Encoder-RFID positioning inspection robot control method, system, equipment and medium

Abstract

The invention relates to an encoder-RFID positioning inspection robot control method, system and device and a medium. The method comprises the following steps: firstly, acquiring a multi-view image, temperature and environmental gas concentration data of pump room equipment, performing time-space synchronization calibration by combining encoder-RFID positioning data, and fusing into multi-source preliminary fusion data; secondly, robot track data are extracted, the real-time position is calculated, RFID calibration is carried out, and features are extracted and integrated into a comprehensive data set in combination with historical data; analyzing the comprehensive data set to determine a normal mode of the equipment, and comparing parameters to generate an abnormal identifier and a fault early warning signal; and finally, wirelessly transmitting an early warning signal, receiving an instruction, adjusting the state of the robot, returning data and generating monitoring data. By adopting the method, the monitoring and positioning precision can be improved, faults can be early warned, the inspection efficiency is improved, and the operation risk is reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot control, and particularly relates to an encoder-RFID positioning and inspection robot control method, system, device and medium. BACKGROUND

[0002] In the industrial fields of coal mines, chemical industries and the like, a pump house is a core equipment operation site, and the safe and stable operation of the equipment is crucial to production operations. The traditional manual inspection method has problems of low efficiency, large error, poor real-time performance and high risk of operation in a dangerous environment, and cannot meet the demand of modern industry for intelligent management of equipment. Although the existing inspection robot technology can realize partial automatic inspection, in the aspects of multi-source data acquisition and fusion, there are problems of non-uniformity of data space-time reference and low fusion accuracy, resulting in lack of accuracy in analysis of the equipment state; in the aspect of robot positioning, a single positioning technology is susceptible to environmental interference, and the positioning accuracy cannot meet the demand of accurate operation under complex working conditions; in the aspect of equipment fault early warning, simple threshold judgment is mainly relied on, potential fault risks are difficult to dig out, and early warning and accurate prediction cannot be realized. SUMMARY

[0003] Therefore, it is necessary to provide an encoder-RFID positioning and inspection robot control method, system, device and medium capable of realizing efficient fusion of multi-source data, accurate positioning and intelligent fault early warning, and improving the intelligent operation and maintenance level of industrial equipment.

[0004] In a first aspect, the application provides an encoder-RFID positioning and inspection robot control method, comprising:

[0005] Multi-view image data, equipment temperature data and environmental gas concentration data of equipment in a pump house are acquired, and the data is unified to the same space-time reference through space-time synchronous calibration technology to obtain multi-source preliminary fusion data.

[0006] The position of an inspection robot on a track is determined based on the multi-source preliminary fusion data by using encoder and RFID positioning technology, and a comprehensive data set is obtained by combining feature extraction of historical multi-source data.

[0007] Feature analysis is performed on the comprehensive data set by using a multi-dimensional feature extraction algorithm, an abnormal state of equipment operation is judged according to feature correlation, and a preset early warning signal generation rule library is called to generate an equipment fault early warning signal.

[0008] The wireless mobile communication technology is adopted to transmit the equipment fault early warning signal to the ground, and the instructions of the ground workstation are received to drive the motion control of the inspection robot, and remote monitoring data is generated.

[0009] In one embodiment, data is unified to the same spatiotemporal reference using spatiotemporal synchronization calibration technology to obtain preliminary multi-source fused data, including:

[0010] Acquire multi-view image data of the equipment in the pump house; the multi-view image data includes all-round images covering the overall operating condition of the equipment and local detailed images focusing on key components of the equipment.

[0011] Based on the encoder-RFID positioning data of the inspection robot and combined with multi-view image data of each frame, image processing technology is used to perform spatiotemporal synchronization calibration on the multi-view image data to obtain an image dataset with a unified spatiotemporal reference; the unified spatiotemporal reference is based on the robot's orbit coordinates and the collection timestamp.

[0012] The image dataset is processed using edge detection and corner detection algorithms to extract device feature points, including device outlines, component dimensions, and instrument scales. Combined with the spatiotemporal coordinate information of each device feature point, feature point cloud data containing the three-dimensional spatial features of the device is generated.

[0013] The inspection robot uses an infrared thermal imager to collect real-time temperature data of the equipment in the pump room, and records the spatiotemporal coordinates of each temperature data point. The equipment in the pump room includes key heat-generating components such as motor shafts, reducers, and the front and rear shafts of the water pumps.

[0014] The inspection robot uses methane and smoke sensors to collect ambient gas concentration data in the pump room and records the spatiotemporal coordinates of each gas concentration data collection.

[0015] Using a unified spatiotemporal benchmark as a reference, spatiotemporal synchronization calibration technology is adopted to combine the collected spatiotemporal coordinates to correlate and fuse the equipment temperature data and ambient gas concentration data with the generated feature point cloud data, thereby obtaining multi-source preliminary fused data that includes equipment spatial features, temperature status, and ambient gas status.

[0016] In one embodiment, the position of the inspection robot on the track is determined using encoder and RFID positioning technology based on preliminary multi-source fusion data. A comprehensive dataset is obtained by combining features extracted from historical multi-source data, including:

[0017] The inspection robot's trajectory data is extracted from the multi-source preliminary fusion data; the trajectory data includes the timestamps corresponding to each sampling time and the speed vector of the robot moving along the track.

[0018] Based on the principle of encoder pulse counting, the running trajectory data is combined with the preset track length parameter to calculate the mapping relationship between the number of encoder pulses per unit time and the track displacement, and the real-time position coordinates of the inspection robot on the track are obtained.

[0019] The fixed coordinates of the preset RFID tags along the track are read, and the real-time position coordinates calculated by the encoder are compared with the fixed coordinates. If the deviation exceeds the preset threshold, the real-time position coordinates are corrected based on the RFID tag coordinates, and a corrected position dataset is generated.

[0020] Acquire historical multi-source data; historical multi-source data includes pump room environmental parameters and robot operating status.

[0021] The location fluctuation amplitude and average movement speed are extracted from the corrected location dataset. The mean environmental parameters, peak equipment temperature, and robot failure frequency are extracted from historical multi-source data. The features are aligned by time dimension and then integrated to obtain a comprehensive dataset containing spatiotemporal features, environmental features, and equipment status features.

[0022] In one embodiment, the mapping relationship between the number of encoder pulses per unit time and the track displacement is expressed by the following formula:

[0023]

[0024] Among them, S t N represents the orbital displacement per unit time t. t The encoder pulse count is represented by L, the standard length of a single track segment is represented by θ, the actual tilt angle of the track segment where the robot is located is represented by P0, and the standard total number of encoder pulses is represented by δ. T This represents the correction factor for the encoder pulse output being affected by ambient temperature, where ΔT represents the difference between the actual ambient temperature and the standard calibration temperature within a unit time t.

[0025] In one embodiment, a multi-dimensional feature extraction algorithm is used to perform feature analysis on the comprehensive dataset. Based on feature correlation, an abnormal operating state of the equipment is determined. A preset early warning signal generation rule base is invoked to generate a device fault early warning signal, including:

[0026] Extract device operating status data from the comprehensive dataset; device operating status data includes device parameters and historical records.

[0027] Multi-dimensional feature extraction algorithms are used to perform time-series trend analysis on equipment operation status data, and deviation calculations are performed on the inherent parameters and real-time parameters of the equipment to obtain equipment operation status features that include parameter trend features, deviation features, and statistical features.

[0028] By analyzing the temporal correlation between equipment operating status characteristics and historical faults using time-series correlation mining algorithms, and classifying historical data with similar operating characteristics using clustering analysis algorithms, the normal operating modes of the equipment under different operating conditions are determined; the normal operating modes include high-load operating mode and low-load operating mode.

[0029] The system compares the characteristic parameters of the current operating mode of the equipment with the preset normal operating threshold. If a certain characteristic parameter exceeds the normal threshold range, an abnormal status identifier is generated. The abnormal status identifier includes the name of the abnormal parameter, the current value, the deviation from the threshold, and the timestamp of the abnormality. The normal operating threshold is set based on the range of characteristic parameters of the normal operating mode, combined with industry safety standards and equipment factory requirements.

[0030] The system calls a preset warning signal generation rule base based on the abnormal status identifier to generate a device fault warning signal. The device fault warning signal must carry the device number, abnormal identifier ID, and warning generation time. The warning signal generation rule base contains the mapping relationship between different abnormal identifiers and corresponding warning signals.

