A heterogeneous sensor fusion detection method for intelligent gangue sorting robot in confined space

Abstract

This invention discloses a heterogeneous sensor fusion detection method for intelligent waste rock sorting robots in confined spaces, belonging to the field of heterogeneous sensor fusion technology. It simultaneously collects VOC time-series signals from a bionic electronic nose array, LiDAR cloud data, and the pose information of an inertial measurement unit (IMU). A generative adversarial network (GAN) model is used to denoise the VOC time-series signals, and voxel filtering and outlier removal are performed on the LiDAR cloud data. Based on CFD simulation, the VOC concentration gradient is predicted, and a three-dimensional environmental map is constructed by combining the LiDAR cloud data and the IMU pose, with an odor concentration heatmap annotated. Robust odor fingerprints are extracted from the denoised VOC time-series signals, aligned with the characteristic distributions of different mining areas, and the classification model is updated. This scheme guides the waste rock sorting robot to prioritize the exploration of hidden areas with low VOC concentrations but high CO2 accumulation by using odor concentration gradients, significantly improving the detection capability of inconspicuous waste rock piles.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous sensor fusion technology, specifically a heterogeneous sensor fusion detection method for a confined space intelligent coal sorting robot. Background Technology

[0002] Coal gangue sorting robots are coal mine robots designed to solve the problem of gangue removal during raw coal production and transportation. They are an effective way to achieve clean coal utilization and green production. Coal gangue sorting robots are mainly divided into cable-driven, Cartesian coordinate, and parallel robot types. Among parallel coal gangue sorting robots, the Delta robot has advantages such as high structural rigidity and high operating speed, meeting the needs of coal gangue sorting.

[0003] Traditional gangue sorting robots rely on single-modal sensors such as vision (e.g., cameras) and weight detection, which are easily affected by complex mining environments such as dust and insufficient lighting. They also cannot directly perceive the characteristics of volatile organic compounds (VOCs) released by gangue, and have limited ability to identify gangue with low concentration and hidden accumulation. Traditional methods also miss detections due to the lack of gas sensing capabilities, resulting in low sorting efficiency, which needs to be improved. Summary of the Invention

[0004] The purpose of this invention is to provide a heterogeneous sensor fusion detection method for intelligent coal sorting robots in confined spaces, so as to solve the problems mentioned in the background art.

[0005] The technical solution of this invention is a heterogeneous sensor fusion detection method for intelligent coal sorting robots in confined spaces, comprising the following steps:

[0006] S1: Activate the bionic electronic nose array, lidar, inertial measurement unit and tactile sensor, perform baseline drift correction on the bionic electronic nose array, and align with the lidar coordinate system;

[0007] S2: Simultaneously acquire VOC timing signals from the bionic electronic nose array, LiDAR cloud data, and pose information from the inertial measurement unit. Use a generative adversarial network model to denoise the VOC timing signals, perform voxel filtering and outlier removal on the LiDAR cloud data, and fuse the inertial measurement unit and LiDAR cloud data to estimate the robot pose.

[0008] S3: Based on CFD simulation, predict the VOC concentration gradient, combine LiDAR cloud data and robot pose information from the inertial measurement unit to construct a 3D environmental map, and annotate the odor concentration heat map;

[0009] S4: Construct a classification model, extract robust odor fingerprints from the denoised VOC time series signals, align the feature distributions of different mining areas, and update the classification model;

[0010] S5: Generate a gangue sorting task queue based on the odor concentration heat map, establish a deep reinforcement learning model to plan the robotic arm grasping path, and adjust the grasping robotic arm action in real time with the feedback from the tactile sensor, and update the odor concentration heat map in real time.

[0011] S6: Collect and grasp the results and feed them back to the classification model for incremental learning, and allocate unexplored areas by sharing the progress of multiple robot tasks;

[0012] S7: Stop the operation according to the preset termination conditions. The termination conditions include the probability of gangue in the odor concentration heat map being less than 0.1 or the preset operation time threshold being reached.

[0013] S8: Store the data in the classification model; classify model parameters and sorting logs, and generate a sorting efficiency analysis report.

[0014] As a further optimization of this technical solution, the baseline drift correction of the bionic electronic nose array in S1 is achieved through the following steps:

[0015] S1a: Acquire the raw signal of the bionic electronic nose array in a VOC-free environment and calculate the baseline offset;

[0016] S1b: A polynomial fitting algorithm is used to generate a compensation curve based on the original signal of the bionic electronic nose array and the calculated baseline offset, so as to correct the data output of the bionic electronic nose array in real time.

