A multi-target dynamic weighted hydrogen leakage positioning method based on a quadruped robot

Abstract

The present application relates to hydrogen metallurgy field gas leakage detection technical field, specifically relates to a kind of multi-target dynamic weighted hydrogen leakage positioning method based on quadruped robot.The method is first according to the preset inspection path, and periodic inspection is carried out;Dangerous gas equipment such as blast furnace, pipeline is detected using thermal imager, if abnormal area is found, stop according to the preset inspection path inspection, start to find leakage source, including: based on the current search stage and environmental state, the direction of multi-target weighted source finding is calculated, real-time adjustment constraint weight is issued instruction, so that robot advances to the direction after weighted optimization;Constraint weight includes concentration gradient, safety distance, motion smooth and direction stability weight;When robot reaches the area of maximum preset gas concentration, the area is taken as leakage source area.The present application carries out multi-target dynamic weighted positioning to quadruped robot, improves the degree of automation and processing speed of hydrogen leakage source detection.

Description

Technical Field

[0001] This invention relates to the field of hydrogen metallurgical field gas leak detection technology, specifically to a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot. Background Technology

[0002] Hydrogen metallurgy is a green ironmaking process that uses hydrogen instead of carbon as a reducing agent to reduce iron ore, and it is an important technological path for the steel industry. Hydrogen metallurgy sites involve numerous hydrogen pipelines, storage facilities, and processing devices. Hydrogen is highly flammable and explosive, has a high diffusion coefficient, and a wide explosion limit. Defects in pipeline welds, aging valve seals, and corrosion of furnace equipment can all lead to hydrogen leaks. Hydrogen leaks pose significant safety risks: hydrogen is colorless and odorless, making leaks extremely difficult to detect, and it can ignite or even explode upon contact with an open flame or static electricity.

[0003] The working environment at a hydrogen metallurgical site has the following typical characteristics: dense equipment and crisscrossing pipelines: blast furnaces, storage tanks, and pipelines are arranged in a compact manner, forming a complex three-dimensional spatial structure; numerous obstacles: these can easily cause robots to get stuck in "dead zones" or "stuck" during local navigation, especially at pipe angles and behind equipment; complex airflow field: factors such as high-temperature hot airflow around the blast furnace and eddies generated by pipeline obstruction cause hydrogen to diffuse rapidly, resulting in irregular concentration distribution and drastic fluctuations in gradient direction after leakage.

[0004] Traditional single concentration gradient source tracing methods have the following limitations: they are easily misled by local concentration accumulation or airflow interference, and have poor convergence in complex scenarios; they cannot effectively deal with navigation dead zone problems, and the robot may remain stuck in a certain place for a long time, reducing source tracing efficiency; and they lack unified consideration of multiple safety constraints such as collision avoidance, distance safety, and motion stability.

[0005] In recent years, with the rapid development of robotics and sensing technologies, the use of mobile robots for hazardous gas leak detection has gradually become a research hotspot in the field of industrial safety. Quadruped robots possess advantages such as strong terrain adaptability, excellent obstacle-crossing ability, and stable movement in complex industrial environments. Combined with gas concentration sensor arrays and thermal imaging technology, they can automatically search for and locate hydrogen leak sources without requiring personnel to enter hazardous areas. Therefore, the method of using quadruped robots for multi-target dynamic localization of hydrogen leak sources is a key research focus in this field. Summary of the Invention

[0006] To address the limitations of traditional single-concentration gradient source tracing methods, this invention aims to provide a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot. The specific technical solution adopted is as follows: The robot performs periodic inspections based on existing map information and following a preset inspection path. Thermal imaging devices are used to detect equipment that carries hazardous gases. The real-time data is compared with historical data. If an abnormal area is found, the inspection according to the preset inspection path is stopped and the source of the leak is searched. If no abnormality is found, the periodic inspection continues. The process of stopping the inspection along the preset path and starting the search for the leak source when an abnormal area is detected includes: calculating the multi-objective weighted search direction based on the current search stage and environmental conditions, adjusting the constraint weights in real time, and issuing instructions to make the robot move in the weighted optimized direction; the constraint weights include the weights of concentration gradient, safe distance, motion smoothness, and directional stability; when the robot reaches the area with the maximum preset gas concentration, the area is taken as the leak source area; the robot collects image information, gas concentration information, temperature information, and location information of the leak source area and reports them to the control room.