[0031] In one embodiment, after generating the device fault warning signal, the method further includes:

[0032] Extract dynamic feature factors related to equipment operating status from equipment fault early warning signals; dynamic feature factors include the rate of change of equipment parameters, environmental interference factors, and historical fault correlation factors of equipment during the period of abnormality occurrence.

[0033] The random forest algorithm is used to classify dynamic feature factors, and the currently extracted dynamic feature factors are input into the constructed fault classification model to output the equipment fault risk level. The fault classification model is trained using dynamic feature factors in historical fault data and known fault types.

[0034] High-risk factors are extracted from the equipment failure risk level, and combined with the original equipment parameter data and abnormal status identification information corresponding to the high-risk factors, failure prediction data is generated by integrating them according to the preset data format. The high-risk factors are dynamic feature factors whose contribution to the failure risk level determination exceeds the preset threshold.

[0035] The fault prediction data is input into the fault probability calculation model built based on the Bayesian probability algorithm to calculate the fault occurrence probability distribution. The warning priority is set according to the probability distribution and the fault impact range to obtain the fault warning priority. The fault probability calculation model uses the correlation probability between historical high-risk factors and fault occurrence as training samples.

[0036] In one embodiment, wireless mobile communication technology is used to transmit equipment fault warning signals to the ground, while simultaneously receiving instructions from the ground workstation to drive the motion control of the inspection robot and generate remote monitoring data, including:

[0037] The equipment fault warning signal information is sent to the ground workstation through wireless mobile communication technology, and the command data of the ground workstation is obtained.

[0038] Based on the command data, the inspection robot and the ground workstation can communicate in two directions to obtain voice interaction data.

[0039] Voice interaction data is processed using a speech recognition algorithm to generate robot control commands.

[0040] Adjust the operating status of the inspection robot according to the robot control instructions to obtain operating status data.

[0041] Operational status data is transmitted to the ground workstation via wireless mobile communication technology to obtain feedback control signals.

[0042] Based on the feedback control signal, the current operating status data, and the distribution coordinates of the pump room equipment, the A* path planning algorithm is used to optimize the inspection robot's path and generate optimized path data; the distribution coordinates of the pump room equipment are pre-stored in the robot's local map.

[0043] The motion control of the inspection robot is driven by optimized path data to obtain real-time positioning data.

[0044] Real-time location data is transmitted back to the ground workstation via wireless mobile communication technology to generate remote monitoring data.

[0045] Secondly, this application also provides an encoder-RFID positioning inspection robot control system, the system including:

[0046] The multi-source data fusion module is used to acquire multi-view image data, equipment temperature data, and ambient gas concentration data of equipment in the pump room. The data is unified to the same spatiotemporal reference through spatiotemporal synchronization calibration technology to obtain preliminary multi-source fused data.

[0047] The robot positioning module is used to determine the position of the inspection robot on the track based on the preliminary fusion of multi-source data and the use of encoder and RFID positioning technology. It also integrates features extracted from historical multi-source data to obtain a comprehensive dataset.

[0048] The equipment anomaly analysis module is used to perform feature analysis on a comprehensive dataset using a multi-dimensional feature extraction algorithm, determine the abnormal operating status of the equipment based on feature correlation, and generate equipment fault warning signals by calling a preset warning signal generation rule base.

[0049] The robot motion control module is used to transmit equipment fault warning signals to the ground using wireless mobile communication technology, and at the same time receive instructions from the ground workstation to drive the inspection robot's motion control and generate remote monitoring data.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0052] The aforementioned encoder-RFID positioning inspection robot control method, system, device, computer equipment, and storage medium first acquire multi-view image data (including omnidirectional images of the overall equipment operating condition and detailed images of key components), equipment temperature data (key heat-generating components such as motor shafts and reducers), and ambient gas concentration data of the equipment in the pumping station using visible light cameras, infrared thermal imagers, methane sensors, and smoke sensors mounted on the inspection robot. Based on encoder-RFID positioning data and image processing technology, the multi-source data is spatiotemporally synchronized and calibrated. Using the robot's trajectory coordinates and acquisition timestamps as a unified benchmark, the image data, temperature data, gas concentration data, and equipment feature point cloud data are correlated and fused to form preliminary multi-source fused data. Second, robot trajectory data (timestamps and velocity vectors) is extracted from the preliminary multi-source fused data. Combining the encoder pulse counting principle and the preset track length parameters, real-time position coordinates are calculated, and deviation calibration is performed using fixed coordinates via RFID tags to generate a corrected position dataset. Simultaneously, historical pumping station environmental parameters and robot operating status data are acquired, and features such as position fluctuation amplitude and average environmental parameters are extracted. These features are then integrated with the corrected position data along the time dimension to form a comprehensive dataset. This method utilizes multi-dimensional feature extraction algorithms to analyze a comprehensive dataset. Through time-series trend analysis, deviation calculation, and historical fault correlation mining, it determines the normal operating mode of the equipment. Current operating parameters are compared with preset thresholds; when thresholds are exceeded, an anomaly identifier is generated, and a fault warning signal is generated by calling a rule base. Finally, the warning signal is transmitted to a ground workstation via wireless mobile communication technology. Upon receiving instructions, the workstation drives the robot to adjust its operating status, optimize path planning, and transmit real-time positioning data to generate remote monitoring data. This method solves the analysis error problem caused by inconsistent data benchmarks in traditional inspections through multi-source data spatiotemporal synchronous fusion technology, improving the accuracy of equipment status monitoring. Combined with encoder-RFID dual positioning technology, it improves the robot's track positioning accuracy, meeting the precise inspection needs of complex pump room environments. Based on multi-dimensional feature extraction and intelligent analysis algorithms, it overcomes the limitations of traditional threshold judgment, predicting potential equipment faults 2-4 hours in advance with a fault warning accuracy rate of over 90%. Through wireless communication and remote control technologies, it improves the efficiency of manual inspections and significantly reduces the risks of manual operations in high-risk environments. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart of the encoder-RFID positioning inspection robot control method provided in an embodiment of the present invention;

[0055] Figure 2 This is a structural block diagram of the encoder-RFID positioning inspection robot control system provided in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of the hardware circuit layout design of the intrinsically safe inspection robot for mining provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, such as Figure 1 As shown, this application provides a control method for an encoder-RFID positioning inspection robot, which may include the following steps:

[0059] Step S101: Acquire multi-view image data, equipment temperature data, and ambient gas concentration data of the equipment in the pump room. Unify the data to the same spatiotemporal reference through spatiotemporal synchronization calibration technology to obtain preliminary multi-source fusion data.

[0060] Specifically, the inspection robot uses an intrinsically safe visible light camera (resolution ≥1920×1080) to collect multi-view images of the equipment, including omnidirectional images covering the overall operating conditions and detailed images of key components such as valve stems and instruments. An infrared thermal imager (temperature range -20℃~+150℃) is used to collect real-time temperature data of heat-generating components such as motor shafts and the front and rear shafts of water pumps. Ambient gas concentration data is obtained through a methane sensor (detection range 0-100% LEL) and a smoke sensor. Based on the spatiotemporal coordinates (track coordinates + timestamp) provided by encoder-RFID positioning, feature point matching and timestamp alignment techniques are used to perform spatiotemporal synchronous calibration of the multi-source data. This unifies image data, temperature data, and gas concentration data into a unified framework with the robot's track coordinate system as the spatial reference and the acquisition time as the temporal reference, generating preliminary multi-source fusion data that includes equipment spatial characteristics, temperature status, and environmental parameters.

[0061] Step S102: Based on the preliminary fusion data from multiple sources, the position of the inspection robot on the track is determined using encoder and RFID positioning technology. The feature is extracted and integrated with historical multi-source data to obtain a comprehensive dataset.

[0062] Specifically, robot trajectory data containing timestamps (accurate to milliseconds) and velocity vectors (range 0-0.31 m / s) is extracted from the multi-source preliminary fusion data. Based on the encoder pulse counting principle and combined with the standard length of a single track segment (e.g., 6000 mm for an 8# I-beam track), real-time position coordinates (accuracy ±2 mm) are calculated using a formula. The fixed coordinates of RFID tags along the track (spacing ≤ 10 meters, coordinate accuracy ±1 mm) are read and compared with the coordinates calculated by the encoder. If the deviation exceeds 3 mm, the RFID coordinates are used for correction, generating a corrected position dataset. Simultaneously, historical pump room environmental parameters (humidity, illuminance, etc. for the past 3 months) and robot operating status data are acquired. Features such as position fluctuation amplitude and average environmental parameters are extracted from these data and integrated with the corrected position data by timestamp alignment, forming a comprehensive dataset containing spatiotemporal features, environmental features, and equipment status features.