[0017] S3 specifically includes the following steps:

[0018] S31: CFD simulation establishes a gas diffusion model based on predefined gangue VOC release rate and mine environmental parameters. The model prediction values ​​of VOC concentration distribution at different spatial locations are calculated and predicted through CFD simulation. The measured VOC concentration of the bionic electronic nose array is compared with the model prediction values ​​to calibrate the gangue candidate area. A preset odor fingerprint library stores the VOC components and concentration characteristics released by gangue in different mining areas.

[0019] S32: Construct a three-dimensional geometric map using lidar cloud data and pose information from the inertial measurement unit, and superimpose concentration features as voxel attributes onto the three-dimensional geometric map to form a dynamic three-dimensional map that integrates odor concentration.

[0020] S33: The particle filter algorithm is used to update the probability distribution of gangue in the dynamic three-dimensional map in real time, and the A-Star algorithm is used to plan the robot navigation path, giving priority to exploring areas with low VOC concentration and high CO2 accumulation.

[0021] The conditions for calibrating the gangue candidate region in S31 are as follows:

[0022] When the measured VOC concentration of the bionic electronic nose array is lower than the preset low threshold and the CO2 concentration of the gangue candidate area is higher than the preset high threshold, it is determined to be a gangue candidate area.

[0023] When the deviation between the measured VOC concentration and the model prediction value of the bionic electronic nose array exceeds 20%, the S4 mentioned above is triggered through the following steps:

[0024] S41: Construct a synthetic dataset containing dust and humidity noise, and use an adversarial network model to generate denoised VOC time-series signals;

[0025] S42: An LSTM-Transformer hybrid model is used to extract odor fingerprints from the VOC time-series signals acquired by the bionic electronic nose array. The time-series features output by the LSTM-Transformer hybrid model are dynamically weighted through an attention mechanism to generate an odor fingerprint vector, which is finally mapped to a voxel-level gangue candidate region probability value.

[0026] S5 is implemented through the following steps:

[0027] S51: Based on the probability value of the candidate gangue area and the distance to the target area, a sorting priority queue is generated using the Q-learning algorithm, where areas with a probability value higher than a preset threshold are marked as high priority.

[0028] S52: Generates optimized paths for robotic arm joint angles and end-effector velocities using a deep reinforcement learning model, and performs collision detection based on LiDAR cloud data;

[0029] S53: When the tactile sensor detects an abnormal pressure distribution, it triggers the replanning of the robotic arm joint path and updates the odor concentration heat map.

[0030] The S6 also includes the following steps:

[0031] S61: If the robotic arm fails to grasp continuously for more than the preset number of times, trigger step S2 to rescan the target area;

[0032] S62: Assigns multi-robot collaborative tasks through ant colony optimization algorithm, covering unexplored areas and reducing repetitive work.

[0033] The preset operation time threshold in S7 is dynamically calculated based on the area of ​​the operation area.

[0034] The S7 also includes equipment maintenance steps, specifically including:

[0035] S71: Cleans the filter of the bionic electronic nose array, using compressed air to remove dust deposits;

[0036] S72: Uses LiDAR to scan the joint gaps of the robotic arm to detect whether the wear exceeds a preset threshold;

[0037] S73: If wear of the robotic arm end effector is detected to be ≥0.5mm, a warning signal is triggered and a maintenance log is recorded.

[0038] This invention provides an improved heterogeneous sensor fusion detection method for intelligent coal sorting robots in confined spaces, which has the following improvements and advantages compared with the prior art:

[0039] 1. This invention predicts VOC concentration gradients based on CFD simulations and corrects the gas diffusion model in real time using measured data, thus labeling the probability distribution of gangue. Compared to traditional methods that rely solely on visual features (such as color and shape), this method guides the gangue sorting robot to prioritize exploring hidden areas with low VOC concentrations but high CO2 accumulations through odor concentration gradients, significantly improving the detection capability of inconspicuous gangue piles.

[0040] 2. Odor fingerprint extraction using LSTM-Transformer hybrid model: LSTM captures temporal dependencies, while Transformer extracts global features through self-attention mechanism, generating a 12-dimensional robust odor fingerprint vector. Compared with traditional frequency domain analysis methods such as Fourier transform, it can more accurately characterize the dynamic VOC signal features.

[0041] 3. For the low VOC concentration environment of mines, baseline drift correction using a biomimetic electronic nose array improves the detection accuracy of low-concentration signals. Combined with a particle filtering algorithm, the probability distribution of gangue in the 3D map is updated in real time, achieving millimeter-level positioning in narrow areas. The operation time threshold is dynamically calculated based on the area and integrates equipment maintenance mechanisms (such as cleaning the electronic nose filter and detecting wear on the robotic arm) to ensure stable operation in dusty and humid environments over long periods.