[0007] Furthermore, the method also includes: After identifying the leak source area and reporting the information to the control room, the robot leaves the leak source area and waits for the next instruction, including: if the leak source has been repaired, the robot continues its inspection; if the leak source has not been repaired, the robot changes its inspection route; or the robot changes its posture and waits for further instructions.

[0008] Furthermore, the multi-objective weighting includes: The search phase includes: the initial search phase, the intermediate search phase, and the final search phase; In the initial search phase: the weight is configured as w. grad :w safe :w smooth :w stab =0.3:0.4:0.2:0.1; During the mid-stage search: the weight is configured as w. grad :w safe :w smooth :w stab =0.4:0.3:0.2:0.1; In the final search phase: the weight is configured as w. grad :w safe :w smooth :w stab =0.5:0.25:0.15:0.1; Among them, w grad The weights for the concentration gradient reward; w safe Weighting of the safe distance reward; w smooth Weights for motion smoothing rewards; w stab The weight of the directional stable reward; When the system detects that the robot's speed has been below the preset speed threshold for an extended period, it automatically increases the speed.smooth and w stab The weights; When the fluctuation amplitude along the concentration gradient exceeds the preset fluctuation threshold, w is automatically reduced. grad Weighting, increasing w stab The weighting of the weights causes the robot to adopt a more conservative movement strategy.

[0009] Furthermore, the rewards include concentration gradient rewards, safe distance rewards, motion smoothness rewards, and directional stability rewards, among others: Concentration gradient reward: Encourages the robot to move along the direction of the largest concentration gradient, which is used to quickly converge to the vicinity of the leak source; Safe distance reward: Constrains the minimum safe distance between the robot and obstacles and hazardous areas to prevent collisions and excessive approach; Smooth motion reward: Encourages the robot to adopt a smooth and continuous motion trajectory, avoiding getting stuck due to sharp turns; Directional stability bonus: When environmental airflow disturbances cause fluctuations in the direction of the concentration gradient, a time window is used for smoothing to ensure the stability of the motion command.

[0010] Furthermore, the multi-objective weighting includes: The transition from the initial search phase to the intermediate search phase occurs when the maximum concentration value collected by the robot reaches the preset intermediate transition threshold C. threshold1 Triggered at time; The transition from the mid-stage search to the final stage search is as follows: when the maximum concentration value collected by the robot reaches the preset final stage transition threshold C. threshold2 Triggered at time; If the robot continuously moves a distance exceeding a preset distance threshold during a certain stage but the concentration does not increase, the system determines that it has entered a dead zone and automatically reduces the concentration. grad Weighting, increasing w smooth and w stab The weights are adjusted until the concentration increases again and the original weights are restored.

[0011] Furthermore, the calculation of the source-finding direction for multi-objective weighting includes: The robot is equipped with a gas sensor array, which collects the concentration of hazardous gases in 16 directions in real time. Let the concentration values ​​in 16 directions be C1, C2, C3…C 16 The corresponding direction vectors are d1, d2, d3…d 16 ; (1) Concentration gradient reward The calculation formula is as follows: ; in, Let be the concentration value in the i-th direction. Let i be the direction vector of the i-th direction; (2) Safe distance reward The calculation formula is as follows: ; Where j is S obstacle The index of the obstacle's direction in the set; S obstacle d represents the set of directions corresponding to all obstacles within a safe threshold from the robot; j It is the unit direction vector corresponding to the j-th direction; |S obstacle | represents the number of obstacle directions in the set; (3) Motion smoothing reward The calculation formula is as follows: ; in, The direction of motion at the previous moment. This is the optimal direction calculated so far. For smoothing coefficients; (4) Directional stability reward The calculation formula is as follows: ; Where n is the length of the time window, is the directional gradient of the k-th sampling period; t is the sampling period index at the current time. (5) Final weighted source tracing direction: v final =w grad ×v grad +w safe ×v safe +w smooth ×v smooth +w stab ×v stab ; Among them, v smooth v is obtained by weighted average of the previous motion direction and the current optimal direction. stab This represents the average gradient direction within the time window; (6) Direction of maximum concentration gradient: The robot eventually faces v final Engage in exercise.

[0012] Furthermore, the method also includes: Time window smoothing: The gradient direction and concentration gradient magnitude over the past N sampling periods are smoothed using a sliding window to eliminate abnormal fluctuations caused by instantaneous airflow interference, ensuring that the motion commands issued to the robot are continuous and stable; Smooth weight switching: The weight switching between stages adopts a gradual increase / decrease strategy, rather than a jump switch, to avoid sudden changes in the direction of movement caused by weight changes; Dynamic feedback adjustment: Based on the deviation between the robot's actual execution results and the expected results, the weights are dynamically fine-tuned to form a negative feedback loop, further improving convergence and accuracy; the actual execution results include: position change and concentration change rate.

