Equipment monitoring method for tunnel excavation of three-arm drill jumbo

Abstract

The invention provides an equipment monitoring method for tunnel excavation of a three-arm drill jumbo, which is suitable for the field of tunnel excavation and comprises the steps of installation of acquisition equipment, data acquisition, data preprocessing, extraction of characteristic parameters, establishment of an input and output data set and establishment and application of an identification model. The equipment monitoring method for tunnel excavation of the three-arm drill jumbo has the advantages of being accurate in recognition, high in safety and the like, and can be widely applied to the field of tunnel excavation.

Description

Technical Field

[0001] The invention relates to an equipment monitoring method for tunnel excavation using a three-arm rock drilling trolley, and is applicable to the field of tunnel excavation. Background Art

[0002] In tunnel construction, the three-arm rock drilling rig is the main construction equipment, and its working status directly affects the efficiency and safety of tunnel excavation. However, the tunnel construction environment is complex, and the equipment is prone to failures during long-term, high-intensity operation, resulting in delays in construction progress and even safety accidents. Therefore, it is of great significance to monitor the operating status of the three-arm rock drilling rig in real time, detect equipment abnormalities in a timely manner, and prevent equipment failures. Traditional equipment monitoring methods mostly rely on manual regular maintenance or sensor-based data monitoring, which often have problems such as poor real-time performance and untimely response. Sound, as one of the important representations of the equipment's operating status, can reflect the equipment's operating status under different working conditions. Therefore, equipment status monitoring based on sound data has gradually become a research hotspot.

[0003] With the development of artificial intelligence and machine learning technology, it is possible to establish an intelligent recognition model based on the sound data collected during the operation of the equipment to classify and identify the normal and abnormal states of the equipment. This monitoring method based on sound data can provide real-time, contactless equipment status detection with high accuracy and reliability, and is suitable for equipment monitoring needs in complex tunnel construction environments. To this end, a three-arm drilling rig equipment monitoring method based on sound data analysis is proposed, which can monitor the equipment operation status in real time through sound feature extraction and intelligent algorithm models, timely warn of equipment abnormalities, and ensure the smooth progress of tunnel construction. Summary of the invention

[0004] The purpose of the present invention is to propose a three-arm drilling rig equipment monitoring method based on sound data analysis, which can monitor the equipment operation status in real time through sound feature extraction and intelligent algorithm model, timely warn of equipment abnormalities, and ensure the smooth progress of tunnel construction.

[0005] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0006] A method for monitoring equipment for tunnel excavation using a three-arm drilling trolley, characterized in that the method comprises the following steps:

[0007] S101, installation of acquisition equipment;

[0008] The installation of the collection equipment includes installing the recording collection equipment at a fixed position of the three-arm drilling rig to collect sound data during the operation of the three-arm drilling rig;

[0009] S102, data collection;

[0010] The data collection includes using the recording collection device to record a large number of sound samples, totaling n, at a certain sampling frequency and recording duration during the tunnel excavation of the three-arm rock drilling rig, denoted as S1~S n The sound samples include both normal use samples and abnormal use samples of the instrument and equipment, and the proportion of abnormal use samples of the instrument and equipment to the total sound samples must be ≥10%;

[0011] S103, data preprocessing;

[0012] The data preprocessing includes processing the n sound samples S1 to S n Performing denoising processing, wherein the denoising processing adopts a median filtering method;

[0013] S104, extracting characteristic parameters;

[0014] The feature parameter extraction includes for each sound sample S i , i = 1 ~ n, extract 5 characteristic parameters, namely short-time energy, zero crossing rate, spectrum centroid, spectrum bandwidth and MFCC coefficient, the parameters are marked as Pij, i = 1 ~ n, j = 1 ~ 5;

[0015] S105, establishing an input and output data group;

