Model training and surrounding rock identification method and device and engineering mechanical equipment

Abstract

The invention provides a model training and surrounding rock identification method and device and engineering machinery equipment, and relates to the technical field of engineering machinery. The model training method comprises the following steps: acquiring operation sample data of the engineering machinery equipment for performing operation on sample surrounding rock under a plurality of gears; processing the operation sample data under each gear by using a neural network model corresponding to each gear in a plurality of gears so as to obtain a strength predicted value and an integrity predicted value of the sample surrounding rock; determining a loss function value according to the strength predicted value and the integrity predicted value of the sample surrounding rock and the strength label value and the integrity label value of the sample surrounding rock; and training the neural network model corresponding to each gear in the plurality of gears according to the loss function value. By means of the method, the surrounding rock parameters can be accurately and rapidly determined at low cost, and the refined surrounding rock recognition requirement under the complex construction working condition is met.

Description

Technical Field

[0001] The present disclosure relates to the technical field of engineering machinery, and in particular to a model training and surrounding rock identification method, device, and engineering machinery equipment. Background Art

[0002] During tunnel construction, surrounding rock drilling can be carried out using engineering machinery such as multi-functional drilling rigs, shield machines, drilling rigs, horizontal directional drilling rigs, down-the-hole drilling rigs, or rotary drilling rigs.

[0003] Surrounding rock identification is crucial for tunnel construction. This helps determine optimal operating parameters, leading to efficient drilling. Traditionally, surrounding rock identification is accomplished through sampling and testing. However, this method has drawbacks such as difficulty in obtaining data, high costs, and time consumption.

[0004] In response to the shortcomings of traditional surrounding rock identification methods, relevant technologies have been proposed that the grade of surrounding rock can be quickly determined based on machine learning algorithms such as neural networks. Summary of the Invention

[0005] According to the first aspect of the present disclosure, a model training method is proposed, including: obtaining operation sample data of engineering machinery equipment operating on sample surrounding rock in multiple gears; using a neural network model corresponding to each of the multiple gears to process the operation sample data in each gear to obtain a strength prediction value and an integrity prediction value of the sample surrounding rock; determining a loss function value based on the strength prediction value and the integrity prediction value of the sample surrounding rock, as well as the strength label value and the integrity label value of the sample surrounding rock; and training the neural network model corresponding to each of the multiple gears based on the loss function value.

[0006] In some embodiments, each gear is characterized by at least one of an impact gear and a swing gear.

[0007] In some embodiments, the obtaining of operation sample data of the engineering machinery equipment operating on the sample surrounding rock in multiple gears includes: receiving original measurement data of the engineering machinery equipment operating on the sample surrounding rock in multiple gears from multiple sensors on the engineering machinery equipment, wherein the original measurement data includes at least one of the thrust pressure, impact pressure, rotation pressure, impact flow, rotation speed, displacement of the power head, and speed data of the power head of the engineering machinery equipment; and preprocessing the original measurement data to obtain the operation sample data in the multiple gears.

[0008] In some embodiments, the preprocessing of the original measurement data to obtain the operation sample data under the multiple gears includes: filtering out abnormal operation data from the original measurement data under the multiple gears to obtain normal operation data under the multiple gears, wherein the abnormal operation data includes at least one of standby non-operation data, hole-guiding data, drill stuck data and drill lifting data; denoising, standardization and normalization of the normal operation data under the multiple gears to obtain the operation sample data under the multiple gears.

[0009] In some embodiments, the denoising process for the normal operating data under the multiple gears includes: using a low-pass filter algorithm to remove high-frequency noise in the normal operating data; and using a median moving average filter algorithm to remove pulse noise in the normal operating data.

[0010] In some embodiments, the use of the neural network model corresponding to each of the multiple gears to process the operation sample data under each gear includes: inputting the operation sample data under each gear into the neural network model corresponding to each gear to obtain the strength output value and integrity output value of the sample surrounding rock; and performing denormalization and destandardization on the strength output value and the integrity output value of the sample surrounding rock to obtain the strength prediction value and the integrity prediction value of the sample surrounding rock.

[0011] In some embodiments, the strength of the sample surrounding rock is uniaxial compressive strength, and the integrity of the sample surrounding rock is crack coefficient; and / or, the neural network model includes a gated recurrent neural network module and a fully connected layer.

[0012] According to a second aspect of the present disclosure, a surrounding rock identification method is provided, comprising: obtaining operation data of engineering machinery equipment operating on the surrounding rock to be identified in a specified gear position; processing the operation data using a surrounding rock identification model corresponding to the specified gear position to obtain a strength prediction value and an integrity prediction value of the surrounding rock to be identified.

[0013] In some embodiments, the designated gear is characterized by at least one of an impact gear and a swing gear.

[0014] In some embodiments, the obtaining of operation data of the engineering machinery equipment operating on the surrounding rock to be identified at a specified gear position includes: receiving original measurement data of the engineering machinery equipment operating on the surrounding rock to be identified at the specified gear position from multiple sensors on the engineering machinery equipment, wherein the original measurement data includes at least one of the thrust pressure, impact pressure, rotation pressure, impact flow, rotation speed, displacement of the power head, and speed data of the power head of the engineering machinery equipment; and pre-processing the original measurement data to obtain the operation data at the specified gear position.

[0015] In some embodiments, the preprocessing of the original measurement data to obtain the operating data under the specified gear position includes: filtering out abnormal operating data from the original measurement data under the specified gear position to obtain normal operating data under the specified gear position, wherein the abnormal operating data includes at least one of standby non-operating data, hole-guiding data, stuck drill data, and drill lifting data; and denoising, standardizing, and normalizing the normal operating data under the specified gear position to obtain the operating data under the specified gear position.