[0063] Step S103: Use a multi-dimensional feature extraction algorithm to perform feature analysis on the comprehensive dataset, determine the abnormal operating status of the equipment based on feature association, and call the preset warning signal generation rule base to generate equipment fault warning signals.

[0064] Specifically, a multi-dimensional feature extraction algorithm is used to analyze the comprehensive dataset: a 1-hour sliding window time-series trend analysis is performed on the dynamic parameters of the equipment (such as temperature and vibration) to calculate the slope of change and the amplitude of fluctuation; the deviation rate between the inherent parameters (such as rated speed) and real-time parameters is calculated (deviation rate = (real-time value - rated value) / rated value × 100%); and the monthly average number of abnormalities in historical records is statistically analyzed to obtain three types of features: parameter trend, deviation, and statistics. A time-series correlation mining algorithm is used to analyze the temporal correlation between features and historical faults (such as the probability of faults 2 hours after the temperature exceeds the threshold), and a clustering algorithm (K-means, K = 2) is used to determine the high-load (power ≥ 80% of rated value) and low-load (power ≤ 50% of rated value) operating modes. The current mode feature parameters are compared with thresholds set based on industry standards (such as the "Technical Specifications for Maintenance of Coal Mine Electromechanical Equipment") and equipment factory requirements. When the threshold is exceeded, an anomaly identifier containing the abnormal parameter name, current value, deviation range, and timestamp is generated. The rule base of the pre-stored mapping relationship between anomaly identifiers and warning signals is called to generate a fault warning signal carrying the equipment number, anomaly ID, and generation time.

[0065] Step S104: The equipment fault warning signal is transmitted to the ground using wireless mobile communication technology. At the same time, the instructions from the ground workstation are received to drive the motion control of the inspection robot and generate remote monitoring data.

[0066] By adapting to wireless mobile communication technology in coal mines (supporting 200-meter base station coverage, TCP / IP protocol), fault warning signals (including priority and probability distribution) are transmitted to the ground workstation. Upon receiving command data, two-way voice communication is initiated, and control commands (such as adjusting speed and locking detection points) are generated using a CNN-LSTM speech recognition algorithm (accuracy ≥95%). Based on these commands, the robot's operating status (speed, mode, etc.) is adjusted, and the status data is transmitted back to the ground to obtain feedback control signals. The inspection path is optimized based on the A* path planning algorithm (pre-stored pump room equipment coordinates), avoiding areas with abnormal equipment. The robot is driven to move along the optimized path, and real-time positioning data (coordinate accuracy ±5mm) is collected via encoder-RFID. After being transmitted back to the ground workstation, this data is integrated to generate remote monitoring data including robot position, operating status, and fault warning progress, achieving full-process visualized management and control.

[0067] The aforementioned encoder-RFID positioning inspection robot control method first acquires multi-view image data (including omnidirectional images of the overall equipment operating condition and detailed images of key components), equipment temperature data (motor shafts, reducers, and other key heat-generating components), and ambient gas concentration data of the equipment within the pumping station using visible light cameras, infrared thermal imagers, methane sensors, and smoke sensors mounted on the inspection robot. Based on encoder-RFID positioning data and image processing technology, the multi-source data is spatiotemporally synchronized and calibrated. Using the robot's trajectory coordinates and acquisition timestamps as a unified benchmark, the image data, temperature data, gas concentration data, and equipment feature point cloud data are correlated and fused to form preliminary multi-source fused data. Secondly, robot trajectory data (timestamps and velocity vectors) is extracted from the preliminary multi-source fused data. Combining the encoder pulse counting principle and the preset track length parameters, real-time position coordinates are calculated, and deviation calibration is performed using fixed coordinates via RFID tags to generate a corrected position dataset. Simultaneously, historical pumping station environmental parameters and robot operating status data are acquired, and features such as position fluctuation amplitude and average environmental parameters are extracted. These features are then integrated with the corrected position data along the time dimension to form a comprehensive dataset. This method utilizes multi-dimensional feature extraction algorithms to analyze a comprehensive dataset. Through time-series trend analysis, deviation calculation, and historical fault correlation mining, it determines the normal operating mode of the equipment. Current operating parameters are compared with preset thresholds; when thresholds are exceeded, an anomaly identifier is generated, and a fault warning signal is generated by calling a rule base. Finally, the warning signal is transmitted to a ground workstation via wireless mobile communication technology. Upon receiving instructions, the workstation drives the robot to adjust its operating status, optimize path planning, and transmit real-time positioning data to generate remote monitoring data. This method solves the analysis error problem caused by inconsistent data benchmarks in traditional inspections through multi-source data spatiotemporal synchronous fusion technology, improving the accuracy of equipment status monitoring. Combined with encoder-RFID dual positioning technology, it improves the robot's track positioning accuracy, meeting the precise inspection needs of complex pump room environments. Based on multi-dimensional feature extraction and intelligent analysis algorithms, it overcomes the limitations of traditional threshold judgment, predicting potential equipment faults 2-4 hours in advance with a fault warning accuracy rate of over 90%. Through wireless communication and remote control technologies, it improves the efficiency of manual inspections and significantly reduces the risks of manual operations in high-risk environments.

[0068] In one embodiment, unifying data to the same spatiotemporal reference using spatiotemporal synchronization calibration technology to obtain preliminary multi-source fused data may include the following steps:

[0069] Step S201: Acquire multi-view image data of the equipment in the pump room; the multi-view image data includes an all-round image covering the overall operating condition of the equipment and a local detail image focusing on the key components of the equipment.

[0070] Step S202: Based on the encoder-RFID positioning data of the inspection robot and the multi-view image data of each frame, image processing technology is used to perform spatiotemporal synchronization calibration on the multi-view image data to obtain an image dataset with a unified spatiotemporal reference; the unified spatiotemporal reference is based on the robot's orbit coordinates and the collection timestamp.

[0071] Preferably, encoder-RFID positioning data is the key basis for determining the position of the inspection robot and the spatiotemporal reference of the data. The encoder, using pulse counting principles, converts the robot's motion information into pulse signals. Combined with preset track length parameters, it calculates the robot's displacement on the track, thus obtaining its real-time position coordinates. RFID positioning, by reading RFID tags deployed along the track (which store fixed coordinate information), calibrates the position calculated by the encoder. When the deviation between the two exceeds a preset threshold, the robot's position is corrected based on the RFID tag coordinates. The encoder provides continuous position change information, and RFID ensures positioning accuracy. Together, they provide accurate spatiotemporal coordinates (track coordinates and acquisition timestamps) for the acquisition of multi-view image data.

[0072] Image feature point matching algorithms (such as SIFT and SURF) are used to extract feature points from images from different perspectives and establish the correspondence between feature points in the images. Then, by combining the coordinate information in the encoder-RFID positioning data, each frame of image is mapped to a unified framework with the robot orbit coordinate system as the spatial reference and the acquisition time as the temporal reference, eliminating the spatiotemporal deviation caused by different shooting angles and acquisition times. Finally, by aligning the timestamps, the consistency of image data in the time dimension is ensured, thereby obtaining an image dataset with a unified spatiotemporal reference.

[0073] Step S203: Extract device feature points, including device outline, component size, and instrument scale, from the image dataset using edge detection and corner detection algorithms. Combine the spatiotemporal coordinate information of each device feature point to generate feature point cloud data containing the three-dimensional spatial features of the device.

[0074] Furthermore, for image datasets with a unified spatiotemporal reference, edge detection algorithms (such as the Canny algorithm) and corner detection algorithms (such as the Harris algorithm) are employed to extract key feature points such as the device's outline, component dimensions, and instrument scales. Edge detection algorithms can identify boundaries with drastic grayscale changes in the image, outlining the device's external contours; corner detection algorithms can locate corners with distinct features in the image, used to characterize key parts of components and details such as instrument scales. After extracting feature points, spatiotemporal coordinate information is assigned to each feature point in conjunction with encoder-RFID positioning data, extending the two-dimensional image feature points into three-dimensional space, constructing feature point cloud data containing the device's three-dimensional spatial features. This data not only records the device's spatial morphology but also preserves accurate spatiotemporal attributes.

[0075] Step S204: Use the infrared thermal imager mounted on the inspection robot to collect real-time temperature data of the equipment in the pump room, and record the spatiotemporal coordinates of each temperature data collection. The equipment in the pump room includes key heat-generating components such as motor shafts, reducers, and the front and rear shafts of the water pump.