[0042] 4. This invention is based on multi-physics coupling modeling, which integrates gas diffusion model and robot kinematics model. It introduces VOC concentration gradient factor through the cost function of A-Star algorithm, and prioritizes the exploration of gangue boundary areas with significant odor concentration changes. Compared with traditional path planning that only considers distance or energy consumption, it significantly improves the detection efficiency in complex terrain. Attached Figure Description

[0043] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0044] Figure 1 This is a flowchart of the heterogeneous sensor fusion detection method for the intelligent coal sorting robot in confined space according to the present invention. Detailed Implementation

[0045] The following will be combined with the appendix Figure 1This invention will be described in detail, and the technical solutions in the embodiments of this invention will be clearly and completely described. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0046] This invention constructs an intelligent system for sorting gangue in complex and confined spaces. The core solution is to address the technical problem of how to achieve high-precision real-time detection, positioning, and efficient sorting of gangue in noisy and dynamically changing mining environments through multi-sensor fusion and intelligent algorithms.

[0047] Example 1:

[0048] This invention provides an improved method for heterogeneous sensor fusion detection in a confined space intelligent coal sorting robot, such as... Figure 1 As shown, it includes the following steps:

[0049] S1: Activate the bionic electronic nose array, lidar, inertial measurement unit, and tactile sensor. Correct the baseline drift of the bionic electronic nose array and align it with the lidar coordinate system. The tactile sensor enables the robot to operate reliably through accurate feedback of tactile data even in extreme environments such as mine dust and insufficient light. The tactile sensor is existing technology and will not be described in detail here.

[0050] S2: Simultaneously acquire VOC timing signals from the bionic electronic nose array, LiDAR cloud data, and pose information from the inertial measurement unit. Use a generative adversarial network model to denoise the VOC timing signals, perform voxel filtering and outlier removal on the LiDAR cloud data, and fuse the inertial measurement unit and LiDAR cloud data to estimate the robot pose.

[0051] The Generative Adversarial Network (GAN) model comprises a generator and a discriminator. The generator receives noisy VOC (Voice-Oriented Content) time-series signals and outputs a denoised signal; the discriminator determines whether the signal is true, noise-free data. Through adversarial training, the generator gradually improves its denoising capabilities, eventually outputting a VOC signal with a signal-to-noise ratio improved to 30 dB. An inertial measurement unit (IMU) provides high-frequency but drift-prone pose data, while a lidar system provides high-precision but low-frequency spatial information. Kalman filtering fuses the two data sets to estimate the robot's pose in real time.

[0052] S3: Based on CFD simulation, predict the VOC concentration gradient, combine LiDAR cloud data and robot pose information from the inertial measurement unit to construct a 3D environmental map, and annotate the odor concentration heat map;

[0053] S4: Construct a classification model, extract robust odor fingerprints from the denoised VOC time series signal, align the feature distributions of different mining areas, and update the classification model. Use a domain adversarial neural network to align the feature distributions of different mining areas and update the classification model. The domain adversarial neural network is used in transfer learning to align the feature distributions of different mining areas (source domain and target domain) and reduce domain differences.

[0054] By inputting mine environmental parameters (wind speed, temperature, VOC release rate from gangue), the VOC concentration gradient distribution is predicted, and the model is calibrated by comparing it with measured data.

[0055] S5: Generate a gangue sorting task queue based on the odor concentration heat map, establish a deep reinforcement learning model to plan the robotic arm grasping path, and adjust the grasping robotic arm action in real time with the feedback from the tactile sensor, and update the odor concentration heat map in real time.

[0056] S6: Collect and grasp the results and feed them back to the classification model for incremental learning, and allocate unexplored areas by sharing the progress of multiple robot tasks;

[0057] The A-Star algorithm is a heuristic pathfinding algorithm used for robot navigation path planning. It introduces an odor concentration gradient factor to the traditional A-Star algorithm, balancing path distance with gangue detection efficiency. It prioritizes regions with significant VOC concentration gradients (such as gangue boundaries) to improve exploration efficiency. The cost function is:

[0058] g(n): Actual cost (path length, energy consumption)

[0059] h(n): Heuristic cost (Euclidean distance to the target);

[0060] VOC concentration gradient (guides the robot to prioritize exploration of rock boundaries).

[0061] β: Weighting coefficient (balancing path efficiency and detection sensitivity).

