A Multi-Sensor Fusion Method and System for Rock Hardness Identification in Tunneling Based on Geometric Consistency Maps
By using a multi-sensor fusion method based on geometric consistency maps, the problem of low rock hardness identification accuracy of traditional anchor drilling rigs in underground coal mines has been solved, achieving efficient identification in complex environments and improving identification accuracy and equipment adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI UNIV OF SCI & TECH
- Filing Date
- 2026-04-17
- Publication Date
- 2026-06-02
AI Technical Summary
Traditional rock hardness identification methods for rock bolt drilling rigs in coal mines suffer from low accuracy and poor robustness. They are particularly difficult to effectively extract rock hardness features in high-noise environments, and traditional graph construction methods rely on manually set thresholds, making them unsuitable for complex environments.
A multi-sensor fusion method based on geometric consistency graphs is adopted. The signal is mapped to a two-dimensional geometric point cloud through pseudo-time coordinates. The node features are extracted by random projection method, and an adjacency relationship and weight matrix with adaptive threshold are constructed. Deep learning is combined with graph convolutional neural network to achieve end-to-end identification of rock hardness.
By effectively mining the geometric correlation of multi-sensor signals, the accuracy and adaptability of rock hardness identification are improved. It can accurately identify rock hardness in complex environments such as high dust, high humidity, and strong vibration, thereby improving drilling efficiency and equipment life.
Smart Images
Figure CN122132892A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent coal mining and rock hardness identification during drilling, specifically to a multi-sensor fusion method and system for identifying rock hardness in fully mechanized tunneling based on geometric consistency maps. Background Technology
[0002] In coal mine underground roadway support operations, rock bolt drilling rigs have become key equipment for ensuring the stability of surrounding rock in modern high-yield and high-efficiency mines due to their advantages such as strong adaptability to complex geological conditions, high drilling accuracy, and ability to achieve rapid support operations. A rock bolt drilling rig system typically consists of modules such as a power head, propulsion mechanism, hydraulic system, and control system. It completes the drilling task of the anchor bolt hole by driving the drill rod to rotate and applying axial thrust to cut the rock. Quickly and accurately identifying the rock hardness of the longwall face and dynamically adjusting the drill-anchor rotation speed and output torque are of great significance for improving the speed and quality of anchor support.
[0003] However, the underground environment in coal mines is extremely harsh, characterized by high dust levels, high humidity, strong vibrations, limited space, and complex geological structures. These adverse factors cause anchor drilling rigs to face complex working conditions during drilling, such as drastic changes in rock hardness, interbedded bedding, and inclusions of rock, making them highly susceptible to accidents such as low drilling efficiency and abnormal drill bit wear. Therefore, effectively identifying the rock hardness during anchor drilling is of great practical significance and an urgent technical requirement for ensuring safe and efficient coal mine tunneling and achieving adaptive control of drilling parameters.
[0004] In traditional rock hardness identification methods, the most crucial step is feature extraction. The aim is to extract effective features reflecting the physical and mechanical properties of rocks from multi-source sensor signals to improve identification accuracy. Commonly used feature extraction methods include time-domain statistical analysis, spectral analysis, and wavelet transform. Currently, most identification methods require manual feature selection when dealing with non-stationary and nonlinear drilling signals; different feature selections determine the effectiveness of hardness identification. Furthermore, the performance of existing feature extraction methods often relies on expert experience, making it difficult to fully explore the deep correlations between multi-source heterogeneous data. Introducing deep learning theory into the field of rock hardness identification overcomes these shortcomings. Compared to traditional shallow models, deep learning constructs an end-to-end model. It can directly obtain rock hardness state information from acquired vibration, current, voltage, and oil pressure signals, while simultaneously meeting the high requirements for accuracy and real-time performance in the era of intelligent mining.
[0005] Graph neural networks, as an important branch of deep learning, have shown significant advantages in processing non-Euclidean structured data and are widely used in recommender systems, link prediction, node classification, and protein structure prediction. Their core capability lies in effectively modeling and mining the complex relationships between nodes, thereby achieving deep feature extraction and providing a new technical path and research opportunity for rolling bearing fault diagnosis. However, traditional graph neural network methods struggle to effectively extract features in noisy environments, and the fixed receptive field results in insufficient feature transmission between nodes.