[0016] In some embodiments, the processing of the operation data using the neural network model corresponding to the designated gear includes: inputting the operation data under the designated gear into the neural network model corresponding to the designated gear to obtain the strength output value and the integrity output value of the surrounding rock to be identified; performing denormalization and destandardization on the strength output value and the integrity output value of the surrounding rock to be identified to obtain the strength prediction value and the integrity prediction value of the surrounding rock to be identified.

[0017] According to a third aspect of the present disclosure, a device is provided, comprising: a module for executing the aforementioned model training method, or a module for executing the aforementioned surrounding rock identification method.

[0018] According to a fourth aspect of the present disclosure, a device is provided, comprising: a memory; and a processor coupled to the memory, wherein the processor is configured to execute the aforementioned model training method or the aforementioned surrounding rock identification method based on instructions stored in the memory.

[0019] According to a fifth aspect of the present disclosure, there is provided an engineering machinery equipment, comprising: the device as described above.

[0020] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided, on which computer instructions are stored. When the instructions are executed by a processor, the model training method as described above or the surrounding rock identification method as described above is implemented.

[0021] According to a seventh aspect of the present disclosure, a computer program product is provided, on which computer instructions are stored. When the instructions are executed by a processor, the model training method as described above or the surrounding rock identification method as described above is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0023] The present disclosure can be more clearly understood from the following detailed description with reference to the accompanying drawings.

[0024] Figure 1 Schematic diagram of a flow chart of a model training method according to some embodiments of the present disclosure.

[0025] Figure 2 Schematic diagram of the flow of model training methods according to other embodiments of the present disclosure.

[0026] Figure 3 Schematic diagram of the flow of the surrounding rock identification method according to some embodiments of the present disclosure.

[0027] Figure 4 Schematic diagram of the flow of surrounding rock identification methods according to other embodiments of the present disclosure.

[0028] Figure 5 Schematic diagram of the structure of a model training device according to some embodiments of the present disclosure.

[0029] Figure 6 Schematic diagram of the structure of a surrounding rock identification device according to some embodiments of the present disclosure.

[0030] Figure 7 Schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure.

[0031] Figure 8 Schematic diagram of the structure of a surrounding rock identification system according to some embodiments of the present disclosure.

[0032] Figure 9 Schematic diagram of the structure of a signal acquisition module according to some embodiments of the present disclosure.

[0033] Figure 10 Schematic diagram of the structure of engineering machinery equipment according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0034] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure.

[0035] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0036] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0037] Technologies, methods and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, such technologies, methods and equipment should be considered part of the authorization specification.

[0038] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0039] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0040] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0041] Conventional rock identification methods often only roughly classify surrounding rock, failing to obtain specific, physically meaningful rock parameters. This makes it difficult to accurately determine optimal operating parameters. Furthermore, conventional rock identification methods cannot meet the demand for refined rock identification in complex construction conditions.

[0042] In view of this, the present disclosure proposes a model training and surrounding rock identification method, device and engineering machinery equipment, which can improve the performance of the trained surrounding rock identification model, thereby facilitating the subsequent accurate, rapid and low-cost determination of surrounding rock parameters, and meeting the refined surrounding rock identification needs under complex construction conditions.

[0043] Figure 1 FIG. 1 is a flow chart of a model training method according to some embodiments of the present disclosure. Figure 1 As shown, the model training method includes steps 11 to 14.

[0044] In step 11, operation sample data of the engineering machinery equipment operating on sample surrounding rocks at multiple gears is obtained.

[0045] In some embodiments, step 11 includes: receiving original measurement data of the engineering machinery equipment operating on sample surrounding rocks (or surrounding rock samples) at multiple gears from multiple sensors on the engineering machinery equipment; and preprocessing the original measurement data to obtain operation data at multiple gears.

[0046] The gear can be characterized by at least one of an impact gear and a rotary gear. In some examples, the multiple gears refer to multiple impact gears. In these examples, step 11 includes: obtaining operation sample data of the engineering machinery equipment drilling into the sample surrounding rock at different impact gears. In other examples, the multiple gears refer to multiple rotary gears. In these examples, step 11 includes: obtaining operation sample data of the engineering machinery equipment drilling into the sample surrounding rock at different rotary gears. In still other examples, the multiple gears refer to a combination of multiple impact gears and rotary gears. In these examples, step 11 includes: obtaining operation sample data of the engineering machinery equipment drilling into the sample surrounding rock at different combinations of impact gears and rotary gears.

[0047] Typically, the engineering machinery and equipment used for surrounding rock operations include shield machines, multi-purpose drilling rigs, drilling rigs, horizontal directional drills, down-the-hole drills, and rotary drilling rigs. During implementation, the required raw measurement data and pre-processed operational sample data may vary for different types of engineering machinery and equipment.

[0048] In some examples, the raw measurement data includes at least one of propulsion pressure, impact pressure, rotation pressure, impact flow, rotation speed, displacement of the power head, and speed data of the power head of the engineering machinery equipment.

[0049] For example, when the engineering machinery is a multi-functional drilling rig or other similar drilling engineering machinery, sensors can be used to collect propulsion pressure, impact pressure, rotary pressure, rotary speed, and displacement of the power head at different times. After obtaining the displacement of the power head, the drilling speed can be obtained by differentiating it with respect to time. For example, at time t1, the propulsion pressure, impact pressure, rotary pressure, rotary speed, and displacement of the power head are collected; at time t2, the propulsion pressure, impact pressure, rotary pressure, rotary speed, and displacement of the power head are collected. In this way, time series data consisting of the propulsion pressure, impact pressure, rotary pressure, rotary speed, and drilling speed of the power head collected at different times can be obtained, and this data can be used as the original measurement data.

