Support action posture detection method based on neural network and multi-source tilt angle sensor
By using neural networks and multi-source inclination sensors to detect the movement posture of the bracket in coal mining, the problem that the existing technology cannot detect in real time and is affected by the environment is solved, and higher detection accuracy and safety are achieved, which promotes coal mining. Automated and intelligent management.
Patent Information
- Application Number
- CN202411250164.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2025-01-07
AI Technical Summary
The prior art cannot detect the stent movement in real time or the detection process is greatly affected by the underground environment, resulting in insufficient accuracy and reliability of the stent movement detection, which poses safety hazards.
The bracket action posture detection method based on neural network and multi-source inclination sensor is adopted. By installing multiple inclination sensors on the bracket to collect data, a neural network model is built for training and learning, real-time and accurate detection of the bracket action posture is achieved.
Improve the accuracy and reliability of bracket action detection, improve the safety of coal mining operations, reduce accident risks, provide strong guarantees for the safety of coal mining working faces, and help achieve automated control of brackets and Intelligent management.
Smart Images
Figure CN119268644A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of coal mining, and in particular to a support action posture detection method based on a neural network and a multi-source tilt sensor. Background Art
[0002] At a time when energy resources such as oil and natural gas are relatively scarce and the safety risks of imports are becoming increasingly prominent, coal, as an important component of the energy structure, has become increasingly prominent in terms of its status and role. However, in the process of coal mining, the underground environment is complex and changeable, and there are many safety hazards, which pose a serious threat to the safety of workers. Therefore, it is particularly important to accurately detect the status of various equipment during coal mining.
[0003] At present, there are two main technical means for detecting the motion of the support. The first is to collect the data of the support inclination sensor in real time and set the corresponding threshold to roughly judge the posture of the support. However, this method can only provide static information of the support and cannot capture the dynamic motion process of the support in real time. The second is to use the camera to capture the real-time video data of the support and combine image processing and recognition technology to judge the motion of the support. Although this method can provide intuitive visual information, the interference of underground dust is serious and the video shooting range is limited. It is mainly suitable for detecting specific actions such as retracting and extending the guardrail.
[0004] Therefore, it is necessary to study a bracket motion posture detection method based on neural network and multi-source tilt sensor to overcome the limitations of the existing technology and improve the accuracy and real-time performance of bracket motion detection. Summary of the invention
[0005] The purpose of the present invention is to solve the above problems and provide a method for detecting the motion posture of a bracket based on a neural network and a multi-source tilt sensor, which is simple to operate and has high accuracy.
[0006] In order to achieve the above object, the technical solution of the present invention is:
[0007] The method for detecting the motion posture of a bracket based on a neural network and a multi-source tilt sensor comprises the following steps:
[0008] S1. Install inclination sensors at multiple locations on the bracket;
[0009] S2, collecting the angle monitoring data of the tilt sensor, the monitoring time, and the action posture data of the bracket in this state to construct a data sample, and dividing a number of data samples formed by various actions of the bracket into a training set and a test set;
[0010] S3, constructing a neural network model, inputting the training set into the neural network model for training, and testing the trained neural network model with the test set, and finally obtaining a deep learning model for bracket action detection based on the neural network architecture;
[0011] S4, optimizing the bracket motion detection deep learning model, saving and loading the optimized bracket motion detection deep learning model into the controller system;
[0012] S5. Obtain the angle monitoring data of the inclination sensor in real time through the controller system, and input the angle monitoring data into the support motion detection deep learning model. The support motion detection deep learning model outputs the current posture of the support after analysis and calculation, so as to realize the detection of the support motion posture in coal mining operations.
[0013] Furthermore, in step S1, inclination sensors are installed on the top beam, connecting rod, base, tail beam, primary guard rail and secondary guard rail of the bracket.
[0014] Furthermore, in step S2, the angle monitoring data includes roll angle data, pitch angle data, and yaw angle data of the inclination sensor; the monitoring time includes multiple continuous time points, and the interval between multiple continuous time points is 10ms; a data sample is constructed by combining the angle monitoring data of multiple time points and the motion posture data of the bracket in this state.