[0013] The embodiments of the present invention have at least the following beneficial effects: This invention utilizes a combined application of a thermal imager and a gas sensor array, along with a dynamic reward-weighted strategy, to guide a quadruped robot to autonomously search for and locate hydrogen leak sources. Specifically designed to address the environmental characteristics of hydrogen metallurgical sites, such as navigation dead zones caused by dense equipment and gradient instability due to the easy diffusion of hydrogen, this invention eliminates the need for personnel to enter hazardous areas, solving the problems of high safety risks, slow response times, and limited coverage associated with traditional manual inspections. Its advantages include high automation, high source-finding efficiency, strong robustness, and strong adaptability. The quadruped robot itself has strong terrain adaptability, enabling it to adapt to the complex working environment of hydrogen metallurgical sites. The system is also highly scalable, allowing for flexible adjustment of weight and sensor configurations according to on-site requirements, making it of significant engineering application value.

[0014] This invention exhibits high robustness in complex industrial environments, avoiding problems such as getting stuck in dead zones and being misled by airflow compared to the single gradient method, thus improving the success rate and efficiency of source finding. It also possesses strong adaptability: dynamically adjusting weight configuration based on real-time conditions during the search process, such as concentration increase, robot jamming, and gradient fluctuations, without manual intervention. Furthermore, it demonstrates good engineering applicability: unifying multiple safety constraints such as collision avoidance, distance safety, and motion stability, it meets the needs of engineering practice, and the weight parameters can be flexibly adjusted according to on-site conditions. Finally, it exhibits strong scalability: the framework is easily extended, allowing the addition of new objective functions, such as energy consumption optimization and time optimization, without changing the core architecture. Finally, it boasts a high degree of automation: the entire system requires no personnel to enter hazardous areas, exhibiting a high degree of automation, fast processing speed, and simple deployment. Attached Figure Description

[0015] Figure 1 This is a flowchart of a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot, provided in one embodiment of the present invention. Figure 2 This is a process flow diagram of a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot, provided in one embodiment of the present invention. Figure 3 This is a system block diagram of a quadruped robot based on a hydrogen metallurgical field, provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the arrangement of a gas sensor array provided in one embodiment of the present invention; Figure 5 This is a schematic diagram of a dynamic weighted source tracing path provided in one embodiment of the present invention; Figure 6 This is a flowchart of a method for finding a leakage source provided in an embodiment of the present invention. Detailed Implementation

[0016] This invention provides a specific implementation method for a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot. This method is applicable to a quadruped robot used for locating hydrogen leak sources. The quadruped robot includes a robot body, a motion control motherboard, a thermal imager, and a gas concentration sensor array. To address the limitations of traditional single-concentration gradient source-finding methods, this invention employs a dynamic reward weighted strategy instead of the traditional single-concentration gradient method when searching for leak sources. This dynamic reward weighted strategy dynamically adjusts the weights of multiple targets, such as concentration gradient, safe distance, motion smoothness, and directional stability, based on the real-time state during the search process. This enables the robot to stably approach the leak source in complex industrial environments, effectively avoiding navigation dead zones and coping with airflow interference. Please refer to [link to relevant documentation]. Figure 3 , Figure 3 This is a system block diagram of a quadruped robot based on a hydrogen metallurgical field, according to one embodiment of the present invention. Figure 3 It includes a quadruped robot body 100, a thermal imager 200, a hydrogen leak source 300, a gas concentration sensor array 400, and a control room 500.

[0017] The following description, in conjunction with the accompanying drawings, details a specific scheme for a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot provided by the present invention.

[0018] Please see Figure 1 The diagram illustrates a flowchart of a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot, according to an embodiment of the present invention. The method includes the following steps: In step S100, the robot performs periodic inspections according to the existing map information and the preset inspection path.

[0019] First, the robot performs periodic inspections based on a pre-established map of the hydrogen metallurgical site, following a preset inspection path set by the operator. The preset inspection path covers the main hydrogen pipelines, hydrogen storage tank area, and the area surrounding hydrogen-using equipment. The operator can remotely adjust the path planning and inspection cycle through the control room.