[0016] The establishment of the input and output data set includes establishing input and output data. For the sound sample Si, the input is the parameter P ij , i = 1 ~ n, j = 1 ~ 5, the output is 0 or 1, when the sound sample Si represents that the instrument is working normally, the output is 0, and when the sound sample Si represents that the instrument is working abnormally, the output is 1, and the final input and output data group is C;

[0017] S106, establishment and application of recognition model;

[0018] The establishment and application of the recognition model includes constructing a support vector machine recognition model for the recognition of sound samples to achieve the purpose of equipment monitoring of tunnel excavation of a three-arm drilling rig;

[0019] The establishment and application of the recognition model, including the steps of constructing the support vector machine recognition model, are as follows: a) constructing a support vector machine model, b) dividing the data set, dividing the input and output data group C into a training set and a test set, with a division ratio of 8:2, the training set is used to train the model, and the test set is used to verify the performance of the model, c) inputting the training set data into the selected model, using the loss function and optimization algorithm to train the model, and gradually adjusting the model parameters to minimize the loss function value and optimize the classification performance of the model, d) model verification and evaluation, after the training is completed, inputting the test set into the model for evaluation, and using the classification accuracy index to evaluate the performance of the model in identifying the normal and abnormal states of the device. If the model performance does not meet the requirements, it can be further optimized by adjusting the model structure or hyperparameters (such as learning rate, number of hidden layer neurons, etc.), and the optimization measures are to increase the number of samples, adjust the model structure, introduce regularization methods, and finally obtain the optimized model;

[0020] Furthermore, in the above S103, the steps of the median filter algorithm are: a) defining the window size, b) taking the median value of the surrounding samples of each sample point as a new sample value, and repeating this process until the entire signal is processed;

[0021] Further, in the above S105, the MFCC coefficient extraction step is: a) dividing the sound sample into a plurality of frames, each frame containing a plurality of sample points, b) applying Fourier transform to each frame to obtain a spectrum, c) mapping the spectrum energy to a Mel-scale filter bank, d) calculating the logarithmic energy of each Mel filter output, e) applying discrete cosine transform to the logarithmic energy, and extracting the first plurality of coefficients as MFCC feature parameters;

[0022] Furthermore, in the above S105, the input value Pij of the characteristic parameter is subjected to standardization processing before being input into the recognition model, so that the value ranges of different characteristic parameters are normalized to improve the accuracy of the recognition model;

[0023] Furthermore, in the above S106, the kernel function of the support vector machine recognition model is a radial basis function, and the model is optimized and adjusted by a cross-validation method;

[0024] Furthermore, in the above S106, when training the recognition model, the proportion of abnormal sound samples is weighted, and the weight of the weighted processing is set as follows: the weight of the normal working sample of the instrument is set to 1, and the weight of the abnormal working sample of the instrument is calculated according to formula (1),

[0025]

[0026] Where N 总 is the total number of samples, N 异 is the number of abnormal samples.

[0027] Furthermore, in the above S106, the loss function adopts a weighted loss function, and the weighted loss function is formula (2):

[0028]

[0029] In the formula, ω i is the weight of the i-th sample, y i The actual output result is: is the model predicted probability.

[0030] The beneficial effects of the present invention are:

[0031] The beneficial effects of the present invention are: (1) Extraction of multi-dimensional feature parameters: By extracting multi-dimensional sound features such as short-time energy, zero-crossing rate, spectrum centroid, spectrum bandwidth and MFCC coefficient, various aspects of the equipment operation status are fully characterized, providing comprehensive information support for the accurate classification of the model; (2) Weighted processing of abnormal samples: During model training, abnormal sound samples are weighted to ensure that the model can more sensitively identify abnormal conditions, reduce the possibility of underreporting, and improve the reliability and safety of equipment monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 : A flow chart of an equipment monitoring method for tunnel excavation of a three-arm rock drilling rig of the present invention. DETAILED DESCRIPTION

[0033] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings; it should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0034] The following is a specific embodiment of a method for monitoring equipment for tunnel excavation using a three-arm drilling rig.