[0006] In summary, traditional graph construction methods require manual setting of numerous parameters, making it difficult to adapt to the dynamic changes in sensor data from underground coal mines, resulting in weak generalization ability of the graph structure. Traditional methods often only focus on the numerical similarity of sensor signals, failing to delve into the geometric structural consistency of multiple sensors in the spatiotemporal distribution within a single sample, leading to insufficient feature discrimination and ultimately low recognition accuracy. Summary of the Invention
[0007] To overcome the technical difficulties of low accuracy and poor robustness in existing methods for identifying rock hardness during coal mine drilling in noisy environments, this invention proposes a multi-sensor fusion method and system for identifying rock hardness in coal mines based on geometric consistency maps. The invention achieves these technical problems through the following technical solutions: Option 1: This invention proposes a multi-sensor fusion method for identifying rock hardness in tunneling based on geometric consistency maps. The method includes the following steps: Step 1: Acquire multi-sensor time series signals, map the multi-sensor time series signals into a two-dimensional geometric point cloud by introducing pseudo-time coordinates, and extract the geometric fingerprint features of the multi-sensor nodes using the random projection method. Step 2: Construct the adjacency relationship and weight matrix among multiple sensors based on the geometric consistency metric, and generate graph data containing topological structure and attribute information by combining the temporal statistical characteristics of the nodes themselves; Step 3: Input the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning, and realize the identification of rock hardness categories through end-to-end training.
[0008] Furthermore, a preferred embodiment is provided, wherein the method for mapping multi-sensor time series signals into a two-dimensional geometric point cloud by introducing pseudo-time coordinates in step 1 is as follows:
[0009] in, For the first i The first of each sensor m The actual sequence signal of each sample, L This is the signal length.
[0010] Furthermore, a preferred embodiment is provided, wherein the method for extracting the geometric fingerprint features of multiple sensor nodes using random projection in step 1 is as follows:
[0011] in, Let be a random projection matrix whose elements follow a Gaussian distribution with mean 0 and variance 0.1, and the projection dimension is . .
[0012] Furthermore, a preferred embodiment is provided, wherein the method for constructing the adjacency relationship and weight matrix among multiple sensors based on the geometric consistency metric in step 2 is as follows: Step 2.1: For any two sensor nodes in the figure, calculate the Euclidean distance between the geometric consistency vectors of the two nodes; Step 2.2: Convert the distance metric into a similarity metric using a Gaussian kernel function; Step 2.3: Using the adaptive thresholding method, define a dynamic threshold and construct the adjacency matrix and weight matrix of the graph based on the dynamic threshold.
[0013] Furthermore, a preferred embodiment is provided, wherein step 2 specifically includes the following steps: The method for calculating the Euclidean distance between the geometric consistency vectors of two nodes in step 2.1 is as follows:
[0014]
[0015] in, m i and s i Let be the mean vector and standard deviation vector of the projected point set, respectively. i i For the two principal directions of the projection point set v 1 and v The included angle of 2 is calculated using the following formula:
[0016]
[0017] Step 2.2 uses a Gaussian kernel function to convert the distance metric into a similarity metric as follows:
[0018] in, c This is the median distance between all nodes; Step 2.3 employs an adaptive thresholding method, defining a dynamic threshold. The method for constructing the adjacency matrix and weight matrix of the graph based on this dynamic threshold is as follows: Define the dynamic threshold as:
[0019] The adjacency matrix of the graph is:
[0020] in, m s and s s Let S be the mean and standard deviation of set S, and let S be the upper triangular elements of similarity matrix C.
[0021] Furthermore, a preferred embodiment is provided, wherein the method for inputting the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning in step 3 is as follows: For nodes i Time series signals Extracting six-dimensional statistical feature vectors :
[0022] in, f 1= m The sample mean. f 2= s The standard deviation of the sample is 1. f 3= s 2 For sample variance, For signal skewness, For signal kurtosis, The energy is the sum of the squares of the signal amplitude.