[0020] This quadruped robot incorporates a joint detection mechanism combining a thermal imager and a gas concentration sensor array: the thermal imager handles the initial screening of anomalous areas at the macroscopic level. When hydrogen leaks from a pipeline or equipment, the throttling expansion effect (Joule-Thomson effect) causes a localized temperature drop near the leak point, allowing the thermal imager to quickly identify areas with abnormally low temperatures. The gas sensor array, on the other hand, performs precise measurements of gas concentration at the microscopic level, accurately calculating the concentration gradient with spatial resolution in 16 directions. The two sensing methods corroborate each other, effectively reducing false alarm rates and improving the reliability of leak source location.

[0021] Step S200: Use a thermal imager to detect equipment that carries hazardous gas, compare the detected real-time data with historical data, if an abnormal area is found, stop the inspection according to the preset inspection path and start looking for the leak source, if no abnormality is found, continue the periodic inspection.

[0022] The inspection process utilizes thermal imaging to detect abnormal areas: During the inspection, the thermal imager continuously scans equipment carrying hydrogen, such as blast furnaces and pipelines. The thermal imager continuously monitors the temperature of the blast furnace and pipelines, establishing a historical temperature database and generating baseline data. Real-time data is then compared with this baseline data. If an abnormal temperature distribution is detected in a localized area, such as a temperature deviation of more than 5°C from the historical baseline temperature at a pipeline interface (primarily detecting an abnormal temperature decrease), and based on the Joule-Thomson effect indicating a hydrogen leak, a suspected leak area is identified. The robot immediately stops its pre-set path inspection and enters leak source search mode. If no abnormality is found, the inspection task continues. In other words, when the temperature data in the real-time thermal image is lower than the historical baseline temperature data, an abnormal area is identified, and the robot immediately stops its pre-set path inspection and enters leak source search mode.

[0023] After identifying the leak source area and reporting the information to the control room, the robot leaves the leak source area and waits for the next instruction, including: if the leak source has been repaired, the robot continues its inspection; if the leak source has not been repaired, the robot changes its inspection route; or the robot changes its posture and waits for further instructions.

[0024] After entering leak source search mode, that is, after discovering an abnormal area, the inspection according to the preset inspection path is stopped and the search for the leak source begins, please refer to [link to relevant instructions]. Figure 6 , Figure 6 A flowchart illustrating the method for locating leakage sources provided in this embodiment of the invention: Step S210: Based on the current search stage and environmental state, calculate the source-finding direction of the multi-objective weighted search, adjust the constraint weights in real time, and issue instructions to make the robot move in the weighted optimized direction; the constraint weights include the weights of concentration gradient, safe distance, motion smoothness, and directional stability.

[0025] The gas concentration sensor array begins operation first, collecting real-time hydrogen concentration data in 16 directions around the robot to form a directional concentration distribution map centered on the robot. This provides the data foundation for subsequent weighted source-finding calculations. Please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram of the arrangement of a gas sensor array provided in one embodiment of the present invention; Figure 4 It includes a robot body 100 and orientation sensor units 301-316; N represents north, S represents south, W represents west, and E represents east.

[0026] Then, the source-finding direction is dynamically weighted, specifically: Step 1: Initialize the search phase and weight configuration: After the robot enters the leak source search mode, the system determines the current search stage based on the maximum concentration value collected and initializes the corresponding weight configuration. The search stages include: initial search stage, intermediate search stage, and final search stage. In the initial search phase (when the source of the leak is relatively far away): the weight is configured as w. grad :w safe :w smooth :w stab =0.3:0.4:0.2:0.1; Prioritize safety and stability to avoid getting stuck in dead zones in complex obstacle environments.

[0027] During the intermediate search phase (at which point the distance to the leak source is moderate): the weight is configured as w. grad :w safe :w smooth :w stab =0.4:0.3:0.2:0.1; Balance gradient tracing and safety constraints to enhance the tracking of leakage sources.

[0028] In the final search phase (when the source of the leak is very close): the weight is configured as w. grad :w safe :w smooth :w stab =0.5:0.25:0.15:0.1; Strengthen gradient tracing to accurately locate the leak source while maintaining necessary safety constraints.

[0029] Among them, w grad The weights for the concentration gradient reward; w safe Weighting of the safe distance reward; w smooth Weights for motion smoothing rewards; w stab The weight of the direction-based stable reward.