[0062] Odor gradient factor: Add the VOC concentration change rate to g(n) to prioritize the exploration of regions with significant concentration gradients (such as gangue boundaries).

[0063] The technical advantage lies in the ability to prioritize the exploration of hidden areas: the boundaries of gangue can be identified through concentration gradients, and even when the VOC concentration is below a preset threshold, the gangue pile can still be located by gradient changes (e.g., a gradient of -0.1ppm / m in a 0.2ppm concentration area indicates the existence of a boundary).

[0064] It can also improve detection efficiency: compared with traditional path planning that only considers distance, the detection efficiency in complex terrain is improved by more than 30%.

[0065] The Q-value update formula of Q-learning is used to generate a priority queue for sorting tasks, optimize the sorting order of gangue through reinforcement learning, balance the sorting success rate and the energy consumption of movement, and dynamically adjust the task priority (areas with probability values ​​> 0.8 are given priority).

[0066] Q value update:

[0067]

[0068] Q(s,a): The Q value (expected cumulative reward) of performing action a in state s.

[0069] s: State, defined as the probability value of gangue in the odor map and the distance between the target and the robot.

[0070] a: Action, namely, selecting the next area of ​​gangue to be sorted.

[0071] α: Learning rate, which controls the rate at which new information updates the old Q value.

[0072] R: Reward function, formula: Sorting success rate, Mobile energy consumption R = Sorting success rate - λ·

[0073] Mobile energy consumption (λ is the energy consumption weighting coefficient; the document does not specify the value, but it reflects the goal of "minimizing energy consumption").

[0074] γ: Discount factor, used to balance the importance of current rewards and future rewards (0≤γ≤1).

[0075] s': The next state after performing action a.

[0076] a': All possible actions in the next state s'. This represents the maximum Q value of the next state.

[0077] Task queue generation: Priority is dynamically adjusted based on Q value, with areas with probability value > 0.8 being processed first to minimize movement energy consumption.

[0078] The LSTM-Transformer hybrid model refers to: LSTM: adept at capturing temporal dependencies, used to extract local temporal features of VOC signals.

[0079] The practical significance lies in the fact that LSTM captures the temporal dependence of VOC signals, while Transformer extracts global features through a self-attention mechanism to generate a 12-dimensional odor fingerprint vector.

[0080] S7: Stop the operation according to the preset termination conditions. The termination conditions include the probability of gangue in the odor concentration heat map being less than 0.1 or the preset operation time threshold being reached.

[0081] S8: Store the data in the classification model and generate a sorting efficiency analysis report.

[0082] In this embodiment, after the robot starts up, it first performs system initialization and sensor calibration, turns on the bionic electronic nose array, lidar, inertial measurement unit, tactile sensor and robotic arm control system, corrects the baseline drift of the bionic electronic nose array, aligns the lidar coordinate system, loads the confined space map framework and sets the initial parameters of the gas diffusion model, establishes low-latency communication between the edge computing unit and the cloud and the robotic arm controller, and outputs the calibrated data streams of the bionic electronic nose array, lidar, inertial measurement unit and tactile sensor and the initial environmental parameter configuration.

[0083] Entering the multimodal data real-time acquisition and preprocessing stage, the bionic electronic nose array acquires VOC concentration time-series signals at a frequency of ≥100Hz, the lidar generates high-precision point cloud data at 10Hz, and the inertial measurement unit provides robot pose information at 100Hz. Then, these data are preprocessed, the odor signal is denoised using a pre-trained GAN model, the lidar cloud data is optimized through voxel filtering and outlier removal, the inertial measurement unit and lidar cloud data are fused, and the robot's precise pose is estimated through Kalman filtering. The denoised VOC signal, optimized point cloud, and high-precision pose information are output.

[0084] Dynamic odor SLAM and map construction are performed based on CFD simulation and real-time environmental parameter prediction of VOC concentration gradient distribution. The prediction results are compared with the measured data of the bionic electronic nose array to dynamically correct the model parameters. Then, the lidar cloud data and the pose of the inertial measurement unit are input into ORB-SLAM3 to construct a 3D environmental map, and the odor concentration heat map is marked on the map. The odor map is updated in real time using a particle filter algorithm, and hotspot localization is optimized by combining Gaussian process regression. Explored and unexplored areas are marked, and the A-Star algorithm is used to plan the robot navigation path. The output is a dynamic 3D map with fused odor concentration and the robot navigation path. In this process, environmental parameters are collected and preprocessed in real time. Temperature and humidity sensors and barometers are deployed to collect environmental parameters. The bionic electronic nose array collects raw VOC signals at 10Hz and records the lidar cloud data timestamps simultaneously.