[0076] Step S205: Collect ambient gas concentration data in the pump room using the methane sensor and smoke sensor mounted on the inspection robot, and record the spatiotemporal coordinates of each gas concentration data collection.

[0077] Step S206: Using a unified spatiotemporal benchmark as a reference, spatiotemporal synchronization calibration technology is adopted to combine the collected spatiotemporal coordinates to correlate and fuse the equipment temperature data and ambient gas concentration data with the generated feature point cloud data, respectively, to obtain multi-source preliminary fused data containing equipment spatial features, temperature status, and ambient gas status.

[0078] Preferably, the spatiotemporal synchronization calibration technology first performs spatial coordinate mapping and timestamp alignment on multi-view image data based on encoder-RFID positioning data to establish an image dataset with a unified spatiotemporal reference. Second, when collecting equipment temperature data and ambient gas concentration data, the spatiotemporal coordinates of the collection are recorded simultaneously. Finally, in the data fusion stage, using the unified spatiotemporal reference established by the image data as a reference, the temperature data, gas concentration data and feature point cloud data are associated. Through spatiotemporal coordinate matching, each temperature value and gas concentration value is mapped to a specific location in the three-dimensional space of the equipment, realizing accurate fusion of multi-source data in the same spatiotemporal dimension, ensuring data consistency and availability, and providing reliable data support for equipment status analysis and fault diagnosis.

[0079] Specifically, the inspection robot, equipped with a visible light camera, collects multi-view images of the equipment within the pumping station, encompassing comprehensive images reflecting the overall operating condition of the equipment as well as detailed images focusing on key components such as valves and instruments. Based on the track coordinates and timestamps output by the robot's encoder-RFID positioning system, image processing techniques such as feature point matching and timestamp alignment are employed to perform spatiotemporal synchronization calibration on the multi-view image data. This unifies each frame of the image into a framework with the track coordinate system as the spatial reference and the acquisition time as the temporal reference, forming an image dataset with a unified spatiotemporal reference. Furthermore, edge detection and corner detection algorithms are used to extract feature points such as equipment outlines, component dimensions, and instrument scales from the image dataset, and combined with their corresponding spatiotemporal coordinate information to construct feature point cloud data containing the three-dimensional spatial features of the equipment. Secondly, an infrared thermal imager is used to collect temperature data from key heat-generating components such as the motor shaft, reducer, and front and rear shafts of the water pump, simultaneously recording the spatiotemporal coordinates corresponding to each temperature data point. Methane and smoke sensors are used to acquire ambient gas concentration data in real time, and the corresponding spatiotemporal coordinates are marked. Finally, using the previously established unified spatiotemporal benchmark as a reference, the spatiotemporal synchronization calibration technology is used again to correlate and fuse temperature data, ambient gas concentration data and feature point cloud data, so that each parameter accurately corresponds to the specific location in the three-dimensional space of the equipment, and finally forms multi-source preliminary fused data containing equipment spatial features, temperature status and ambient gas status.

[0080] This embodiment's data acquisition and fusion method achieves precise alignment of multiple data types, including images, temperature, and gas concentration, by constructing a unified spatiotemporal reference. This avoids analysis errors caused by spatiotemporal misalignment in traditional multi-source data, improving the spatial positioning accuracy of equipment status monitoring to ±5mm and controlling the time synchronization error within ±100ms. Employing a fusion strategy combining feature point clouds and spatiotemporal calibration, it effectively integrates key information such as equipment spatial morphology, operating temperature, and environmental parameters, providing a complete and structured data foundation for subsequent equipment fault diagnosis. Compared to traditional single data acquisition methods, this method can comprehensively capture multidimensional features during equipment operation, supporting early identification of abnormal equipment states and improving fault warning accuracy by approximately 30%. It also provides data support for the precise positioning and intelligent path planning of inspection robots, significantly enhancing the automation and intelligence level of industrial equipment operation and maintenance.

[0081] In one embodiment, the position of the inspection robot on the track is determined using encoder and RFID positioning technology based on preliminary multi-source fusion data. A comprehensive dataset is then obtained by combining features extracted from historical multi-source data. This process may include the following steps:

[0082] Step S301: Extract the inspection robot's trajectory data from the multi-source preliminary fusion data; the trajectory data includes the timestamps corresponding to each sampling time and the speed vector of the robot moving along the track.

[0083] Step S302: Based on the pulse counting principle of the encoder, the running trajectory data is combined with the preset length parameter of the track to calculate the mapping relationship between the number of encoder pulses per unit time and the track displacement, and the real-time position coordinates of the inspection robot on the track are obtained.

[0084] Preferably, the encoder generates pulse signals through the rotation of an internal grating disk. Its pulse counting principle converts the robot's displacement along the track into a number of pulses. Specifically, when the robot moves on the track, it drives the encoder shaft to rotate. The light-transmitting and opaque areas on the grating disk alternately block the light source, causing the photosensitive element to generate pulse signals. The number of pulses per unit time is directly related to the robot's moving speed and the track length. By using a preset standard length for a single track segment (e.g., 6000mm for an 8# I-beam track segment), a mathematical mapping relationship between the number of pulses and the displacement can be established, thereby enabling the calculation of the robot's position.

[0085] Step S303: Read the fixed coordinates of the preset RFID tags along the track and compare the real-time position coordinates calculated by the encoder with the fixed coordinates. If the deviation exceeds the preset threshold, correct the real-time position coordinates based on the RFID tag coordinates and generate a corrected position dataset.

[0086] Furthermore, RFID tags are deployed along the track at intervals of ≤10 meters, with each tag storing fixed coordinate information (accuracy ±1mm). The real-time position coordinates calculated by the encoder are compared with the fixed coordinates of the RFID tags. If the deviation exceeds a preset threshold (e.g., 3mm), it is determined that the encoder positioning has a large error. In this case, the real-time position coordinates are corrected based on the RFID tag coordinates. This mechanism utilizes the high-precision characteristics of RFID positioning to compensate for the cumulative errors that may occur in encoder positioning due to factors such as mechanical wear and environmental interference. By dynamically calibrating and generating a corrected position dataset, the accuracy of the robot's position data is ensured.

[0087] Step S304: Obtain historical multi-source data; historical multi-source data includes pump room environmental parameters and robot operating status.

[0088] Step S305: Extract the position fluctuation amplitude and average movement speed from the corrected position dataset, extract the mean environmental parameters, peak equipment temperature, and robot failure frequency from historical multi-source data, align the features by time dimension and integrate them to obtain a comprehensive dataset containing spatiotemporal features, environmental features, and equipment status features.

[0089] Spatiotemporal features such as position fluctuation amplitude and average movement speed are extracted from the corrected position dataset to reflect the robot's motion stability and trajectory patterns on the track. Simultaneously, features such as average environmental parameters (e.g., average methane concentration), peak equipment temperature, and robot failure frequency are extracted from historical multi-source data (pump room environmental parameters and robot operating status). These features are aligned using timestamps, that is, position features, environmental features, and equipment status features at the same point in time are correlated and integrated to form a comprehensive dataset containing spatiotemporal, environmental, and equipment status features. This integration logic ensures consistency across different types of data in the temporal dimension.

[0090] First, the robot's trajectory data is extracted from the multi-source preliminary fusion data, including sampling timestamps accurate to milliseconds and velocity vectors along the track (velocity range 0-0.31 m / s). Based on the encoder pulse counting principle and combined with the standard length parameter of a single track segment (e.g., 6000 mm for an 8# I-beam track), the mapping relationship between the number of encoder pulses per unit time and track displacement is calculated using a formula to determine the robot's real-time position coordinates on the track (accuracy ±2 mm). Then, the fixed coordinates of preset RFID tags with a spacing ≤10 meters along the track (coordinate accuracy ±1 mm) are read and compared with the encoder-calculated coordinates. When the deviation exceeds a preset threshold of 3 mm, the real-time position is corrected based on the RFID tag coordinates, generating a corrected position dataset. Second, historical multi-source data covering nearly three months is acquired, including pump room environmental parameters (humidity, methane concentration, etc.) and robot operating status (historical trajectory, fault records). Spatiotemporal features such as position fluctuation amplitude and average movement speed are extracted from the corrected position dataset; data features such as the average environmental parameters, peak equipment temperature, and robot fault frequency are extracted from the historical data. By using a timestamp alignment mechanism, the above features are integrated along the time dimension to form a comprehensive dataset that includes spatiotemporal features, environmental features, and device status features.