[0035] The specific project based on this embodiment is a tunnel of the eighth section of the Malaysia East Rail Project. The total length of the tunnel is 959m, of which the entrance mileage is CH516+955 and the open tunnel is 5m long. The exit mileage is CH517+914 and the open tunnel is 25m long. The section from the starting point of the tunnel to CH517+000 is -3‰, and the section from CH517+000 to the end of the tunnel is 5.8‰. The maximum buried depth is 74m and the minimum buried depth is 9m. The section CH517+160-CH517+260 is a shallow buried section of the tunnel body with a shallowest buried depth of 10.5m. The tunnel is designed as a composite lining structure with a standard line spacing of 4.2m. The entrance portal is an end wall type and the exit portal is a bevel type.

[0036] The geological lithology of the tunnel is mainly medium-weathered granite, and the rock mass is relatively broken. The surface vegetation of the Ema Tunnel is well developed, all of which are mature tree forests. The entrance is located under the high-voltage line corridor, and the high-voltage line needs to be relocated; the exit has convenient transportation conditions, close to the E8 Expressway and adjacent to the existing local roads.

[0037] like Figure 1 FIG. 1 is a flow chart of a method for monitoring equipment for tunnel excavation using a three-arm drilling rig according to the present invention.

[0038] A method for monitoring equipment for tunnel excavation using a three-arm drilling trolley, characterized in that the method comprises the following steps:

[0039] S101, installation of acquisition equipment;

[0040] The installation of the collection equipment includes installing the recording collection equipment at a fixed position of the three-arm drilling rig to collect sound data during the operation of the three-arm drilling rig;

[0041] In this embodiment, a recording and acquisition device is installed at a fixed position at the three-arm drilling rig tunnel excavation construction site. The acquisition device is placed at a suitable distance from the drilling rig so that the sound of the device in operation can be clearly recorded. The acquisition device selects a recording device with anti-noise and high sampling rate functions to ensure that the recorded sound data is of high quality and can clearly reflect the operating status of the device.

[0042] S102, data collection;

[0043] The data collection includes using the recording collection device to record a large number of sound samples, totaling n, at a certain sampling frequency and recording duration during the tunnel excavation of the three-arm rock drilling rig, denoted as S1~S n The sound samples include both normal use samples of the instrument and equipment and abnormal use samples of the instrument and equipment. At the same time, it is required that the proportion of abnormal working samples of the instrument and equipment in the total sound samples must be ≥10%.

[0044] In this embodiment, when the drilling rig is performing tunnel excavation operations, the sound data of the equipment is recorded using a recording acquisition device according to the set sampling frequency (e.g., 44.1kHz) and recording duration (e.g., 1 minute per recording). In order to ensure the diversity of samples, the recorded data should cover the normal working state and abnormal working state of the equipment. A total of 10,000 sound samples were collected, including 1,000 samples of abnormal working state, and abnormal sound samples accounted for 10% of the total number of samples.

[0045] S103, data preprocessing;

[0046] The data preprocessing includes processing the n sound samples S1 to S n Performing denoising processing, wherein the denoising processing adopts a median filtering method;

[0047] Furthermore, in the above S103, the steps of the median filter algorithm are: a) defining the window size, b) taking the median value of the surrounding samples of each sample point as a new sample value, and repeating this process until the entire signal is processed;

[0048] S104, extracting characteristic parameters;

[0049] The feature parameter extraction includes for each sound sample S i , i = 1 ~ n, extract 5 characteristic parameters, namely short-time energy, zero crossing rate, spectrum centroid, spectrum bandwidth and MFCC coefficient, the parameters are marked as Pij, i = 1 ~ n, j = 1 ~ 5;

[0050] Furthermore, in the above S105, the input value Pij of the characteristic parameter is subjected to standardization processing before being input into the recognition model, so that the value ranges of different characteristic parameters are normalized to improve the accuracy of the recognition model;