[0023] Furthermore, a preferred embodiment is provided in which a graph convolutional neural network is used as a feature learner and classifier in step 3 to achieve end-to-end identification of rock hardness.
[0024] Option 2: A multi-sensor fusion rock hardness identification system based on geometric consistency maps, the system comprising: The data acquisition module is used to acquire time series signals from multiple sensors. By introducing pseudo-time coordinates, the time series signals from multiple sensors are mapped into a two-dimensional geometric point cloud, and the geometric consistency features of the nodes of multiple sensors are extracted using the random projection method. The training module is used to construct the adjacency relationships and weight matrices between multiple sensors by geometric consistency measurement, and to generate graph data containing topological structure and attribute information by combining the temporal statistical characteristics of the nodes themselves. The recognition module is used to input the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning, and to achieve the recognition of rock hardness categories through end-to-end training.
[0025] Option 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Option 1.
[0026] Option 4: A computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in Option 1.
[0027] The advantages of this invention are: The multi-sensor fusion rock hardness identification method and system based on geometric consistency maps described in this invention addresses the problem that existing technologies typically use simple feature stitching or direct vector superposition to process multi-sensor data, neglecting the differences between different physical dimensions and their inherent geometric correlations. It proposes an adaptive geometric consistency map construction algorithm, which maps a one-dimensional time series to a two-dimensional geometric point cloud and uses random projection technology to unify heterogeneous signals into a common geometric space for measurement. This method effectively eliminates dimensional differences caused by sensor types and can deeply explore the potential geometrical coupling relationships between multi-source signals during drilling. It can deeply explore the potential geometrical relationships between multi-source signals during drilling, thereby constructing graph-structured data that better characterizes rock hardness features.
[0028] The multi-sensor fusion rock hardness identification method and system based on geometric consistency graphs described in this invention addresses the problem that existing graph construction methods often rely on manually setting fixed thresholds (such as Euclidean distance thresholds) or a fixed number of neighboring nodes to determine node connections, which is difficult to adapt to rock identification in complex environments. This application designs an adaptive threshold calculation mechanism based on statistical characteristics, where the threshold value is automatically adjusted according to the consistency distribution within the current sample. When sensor coupling is high, the threshold automatically increases to retain strong correlations; when consistency dispersion is large, the threshold is appropriately decreased to retain key connections.
[0029] This invention is also applicable to rock hardness identification in coal mining operations with high dust, high humidity, strong vibration, limited space, and complex geological structures, as well as rock hardness identification in non-coal tunneling. Attached Figure Description
[0030] Figure 1 This is a schematic diagram of the multi-sensor fusion rock hardness identification method based on geometric consistency maps described in Implementation Method 1.
[0031] Figure 2 This is a schematic diagram of the geometric consistency graph data construction framework described in Implementation Method 1. Detailed Implementation
[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.
[0033] Implementation Method 1, see [link] Figure 1 to Figure 2 This embodiment describes a multi-sensor fusion rock hardness identification method based on geometric consistency graphs, which includes the following steps: First, time-series signals from multiple sensors are collected and mapped into a two-dimensional geometric point cloud by introducing pseudo-time coordinates. Then, the geometric consistency features of each sensor node are extracted using random projection technology. Next, adjacency relationships and weight matrices between sensors are constructed based on geometric consistency metrics, and graph data containing topological structure and attribute information is generated by combining the temporal statistical features of the nodes themselves. Finally, the constructed graph data is input into a graph convolutional neural network for feature aggregation and deep learning, achieving accurate identification of rock hardness categories through end-to-end training.
[0034] To address the challenges of weak correlation features among multi-source signals, scarce samples, and reliance on manually set thresholds in traditional graph construction methods for rock hardness identification during anchor drilling, this invention proposes an adaptive geometric consistency graph construction algorithm. This algorithm aims to uncover the potential geometric structural correlations between multi-source signals such as vibration, current, voltage, and oil pressure. It constructs a topological graph structure to characterize the coupling features of the drilling system and ultimately uses a graph convolutional network to identify rock hardness.