[0091] This embodiment utilizes dual positioning technology combining encoders and RFID to improve robot trajectory positioning accuracy by 40%, effectively eliminating errors caused by environmental interference in single-encoder positioning and meeting the precise positioning requirements under complex pump room conditions. The data feature extraction and integration strategy cross-dimensionally correlates robot operating status, environmental parameters, and historical fault data, improving the completeness of equipment status analysis data by over 60%. The construction of the comprehensive dataset provides multi-dimensional data support for subsequent equipment anomaly diagnosis. Compared to traditional single-source data analysis, the accuracy of fault warnings is improved by 30%, while providing real-time position references for robot path planning, reducing the risk of blind spots in inspections due to positioning deviations, and significantly improving the automation and intelligence level of industrial equipment operation and maintenance.

[0092] In one embodiment, the mapping relationship between the number of encoder pulses per unit time and the track displacement can be expressed by the following formula:

[0093]

[0094] Among them, S t N represents the orbital displacement per unit time t. t The encoder pulse count is represented by L, the standard length of a single track segment is represented by θ, the actual tilt angle of the track segment where the robot is located is represented by P0, and the standard total number of encoder pulses is represented by δ. T This represents the correction factor for the encoder pulse output being affected by ambient temperature, where ΔT represents the difference between the actual ambient temperature and the standard calibration temperature within a unit time t.

[0095] Preferably, N t The encoder pulse count per unit time t refers to the total number of valid pulses output by the encoder per unit time.

[0096] The standard length of a single section of track L is the preset fixed length of a single section of the inspection robot's running track. For example, when using an 8# I-beam track, the length of a single section is 6000mm, which serves as the length benchmark parameter for displacement calculation.

[0097] The standard total pulse count of the P0 encoder is the total number of pulses output by the encoder when the robot passes through a single standard length track completely under standard operating conditions (25℃, horizontal track). It needs to be determined by factory and on-site calibration and is used as the standard coefficient for pulse-to-displacement conversion.

[0098] δ T The correction factor for the encoder pulse output affected by ambient temperature is a parameter that reflects the degree to which changes in ambient temperature affect the accuracy of the encoder pulse output. The unit is ℃. It is determined by the encoder technical parameters in combination with field calibration and is used to compensate for measurement errors caused by temperature.

[0099] This embodiment, by comprehensively considering actual influencing factors such as track tilt angle and ambient temperature, significantly improves the positioning accuracy of the inspection robot compared to the traditional displacement calculation method based solely on pulse counting. By introducing cosine correction for the tilt angle, projection errors in displacement calculation on inclined tracks are eliminated; and by combining the product of temperature coefficient and temperature deviation, the interference of ambient temperature changes on encoder pulse accuracy is effectively compensated, keeping the positioning error within ±0.5mm. This provides a more accurate foundation of robot position data for subsequent equipment status analysis, ensures the spatiotemporal consistency of multi-source data fusion, and thus improves the accuracy and reliability of equipment fault early warning, meeting the high-precision positioning requirements of inspection robots under complex industrial conditions.

[0100] In one embodiment, a multi-dimensional feature extraction algorithm is used to perform feature analysis on the comprehensive dataset. Based on feature correlation, the abnormal operating state of the equipment is determined, and a pre-set warning signal generation rule base is invoked to generate a device fault warning signal. This may include the following steps:

[0101] Step S401: Extract device operating status data from the comprehensive dataset; device operating status data includes device parameters and historical records.

[0102] Step S402: Use a multi-dimensional feature extraction algorithm to perform time-series trend analysis on the equipment operation status data, calculate the deviation between the inherent parameters and real-time parameters of the equipment, and obtain equipment operation status features that include parameter trend features, deviation features, and statistical features.

[0103] Step S403: Analyze the temporal correlation between equipment operating status characteristics and historical faults using a time-series correlation mining algorithm, and classify historical data with similar operating characteristics using a clustering analysis algorithm to determine the normal operating mode of the equipment under different operating conditions; the normal operating mode includes high-load operating mode and low-load operating mode.

[0104] Step S404: Compare the characteristic parameters of the current operating mode of the equipment with the preset normal operating threshold. If a certain characteristic parameter exceeds the normal threshold range, an abnormal status identifier is generated. The abnormal status identifier includes the abnormal parameter name, current value, deviation from the threshold, and abnormal occurrence timestamp. The normal operating threshold is set based on the range of characteristic parameters of the normal operating mode, combined with industry safety standards and equipment factory requirements.

[0105] Step S405: Based on the abnormal status identifier, call the preset warning signal generation rule library to generate a device fault warning signal; the device fault warning signal must carry the device number, abnormal identifier ID and warning generation time; the warning signal generation rule library contains the mapping relationship between different abnormal identifiers and corresponding warning signals.

[0106] First, equipment operating status data is extracted from a comprehensive dataset, covering inherent equipment parameters (such as motor rated power and pump rated speed), real-time parameters (temperature and vibration frequency), and historical records (normal operation data and fault maintenance logs for the past 3 months). A multi-dimensional feature extraction algorithm is used to perform time-series trend analysis on the equipment operating status data, calculating the slope of parameter changes through a 1-hour sliding window; simultaneously, the inherent equipment parameters and real-time parameters are compared to calculate the deviation rate. Combined with information such as the number of equipment anomalies in historical records, equipment operating status features including parameter trend characteristics, deviation characteristics, and statistical characteristics are obtained. Second, a time-series association mining algorithm is used to analyze the temporal correlation between equipment operating status features and historical faults, for example, identifying high-incidence periods of faults after the temperature has continuously exceeded the threshold for 2 hours. A clustering analysis algorithm (K-means, K=2) is used to classify historical data with similar operating characteristics, determining high-load operating modes (power ≥ 80% of rated value) and low-load operating modes (power ≤ 50% of rated value). The system compares the characteristic parameters of the current operating mode of the equipment with normal operating thresholds set based on industry standards (such as the "Maintenance Specifications for Coal Mine Electromechanical Equipment") and the equipment's factory requirements. When a threshold is exceeded, an abnormal status identifier is generated, which includes the abnormal parameter name, value, deviation range, and timestamp. Finally, based on the abnormal status identifier, a pre-stored mapping relationship-based early warning signal generation rule base is invoked to generate an equipment fault early warning signal carrying the equipment number, abnormal identifier ID, and generation time.

[0107] This embodiment achieves a comprehensive quantitative assessment of equipment operating status through multi-dimensional feature extraction and multi-algorithm collaborative analysis. Compared to traditional single-threshold judgment, multi-dimensional feature extraction incorporates equipment parameter trends and historical statistical patterns into the analysis, improving the accuracy of anomaly detection. The application of time-series correlation mining and clustering algorithms can accurately identify the normal operating mode and its boundary conditions of the equipment, effectively distinguishing between normal fluctuations and abnormal states, and reducing the false alarm rate. By combining industry standards and equipment characteristics to set thresholds and establishing standardized early warning signal mapping rules, the standardization and traceability of fault early warning are ensured, providing a reliable basis for preventive maintenance of industrial equipment, reducing downtime losses caused by sudden failures, and improving equipment operation and maintenance efficiency and safety.

[0108] In one embodiment, after generating the device fault warning signal, the following steps may also be included:

[0109] Step S501: Extract dynamic feature factors related to the equipment operating status from the equipment fault early warning signal; the dynamic feature factors include the rate of change of equipment parameters, environmental interference factors, and historical fault correlation factors of the equipment during the abnormality occurrence period.

[0110] Preferably, the dynamic characteristic factor is a quantitative parameter extracted from the equipment fault early warning signal that is directly related to the equipment's operating status, used to accurately depict the dynamic change pattern before and after the fault occurs. Among them, the rate of change of equipment parameters during the abnormal occurrence period (such as the hourly rate of temperature rise and pressure fluctuation gradient) reflects the deterioration trend of the equipment status; environmental interference factors (such as the amplitude of methane concentration fluctuation and humidity change rate) characterize the degree of influence of environmental factors on equipment operation; and equipment historical fault correlation factors (such as the number of similar abnormalities in the past six months and the fault occurrence rate under the same operating conditions) reflect the historical correlation of equipment faults.

[0111] Step S502: The random forest algorithm is used to classify the dynamic feature factors, and the currently extracted dynamic feature factors are input into the constructed fault classification model to output the equipment fault risk level; the fault classification model is trained using the dynamic feature factors in historical fault data and known fault types.