[0030] The design principle of this weight configuration is: in the early search phase, priority is given to ensuring safety and smooth motion, avoiding overly aggressive gradient tracking in complex obstacle environments that could lead to dead zones; as the source of leakage approaches, the guiding force of the concentration gradient is gradually increased, while maintaining necessary stability and safety constraints.

[0031] Step 2, Multi-objective reward calculation: The motion control motherboard processes the concentration data from 16 directions collected by the sensor array in real time and calculates four core reward items: (1) Concentration gradient reward: Based on the concentration values ​​in each direction, calculate the concentration-weighted average direction to encourage the robot to move along the direction with the largest concentration gradient, so as to quickly converge to the vicinity of the leakage source; Concentration gradient reward The calculation formula is as follows: ; in, Let be the concentration value in the i-th direction. Let be the direction vector for the i-th direction.

[0032] The direction of the concentration-weighted average is preferentially pointed towards high concentration areas, causing the robot to tend to move towards high concentration areas and quickly converge to the vicinity of the leak source.

[0033] (2) Safety distance reward: Based on the location of obstacles detected by the lidar, calculate the direction away from the obstacles, constrain the minimum safe distance between the robot and obstacles and dangerous areas, and prevent collisions and excessive approach; Safe distance reward The calculation formula is as follows: ; Where j is S obstacle The index of the obstacle's direction in the set; S obstacle S is the set of directions corresponding to all obstacles within a safe threshold from the robot (e.g., if the lidar detects obstacles in the north, northwest, and west, then S). obstacle ={North, Northwest, West}); d j It is the unit direction vector corresponding to the j-th direction (e.g., north dj=(0,1), west dj=(0,1)). j =(-1,0);|S obstacle | represents the number of obstacle directions in the set; the negative sign indicates a reverse value, that is, reversing the resultant vector of all obstacle directions, so that V safe Points in the most open direction away from obstacles. The final result of this item is a direction vector pointing to a safe area (an area without obstacles), guiding the robot to automatically avoid surrounding obstacles.

[0034] This constraint points away from obstacles to prevent the robot from colliding with them or getting too close to hazardous areas. This constraint is crucial in the obstacle-rich environment of a hydrogen metallurgical site.

[0035] (3) Motion smoothing reward: The current optimal direction is smoothly interpolated with the previous motion direction to encourage the robot to adopt a smooth and continuous motion trajectory and avoid getting stuck due to sharp turns; Motion smoothing reward The calculation formula is as follows: ; in, The direction of motion at the previous moment. This is the optimal direction calculated so far. This is the smoothing coefficient.

[0036] This allows the robot to use a continuous motion trajectory, avoiding sharp turns.

[0037] (4) Directional stability reward: The gradient direction of the past sampling period is smoothed by time window. When the concentration gradient direction fluctuates due to environmental airflow disturbance, the motion command is ensured by smoothing through time window.

[0038] Motion smoothing reward The calculation formula is as follows: ; Where n is the length of the time window, Let be the gradient direction of the k-th sampling period; t is the sampling period index at the current moment. If t=50, that is, the current sampling period is the 50th sampling period, and n=5, then the gradient directions of the 46th to 50th sampling periods are averaged.

[0039] This method smooths the gradient direction over a time window, effectively eliminating transient airflow interference caused by hydrogen diffusion and improving convergence stability in irregular concentration fields.

[0040] The third step is the final weighted source tracing direction: v final =w grad ×V grad +w safe ×V safe +w smooth ×V smooth +w stab ×V stab ; Wherein, the weights satisfy w grad +w safe +w smooth +w stab=1, and is set during initialization according to the current search stage, and adjusted in real time as necessary.

[0041] The final weighted source-finding direction integrates concentration gradient guidance, safety constraints, motion continuity, and directional stability. Compared to the traditional single gradient method, the multi-objective weighted strategy can more robustly guide the robot to approach the leak source in complex industrial environments. Specifically, concentration gradient guidance points the robot towards high concentrations, safety constraints help the robot stay away from obstacles, motion continuity prevents the robot from making sharp turns, and directional stability avoids interference from airflow.

[0042] Step 4: Dynamic weight adaptive adjustment: During the movement, the system continuously monitors the robot's status and adjusts the weights in real time based on the following conditions: (1) Automatic switching of search phases: When the maximum concentration value reaches the conversion threshold, the weights are automatically switched from one phase to the next. The transition from the initial search phase to the intermediate search phase: when the robot collects the maximum concentration value C max Reaching the preset mid-term conversion threshold C threshold1 When the weight is switched from the initial configuration to the intermediate configuration, the preset intermediate conversion threshold is 100ppm. 100ppm is approximately 0.25% of the lower explosive limit of hydrogen (4%), which usually indicates that a leak has been confirmed and the leak source is 5-10 meters away.