[0015] Furthermore, in step S2, the plurality of data samples are divided into five mutually exclusive subsets, four of which are used as training sets and the other one is used as a test set.
[0016] Furthermore, in step S3, the neural network model includes an input layer, a convolution layer, a normalization layer, an activation layer, and an output layer; the data shape of the input layer is 3x×n×1; wherein x is the number of all inclination sensors on a single bracket at a certain time point, and n is the number of time points; a filter V is provided in the convolution layer, and the filter V=[v i ,v2,...,v c' ]; the output after convolution is U = [u i ,u2,...,u c' ]; the normalized calculation result is: The activation layer is calculated as: Among them, δ is the relu activation function, W is the weight matrix, and Z is the output result of the normalization layer; the output layer adopts a softmax structure, and the number of nodes corresponds to the number of types of bracket movements to be detected; the angle monitoring data and monitoring time are used as the input layer, and the bracket's action posture data is used as the output layer to train the neural network model.
[0017] Furthermore, in step S3, when the neural network model is trained and tested with the training set and the test set, each subset is used as a test set for five iterations of training.
[0018] Furthermore, in step S3, the calculation result of the bracket action detection deep learning model finally obtained is S.
[0019] Furthermore, in step S4, when optimizing the deep learning model for bracket motion detection, the weight and bias parameters of the deep learning model for bracket motion detection are adjusted using the Adam optimization algorithm with an initial learning rate of 0.0001 to minimize the loss function in the deep learning model for bracket motion detection.
[0020] Furthermore, in step S4, the optimized bracket action detection deep learning model is saved as a .h5 format file.
[0021] Compared with the prior art, the present invention has the following advantages and positive effects:
[0022] The present invention discloses a method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor, which can comprehensively utilize the angle monitoring data of the multi-source tilt sensor, and finally realize real-time and accurate detection of the motion posture of the support by constructing a neural network model for training and learning; it overcomes the problem that the prior art cannot detect the motion of the support in real time or the detection process is greatly affected by the underground environment, improves the accuracy and reliability of the motion detection of the support, enhances the safety of coal mining operations, reduces the risk of accidents in the coal mining process, and provides a strong guarantee for the safe production of the fully mechanized mining face of the coal mine; at the same time, it can understand the motion information of the support in real time, which is helpful to realize the automatic control and intelligent management of the support, thereby effectively improving the efficiency of coal mining and promoting the sustainable development of the coal industry. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0024] Figure 1 It is the overall architecture diagram of the present invention;
[0025] Figure 2 It is a flow logic diagram of the present invention. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work, any modifications, equivalent substitutions, improvements, etc., should be included in the protection scope of the present invention.
[0027] like Figure 1 , Figure 2 As shown, this embodiment proposes a method for detecting the motion posture of a bracket based on a neural network and a multi-source tilt sensor, and the detection process mainly includes data collection, neural network construction, model training, model storage and import, real-time detection, etc.;
[0028] The data acquisition is dedicated to accurately capturing the tilt angle change information of the bracket during multiple actions. These actions include but are not limited to key actions such as lifting columns, retracting and extending guardrails, and left and right tilting. To achieve this goal, a multi-source tilt sensor simultaneous acquisition strategy is adopted, including tilt sensors such as top beam, connecting rod, base, tail beam, first-level guardrail, and second-level guardrail. Through the collaborative work of these sensors, the posture change data of the bracket during the action process can be fully and accurately obtained, providing a solid foundation for subsequent data analysis and model training.
[0029] The neural network construction refers to the careful construction of a classification model using a deep learning framework, which includes a series of key steps. The first task is to determine the network structure, which is a crucial step. It involves the number of neurons, the number of network layers, and the connection between layers. In order to build the most suitable network structure for bracket action detection, the nature and complexity of the problem are deeply analyzed to ensure that the network structure can fully capture the inherent laws and characteristics of the data. At the same time, the network structure is designed and adjusted to improve the classification performance of the model and ensure the accuracy and reliability of bracket action detection.
[0030] The model training is a complex and critical process. Model training refers to the use of a series of data and algorithms to train a specific model so that it can make corresponding predictions or classifications of input data. In the field of machine learning, model training is an important step in achieving automated learning. By inputting a large amount of sample data into the algorithm model for training, the model can extract useful features and knowledge by learning the patterns and relationships between the data, and predict or classify unknown data.