[0050] For example, if the engineering machinery is a multi-function drilling rig or other similar drilling machinery, sensors can be used to collect thrust pressure, impact pressure, rotary pressure, impact flow rate, rotary speed, and the drilling speed of the power head at different times. This can generate time series data consisting of the thrust pressure, impact pressure, rotary pressure, impact flow rate, rotary speed, and drilling speed of the power head collected at different times, and this data can be used as the raw measurement data.

[0051] After acquiring the raw measurement data, preprocessing the raw measurement data may include at least one of the following: filtering out abnormal operation data from the raw measurement data; performing denoising on the raw measurement data; and performing standardization and normalization on the raw measurement data. The abnormal operation data filtered out from the raw measurement data may include at least one of idle data, hole-guiding data, stuck drill data, and drill-lifting data.

[0052] In some examples, the raw measurement data is preprocessed as follows: abnormal operation data is filtered out from the raw measurement data under multiple gears to obtain normal operation data under multiple gears; the normal operation data under multiple gears is denoised, standardized, and normalized to obtain operation sample data under multiple gears.

[0053] In the above example, abnormal operation data can be filtered out from the raw measurement data based on its characteristics. For example, standby, non-operation data has impact and thrust pressures close to zero, with slight fluctuations due to noise; pilot hole data has impact and thrust pressures that are intermittent and short-lived; and stuck drill data has rotation pressures exceeding a certain threshold. Based on these data characteristics, abnormal operation data can be filtered out as follows: from the raw measurement data, data with impact and thrust pressures less than a certain value is filtered out to filter out standby, non-operation data; from the raw measurement data, data with rotation pressures greater than a certain value is filtered out to filter out stuck drill data; and from the raw measurement data, data with a continuous duration less than a certain value is filtered out to filter out pilot hole data and drill lifting data. By filtering out abnormal operation data from the raw measurement data, the interference of abnormal operation data on the training of the surrounding rock identification model can be reduced, thereby helping to improve the performance of the trained surrounding rock identification model.

[0054] In the above example, the normal operating data under multiple gears can be denoised using the following exemplary methods: a low-pass filter algorithm is used to remove high-frequency noise from the normal operating data; and a median moving average filter algorithm is used to remove impulse noise from the normal operating data. Combining these two denoising algorithms can better eliminate noise from the raw measurement data, thereby helping to reduce the interference of noisy data on the training of the surrounding rock identification model, thereby improving the performance of the trained surrounding rock identification model. Furthermore, in specific implementations, other algorithms can also be used to denoise the normal operating data. For example, a moving average filter algorithm, a median filter algorithm, etc. can be used to denoise the normal operating data.

[0055] In the above example, the normal operation data under multiple gears can be normalized according to the following exemplary formula:

[0056]

[0057] in, is the normalized data, is the data before normalization, is the minimum value of the data, is the maximum value of the data. Taking the push pressure sample data set as an example, is the normalized thrust pressure sample, is the thrust pressure sample before normalization, is the minimum value of the thrust pressure in the thrust pressure sample data set, is the maximum thrust pressure in the thrust pressure sample data set.

[0058] In the above example, the normal operation data under multiple gears can be standardized according to the following exemplary formula:

[0059]

[0060] in, For the data after standardization, is the data before standardization. is the data mean, is the standard deviation. Taking the push pressure sample data set as an example, is the propulsion pressure sample after standardization. is the propulsion pressure sample before standardization. is the mean thrust pressure calculated based on the thrust pressure sample data set, is the standard deviation calculated based on the propulsion pressure sample data set.

[0061] In the embodiment of the present disclosure, by normalizing and standardizing the original measurement data, the input data and output data can be fixed within a certain range and converted into dimensionless scalars, which can improve the convergence speed of the model, avoid gradient vanishing or gradient exploding problems, and increase the stability and generalization ability of the model.

[0062] In step 12, the neural network model corresponding to each of the multiple gears is used to process the operation sample data under each gear to obtain the strength prediction value and integrity prediction value of the sample surrounding rock.

[0063] There are multiple implementations of step 12. Two implementations are described below as examples.

[0064] In the first embodiment, the operation sample data under each gear is input into the neural network model corresponding to the gear to obtain the strength output value and integrity output value of the sample surrounding rock, and the strength output value of the sample surrounding rock is used as the strength prediction value, and the integrity output value of the sample surrounding rock is used as the integrity prediction value.

[0065] In the second embodiment, the operation sample data under each gear is input into the neural network model corresponding to the gear to obtain the strength output value and integrity output value of the sample surrounding rock; the strength output value of the sample surrounding rock is denormalized and denormalized to obtain the strength prediction value; the integrity output value of the sample surrounding rock is denormalized and denormalized to obtain the integrity prediction value.

[0066] In the embodiments of the present disclosure, the neural network model used for surrounding rock identification can adopt a variety of network structures. In some examples, the neural network model includes a gated recurrent neural network module and a fully connected layer. For example, the gated neural network module is a gated recurrent unit (GRU). GRU is simplified based on the long short-term memory network, introducing fewer parameters and structural complexity. In these examples, using the neural network model corresponding to each gear to process the operation sample data includes: using the gated recurrent neural network module to process the operation sample data under the gear to obtain a hidden state vector; using the fully connected layer to process the hidden state vector to obtain a strength prediction value of the sample surrounding rock and a integrity prediction value of the sample surrounding rock.