[0043] The transition from the mid-stage search to the final stage search is as follows: when the maximum concentration value collected by the robot reaches the preset final stage transition threshold C. threshold2 At that time, the weights automatically switch from the mid-term configuration to the final configuration; the preset final conversion threshold is 500ppm, which is approximately 1.25% of the lower explosive limit of hydrogen, indicating that it is very close to the leak source (1-2 meters away) and can be located with high confidence.

[0044] Among them, the preset intermediate conversion threshold C threshold1 Less than the preset final conversion threshold C threshold2 .

[0045] In a preferred embodiment of the present invention, the robot further includes an automatic escape function: when the robot's continuous movement distance d exceeds a preset distance threshold but the concentration C is below the threshold, the robot can escape from obstacles. max When there is no increase, or when the robot speed is detected to be below the preset speed threshold for an extended period, the system automatically increases the speed. smooth and w stabThe weights indicate whether the robot has entered a dead zone or reached a localized concentration plateau, triggering an automatic escape strategy. The preset distance threshold is 2m, while the typical step length of a quadruped robot in a complex industrial environment is 0.3-0.5m. If the concentration hasn't increased after 2m (approximately 4-6 steps), it indicates the robot has gone in the wrong direction and entered a dead zone. The preset speed threshold is an empirical value of 0.1 m / s, which can be adjusted by the implementer based on actual conditions.

[0046] Escape strategies include: reducing w grad Weight, increase w smooth and w stab The weighting forces the robot to prioritize smooth movements to escape difficulties. Once the concentration increases again, the system automatically restores the standard weighting configuration for the current stage.

[0047] In a preferred embodiment of the present invention, the robot also includes an airflow interference response function: when the fluctuation amplitude of the robot's concentration gradient direction exceeds a preset fluctuation threshold within a short period of time, indicating strong airflow interference, the system automatically reduces the w grad Weight, increase w stab Weights enable the robot to adopt a more conservative and stable motion strategy. The preset fluctuation threshold is set to 30°. A fluctuation of 30° usually indicates a significant disturbance, and under normal circumstances, the fluctuation between two adjacent samples is less than 10°.

[0048] As a preferred embodiment of the present invention, the robot also includes a weight smoothing switching function: all weight adjustments adopt a gradual increase / decrease strategy rather than a jump switch, to avoid sudden changes in the direction of motion caused by weight changes.

[0049] Please see Figure 5 , Figure 5 This is a schematic diagram of a dynamic weighted source tracing path provided in one embodiment of the present invention.

[0050] The dynamic reward weighting strategy of this invention is specifically designed for the environmental characteristics of hydrogen metallurgical sites: (1) Addressing the problem of dense obstacles and navigation dead zones: In hydrogen metallurgical sites, equipment is densely packed and pipelines are intertwined. When robots perform source-finding tasks in such environments, traditional single gradient methods are prone to getting stuck in local concentration plateaus or geometric dead zones at pipe angles, behind equipment, etc., causing the robot to remain stuck for a long time or fall into oscillations. This method increases the weights of wsafe and wsmooth in the initial search phase, forcing the robot to prioritize safety and stability, and avoiding the risk of getting stuck in dead zones due to aggressive gradient tracking. Once stuck in a dead zone, an escape mechanism is automatically triggered, reducing the weight of wgrad to ensure that the robot can get out of trouble in time.

[0051] (2) Addressing the issues of hydrogen diffusion and gradient instability: Hydrogen has a high diffusion coefficient and is easily mixed with air. In complex airflow fields (hot airflow around the blast furnace, eddies generated by pipe obstructions, etc.), leaked hydrogen diffuses rapidly, resulting in irregular concentration distribution and drastic fluctuations in gradient direction. Traditional gradient methods directly use the real-time gradient direction, which is easily misled by instantaneous airflow interference, leading to unstable robot movement or misjudgment of the leak location. This method introduces a time window smoothing process (Vstab) to smooth the gradient direction, effectively eliminating high-frequency fluctuations caused by airflow interference, enabling the robot to stably converge to the true leak source in irregular concentration fields.