[0035] When anchor drilling rigs drill into rocks of varying hardness, their vibration, current, voltage, and oil pressure signals exhibit significant time-series characteristics. Traditional methods often simply concatenate multidimensional signals into feature vectors, neglecting the internal temporal morphological structure of the signals and the spatial correlation between sensors. To preserve the local morphological features of the signals within the graph structure, each sensor is first treated as a node in the graph data, and the one-dimensional time-series signal of each node is mapped to a two-dimensional geometric point cloud. The mapping process is defined as follows:
[0036] in, For the first i The first of each sensor m The actual sequence signal of each sample, L This represents the signal length. This mapping preserves the signal's amplitude fluctuations and time-dependent structure.
[0037] Because raw point cloud data has high dimensionality and contains noise, directly calculating geometric similarity in the raw space is computationally intensive and lacks robustness, and different sensors have different physical dimensions. To unify the metric space and extract robust geometric features, this invention introduces a random projection mechanism, with the projection formula as follows:
[0038] in, Let be a random projection matrix whose elements follow a Gaussian distribution with mean 0 and variance 0.1, and the projection dimension is . .
[0039] In the same graph data, the geometric consistency vector of the nodes in the graph is defined as:
[0040] in, m i and s i These are the mean vector and standard deviation vector of the projection point set, respectively. i i For the two principal directions of the projection point set v 1 and v The included angle of 2 is calculated using the following formula:
[0041] m i It reflects the average signal strength. s i It characterizes the degree of signal fluctuation, while i It reflects the geometric state structure of the signal.
[0042] After obtaining the geometric consistency characteristics of each sensor node, the edge connections and weights of the graph are constructed by calculating the geometric consistency between nodes.
[0043] For any two sensor nodes in the graph, calculate the Euclidean distance between their geometric consistency vectors:
[0044] A smaller distance indicates greater geometric similarity. The distance metric is transformed into a similarity score using a Gaussian kernel function. i With nodes j The similarity is:
[0045] in, c This is the median distance between all nodes.
[0046] To construct a sparse graph structure to reduce computational complexity and eliminate weakly correlated noise, the algorithm employs an adaptive thresholding mechanism. Let S be the set of the upper triangular elements of the consistency matrix, and define the dynamic threshold as:
[0047] in, m s and s s Let S be the mean and standard deviation of the set. When all sensors have high consistency, the threshold is increased to preserve strong correlations; when the consistency dispersion is large, the threshold is decreased to preserve key links. The adjacency matrix of the graph is:
[0048] To fully utilize the physical information contained within the signal itself, statistical features of the signal are incorporated when constructing graph node features. For nodes... i Time series signals Extracting six-dimensional statistical feature vectors :
[0049] in, f 1= m This is the sample mean, reflecting the average level of the sensor signal; f 2= s The standard deviation is the sample value, reflecting the amplitude of the signal's fluctuation around the mean. f 3= s 2 The sample variance reflects the degree of dispersion of the signal; Signal skewness reflects the asymmetry of signal distribution; Kuroism, or signal kurtosis, reflects the flatness or sharpness of the signal probability density distribution. Energy is the sum of the squares of the signal amplitude, reflecting the total strength of the signal.
[0050] Ultimately, graph data structures have node characteristics. and adjacency matrix A i,j This composition provides structured data support for subsequent graph neural networks to efficiently learn rock hardness features under small sample conditions.
[0051] A graph convolutional neural network is used as a feature learner and classifier to achieve end-to-end identification of rock hardness.
[0052] Let the first l The node feature matrix of the layer is ,in N This represents the number of sensor nodes. dl This represents the feature dimension of this layer. l +1 layer, its feature propagation is defined as
[0053] in, D for A The degree matrix is W, the weight matrix is W, and sig(˙) is the nonlinear activation function.
[0054] After feature extraction via a multi-layer graph convolutional network, the model outputs a deep feature matrix for each sensor node. H A global average pooling strategy is employed to obtain a global representation of the graph by calculating the average of the features of all nodes.
[0055] in, H i,： Indicates the first i The features of each sensor node are analyzed in the last layer. Finally, the global feature vector is input into the classification layer to achieve classification.