[0067] The strength and integrity of the surrounding rock sample can be characterized in a variety of ways. For example, the strength of the surrounding rock sample can be characterized by uniaxial compressive strength, while the integrity of the surrounding rock sample can be characterized by the fracture coefficient. The fracture coefficient is the square of the ratio of the longitudinal wave velocity of the rock mass to the rock mass. Furthermore, in specific implementations, the strength and integrity of the surrounding rock sample can also be characterized by other methods.

[0068] In step 13, the loss function value is determined according to the strength prediction value and the integrity prediction value of the sample surrounding rock, and the strength label value and the integrity label value of the sample surrounding rock.

[0069] In some examples, step 13 includes: calculating a first loss function value based on the predicted strength value and strength label value of the sample surrounding rock; calculating a second loss function value based on the predicted integrity value and integrity label value of the sample surrounding rock; and determining a total loss function value based on the first and second loss function values. In specific implementations, the above loss function values ​​can be calculated based on functions such as mean square error (MSE) or root mean square error.

[0070] In step 14, the neural network model corresponding to each of the multiple gears is trained according to the loss function value.

[0071] Specifically, the parameters of the neural network model can be adjusted according to the loss function value. After multiple rounds of training, the desired surrounding rock recognition model can be obtained.

[0072] For example, assuming there are 20 gears, based on Figure 1 The process shown here trains the neural network model corresponding to each gear. Afterwards, the 20 neural network models obtained through training that meet the required model accuracy are used as the final surrounding rock recognition model.

[0073] In the embodiments disclosed herein, on the one hand, by collecting surrounding rock operation sample data and using surrounding rock strength and integrity as labels for supervised learning, the model can learn the correlation between surrounding rock operation sample data and surrounding rock strength and integrity, so that the trained model can predict surrounding rock strength and surrounding rock integrity with more physical significance, which helps to improve the performance of the trained surrounding rock identification model; on the other hand, by collecting surrounding rock operation sample data under multiple gears and training the neural network models corresponding to multiple gears accordingly, it can meet the needs of refined surrounding rock identification under complex working conditions.

[0074] Figure 2 FIG. 1 is a flow chart of a model training method according to some other embodiments of the present disclosure. Figure 2 As shown, the model training method in the embodiment of the present disclosure includes steps 21 to 28.

[0075] In step 21, original measurement data of operations performed on sample surrounding rocks at multiple gear positions are collected.

[0076] In some embodiments, raw measurement data is collected by sensors deployed on the engineering machinery. For example, if the engineering machinery is a multi-functional drilling rig, the raw measurement data collected by the sensors may include thrust pressure, impact pressure, rotary pressure, displacement of the power head, impact flow rate, and rotary speed at each of multiple moments. After collecting the displacement of the power head, the drilling speed is obtained by taking the time derivative. This results in the following preliminarily processed raw measurement data: thrust pressure, impact pressure, rotary pressure, drilling speed, impact flow rate, and rotary speed.

[0077] In step 22 , abnormal operation data is filtered out from the original measurement data under multiple gear positions to obtain normal operation data under multiple gear positions.

[0078] Raw measurement data includes normal operation data and abnormal operation data (i.e., abnormal operation data) that changes over time. Abnormal operation data includes at least one of the following: idle operation data, hole entry data, stuck drill data, and drill pull data. For details on how to filter abnormal operation data from raw measurement data, please refer to the relevant content in the previous embodiments.

[0079] In step 23 , the normal operation data under multiple gears are filtered.

[0080] In some examples, normal operation data under multiple gears are filtered according to the following methods: a low-pass filter algorithm is used to remove high-frequency noise in the normal operation data; a median moving average filter algorithm is used to remove impulse noise in the normal operation data.

[0081] In some examples, the normal operation data includes multiple signal sequences, such as a propulsion pressure signal sequence (or propulsion pressure sequence), an impact pressure signal sequence, a rotation pressure signal sequence, etc. Each signal sequence includes N samples, where N is an integer greater than 1.

[0082] In the above example, the median moving average filtering algorithm is used to remove the impulse noise in the normal operation data, including: for each signal sequence (n is the index of the sample, ranging from 0 to N-1), slide a length of Window, assuming that the starting data in the window is , the termination data in the window is , sort the data in the window in ascending order to obtain the sorted sequence ,in, is the index of the sample (the value of m is from arrive ); Take the proportion of the sorted sequence as The average value of the data is calculated and used as the The value of the signal sampling point For example, the value of the filtered signal sampling point can be calculated according to the following formula .

[0083]

[0084] in, is the value of the signal sampling point after filtering; M is the sampling window size, which is a positive integer. In specific implementation, the same or different window sizes can be used to process different signal sequences. For example, when processing thrust pressure, M can be set to 100, and when processing thrust speed, M can be set to 300; k is the number of samples averaged in the window, which is a positive integer less than M. In specific implementation, the value of k can be the same or different when processing different signal sequences. For example, when processing thrust pressure, k can be set to 40, and when processing thrust speed, k can be set to 60.

[0085] After calculating according to the above formula Afterwards, slide to select the next window data and determine the first window after filtering based on the next window data. The value of each signal sampling point is obtained by looping in this way to realize data filtering.