[0031] The model saving and importing are crucial to achieve model persistence and reuse. Specifically, the trained stent motion detection model is saved, which can be transplanted and used in other terminals later, such as stent electro-hydraulic controller. Moreover, by saving and importing the model, the cross-platform application of stent motion detection technology can be realized.
[0032] The real-time detection usually requires importing a saved model when deploying the trained support motion detection deep learning model to the production environment. In this way, the system in the production environment can load and use the model to provide users with real-time support motion detection services. Through this process, a strong guarantee is provided for the safe production of coal mine fully mechanized mining working faces.
[0033] The following is a detailed description of the method for detecting the motion posture of the bracket in the present invention, and the steps are as follows:
[0034] Step 1: First, use the scaled-down physical model of the bracket to collect data; during the data collection process, the tilt sensor data of the bracket during action is accurately collected at a time interval of 10ms to ensure the real-time and accuracy of the data. Each sample contains the roll angle, pitch angle and yaw angle data of X tilt sensors, and a sample is constructed by combining the tilt sensor data at multiple time points n. The data structure of a single sample is in the shape of 3x×n×1, which fully reflects the posture changes during the bracket action. Then, let the bracket model perform key actions such as lifting columns, retracting and extending guardrails, and moving racks, and record the tilt sensor data during these actions in real time. According to the above data collection rules, we make these data into independent samples to provide raw data for subsequent data processing and model training. In order to conduct supervised neural network model training, corresponding labels are made for each sample. These labels accurately identify the action type of the bracket, providing an important basis for model training and verification. Finally, the prepared bracket action data set is scientifically divided. According to the rule of five-fold cross-validation, all the collected tilt sensor data are divided into five mutually exclusive subsets, each of which is equal in size. Make sure each subset contains a variety of different categories of data to keep the data distribution balanced. In training, four subsets are used as training sets and the remaining one subset is used as a test set. After five iterations of training, each subset is evaluated once as a test set.
[0035] Step 2: Extract the change information of multiple tilt sensors during the bracket movement by constructing a neural network; design a multi-layer convolutional neural network, whose structure is optimized to meet the needs of the bracket movement detection task. The data shape of the data layer of the network is 3x×n×1 to match the data structure of each sample. The output layer adopts a softmax structure, and the number of nodes corresponds to the number of action types to be detected. For each bracket movement sample, a 3x×n×1 matrix is constructed as input data according to its time point n and the number of sensors x. Next, the training set data is input into the neural network in batches for training.
[0036] When a single data is input into the network, a convolution operation is performed first, using V = [v i ,v2,...,v c' ] represents a filter in a convolutional layer, where v i Indicates the i th filters; the output after convolution is U = [u i ,u2,...,u c' ], where u i =v i *X,* represents the convolution operation. The filter can learn the spatial and temporal information of multi-source tilt sensor time series data; then the output after the batch normalization layer is Batch normalization can effectively prevent gradient explosion and gradient disappearance in the network and accelerate the convergence of the network. Finally, the nonlinear interaction between feature map channels is learned through the activation layer to obtain the complete dependency relationship between channels. It is expressed by the following formula: Among them, δ refers to the relu activation function, W is the weight matrix, and Z is the output result of the normalization layer. After multiple convolutions and activation calculations, the final multi-source tilt sensor time series data feature is expressed as
[0037] After the forward propagation, the network parameters are continuously adjusted through the back-propagation algorithm and gradient descent optimization method, so that the model can gradually learn the characteristics and rules of the bracket's movements. During the back-propagation process, the Adam optimization algorithm with an initial learning rate of 0.0001 is used to adjust the model's weights and biases parameters to minimize the loss function, that is, the difference between the output and the label of the sample after passing through the neural network is as small as possible. During the training process, pay close attention to the performance of the model, and evaluate the performance through the test set to ensure that the model has good generalization ability. Finally, save the trained neural network model as a .h5 file so that the model can be easily ported to other platforms or systems for use, realizing the persistence and reuse of the model.