[0051] Further, in the above S105, the MFCC coefficient extraction step is: a) dividing the sound sample into a plurality of frames, each frame containing a plurality of sample points, b) applying Fourier transform to each frame to obtain a spectrum, c) mapping the spectrum energy to a Mel-scale filter bank, d) calculating the logarithmic energy of each Mel filter output, e) applying discrete cosine transform to the logarithmic energy, and extracting the first plurality of coefficients as MFCC feature parameters;

[0052] In this embodiment, the denoising process uses the median filter method to denoise the 10,000 recorded sound samples. The MFCC coefficients are further extracted from the denoised sound samples, and the steps are: a) divide each sound sample into several frames, each frame contains 25ms of sound data, b) apply Fourier transform to each frame to obtain the spectrum, c) map the spectrum energy to 26 Mel-scale filter groups, d) calculate the logarithmic energy of each Mel filter output, e) apply discrete cosine transform to the logarithmic energy, and extract the first 13 coefficients as MFCC feature parameters.

[0053] In this embodiment, the following five characteristic parameters are extracted for each sound sample Si: short-time energy: measures the magnitude of instantaneous energy in the sound sample; zero crossing rate: indicates the frequency at which the signal passes through the zero axis, reflecting the frequency characteristics of the sound signal; spectrum centroid: reflects the central position of the signal spectrum, representing the "center of gravity" of the sound; spectrum bandwidth: describes the width of the spectrum, indicating the distribution of frequency components in the sound; MFCC coefficient: uses the average and standard deviation of the 13 MFCC coefficients extracted in the previous steps as features.

[0054] S105, establishing an input and output data group;

[0055] The establishment of the input and output data set includes establishing input and output data. For the sound sample Si, the input is the parameter P ij , i = 1 ~ n, j = 1 ~ 5, the output is 0 or 1, when the sound sample Si represents the normal operation of the instrument and equipment, the output is 0, when the sound sample Si represents the abnormal operation of the instrument and equipment, the output is 1, and the final input and output data group is C;

[0056] S106, establishment and application of recognition model;

[0057] The establishment and application of the recognition model includes constructing a support vector machine recognition model for the recognition of sound samples to achieve the purpose of equipment monitoring of tunnel excavation of a three-arm drilling rig;

[0058] The establishment and application of the recognition model, including the steps of constructing the support vector machine recognition model, are as follows: a) constructing a support vector machine model, b) dividing the data set, dividing the input and output data group C into a training set and a test set, with a division ratio of 8:2, the training set is used to train the model, and the test set is used to verify the performance of the model, c) inputting the training set data into the selected model, using the loss function and optimization algorithm to train the model, and gradually adjusting the model parameters to minimize the loss function value and optimize the classification performance of the model, d) model verification and evaluation, after the training is completed, inputting the test set into the model for evaluation, and using the classification accuracy index to evaluate the performance of the model in identifying the normal and abnormal states of the device. If the model performance does not meet the requirements, it can be further optimized by adjusting the model structure or hyperparameters (such as learning rate, number of hidden layer neurons, etc.), and the optimization measures are to increase the number of samples, adjust the model structure, introduce regularization methods, and finally obtain the optimized model;

[0059] Furthermore, in the above S106, the kernel function of the support vector machine recognition model is a radial basis function, and the model is optimized and adjusted by a cross-validation method;

[0060] Furthermore, in the above S106, when training the recognition model, the proportion of abnormal sound samples is weighted, and the weight of the weighted processing is set as follows: the weight of the normal working sample of the instrument is set to 1, and the weight of the abnormal working sample of the instrument is calculated according to formula (1),

[0061]

[0062] Where N 总 is the total number of samples, N 异 is the number of abnormal samples.

[0063] Furthermore, in the above S106, the loss function adopts a weighted loss function, and the weighted loss function is formula (2):

[0064]

[0065] In the formula, ω i is the weight of the i-th sample, y i The actual output result is: is the model predicted probability.