[0112] The random forest algorithm is used to classify the extracted dynamic feature factors. This algorithm constructs multiple decision trees (e.g., 100 trees), each trained based on randomly sampled samples and features, and finally outputs the classification result through a voting mechanism. Specifically, firstly, a fault classification model is trained using historical fault data (containing dynamic feature factors and corresponding fault types, such as motor overheating and valve leakage). The model establishes classification rules by learning the mapping relationship between feature factors and fault types. Then, the currently extracted dynamic feature factors are input into the trained model. The model outputs three levels of fault risk: "low," "medium," and "high," based on the feature matching degree, and also provides the confidence level for each level (e.g., 85% confidence for the high-risk level).

[0113] Step S503: Extract high-risk factors from the equipment failure risk level, and combine the original equipment parameter data and abnormal status identification information corresponding to the high-risk factors to generate failure prediction data according to the preset data format; the high-risk factors are dynamic feature factors whose contribution to the failure risk level determination exceeds the preset threshold.

[0114] High-risk factors are dynamic feature factors whose contribution to the fault risk level determination exceeds a preset threshold (e.g., 30%). Their selection is based on a feature importance assessment mechanism using a random forest algorithm (e.g., Gini coefficient gain, reduction in out-of-bag data error). The algorithm calculates the weight of each dynamic feature factor's impact on the risk level determination result, selecting factors with weights exceeding the threshold (e.g., "temperature change rate > 0.8℃ / min" contributes 42% to the high-risk level). Subsequently, the corresponding original sampled data of equipment parameters (e.g., original temperature curves, real-time pressure values) and abnormal status identification information (e.g., abnormal ID, occurrence time) are linked and integrated according to a preset format (e.g., "factor name - original value - timestamp - abnormal ID") to generate structured fault prediction data. This selection logic focuses on the key factors that have the greatest impact on fault risk.

[0115] Step S504: Input the fault prediction data into the fault probability calculation model constructed based on the Bayesian probability algorithm, calculate the fault occurrence probability distribution, set the early warning priority according to the probability distribution and the fault impact range, and obtain the fault early warning priority; the fault probability calculation model uses the correlation probability between historical high-risk factors and fault occurrence as training samples.

[0116] Furthermore, the model uses historical high-risk factors (such as "humidity > 90% and vibration frequency > 50Hz") and the probability of failure occurrence (e.g., a failure rate of 72% under this combination) as training samples to construct a prior probability distribution. After inputting current failure prediction data, the model updates the probability distribution using Bayes' theorem (posterior probability = prior probability × likelihood / evidence) to calculate the probability of failure occurrence at different future time points (e.g., 1 hour, 6 hours). Finally, it sets the warning priority based on the scope of failure impact (e.g., whether it causes the main system to shut down) (e.g., a probability > 80% and impact on the main system is a first-level priority). This algorithm effectively integrates historical experience with real-time data, maintaining high probability prediction accuracy even in small sample scenarios.

[0117] First, for the generated equipment fault warning signals, dynamic feature factors are extracted within one hour before and after the anomaly occurs. These include the rate of change of equipment parameters (such as the rate of temperature rise and the frequency of pressure fluctuations), environmental interference factors (changes in methane concentration and humidity fluctuations), and historical fault correlation factors (the number of similar anomalies in the past six months). A random forest algorithm is used to classify these dynamic feature factors. This algorithm constructs multiple decision trees and uses voting to make decisions. A fault classification model is trained using dynamic feature factors from historical fault data and known fault types (such as motor overheating and valve jamming) (sample size ≥ 1000 groups, cross-validation accuracy ≥ 90%). The current dynamic feature factors are input into the model, which outputs three levels of fault risk: "low," "medium," and "high," along with the probability of each level. Second, high-risk factors with a contribution exceeding 30% are extracted from the fault risk levels (contribution is calculated using the random forest feature importance evaluation function). These factors are then combined with the corresponding original sampled values ​​of equipment parameters and anomaly status identification information, and integrated into a fault prediction data set in the format of "equipment number-factor name-value-anomaly ID." The data is input into a failure probability calculation model based on a Bayesian probability algorithm. This model uses the correlation probability between historical high-risk factors and failure occurrence as training samples (e.g., "failure probability is 65% when temperature rise rate > 0.5℃ / min") to calculate the failure probability distribution for time nodes such as 1 hour and 24 hours in the future. Finally, warning priorities are set according to the probability distribution and the scope of failure impact (e.g., whether it affects the main drainage system), forming a hierarchical warning system with "high probability + high impact" corresponding to the first-level priority.

[0118] This embodiment overcomes the limitations of traditional threshold-based early warning by integrating dynamic feature factor extraction with intelligent algorithms, achieving quantitative assessment and accurate prediction of fault risks. The introduction of the random forest algorithm improves the accuracy of fault risk level classification, effectively distinguishing the severity of potential faults. The Bayesian probability model, combined with historical correlation probability and multi-factor fusion calculation, controls the fault occurrence probability prediction error within ±5%, improving prediction accuracy compared to traditional statistical methods. By screening high-risk factors based on contribution and constructing probability distributions, the early warning priority reflects both the probability of fault occurrence and optimizes the handling sequence based on the scope of impact, providing maintenance personnel with a scientific basis for decision-making. This improves fault handling response efficiency by 50%, while reducing unnecessary manual intervention and lowering industrial equipment maintenance costs and safety risks.

[0119] In one embodiment, wireless mobile communication technology is used to transmit equipment fault warning signals to the ground, while simultaneously receiving instructions from the ground workstation to drive the inspection robot's motion control and generate remote monitoring data. This may include the following steps:

[0120] Step S601: Send equipment fault warning signal information to the ground workstation via wireless mobile communication technology and obtain instruction data from the ground workstation.

[0121] Step S602: Based on the instruction data, realize two-way voice communication between the inspection robot and the ground workstation to obtain voice interaction data.

[0122] Step S603: Use a speech recognition algorithm to process the voice interaction data and generate robot control commands.

[0123] Step S604: Adjust the operating status of the inspection robot according to the robot control instructions to obtain operating status data.

[0124] Step S605: Transmit operational status data to the ground workstation via wireless mobile communication technology to obtain feedback control signals.

[0125] Step S606: Based on the feedback control signal, the current operating status data and the distribution coordinates of the pump room equipment are used to optimize the inspection robot path using the A* path planning algorithm to generate optimized path data; the distribution coordinates of the pump room equipment are pre-stored in the robot's local map.

[0126] Preferably, based on the control signals fed back by the ground workstation (including path adjustment instructions and key inspection area requirements), combined with the current operating status data of the inspection robot (real-time position coordinates, moving speed, and remaining power) and the pump room equipment distribution coordinates pre-stored in the local map (including the three-dimensional spatial coordinates of key equipment such as motors, water pumps, and valves, with an accuracy of ±5mm), the A* path planning algorithm is used for path optimization. The specific process is as follows: First, starting from the robot's current position and ending at the target detection point specified by the feedback control signal (such as the location of abnormal equipment), a grid map (50mm×50mm resolution) containing the equipment obstacle area (the spatial range occupied by the equipment entity) is constructed based on the local map. Second, the A* algorithm calculates the path cost using the evaluation function f(n) = g(n) + h(n) (g(n) is the actual cost from the starting point to the current node, and h(n) is the estimated cost from the current node to the destination, calculated using Manhattan distance), prioritizing the search for the path node with the lowest cost. Finally, during path generation, the robot automatically avoids equipment obstacle areas and marked abnormal equipment areas (such as the area within 1 meter around a faulty motor), generating optimized path data including waypoint coordinates, turning angles, and estimated arrival times, ensuring that the path meets the requirements of "shortest distance, obstacle avoidance safety, and coverage of key detection points." This process utilizes pre-stored equipment coordinates to achieve precise spatial positioning and, through the heuristic search characteristics of the A* algorithm, quickly generates the optimal inspection path in complex pump room environments, providing precise path guidance for robot motion control.

[0127] Step S607: The motion control of the inspection robot is driven by optimized path data to obtain real-time positioning data.

[0128] Step S608: Real-time positioning data is transmitted back to the ground workstation via wireless mobile communication technology to generate remote monitoring data.