[0085] The probability distribution of gangue is updated in real time using a particle filter algorithm. The particle weight is calculated based on the following formula:

[0086]

[0087] in, To observe the likelihood function, the weights are dynamically adjusted based on the deviation between the measured VOC concentration and the CFD prediction. Particle aggregation areas with a weight > 0.8 are marked as high priority. Multi-source data (VOC concentration, lidar reflectivity) are fused using a particle filtering algorithm to improve the accuracy of gangue location, which is superior to traditional visual positioning errors. When changes in the mine ventilation system cause changes in the VOC diffusion pattern, the model automatically corrects the path to avoid missed detections caused by static planning.

[0088] Considering the low VOC concentration released by gangue, the biomimetic electronic nose array is switched to high gain mode to improve the detection accuracy of low concentration. CFD gas diffusion modeling is performed based on gangue characteristics. Environmental parameters, gangue VOC release rate and prior knowledge of gangue distribution are input. The k-ε turbulence model is used to simulate gas diffusion and mark the possible gangue areas with low concentration VOC + high concentration CO2.

[0089] By calibrating the gangue characteristics of the model with measured data, the model is triggered when the deviation between the measured concentration of the bionic electronic nose array and the CFD prediction value is >20%. The focus is on calibrating areas with low VOC concentration but high CO2 concentration as gangue pile candidate areas. The gangue VOC release rate is corrected and the gangue odor fingerprint database is updated. Multimodal SLAM fusion and semantic annotation are performed. A 3D map with an accuracy of ±5cm is constructed using the LOAM algorithm and ORB-SLAM3 to mark coal piles and gangue piles. The VOC concentration is standardized and superimposed on the map as a voxel attribute to mark coal and gangue candidate areas. Particle filtering is updated based on gangue distribution, and particle sets are evenly distributed in unexplored areas.

[0090] Particle weights are assigned based on the probability of gangue predicted by CFD. If the measured VOC concentration at the particle location is <0.3 and the lidar detects a low reflectivity surface, the weight is multiplied by 2. Particles with a weight <0.1 are excluded. The system focuses on exploring high-probability gangue areas such as the corners of the tunnel, and performs dynamic path planning and hotspot tracking. An odor concentration gradient factor is added to the A-Star algorithm cost function of the robot planning system. The system prioritizes the path with the fastest decrease in VOC concentration, sets obstacle avoidance constraints, and outputs a dynamic 3D map that integrates odor concentration and a robot navigation path with waypoints.

[0091] Next, anti-interference feature extraction and transfer learning are performed. The denoised VOC signal is input into the LSTM-Transformer hybrid model, and robust odor fingerprints are extracted through the attention mechanism. The feature distribution of the new mining area and the source mining area is aligned using a domain adversarial neural network. Under the federated learning framework, the parameters of each mining area are uploaded to the cloud to aggregate the global model. Based on the transfer learning model, the gangue probability is output and the odor map classification label is updated. The gangue probability distribution map and the optimized classification model are output.

[0092] Based on the probability and distance of gangue in the odor map, a sorting task queue is generated by combining the Q-learning algorithm. The target location is input into the deep reinforcement learning DDPG model to generate the optimal grasping path, rehearse the path and avoid obstacles. When the robotic arm grasps, the pressure distribution is fed back in real time through tactile sensors. If abnormal pressure is detected, path replanning is triggered, and the robotic arm grasping path and real-time tactile feedback signal are output.

[0093] Then, the closed-loop execution and dynamic optimization phase begins. The robotic arm grabs the target gangue according to the planned path and places it in the designated area. The odor map is updated in real time, and the grabbing results are collected and fed back to the transfer learning model to incrementally update the classification weights. If there are 3 consecutive failures, the SLAM module is triggered to rescan the area. The updated odor map and task progress are shared through the blockchain. The unexplored area is allocated to the idle robot using the swarm intelligence algorithm. The sorted gangue pile and the optimized global map and model are output.

[0094] The task is terminated when the probability of gangue in the odor map is less than 0.1 or the preset operation time threshold is reached. The complete SLAM map, classification model parameters and sorting log are stored, and a sorting efficiency report is generated. The sensor surface also needs to be cleaned and the wear status of the robotic arm is checked. The archived data package and sorting efficiency analysis report are output.