[0038] Step 3: In actual use of the bracket, load the saved file into the controller system and establish a data pipeline. After the data pipeline is established, obtain the data of the multi-source tilt sensor in real time and use it as the input of the model. After processing and analysis by the model, these data can accurately identify the current action state of the bracket. Finally, the model will output the action recognition result of the bracket. This result not only provides the operator with real-time bracket action information, but also helps to realize the automatic control and intelligent management of the bracket.
[0039] The present invention discloses a method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor, which can comprehensively utilize the angle monitoring data of the multi-source tilt sensor, and finally realize real-time and accurate detection of the motion posture of the support by constructing a neural network model for training and learning; it overcomes the problem that the prior art cannot detect the motion of the support in real time or the detection process is greatly affected by the underground environment, improves the accuracy and reliability of the motion detection of the support, enhances the safety of coal mining operations, reduces the risk of accidents in the coal mining process, and provides a strong guarantee for the safe production of the fully mechanized mining face of the coal mine; at the same time, it can understand the motion information of the support in real time, which is helpful to realize the automatic control and intelligent management of the support, thereby effectively improving the efficiency of coal mining and promoting the sustainable development of the coal industry.
Claims
1. A method for detecting the motion posture of a bracket based on a neural network and a multi-source tilt sensor, characterized in that: The following steps are involved: S1. Install inclination sensors at multiple locations on the bracket; S2, collecting the angle monitoring data of the tilt sensor, the monitoring time, and the action posture data of the bracket in this state to construct a data sample, and dividing a number of data samples formed by various actions of the bracket into a training set and a test set; S3, constructing a neural network model, inputting the training set into the neural network model for training, and testing the trained neural network model with the test set, and finally obtaining a deep learning model for bracket action detection based on the neural network architecture; S4, optimizing the bracket motion detection deep learning model, saving and loading the optimized bracket motion detection deep learning model into the controller system; S5. Obtain the angle monitoring data of the inclination sensor in real time through the controller system, and input the angle monitoring data into the support motion detection deep learning model. The support motion detection deep learning model outputs the current posture of the support after analysis and calculation, so as to realize the detection of the support motion posture in coal mining operations.
2. The method for detecting the motion posture of a bracket based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In the step S1, inclination sensors are installed on the top beam, connecting rod, base, tail beam, primary guard rail and secondary guard rail of the bracket.
3. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S2, the angle monitoring data includes the roll angle data, pitch angle data, and yaw angle data of the inclination sensor; the monitoring time includes multiple continuous time points, and the interval between multiple continuous time points is 10ms; a data sample is constructed by combining the angle monitoring data of multiple time points and the action posture data of the bracket in this state.
4. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S2, a number of data samples are divided into five mutually exclusive subsets, four of which are used as training sets and the other one is used as a test set.
5. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S3, the neural network model includes an input layer, a convolution layer, a normalization layer, an activation layer, and an output layer; the data shape of the input layer is 3x×n×1; wherein x is the number of all inclination sensors on a single bracket at a certain time point, and n is the number of time points; a filter V is provided in the convolution layer, and the filter V=[v i ,v2,...,v c' ]; the output after convolution is U = [u i ,u2,...,u c' ]; the normalized calculation result is: The activation layer is calculated as: Among them, δ is the relu activation function, W is the weight matrix, and Z is the output result of the normalization layer; the output layer adopts a softmax structure, and the number of nodes corresponds to the number of types of bracket movements to be detected; the angle monitoring data and monitoring time are used as the input layer, and the bracket's action posture data is used as the output layer to train the neural network model.
6. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S3, when the neural network model is trained and tested with the training set and the test set, each subset is used as a test set for five iterations of training.
7. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S3, the calculation result of the bracket action detection deep learning model is finally obtained as S.
8. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S4, when optimizing the deep learning model for bracket action detection, the weight and bias parameters of the deep learning model for bracket action detection are adjusted using the Adam optimization algorithm with an initial learning rate of 0.0001 to minimize the loss function in the deep learning model for bracket action detection.
9. The method for detecting the motion posture of a support based on a neural network and a multi-source tilt sensor as claimed in claim 1, characterized in that: In step S4, the optimized bracket action detection deep learning model is saved as a .h5 format file.