[0129] First, using wireless mobile communication technology adapted to the underground coal mine environment (supporting 200-meter base station coverage and TCP / IP protocol transmission), equipment fault warning signals (including warning priority and abnormal parameter details) are sent to the ground workstation, and instruction data (control commands and interaction requests) are received from the workstation. Based on the instruction data, the inspection robot's voice interaction module is activated, establishing a two-way communication channel using a high-gain microphone and speaker, collecting voice interaction data (16kHz sampling rate, WAV format), and processing it using a CNN-LSTM-based speech recognition algorithm to generate structured robot control commands (such as speed adjustment and point detection). Second, the robot's operating status (speed, detection mode, etc.) is adjusted according to the control commands, and the operating status data (including real-time speed and task progress) is transmitted back to the ground workstation to obtain feedback control signals. Based on the feedback signals, combined with the pump room equipment distribution coordinates (accuracy ±5mm) pre-stored on the local map, the A* path planning algorithm is used to optimize the inspection path, avoiding abnormal equipment areas, and generating optimized path data including path node coordinates and estimated arrival times. The robot is driven to move along an optimized path. The encoder-RFID positioning system collects positioning data in real time (positioning accuracy ±3mm), and then transmits it back to the ground via wireless communication. The data is then integrated to generate remote monitoring data that includes the robot's position, operating status, and early warning processing progress.

[0130] This embodiment constructs a closed-loop remote control system, enabling the inspection robot to respond to ground commands with a delay of less than 0.5 seconds, thus improving fault handling efficiency. Two-way voice intercom and voice recognition technology achieve efficient human-machine interaction, avoiding operational deviations caused by misreading commands; the A* path planning algorithm, combined with real-time operating status and equipment distribution, improves inspection path planning efficiency and reduces repetitive inspection mileage. Through real-time data feedback via wireless communication, the ground workstation can obtain multi-dimensional information such as robot positioning and equipment warnings, forming visualized remote monitoring data, providing a basis for operation and maintenance decisions, effectively reducing the frequency of manual downhole operations, and improving the safety and intelligence level of equipment inspection in high-risk environments.

[0131] In one embodiment, such as Figure 2 As shown, this application also provides an encoder-RFID positioning inspection robot control system, which may include:

[0132] The multi-source data fusion module 701 is used to acquire multi-view image data, equipment temperature data, and ambient gas concentration data of equipment in the pump room. The data is unified to the same spatiotemporal reference through spatiotemporal synchronization calibration technology to obtain preliminary multi-source fusion data.

[0133] The robot positioning module 702 is used to determine the position of the inspection robot on the track based on the multi-source preliminary fusion data using encoder and RFID positioning technology, and to obtain a comprehensive dataset by combining features extracted from historical multi-source data.

[0134] The equipment anomaly analysis module 703 is used to perform feature analysis on a comprehensive dataset using a multi-dimensional feature extraction algorithm, determine the abnormal operating status of the equipment based on feature correlation, and generate equipment fault warning signals by calling a preset warning signal generation rule base.

[0135] The robot motion control module 704 is used to transmit equipment fault warning signals to the ground using wireless mobile communication technology, and at the same time receive instructions from the ground workstation to drive the inspection robot's motion control and generate remote monitoring data.

[0136] The aforementioned encoder-RFID positioning inspection robot control system utilizes a multi-source data fusion module. This module employs devices mounted on the inspection robot, such as a visible light camera, infrared thermal imager, and methane sensor, to collect multi-view image data, temperature data, and ambient gas concentration data from equipment within the pumping station. Using spatiotemporal synchronization calibration technology, and with the robot's trajectory coordinates and acquisition timestamps as a benchmark, the various data types are unified into a common spatiotemporal framework: first, feature point matching and time alignment are performed on the multi-view images to generate a unified benchmark image dataset; then, edge detection and corner detection algorithms are used to extract equipment feature points, constructing a 3D feature point cloud; finally, temperature and gas concentration data are correlated with the feature point cloud to form preliminary multi-source fused data. Based on this preliminary multi-source fused data, the robot positioning module calculates the robot's real-time position coordinates using the encoder pulse counting principle and calibrates the coordinates using RFID tags, generating a corrected position dataset. Combining historical environmental parameters, robot operating status, and other multi-source data, features such as position fluctuation amplitude and average environmental parameters are extracted and integrated along the time dimension to form a comprehensive dataset. The equipment anomaly analysis module employs a multi-dimensional feature extraction algorithm to perform time-series trend analysis and parameter deviation calculation on the comprehensive dataset. Combined with time-series correlation mining algorithms and cluster analysis, it determines the normal operating mode of the equipment. The current operating parameters are compared with preset thresholds; if the threshold is exceeded, an anomaly identifier is generated, and a rule base is invoked to output an equipment fault warning signal. The robot motion control module transmits the warning signal to the ground workstation via wireless mobile communication technology. Upon receiving instructions, it drives the robot to adjust its operating state and optimizes the path using the A* path planning algorithm based on feedback control signals. It also transmits positioning data in real time to generate remote monitoring data.

[0137] This embodiment achieves closed-loop management of the entire process—multi-source data acquisition, precise positioning, intelligent analysis, and remote control—through modular design. The multi-source data fusion module solves the problem of inconsistent data acquisition benchmarks in traditional systems, improving the spatiotemporal consistency of equipment status data and providing an accurate data foundation for fault diagnosis. The robot positioning module, combining encoder and RFID dual positioning, improves positioning accuracy from ±10mm to ±3mm, ensuring the accuracy of inspection paths. The equipment anomaly analysis module, based on multi-dimensional feature extraction and algorithm fusion, achieves a fault warning accuracy rate of 92%. The robot motion control module enables real-time human-machine interaction and dynamic path optimization, improving inspection efficiency and reducing manual intervention costs by 40%. Overall, the system provides an efficient and intelligent solution for industrial equipment operation and maintenance, significantly enhancing the safety and reliability of equipment operation under complex conditions.

[0138] In one embodiment, such as Figure 3 As shown, this application also provides a hardware circuit layout design for an intrinsically safe inspection robot for mining applications, which may include:

[0139] Power supply unit: Includes two sets of No. 1 battery packs, providing 12V power supply. The voltage conversion module outputs different voltages such as 5V and 12V to power the various functional modules respectively. A power switch is configured to realize the overall power supply control, and a main stop switch is used to cut off the power supply in an emergency to ensure the safety of equipment and personnel.

[0140] Drive control unit: Equipped with a left driver and a right driver, each connected to two motors, it receives commands from the main control board via the CAN communication protocol to control the robot to move along the track; indicator lights display the driver's working status in real time (such as normal operation, fault alarm), facilitating on-site troubleshooting.

[0141] Core Control and Communication Unit: The main control board serves as the core, connecting to intrinsically safe PTZ, obstacle avoidance sensor, smoke sensor, methane sensor and other peripherals via a 485 communication interface to achieve data acquisition and command issuance; it is equipped with a CAN communication interface to interact with the driver, and an RJ45 network cable interface to connect to the switch. The switch is connected to a wireless CPE (intrinsically safe type) and two antennas to build a wireless communication link and ensure data transmission with the ground workstation.

[0142] Sensing and monitoring unit: Includes 4 obstacle avoidance sensors (ultrasonic type) and 2 laser obstacle avoidance sensors to achieve all-round detection of obstacles along the track; smoke sensor (photoelectric type) and methane sensor collect pump room environmental parameters in real time and trigger alarm when the parameters exceed the standard; RFID module is built-in and works with encoder to achieve precise robot positioning with a positioning accuracy of ±3mm.

[0143] Audio, video and interaction unit: The power amplifier board connects the speaker and microphone to enable two-way voice communication between the ground workstation and the site; the intrinsically safe PTZ is equipped with a visible light camera and an infrared thermal imager, along with a lighting lamp (white light auxiliary) to meet the image acquisition needs in low-light environments underground; the I2C interface is used to expand peripherals and improve equipment compatibility.

[0144] Auxiliary function unit: The feeder is used to enhance signal transmission and ensure data stability between the sensor and the main control board; the overall circuit adopts an intrinsically safe design with a protection level of IP54, which is suitable for the complex environment of explosion-proof, high humidity and dust in coal mines, ensuring long-term reliable operation of the equipment.

[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0146] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the encoder-RFID positioning inspection robot control method, system, device, and medium as described above.

[0147] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0148] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0149] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An encoder-RFID positioning and inspection robot control method, characterized in that, The method comprises: Obtaining multi-view image data, equipment temperature data and environmental gas concentration data of equipment in the pump house, and unifying the data to the same space-time reference through space-time synchronous calibration technology to obtain multi-source preliminary fusion data; Based on the multi-source preliminary fusion data, encoder and RFID positioning technology is used to determine the position of the inspection robot on the track, and combined with the historical multi-source data, the characteristics are extracted and integrated to obtain a comprehensive data set; Using a multi-dimensional feature extraction algorithm to analyze the characteristics of the comprehensive data set, judging the abnormal state of the equipment according to the characteristic correlation, and calling a preset warning signal generation rule library to generate an equipment fault warning signal; Using wireless mobile communication technology to transmit the equipment fault warning signal to the ground, and receiving the instructions of the ground workstation to drive the movement control of the inspection robot, and generating remote monitoring data.