[0095] Example 2 is an explanation of Example 1. Specifically,

[0096] The baseline drift correction of the bionic electronic nose array in S1 is achieved through the following steps:

[0097] S1a: Acquire the raw signal of the bionic electronic nose array in a VOC-free environment and calculate the baseline offset;

[0098] S1b: A polynomial fitting algorithm is used to generate a compensation curve based on the original signal of the bionic electronic nose array and the calculated baseline offset, so as to correct the data output of the bionic electronic nose array in real time.

[0099] As a further optimization of this technical solution, step S3 specifically includes the following steps:

[0100] S31: CFD simulation establishes a gas diffusion model based on predefined gangue VOC release rate and mine environmental parameters. The model prediction values ​​of VOC concentration distribution at different spatial locations are calculated and predicted through CFD simulation. The measured VOC concentration of the bionic electronic nose array is compared with the model prediction values ​​to calibrate the gangue candidate area. A preset odor fingerprint library stores the VOC components and concentration characteristics released by gangue in different mining areas.

[0101] S32: Construct a three-dimensional geometric map using lidar cloud data and pose information from the inertial measurement unit, and superimpose concentration features as voxel attributes onto the three-dimensional geometric map to form a dynamic three-dimensional map that integrates odor concentration.

[0102] S33: The particle filter algorithm is used to update the probability distribution of gangue in the dynamic three-dimensional map in real time, and the A-Star algorithm is used to plan the robot navigation path, giving priority to exploring areas with low VOC concentration and high CO2 accumulation.

[0103] Particle filter algorithm formula:

[0104]

[0105] Parameter explanation: The state of the i-th particle (position + probability of being hit by gangue); w: particle weight; t:

[0106] Time step.

[0107] Observational likelihood (based on VOC concentration and lidar reflectivity).

[0108] Used to update the probability distribution of gangue in a 3D map in real time. Areas with a particle weight > 0.1 are marked as high-probability candidate areas, with a positioning accuracy of ±0.5m.

[0109] Polynomial fitting is a mathematical method that approximates data points using a polynomial function. Its goal is to find a set of polynomial coefficients such that the polynomial curve minimizes the error from the given data points (typically using the least squares method). For baseline drift correction in biomimetic electronic nose arrays, polynomial fitting is used to model the baseline shift of the sensor in a VOC-free environment, generating compensation curves to eliminate drift noise.

[0110] In Example 2, a cubic polynomial is used for dynamic compensation:

[0111] y = a0 + a1x + a2x 2 +a3x 3

[0112] Parameter explanation:

[0113] y: The compensated sensor output value.

[0114] x: Time or the original output sequence of the sensor.

[0115] a0, a1, a2, a3: Polynomial coefficients, determined by fitting baseline offset data.

[0116] The practical significance lies in the fact that cubic polynomials can flexibly fit the nonlinear trend of sensor baseline drift.

[0117] (Such as slow exponential decay or periodic fluctuations), while avoiding overfitting problems that may be caused by higher-order polynomials.

[0118] In Example 2, regarding the specific implementation of baseline drift correction for the bionic electronic nose array, the system employs a cubic polynomial fitting algorithm for dynamic compensation. The specific steps are as follows: In a VOC-free enclosed environment, the bionic electronic nose array continuously acquires raw signals for 10 minutes at a sampling frequency of 10Hz. By calculating the standard deviation of the baseline offset (typical value ≤ 0.8mV), the coefficients of the cubic polynomial are determined, and a compensation curve is generated to correct the output of the bionic electronic nose array in real time.

[0119] Example 3, this example is an explanation of Example 2:

[0120] The conditions for calibrating the gangue candidate region in S31 are as follows:

[0121] When the measured VOC concentration of the bionic electronic nose array is lower than the preset low threshold and the CO2 concentration of the gangue candidate area is higher than the preset high threshold, it is determined to be a gangue candidate area.

[0122] When the deviation between the measured VOC concentration of the bionic electronic nose array and the model prediction exceeds 20%, dynamic correction of the VOC release rate of gangue and the odor fingerprint database is triggered.

[0123] In Example 3, the calibration conditions for the gangue candidate area were achieved through multi-sensor data fusion. A preset low threshold for VOC concentration was 0.3 ppm (based on typical gangue release characteristics), and a high threshold for CO2 concentration was 1000 ppm (reflecting the anoxic environment of the gangue accumulation area). When the measured VOC concentration of the bionic electronic nose array was below 0.3 ppm and the CO2 concentration was above 1000 ppm, it was identified as a gangue candidate area. If the deviation between the measured VOC concentration and the CFD model prediction exceeded 20% (e.g., predicted value 1.0 ppm, measured value ≤0.8 ppm or ≥1.2 ppm), a dynamic correction mechanism was triggered. The correction process included adjusting the gangue VOC release rate model parameters (e.g., correcting the release coefficient from 0.05 g / s to 0.04 g / s) and updating the odor fingerprint database.