[0086] In the embodiment of the present disclosure, by adopting a low-pass filtering algorithm, the high-frequency noise generated by the impact of the power head in the signal can be removed; by adopting a median moving average filtering algorithm, the pulse noise in the signal can be filtered out. By combining the above two denoising algorithms, the noise in the original measurement data can be better eliminated, thereby helping to reduce the interference of noise data on the training of the surrounding rock identification model, and further helping to improve the performance of the trained surrounding rock identification model. Moreover, compared with adopting the median filtering algorithm and the moving average filtering algorithm, by adopting the median moving average filtering algorithm to filter the normal operation data, it is possible to achieve a better filtering effect on rapidly changing signals, and to better eliminate the sampling value deviation caused by interference, thereby overcoming the shortcomings of the median filtering algorithm that the filtering effect on rapidly changing parameters is poor, and overcoming the shortcomings of the moving average filtering algorithm that it is difficult to eliminate the sampling value deviation caused by interference.

[0087] In step 24 , normal operation data under multiple gears are normalized and standardized to obtain operation sample data under multiple gears.

[0088] In some examples, when the engineering machinery is a multi-function drilling rig, the multiple gears are represented by impact gears and swing gears. Assuming there are a impact gears and b swing gears, in step 24, the normal operation data for a*b gears is normalized and standardized to obtain sample operation data for a*b gears. For details on how to normalize and standardize the normal sample data, please refer to the relevant content in the previous embodiment.

[0089] The operation sample data at each gear position are all time series data. In some examples, the operation sample data at each gear position include a thrust pressure sequence, a percussion pressure sequence, a rotary pressure sequence, a drilling speed sequence, a percussion flow sequence, and a rotary speed sequence.

[0090] In step 25, the true strength and true integrity of the sample surrounding rock are determined.

[0091] In some embodiments, the uniaxial compressive strength of the sample surrounding rock is obtained through in-situ sampling and used as the true strength of the sample surrounding rock. The fracture coefficient of the sample surrounding rock is measured through dynamic methods and used as the true integrity of the sample surrounding rock. In addition, other methods can also be used to determine the true strength and true integrity of the sample surrounding rock during specific implementation.

[0092] In step 26, the true strength and true integrity of the sample surrounding rock are normalized and standardized, and the processing results are used as labels for the operation sample data.

[0093] In step 27, the labeled operation sample data of each gear among the multiple gears is used to train the neural network model corresponding to the gear.

[0094] In some examples, the neural network model includes a GRU and a fully connected layer. The GRU is a simplified version of the long short-term memory network, introducing fewer parameters and structural complexity, thus better processing operational sample data. By adding a fully connected layer after the GRU, the model can learn the operational sample data and the correlation between surrounding rock strength and fracture coefficient.

[0095] In the above example, step 27 may include: inputting the operation sample data into the neural network model to obtain the strength output value and integrity output value of the sample surrounding rock; denormalizing and denormalizing the strength output value and integrity output value of the sample surrounding rock to obtain the strength prediction value and integrity prediction value of the sample surrounding rock; calculating the loss function value based on the strength prediction value and integrity prediction value of the sample surrounding rock, as well as the true strength and true integrity of the sample surrounding rock; and training the neural network model based on the loss function value until the training cutoff condition is met.

[0096] In the above example, step 27 may also include: inputting the operation sample data into the neural network model to obtain the strength output value and integrity output value of the sample surrounding rock; calculating the loss function value based on the strength output value and integrity output value of the sample surrounding rock, as well as the true strength and true integrity of the sample surrounding rock; and training the neural network model based on the loss function value until the training cutoff condition is met.

[0097] In some examples, the training cutoff condition is that the accuracy of the trained neural network model reaches a set accuracy threshold. In addition, in specific implementations, the training cutoff condition can also be set based on other model evaluation indicators.

[0098] In step 28, the multiple neural network models finally obtained by training are stored.

[0099] In the disclosed embodiment, the above process can improve the performance of the trained surrounding rock identification model, thereby facilitating the subsequent accurate, rapid and low-cost determination of surrounding rock parameters, thereby meeting the refined surrounding rock identification requirements under complex construction conditions.

[0100] Figure 3 FIG. 1 is a flow chart of a surrounding rock identification method according to some embodiments of the present disclosure. Figure 3 As shown, the surrounding rock identification method includes steps 31 and 32.

[0101] In step 31 , operation data of the engineering machinery equipment operating on the surrounding rock to be identified at a specified gear position is obtained.

[0102] The gear position may be characterized by at least one of an impact gear position and a rotation gear position. For example, the gear position may be characterized by a combination of an impact gear position and a rotation gear position.

[0103] In some examples, step 31 includes: receiving raw measurement data of the engineering machinery operating on the surrounding rock to be identified at a specified gear position from multiple sensors on the engineering machinery; and preprocessing the raw measurement data to obtain operation data at the specified gear position.

[0104] In the above example, the original measurement data may include at least one of propulsion pressure, impact pressure, rotation pressure, impact flow, rotation speed, displacement of the power head, and speed data of the power head of the engineering machinery equipment.

[0105] In the above example, preprocessing the raw measurement data may include at least one of the following: filtering out abnormal operation data from the raw measurement data; performing denoising on the raw measurement data; and performing standardization and normalization on the raw measurement data. The abnormal operation data filtered out from the raw measurement data includes at least one of idle data, hole-starting data, stuck drill data, and drill-lifting data.

[0106] In step 32, the operating data is processed using the surrounding rock identification model corresponding to the designated gear position to obtain the strength prediction value and integrity prediction value of the surrounding rock to be identified.