[0056] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
[0057] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.
Claims
1. A multi-sensor fusion method for identifying rock hardness in tunneling based on geometric consistency maps, characterized in that, The method includes the following steps: Step 1: Acquire multi-sensor time series signals, map the multi-sensor time series signals into a two-dimensional geometric point cloud by introducing pseudo-time coordinates, and extract the geometric fingerprint features of the multi-sensor nodes using the random projection method. Step 2: Construct the adjacency relationship and weight matrix among multiple sensors based on the geometric consistency metric, and generate graph data containing topological structure and attribute information by combining the temporal statistical characteristics of the nodes themselves; Step 3: Input the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning, and realize the identification of rock hardness categories through end-to-end training.
2. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 1, characterized in that, The method in step 1 for mapping multi-sensor time series signals into two-dimensional geometric point clouds by introducing pseudo-time coordinates is as follows: in, For the first i The actual sequence signal of the m-th sample in each sensor. L This is the signal length.
3. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 1, characterized in that, The method for extracting geometric fingerprint features of multi-sensor nodes using random projection as described in step 1 is as follows: in, Let be a random projection matrix whose elements follow a Gaussian distribution with mean 0 and variance 0.1, and the projection dimension is . .
4. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 1, characterized in that, The method for constructing the adjacency relationships and weight matrices among multiple sensors based on geometric consistency metrics in step 2 is as follows: Step 2.1: For any two sensor nodes in the figure, calculate the Euclidean distance between the geometric consistency vectors of the two nodes; Step 2.2: Convert the distance metric into a similarity metric using a Gaussian kernel function; Step 2.3: Using the adaptive thresholding method, define a dynamic threshold and construct the adjacency matrix and weight matrix of the graph based on the dynamic threshold.
5. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 4, characterized in that, Step 2 specifically includes the following steps: The method for calculating the Euclidean distance between the geometric consistency vectors of two nodes in step 2.1 is as follows: in, μ i and σ i Let be the mean vector and standard deviation vector of the projected point set, respectively. θ i For the two principal directions of the projection point set v 1 and v The included angle of 2 is calculated using the following formula: Step 2.2 uses a Gaussian kernel function to convert the distance metric into a similarity metric as follows: in, γ This is the median distance between all nodes; Step 2.3 employs an adaptive thresholding method, defining a dynamic threshold. The method for constructing the adjacency matrix and weight matrix of the graph based on this dynamic threshold is as follows: Define the dynamic threshold as: The adjacency matrix of the graph is: in, μ s and σ s Let S be the mean and standard deviation of set S, and let S be the upper triangular elements of similarity matrix C.
6. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 5, characterized in that, The method for inputting the constructed graph data into the graph convolutional neural network for feature aggregation and deep learning in step 3 is as follows: For nodes i Time series signals Extracting six-dimensional statistical feature vectors : in, f 1= μ The sample mean. f 2= σ The standard deviation of the sample is 1. f 3= σ 2 For sample variance, For signal skewness, For signal kurtosis, The energy is the sum of the squares of the signal amplitude.
7. The multi-sensor fusion rock hardness identification method based on geometric consistency maps according to claim 1, characterized in that, In step 3, a graph convolutional neural network is used as a feature learner and classifier to achieve end-to-end identification of rock hardness.
8. A multi-sensor fusion rock hardness identification system based on geometric consistency maps, characterized in that, The system includes: The data acquisition module is used to acquire time series signals from multiple sensors. By introducing pseudo-time coordinates, the time series signals from multiple sensors are mapped into a two-dimensional geometric point cloud, and the geometric fingerprint features of the multiple sensor nodes are extracted using the random projection method. The training module is used to construct the adjacency relationships and weight matrices between multiple sensors by geometric consistency measurement, and to generate graph data containing topological structure and attribute information by combining the temporal statistical characteristics of the nodes themselves. The recognition module is used to input the constructed graph data into a graph convolutional neural network for feature aggregation and deep learning, and to achieve the recognition of rock hardness categories through end-to-end training.
9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.
10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1-7.