[0124] Example 4, this example is an explanation of Example 1, and S4 is specifically implemented through the following steps:

[0125] S41: Construct a synthetic dataset containing dust and humidity noise, and use an adversarial network model to generate denoised VOC time-series signals;

[0126] S42: An LSTM-Transformer hybrid model is used to extract odor fingerprints from the VOC time-series signals acquired by the bionic electronic nose array. The time-series features output by the LSTM-Transformer hybrid model are dynamically weighted through an attention mechanism to generate an odor fingerprint vector, which is finally mapped to a voxel-level gangue candidate region probability value.

[0127] Example 5, this example is an explanation of Example 1, and S5 is specifically implemented through the following steps:

[0128] S51: Based on the probability value of the candidate gangue area and the distance to the target area, a sorting priority queue is generated using the Q-learning algorithm, where areas with a probability value higher than a preset threshold are marked as high priority.

[0129] S52: Generates optimized paths for robotic arm joint angles and end-effector velocities using a deep reinforcement learning model, and performs collision detection based on LiDAR cloud data;

[0130] S53: When the tactile sensor detects an abnormal pressure distribution, it triggers the replanning of the robotic arm joint path and updates the odor concentration heat map.

[0131] During the robotic arm's grasping process, tactile sensors continuously monitor the pressure distribution. If an abnormal pressure is detected (such as excessively high / low local pressure or uneven pressure distribution), it indicates that the grasping posture or force is inappropriate, triggering a path replanning algorithm to adjust the robotic arm joint angles and end effector speed in real time to ensure stable grasping.

[0132] The S6 also includes the following steps:

[0133] S61: If the robotic arm fails to grasp continuously for more than the preset number of times, trigger step S2 to rescan the target area;

[0134] S62: Assigns multi-robot collaborative tasks through ant colony optimization algorithm, covering unexplored areas and reducing repetitive work.

[0135] Example 6 is an explanation of Example 1. In S7, the preset operation time threshold is dynamically calculated based on the area of ​​the operation area.

[0136] As one of the criteria for determining grasping failure: if the robotic arm fails to grasp continuously more than a preset number of times, it triggers the dynamic odor SLAM module to rescan the target area and correct the gangue positioning information, based on the pressure data from the tactile sensor.

[0137] The S7 also includes equipment maintenance steps, specifically including:

[0138] S71: Cleans the filter of the bionic electronic nose array, using compressed air to remove dust deposits;

[0139] S72: Uses LiDAR to scan the joint gaps of the robotic arm to detect whether the wear exceeds a preset threshold;

[0140] S73: If wear of the robotic arm end effector is detected to be ≥0.5mm, a warning signal is triggered and a maintenance log is recorded.

[0141] The high-speed airflow of compressed air is used to blow away dust particles adhering to the filter screen, restoring the filter's permeability and sensor sensitivity. The compressed air nozzle is positioned on the robot, with the airflow direction directed towards the filter screen of the bionic electronic nose array, and the pressure is set to 0.5 MPa.

[0142] It should be noted that all calculation formulas in this application employ, but are not limited to, regression analysis from machine learning algorithms to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikit-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their validity and accuracy, and ensuring that the calculation process conforms to the constraints of natural laws, rather than being based on artificially set rules.

[0143] The technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of this invention.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting a limited space intelligent gangue sorting robot heterogeneous sensor fusion, characterized in that: The method comprises the following steps: S1: start the bionic electronic nose array, laser radar, inertial measurement unit and tactile sensor, correct the baseline drift of the bionic electronic nose array, and align the coordinate system of the laser radar; S2: synchronously collect the VOC time sequence signal of the bionic electronic nose array, the cloud data of the laser radar and the pose information of the inertial measurement unit, denoise the VOC time sequence signal through a generative adversarial network model, perform voxel filtering and outlier elimination on the cloud data of the laser radar, and estimate the robot pose by fusing the inertial measurement unit and the cloud data of the laser radar; S3: predict the VOC concentration gradient based on CFD simulation, construct a three-dimensional environment map in combination with the cloud data of the laser radar and the robot pose information of the inertial measurement unit, and label the odor concentration heat map; S4: construct a classification model, extract a robust odor fingerprint from the denoised VOC time sequence signal, align the characteristic distribution of different mining areas, and update the classification model; S5: generate a gangue sorting task queue according to the odor concentration heat map, establish a deep reinforcement learning model to plan the grabbing path of the mechanical arm of the robot, adjust the grabbing arm action in combination with the real-time feedback of the tactile sensor, and update the odor concentration heat map in real time; S6: collect the grabbing result and feed it back to the classification model for incremental learning, and distribute the unexplored area by sharing the multi-robot task progress; S7: stop the work according to the preset termination condition, and the termination condition includes that the gangue probability in the odor concentration heat map is all lower than 0.1 or the preset work time threshold is reached; S8: store to the classification model.