[0107] In the embodiment of the present disclosure, multiple rock identification models corresponding to gear positions are pre-stored. In step 32, a rock identification model corresponding to a specified gear position is selected from the multiple rock identification models corresponding to gear positions, and the operation data is processed based on the selected model.

[0108] Step 32 can be implemented in a variety of ways. Two implementations are described below as examples. In the first implementation, sample data from an operation at a specified gear position is input into a surrounding rock identification model corresponding to that gear position to obtain strength output values ​​and integrity output values ​​for the sample surrounding rock. The strength output value of the sample surrounding rock is used as a strength prediction value, and the integrity output value of the sample surrounding rock is used as an integrity prediction value.

[0109] In a second embodiment, sample data from operations at a specified gear position is input into a surrounding rock identification model corresponding to that gear position to obtain strength and integrity output values ​​for the sample surrounding rock. The strength output values ​​of the sample surrounding rock are denormalized and denormalized to obtain a strength prediction value; and the integrity output values ​​of the sample surrounding rock are denormalized and denormalized to obtain an integrity prediction value. The surrounding rock identification model can be trained using the model training method described above.

[0110] Surrounding rock identification models can employ a variety of network structures. In some examples, the surrounding rock identification model includes a gated recurrent neural network module and a fully connected layer. For example, the gated neural network module is a gated recurrent unit (GRU). In these examples, processing the operating sample data using the neural network model corresponding to a specific gear position includes: processing the operating sample data for that gear position using the gated recurrent neural network module to obtain a hidden state vector; and processing the hidden state vector using the fully connected layer to obtain a predicted value for the strength and integrity of the sample surrounding rock.

[0111] In the disclosed embodiment, through the above process, the surrounding rock parameters can be determined accurately, quickly and at low cost, meeting the demand for refined surrounding rock identification under complex construction conditions.

[0112] Figure 4 FIG. 1 is a flow chart of a surrounding rock identification method according to other embodiments of the present disclosure. Figure 4 As shown, the surrounding rock identification method includes steps 41 to 46.

[0113] In step 41, original measurement data of the operation performed on the surrounding rock to be identified at a specified gear position is collected.

[0114] In some examples, the raw measurement data includes thrust pressure, impact pressure, rotary pressure, impact flow rate, rotary speed, and drilling speed of the engineering machinery equipment.

[0115] In step 42 , abnormal operation data is filtered out from the original measurement data at the designated gear position to obtain normal operation data at the designated gear position.

[0116] Regarding how to filter out abnormal operation data, reference may be made to the relevant content in the aforementioned embodiment.

[0117] In step 43 , the normal operation data under the designated gear is filtered.

[0118] Regarding how to filter the normal operation data, reference may be made to the relevant content in the aforementioned embodiment.

[0119] In step 44 , the normal operation data at the designated gear position is normalized and standardized to obtain the operation data at the designated gear position.

[0120] Regarding how to normalize and standardize the normal operation data, reference may be made to the relevant content in the aforementioned embodiment.

[0121] In the embodiment of the present disclosure, preprocessing the original measurement data through steps 42 to 44 can reduce the interference of abnormal operation data and noise data on the subsequent surrounding rock identification model for surrounding rock identification, thereby improving the accuracy and reliability of the surrounding rock identification results.

[0122] In step 45, the operating data at the designated gear position is processed using the surrounding rock identification model corresponding to the designated gear position to obtain the strength output value and integrity output value of the surrounding rock to be identified.

[0123] In step 46 , the strength output value and the integrity output value of the surrounding rock to be identified are subjected to denormalization and destandardization processing to obtain the strength prediction value and the integrity prediction value of the surrounding rock to be identified.

[0124] In some embodiments, the surrounding rock identification method further includes: after step 46, grading the strength prediction value and the integrity prediction value of the surrounding rock to be identified to obtain the strength grade and the integrity grade of the surrounding rock to be identified.

[0125] In the disclosed embodiment, the above process can accurately, quickly and cost-effectively determine surrounding rock parameters, meeting the demand for refined surrounding rock identification under complex construction conditions.

[0126] Figure 5FIG. 1 is a schematic diagram of the structure of a model training device according to some embodiments of the present disclosure. Figure 5 As shown, the model training device 50 is used to execute the model training method as described above, including an acquisition module 51, a prediction module 52, a determination module 53 and a training module 54.

[0127] The acquisition module 51 is configured to acquire operation sample data of the engineering machinery equipment operating on sample surrounding rocks in multiple gears.

[0128] The prediction module 52 is configured to process the operation sample data under each gear position using a neural network model corresponding to each gear position in the plurality of gear positions to obtain a strength prediction value and an integrity prediction value of the sample surrounding rock.

[0129] The determination module 53 is configured to determine a loss function value according to the strength prediction value and the integrity prediction value of the sample surrounding rock, and the strength label value and the integrity label value of the sample surrounding rock.

[0130] The training module 54 is configured to train the neural network model corresponding to each gear in the plurality of gears according to the loss function value.

[0131] In the disclosed embodiment, the above device can improve the performance of the trained surrounding rock identification model, thereby facilitating the subsequent accurate, rapid and low-cost determination of surrounding rock parameters, and meeting the refined surrounding rock identification needs under complex construction conditions.

[0132] Figure 6 FIG. 1 is a schematic diagram of the structure of a surrounding rock identification device according to some embodiments of the present disclosure. Figure 6 As shown, the surrounding rock identification device 60 is used to execute the surrounding rock identification method as described above, and includes an acquisition module 61 and a prediction module 62.

[0133] The acquisition module 61 is configured to acquire operation data of the engineering machinery equipment operating on the surrounding rock to be identified at a specified gear position.

