspection statistical reports and alarm event reports; it receives alarm events reported by the vehicle edge computing module and realizes two-level alarm push: WEB pop-up and in-site message.
6. The environmental monitoring system for a track-mounted robot according to claim 1, characterized in that, The system has an offline autonomous operation mode; when the wired Ethernet communication link of the tow cable is interrupted, the rail-mounted patrol robot continues to execute the local preset inspection plan, and the multi-source environmental perception and acquisition kit continues to carry out environmental sampling. The vehicle-mounted edge computing module independently completes bidirectional exponential moving median noise reduction, coefficient of variation weighted multi-parameter risk contribution estimation, and local audible and visual alarms; all original monitoring and alarm records are cached on the vehicle-mounted local storage medium; after the communication link is restored, the cached data is automatically and batch-transmitted to the remote management and control backend of the site.
7. The environmental monitoring system for a track-mounted robot according to claim 1, characterized in that, The rack and pinion servo drive of the rail-mounted patrol robot has a parking repeatability accuracy of ≤±2mm; after the robot completes the point braking, an editable static delay parameter is configured, and the sensor sampling is only started after the static delay timer ends, thereby avoiding the sampling distortion problem caused by the residual mechanical vibration of the vehicle body.
8. The environmental monitoring system for a track-mounted robot according to claim 2, characterized in that, The remote management backend of the station is configured with a dynamic weight adaptive optimization module, which performs the following parameter iterative optimization steps: S1. Retrieve the historical labeled sample set with manual risk labels stored in the background, and read the dynamic weights of each parameter in the current version; S2. Input historical samples into the risk deduction process to obtain the sample prediction risk results, compare the prediction results with the manual real labels for error, and use robust loss to calculate the overall sample error. Add regularization constraints to the weight parameters to suppress parameter overfitting and parameter shifts caused by outlier labeled samples; S3. Update the dynamic weights of each environmental parameter according to the gradient descent direction; S4. Determine if the iteration has reached the convergence termination condition; if it has not converged, return to S2 to continue the iteration. After iterative convergence, the updated set of dynamic weights will be distributed and synchronized to the vehicle edge computing module for subsequent location risk assessment.
9. According to claim 1 The environmental monitoring system for the rail-mounted robot according to any one of the following eight claims is characterized in that, The remote control backend of the station is configured with a multi-machine collaborative scheduling module; the multi-machine collaborative scheduling module divides the inspection points for multiple rail-mounted patrol robots, generates non-overlapping sub-tasks and sends them to each robot; the on-board edge computing module of each robot interacts with the occupancy status of the points based on the station's local area network to avoid multiple robots sampling the same monitoring point repeatedly.
10. The environmental monitoring system for a track-mounted robot according to claim 9, characterized in that, The remote control backend of the site is configured with a full-domain report merging module; the full-domain report merging module receives monitoring datasets transmitted from multiple robots, completes the fusion and merging of multi-source monitoring data, and automatically outputs a complete prefabricated cabin full-domain environmental inspection report. The report supports online preview and local file export.