2. The method according to claim 1, wherein, The baseline drift correction of the bionic electronic nose array in S1 is realized by the following steps: S1a: collect the original signal of the bionic electronic nose array in a VOC-free environment, and calculate the baseline offset; S1b: generate a compensation curve according to the original signal of the bionic electronic nose array and the calculated baseline offset by using a polynomial fitting algorithm, and correct the data output of the bionic electronic nose array in real time.

3. The method of claim 1, wherein the method further comprises: S3 specifically comprises the following steps: S31: the CFD simulation establishes a gas diffusion model based on a predefined gangue VOC release rate and mine environment parameters, calculates and predicts the model prediction value of the VOC concentration distribution at different spatial positions, compares the measured VOC concentration of the bionic electronic nose array with the model prediction value to calibrate the gangue candidate area, and stores the VOC composition and concentration characteristics of the gangue release in different mining areas in a preset odor fingerprint library; S32: construct a three-dimensional geometric map through the cloud data of the laser radar and the pose information of the inertial measurement unit, superimpose the concentration characteristics as voxel attributes to the three-dimensional geometric map to form a dynamic three-dimensional map fused with odor concentration; S33: update the gangue probability distribution in the dynamic three-dimensional map in real time by using a particle filtering algorithm, and plan the robot navigation path by using an A-Star algorithm, and preferentially explore the low VOC concentration and high CO2 accumulation area.

4. The method according to claim 3, wherein, The condition for calibrating the gangue candidate area in S31 is: When the measured VOC concentration of the bionic electronic nose array is lower than a preset low threshold and the CO2 concentration of the gangue candidate area is higher than a preset high threshold, the gangue candidate area is determined. When the deviation between the measured VOC concentration of the biomimetic electronic nose array and the predicted value of the model exceeds 20%, the dynamic correction of the gangue VOC release rate and the odor fingerprint library is triggered.

5. The method according to claim 1, wherein, The S4 is specifically implemented by the following steps: S41: Construct a synthetic data set containing dust and humidity noise, and use an adversarial network model to generate a denoised VOC time series signal; S42: Use an LSTM-Transformer hybrid model to extract the odor fingerprint of the VOC time series signal collected by the biomimetic electronic nose array, dynamically weight the time series features output by the LSTM-Transformer hybrid model through an attention mechanism, generate an odor fingerprint vector, and finally map it to a voxel-level gangue candidate area probability value.

6. The method of claim 1, wherein the method is a method of detecting a limited space intelligent gangue sorting robot heterogeneous sensor fusion. The S5 is specifically implemented by the following steps: S51: According to the probability value of the gangue candidate area and the distance of the target area, generate a sorting priority queue through the Q-learning algorithm, where the area with a probability value higher than a preset threshold is marked as high priority; S52: Generate an optimized path for the joint angle and end speed of the mechanical arm through a deep reinforcement learning model, and perform collision detection based on laser radar cloud data; S53: When the tactile sensor detects abnormal pressure distribution, trigger the mechanical arm joint path re-planning and update the odor concentration thermodynamic map.

7. The method of claim 1, wherein the method is a method of detecting a limited space intelligent gangue sorting robot heterogeneous sensor fusion. The S6 also has the following steps: S61: If the mechanical arm fails to continuously grasp for more than a preset number of times, trigger the S2 step to rescan the target area; S62: Distribute multi-robot collaborative tasks through an ant colony optimization algorithm to cover unexplored areas and reduce repetitive work.

8. The method of claim 1, wherein the method is a method of detecting a limited space intelligent gangue sorting robot heterogeneous sensor fusion. The preset operation time threshold in S7 is dynamically calculated based on the area of the operation region.

9. The method of claim 1, wherein the method further comprises: The S7 also includes a device maintenance step, specifically including: S71: Clean the filter screen of the biomimetic electronic nose array, and use compressed air to remove dust and other attachments; S72: Scan the joint clearance of the mechanical arm through the laser radar to detect whether the wear exceeds the preset threshold; S73: If the wear of the end effector of the mechanical arm is detected to be ≥0.5mm, trigger an early warning signal and record a maintenance log.

Images