[0134] The prediction module 62 is configured to process the operation data using the surrounding rock identification model corresponding to the designated gear position to obtain the strength prediction value and integrity prediction value of the surrounding rock to be identified.

[0135] In the disclosed embodiment, the above device can accurately, quickly and cost-effectively determine surrounding rock parameters, meeting the demand for refined surrounding rock identification under complex construction conditions.

[0136] Figure 7 FIG. 1 is a schematic diagram of the structure of an electronic device according to some embodiments of the present disclosure. Figure 7As shown, electronic device 70 includes a memory 71 and a processor 72 coupled to memory 71. Memory 71 is used to store instructions for executing the aforementioned model training method or the aforementioned surrounding rock identification method. Processor 72 is configured to execute the model training method or surrounding rock identification method according to any of the embodiments of the present disclosure based on the instructions stored in memory 71.

[0137] Figure 8 FIG. 1 is a schematic diagram of the structure of a surrounding rock identification system according to some embodiments of the present disclosure. Figure 8 As shown, the surrounding rock identification system 80 includes a signal acquisition module 81 , a model training device 50 and a surrounding rock identification device 60 .

[0138] The signal acquisition module is used to collect the operating sample data under multiple gears required in the training phase, and to collect the operating data under the specified gear required in the surrounding rock identification phase.

[0139] In some examples, the signal acquisition module uses Figure 9 The structure shown. Figure 9 As shown, the signal acquisition module 81 includes a propulsion pressure sensor 811 , an impact pressure sensor 812 , a rotary pressure sensor 813 , an encoder 814 , an impact flow sensor 815 , a Hall sensor 816 , a hub 817 , a controller 818 , an onboard computer 819 and a memory 820 .

[0140] Among them, the thrust pressure sensor 811 is used to collect thrust pressure in multiple gears; the impact pressure sensor 812 is used to collect impact pressure; the rotary pressure sensor 813 is used to collect rotary pressure; the encoder 814 is used to collect the displacement of the power head; the impact flow sensor 815 is used to collect the impact flow; and the Hall sensor 816 is used to collect the rotation speed signal. In specific implementations, the thrust pressure sensor, impact pressure sensor, and rotary pressure sensor can be installed on the thrust hydraulic oil pipeline, impact hydraulic oil pipeline, and rotation hydraulic oil pipeline of engineering machinery equipment (such as a multi-functional drilling rig). The encoder can be installed on the coupling of the mast hydraulic motor to test the displacement of the power head; the impact flow sensor can be installed on the impact hydraulic oil pipeline; and the Hall sensor can be installed below the power head shaft to measure the rotation speed of the drill pipe.

[0141] The hub 817 is used to receive the propulsion pressure, impact pressure, rotation pressure, displacement of the power head, impact flow and rotation speed from the above-mentioned multiple sensors, and transmit the received data to the controller 818.

[0142] Controller 818 is configured to transmit the received data to onboard computer 819 via the CAN bus protocol. Onboard computer 819 is configured to derivate the displacement data with respect to time to obtain the drilling speed, and to store the thrust pressure, impact pressure, rotary pressure, drilling speed, impact flow rate, and rotary speed data in memory 820.

[0143] Furthermore, in specific implementations, the signal acquisition module may employ other structures. For example, depending on the data to be collected, the signal acquisition module may include different sensors. For example, the signal acquisition module may include a portion of a propulsion pressure sensor, an impact pressure sensor, a rotational pressure sensor, an encoder, an impact flow sensor, or a Hall effect sensor. Alternatively, the signal acquisition module may include other sensors.

[0144] After the signal acquisition module collects sample operating data for multiple gear positions required for the training phase, the model training device 50 trains neural network models corresponding to the multiple gear positions based on the sample operating data to obtain surrounding rock identification models corresponding to the multiple gear positions. After the signal acquisition module collects operating data for a specified gear position required for the surrounding rock identification phase, the surrounding rock identification device 60 calls the surrounding rock identification model corresponding to the specified gear position to process the operating data to obtain an identification result for the surrounding rock to be identified.

[0145] Figure 10 FIG. 1 is a schematic diagram of the structure of engineering machinery equipment according to some embodiments of the present disclosure. Figure 10 As shown, the engineering machinery equipment 100 includes at least one of a model training device 50 and a surrounding rock identification device 60 .

[0146] The engineering machinery equipment 100 may be a multifunctional drilling rig or other equipment that can be used for surrounding rock operations.

[0147] The model training device 50 is configured to execute the model training method as described above.

[0148] The surrounding rock identification device 60 is configured to execute the surrounding rock identification method as described above.

[0149] In the disclosed embodiments, the above-mentioned engineering machinery and equipment can accurately, quickly and cost-effectively determine surrounding rock parameters, thereby meeting the demand for refined surrounding rock identification under complex construction conditions.

[0150] Here, various aspects of the present disclosure are described with reference to flowcharts and / or block diagrams of methods, devices, and computer program products according to embodiments of the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks, can be implemented by computer-readable program instructions.

[0151] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, so that the processor executes the instructions to produce means for implementing the functions specified in one or more blocks in the flowcharts and / or block diagrams.

[0152] These computer-readable program instructions may also be stored in a computer-readable memory, which cause the computer to operate in a specific manner to produce an article of manufacture, including instructions for implementing the functions specified in one or more blocks in the flowcharts and / or block diagrams.

[0153] The present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.

[0154] The model training and surrounding rock identification method, apparatus, and engineering machinery disclosed herein have been described in detail. To avoid obscuring the concepts of the present disclosure, some details known in the art have been omitted. Based on the above description, those skilled in the art will fully understand how to implement the technical solutions disclosed herein.