[0052] (3) Unified handling of multiple safety constraints: In industrial environments, robots need to avoid collisions with equipment, stay away from high-concentration hazardous areas, and maintain stable movement. Multi-objective weighted strategies integrate these constraints into a weighted optimization framework, which is more in line with actual engineering needs than single-objective optimization and makes it easier to adjust the weight configuration according to the actual situation on site.

[0053] Step S220: When the robot reaches the area with the maximum preset gas concentration, the area is designated as the leakage source area; the robot collects image information, gas concentration information, temperature information, and location information of the leakage source area and reports them to the control room.

[0054] Then, guided by the weighted source-finding direction, the robot continues to move towards the leak source. When the hydrogen concentration detected by the sensor array reaches the preset maximum gas concentration, it is determined that the robot has approached the hydrogen leak source sufficiently and is now in the leak source area, thus achieving its goal of approaching the leak source. The robot then stops moving forward and enters data acquisition mode. The maximum preset gas concentration is set at 800 ppm, which is approximately 2% of the lower explosive limit of hydrogen, providing a sufficient safety margin to accurately locate the leak without getting too close.

[0055] The robot collects a complete set of information about the leak source at its current location, including: images of the leak source taken by the thermal imager, hydrogen concentration information measured by the gas sensor array, temperature information obtained by the thermal imager, and position coordinates provided by the robot's own positioning system. The robot then reports this information to the control room in real time via a wireless network for analysis and handling by the operators.

[0056] In a preferred embodiment of the present invention, after completing the information reporting, the robot actively withdraws from the leak source area, leaving sufficient operating space for subsequent manual or robotic maintenance operations. The robot enters a standby state, awaiting the next instruction from the control room. The instructions include the following three scenarios: the leak source has been repaired, and the robot resumes the preset path to continue performing the inspection task; the leak source has not yet been repaired, and the robot changes its inspection route, avoids dangerous areas, and continues the inspection; or it changes its posture on-site and remains in standby, continuously monitoring changes in gas concentration in the leak area.

[0057] In a preferred embodiment of the present invention, the motion control motherboard of the quadruped robot also needs to ensure the stability of motion control: (1) Time window smoothing: A sliding window is used to smooth key quantities such as gradient direction and concentration gradient magnitude over the past N sampling periods to eliminate abnormal fluctuations caused by instantaneous airflow interference. This method is equivalent to low-pass filtering in the frequency domain, effectively improving resistance to high-frequency interference.

[0058] (2) Smooth weight switching: The weight switching between stages adopts a gradual increase / decrease strategy. The weight is assumed to switch from the old value to the new value linearly within a set time to avoid sudden changes in the direction of motion caused by weight changes and to ensure the continuity of robot motion.

[0059] (3) Dynamic feedback adjustment: The weights are dynamically fine-tuned based on the deviation between the robot's actual performance results (position change, concentration change rate) and the expected results. For example, if the actual concentration growth rate is lower than expected, the weights are automatically increased. grad Weight; if frequent robot turns are detected, w is automatically increased. smooth Weights. This forms a negative feedback loop, further improving convergence and accuracy.

[0060] Please see Figure 2 , Figure 2 This is a process flow diagram of a multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot, provided as an embodiment of the present invention.

Claims

1. A multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot, characterized in that, The method includes the following steps: The robot performs periodic inspections based on existing map information and following a preset inspection path. Thermal imagers are used to inspect equipment that carries hazardous gases. The real-time data is compared with historical data. If an abnormal area is found, the inspection according to the preset inspection path is stopped and the source of the leak is searched. If no abnormality is found, the periodic inspection continues. The process of stopping the inspection along the preset path and starting the search for the leak source when an abnormal area is detected includes: calculating the multi-objective weighted search direction based on the current search stage and environmental conditions, adjusting the constraint weights in real time, and issuing instructions to make the robot move in the weighted optimized direction; the constraint weights include the weights of concentration gradient, safe distance, motion smoothness, and directional stability; when the robot reaches the area with the maximum preset gas concentration, the area is taken as the leak source area; the robot collects image information, gas concentration information, temperature information, and location information of the leak source area and reports them to the control room.

2. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 1, characterized in that, The method also includes: After identifying the leak source area and reporting the information to the control room, the robot leaves the leak source area and waits for the next instruction, including: if the leak source has been repaired, the robot continues its inspection; if the leak source has not been repaired, the robot changes its inspection route; or the robot changes its posture and waits for further instructions.

3. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 1, characterized in that, The multi-objective weighting includes: The search phase includes: the initial search phase, the intermediate search phase, and the final search phase; In the initial search phase: the weight is configured as w. grad :w safe :w smooth :w stab =0.3:0.4:0.2:0.1; During the mid-stage search: the weight is configured as w. grad :w safe :w smooth :w stab =0.4:0.3:0.2:0.1; In the final search phase: the weight is configured as w. grad :w safe :w smooth :w stab =0.5:0.25:0.15:0.1; Among them, w grad The weights for the concentration gradient reward; w safe Weighting of the safe distance reward; w smooth Weights for motion smoothing rewards; w stab The weight of the directional stable reward; When the system detects that the robot's speed has been below the preset speed threshold for an extended period, it automatically increases the speed. smooth and w stab The weights; When the fluctuation amplitude along the concentration gradient exceeds the preset fluctuation threshold, w is automatically reduced. grad Weighting, increasing w stab The weighting of the weights causes the robot to adopt a more conservative movement strategy.

4. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 3, characterized in that, Concentration gradient reward, safe distance reward, motion smoothness reward, and directional stability reward, including: Concentration gradient reward: Encourages the robot to move along the direction of the largest concentration gradient, which is used to quickly converge to the vicinity of the leakage source; Safe distance reward: Constrains the minimum safe distance between the robot and obstacles and hazardous areas to prevent collisions and excessive approach; Smooth motion reward: Encourages the robot to adopt a smooth and continuous motion trajectory, avoiding getting stuck due to sharp turns; Directional stability bonus: When environmental airflow disturbances cause fluctuations in the direction of the concentration gradient, a time window is used for smoothing to ensure the stability of the motion command.

5. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 3, characterized in that, The multi-objective weighting includes: The transition from the initial search phase to the intermediate search phase occurs when the maximum concentration value collected by the robot reaches the preset intermediate transition threshold C. threshold1 Triggered at time; The transition from the mid-stage search to the final stage search is as follows: when the maximum concentration value collected by the robot reaches the preset final stage transition threshold C. threshold2 Triggered at time; If the robot continuously moves a distance exceeding a preset distance threshold during a certain stage but the concentration does not increase, the system determines that it has entered a dead zone and automatically reduces the concentration. grad Weighting, increasing w smooth and w stab The weights are adjusted until the concentration increases again and the original weights are restored.

6. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 1, characterized in that, The calculation of the source-finding direction for multi-objective weighted calculation includes: The robot is equipped with a gas sensor array, which collects the concentration of hazardous gases in 16 directions in real time. Let the concentration values ​​in 16 directions be C1, C2, C3…C 16 The corresponding direction vectors are d1, d2, d3…d 16 ; (1) Concentration gradient reward The calculation formula is as follows: ； in, Let be the concentration value in the i-th direction. Let i be the direction vector of the i-th direction; (2) Safe distance reward The calculation formula is as follows: ； Where j is S obstacle The index of the obstacle's direction in the set; S obstacle d represents the set of directions corresponding to all obstacles within a safe threshold from the robot; j It is the unit direction vector corresponding to the j-th direction; |S obstacle | represents the number of obstacle directions in the set; (3) Motion smoothing reward The calculation formula is as follows: ； in, The direction of motion at the previous moment. This is the optimal direction calculated so far. For smoothing coefficients; (4) Directional stability reward The calculation formula is as follows: ； Where n is the length of the time window, is the directional gradient of the k-th sampling period; t is the sampling period index at the current time. (5) Final weighted source tracing direction: v final =w grad ×v grad +w safe ×v safe +w smooth ×v smooth +w stab ×v stab ; Among them, v smooth v is obtained by weighted average of the previous motion direction and the current optimal direction. stab This represents the average gradient direction within the time window; (6) Direction of maximum concentration gradient: The robot eventually faces v final Engage in exercise.

7. The multi-target dynamic weighted hydrogen leak localization method based on a quadruped robot according to claim 1, characterized in that, The method also includes: Time window smoothing: The gradient direction and concentration gradient magnitude over the past N sampling periods are smoothed using a sliding window to eliminate abnormal fluctuations caused by instantaneous airflow interference, ensuring that the motion commands issued to the robot are continuous and stable; Smooth weight switching: The weight switching between stages adopts a gradual increase / decrease strategy, rather than a jump switch, to avoid sudden changes in the direction of movement caused by weight changes; Dynamic feedback adjustment: Based on the deviation between the robot's actual execution results and the expected results, the weights are dynamically fine-tuned to form a negative feedback loop, further improving convergence and accuracy; the actual execution results include: position change and concentration change rate.

Images