2. The method of claim 1, wherein, The multi-source preliminary fusion data obtained by the space-time synchronous calibration technology comprises: Obtaining multi-view image data of equipment in the pump house; the multi-view image data includes all-around images covering the overall working condition of the equipment, and local detail images focusing on key components of the equipment; Based on the encoder-RFID positioning data of the inspection robot and each frame of multi-view image data, image processing technology is used to perform space-time synchronous calibration on the multi-view image data to obtain an image data set with unified space-time reference; the unified space-time reference takes the robot track coordinates and the collection time stamp as the reference; The image data set is processed by edge detection and corner detection algorithm to extract equipment feature points including equipment contour, component size and instrument scale, and combined with the space-time coordinate information of each equipment feature point, a feature point cloud data containing three-dimensional space features of the equipment is generated; An infrared thermal imager carried by the inspection robot is used to collect the temperature data of the equipment in the pump house in real time, and the collection space-time coordinates corresponding to each temperature data are recorded; the equipment in the pump house includes key heating components such as motor shaft, speed reducer and water pump front and rear shaft; The methane sensor and the smoke sensor carried by the inspection robot are used to collect the environmental gas concentration data in the pump house, and the collection space-time coordinates of each gas concentration data are recorded; With reference to the unified space-time reference, the space-time synchronous calibration technology is used to combine the collection space-time coordinates to associate and fuse the equipment temperature data and the environmental gas concentration data with the generated feature point cloud data, to obtain multi-source preliminary fusion data containing equipment space features, temperature state and environmental gas state.

3. The method of claim 1, wherein, Based on the multi-source preliminary fusion data, encoder and RFID positioning technology is used to determine the position of the inspection robot on the track, and combined with the historical multi-source data, the characteristics are extracted and integrated to obtain a comprehensive data set, which comprises: Extracting the running track data of the inspection robot from the multi-source preliminary fusion data; the running track data includes the time stamp corresponding to each sampling time and the speed vector of the robot moving along the track; Based on the pulse counting principle of the encoder, the running track data is combined with the preset length parameter of the track to calculate the mapping relationship between the number of encoder pulses and the displacement of the track per unit time, and the real-time position coordinates of the inspection robot on the track are calculated. The fixed coordinates of the preset RFID tags along the track are read, the real-time position coordinates calculated by the encoder are compared with the fixed coordinates, if the deviation exceeds a preset threshold, the real-time position coordinates are corrected based on the RFID tag coordinates, and a corrected position data set is generated; Obtain historical multi-source data; the historical multi-source data includes pump house environment parameters and robot running state; From the corrected position data set, the position fluctuation amplitude and the average moving speed are extracted, the environment parameter mean value, the equipment temperature peak value, and the robot failure frequency are extracted from the historical multi-source data, the features are aligned in the time dimension, and then integrated to obtain a comprehensive data set containing space-time features, environment features, and equipment state features.

4. The method of claim 3, wherein, The mapping relationship between the number of encoder pulses per unit time and the track displacement is represented by the following formula: wherein S t represents the amount of track displacement in unit time t, N t represents the number of encoder pulses in unit time t, L represents the standard length of a single track section, θ represents the actual inclination angle of the track section where the robot is located in unit time t, P0 represents the standard total number of encoder pulses, δ T represents a correction coefficient of the encoder pulse output affected by the environmental temperature, and ΔT represents the actual environmental temperature in unit time t and the standard calibration temperature.

5. The method of claim 1, wherein, The multi-dimensional feature extraction algorithm is used for feature analysis on the comprehensive data set, the equipment running abnormal state is judged according to the feature correlation, a preset warning signal generation rule library is called to generate an equipment failure warning signal, including: Extracting equipment running state data from the comprehensive data set; the equipment running state data includes equipment parameters and historical records; Using a multi-dimensional feature extraction algorithm to perform time series trend analysis on the equipment running state data, and performing deviation calculation on the equipment inherent parameters and real-time parameters to obtain equipment running state features including parameter trend features, deviation features, and statistical features; The time correlation relationship between the equipment running state features and historical failures is analyzed by a time series correlation mining algorithm, and the historical data of similar running features are classified by a clustering analysis algorithm to determine the normal running mode of the equipment under different working conditions; the normal running mode includes high-load running mode and low-load running mode; Comparing the feature parameters of the current running mode of the equipment with the preset normal running threshold, if a certain feature parameter exceeds the normal threshold range, an abnormal state identifier is generated; the abnormal state identifier content includes abnormal parameter name, current value, deviation threshold amplitude, and abnormal occurrence timestamp; the normal running threshold is set based on the normal running mode feature parameter range combined with industry safety standards and equipment factory requirements; According to the abnormal state identifier, a preset warning signal generation rule library is called to generate an equipment failure warning signal; the equipment failure warning signal needs to carry the equipment number, the abnormal identifier ID, and the warning generation time; the warning signal generation rule library includes the mapping relationship between different abnormal identifiers and corresponding warning signals.

6. The method of claim 5, wherein, After generating the equipment failure warning signal, it further includes: Extracting dynamic feature factors related to the equipment running state from the equipment failure warning signal; the dynamic feature factors include the equipment parameter change rate in the abnormal occurrence time period, the environmental interference factor, and the equipment historical failure correlation factor; Random forest algorithm is used to classify the dynamic feature factors, and the currently extracted dynamic feature factors are input into the constructed fault classification model to output the equipment failure risk level; the fault classification model is trained using the dynamic feature factors in the historical failure data and the known fault types; extract a high-risk factor from the device failure risk level, integrate the high-risk factor corresponding to the device parameter original data and the abnormal state identification information according to a preset data format to generate failure prediction data; the high-risk factor is the dynamic characteristic factor whose contribution to the failure risk level determination exceeds a preset threshold; input the failure prediction data into a failure probability calculation model constructed based on a Bayesian probability algorithm, calculate a failure probability distribution, set an early warning priority according to the probability distribution and a failure influence range, and obtain a failure early warning priority; the failure probability calculation model takes the correlation probability of historical high-risk factors and failure occurrence as a training sample.

7. The method of claim 1, wherein, The device failure early warning signal is transmitted to the ground by using wireless mobile communication technology, and the movement control of the inspection robot is driven by receiving the instructions of the ground station to generate remote monitoring data, including: The device failure early warning signal information is sent to the ground station by using wireless mobile communication technology, and instruction data of the ground station is obtained; Two-way voice intercom between the inspection robot and the ground station is realized according to the instruction data, and voice interaction data is obtained; The voice interaction data is processed by using a voice recognition algorithm to generate robot control instructions; The running state of the inspection robot is adjusted according to the robot control instructions, and running state data is obtained; The running state data is transmitted to the ground station by using wireless mobile communication technology, and a feedback control signal is obtained; Based on the current running state data and pump house equipment distribution coordinates, an A* path planning algorithm is used to optimize the path of the inspection robot based on the feedback control signal, and optimization path data is generated; the pump house equipment distribution coordinates are pre-stored in the local map of the robot; The movement control of the inspection robot is driven by using the optimization path data, and real-time positioning data is obtained; The real-time positioning data is returned to the ground station by using wireless mobile communication technology, and remote monitoring data is generated.

8. An encoder-RFID positioning and inspection robot control system, characterized by, The system comprises: A multi-source data fusion module is configured to obtain multi-view image data, device temperature data, and environmental gas concentration data of equipment in a pump house, unify the data to the same space-time reference by using space-time synchronization calibration technology, and obtain multi-source preliminary fusion data; A robot positioning module is configured to determine the position of the inspection robot on the track based on the multi-source preliminary fusion data by using an encoder and RFID positioning technology, and obtain a comprehensive data set by integrating features extracted from historical multi-source data; An equipment anomaly analysis module is configured to analyze features of the comprehensive data set by using a multi-dimensional feature extraction algorithm, determine an equipment running abnormal state according to feature correlation, and generate a device failure early warning signal by calling a preset early warning signal generation rule library; A robot movement control module is configured to transmit the device failure early warning signal to the ground by using wireless mobile communication technology, drive the movement control of the inspection robot by receiving instructions of the ground station, and generate remote monitoring data. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images