Claims

1. A model training method, comprising: Obtaining operation sample data of engineering machinery equipment operating on sample surrounding rock at multiple gear positions; Using a neural network model corresponding to each of the plurality of gears, the operation sample data at each gear is processed to obtain a strength prediction value and an integrity prediction value of the sample surrounding rock; Determining a loss function value according to the strength prediction value and the integrity prediction value of the sample surrounding rock, and the strength label value and the integrity label value of the sample surrounding rock; According to the loss function value, a neural network model corresponding to each gear in the multiple gears is trained.

2. The model training method according to claim 1, wherein: Each gear is characterized by at least one of an impact gear and a swing gear.

3. The model training method according to claim 2, wherein: The step of obtaining operation sample data of the engineering machinery operating on sample surrounding rocks at multiple gear positions includes: Receiving, from a plurality of sensors on the engineering machinery, raw measurement data of the engineering machinery operating on sample surrounding rock at a plurality of gears, wherein the raw measurement data includes at least one of propulsion pressure, impact pressure, rotary pressure, impact flow rate, rotary speed, displacement of a power head, and speed data of the power head of the engineering machinery; The original measurement data is preprocessed to obtain operation sample data under the multiple gears.

4. The model training method according to claim 3, wherein: The preprocessing of the original measurement data to obtain the operation sample data under the multiple gears includes: Filtering out abnormal operation data from the raw measurement data under the multiple gear positions to obtain normal operation data under the multiple gear positions, wherein the abnormal operation data includes at least one of standby non-operation data, hole-guiding data, drill stuck data, and drill lifting data; The normal operation data under the multiple gears are subjected to denoising, standardization and normalization processing to obtain operation sample data under the multiple gears.

5. The model training method according to claim 4, wherein: The performing denoising on the normal operation data under the multiple gears includes: Using a low-pass filter algorithm to remove high-frequency noise in the normal operation data; A median moving average filtering algorithm is used to remove impulse noise in the normal operation data.

6. The model training method according to claim 3, wherein: The processing of the operation sample data for each gear position by using the neural network model corresponding to each gear position includes: Inputting the operation sample data under each gear into the neural network model corresponding to each gear to obtain the strength output value and integrity output value of the sample surrounding rock; The strength output value and the integrity output value of the sample surrounding rock are subjected to denormalization and destandardization processing to obtain the strength prediction value and the integrity prediction value of the sample surrounding rock.

7. The model training method according to any one of claims 1 to 6, wherein: The strength of the sample surrounding rock is the uniaxial compressive strength, and the integrity of the sample surrounding rock is the crack coefficient; and / or The neural network model includes a gated recurrent neural network module and a fully connected layer.

8. A surrounding rock identification method comprising: Obtaining operation data of the engineering machinery equipment operating on the surrounding rock to be identified at a specified gear position; The operation data is processed using a surrounding rock identification model corresponding to the designated gear position to obtain a strength prediction value and an integrity prediction value of the surrounding rock to be identified.

9. The surrounding rock identification method according to claim 8, wherein: The designated gear is characterized by at least one of an impact gear and a swing gear.

10. The surrounding rock identification method according to claim 9, wherein: The step of obtaining the operation data of the engineering machinery equipment operating on the surrounding rock to be identified at a specified gear position includes: Receiving, from a plurality of sensors on the engineering machinery, raw measurement data of the engineering machinery operating at the designated gear position on the surrounding rock to be identified, wherein the raw measurement data includes at least one of propulsion pressure, impact pressure, rotation pressure, impact flow rate, rotation speed, displacement of a power head, and speed data of the power head of the engineering machinery; The original measurement data is preprocessed to obtain the operation data under the specified gear.

11. The surrounding rock identification method according to claim 10, wherein: The preprocessing of the original measurement data to obtain the operation data under the specified gear position includes: Filtering abnormal operation data from the raw measurement data at the specified gear position to obtain normal operation data at the specified gear position, wherein the abnormal operation data includes at least one of standby non-operation data, hole-guiding data, drill stuck data, and drill lifting data; The normal operation data under the designated gear position is subjected to denoising, standardization and normalization processing to obtain the operation data under the designated gear position.

12. The surrounding rock identification method according to claim 10, wherein: The processing of the operation data by using the neural network model corresponding to the designated gear position includes: Inputting the operation data under the designated gear into the neural network model corresponding to the designated gear to obtain the strength output value and the integrity output value of the surrounding rock to be identified; The strength output value and the integrity output value of the surrounding rock to be identified are subjected to denormalization and destandardization processing to obtain the strength prediction value and the integrity prediction value of the surrounding rock to be identified.

13. An apparatus comprising: A module for executing the model training method according to any one of claims 1 to 7, or a module for executing the surrounding rock identification method according to any one of claims 8 to 12.

14. An apparatus comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the model training method according to any one of claims 1 to 7, or the surrounding rock identification method according to any one of claims 8 to 12, based on instructions stored in the memory.

15. An engineering machinery device comprising: The device according to claim 13 or 14.

16. A computer-readable storage medium having computer instructions stored thereon, wherein when the instructions are executed by a processor, the model training method according to any one of claims 1 to 7 or the surrounding rock identification method according to any one of claims 8 to 12 is implemented.

17. A computer program product having computer instructions stored thereon, wherein when the instructions are executed by a processor, the model training method according to any one of claims 1 to 7 or the surrounding rock identification method according to any one of claims 8 to 12 is implemented.

Images