[0066] In this embodiment, the support vector machine (SVM) model is constructed by using Python and scikit-learn library to construct the support vector machine (SVM) model, and the radial basis function (RBF) is selected as the kernel function of the SVM.

[0067] Set partitioning: The input and output data set C is divided into a training set and a test set in a ratio of 8:2. Model training: The training set data is input into the SVM model and trained using a weighted loss function. For abnormal samples, the weight coefficient is calculated according to formula (1). Model validation and evaluation: The model performance is evaluated using the test set, the classification accuracy is calculated, the model hyperparameters are further adjusted through cross-validation, and a regularization method is introduced to prevent overfitting.

[0068] In the above embodiments, the present invention discloses a method for monitoring equipment for tunnel excavation using a three-arm drilling rig; including installation of acquisition equipment, data acquisition, data preprocessing, extraction of characteristic parameters, establishment of input and output data groups, and establishment and application of recognition models; the method for monitoring equipment for tunnel excavation using a three-arm drilling rig proposed in the present invention has the advantages of accurate recognition and high safety, and can be widely used in the field of tunnel excavation.

[0069] The above description is a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring equipment for tunnel excavation using a three-arm drilling rig, characterized in that: The equipment monitoring method for tunnel excavation of a three-arm drilling rig comprises the following steps: 1) Installation of acquisition equipment; The installation of the collection equipment includes installing the recording collection equipment at a fixed position of the three-arm drilling rig to collect sound data during the operation of the three-arm drilling rig; 2) Data collection; The data collection includes using the recording collection device to record a large number of sound samples, totaling n, at a certain sampling frequency and recording duration during the tunnel excavation of the three-arm rock drilling rig, denoted as S1~S n The sound samples include both normal working samples of the instrument and equipment and abnormal working samples of the instrument and equipment, and the proportion of abnormal working samples of the instrument and equipment to the total sound samples must be ≥10%; 3) Data preprocessing; The data preprocessing includes processing the n sound samples S1 to S n Performing denoising processing, wherein the denoising processing adopts a median filtering method; 4) Extracting characteristic parameters; The feature parameter extraction includes for each sound sample S i , i = 1 ~ n, extract 5 characteristic parameters, namely short-time energy, zero crossing rate, spectrum centroid, spectrum bandwidth and MFCC coefficient, the parameter is marked as P ij , i = 1 ~ n, j = 1 ~ 5; 5) Establish input and output data groups; The establishing of the input-output data group includes establishing input-output data. For the sound sample Si, the input is the parameter Pij, i=1-n, j=1-5, and the output is 0 or 1. When the sound sample Si represents that the instrument and equipment are working normally, the output is 0, and when the sound sample Si represents that the instrument and equipment are working abnormally, the output is 1. The input-output data group finally constituted is C; 6) Establishment and application of recognition model; The establishment and application of the recognition model includes constructing a support vector machine recognition model for the recognition of sound samples to achieve the purpose of equipment monitoring of tunnel excavation of a three-arm drilling rig; The establishment and application of the recognition model, including the steps of constructing the support vector machine recognition model, are as follows: a) constructing a support vector machine model, b) data set division, dividing the input and output data group C into a training set and a test set, with a division ratio of 8:2, the training set is used to train the model, and the test set is used to verify the performance of the model, c) inputting the training set data into the selected model, using the loss function and optimization algorithm to train the model, and gradually adjusting the model parameters to minimize the loss function value and optimize the classification performance of the model, d) model verification and evaluation, after the training is completed, inputting the test set into the model for evaluation, and using the classification accuracy index to evaluate the performance of the model in identifying the normal and abnormal states of the device. If the model performance does not meet the requirements, it can be further optimized by adjusting the model structure or hyperparameters (such as learning rate, number of hidden layer neurons, etc.), and the optimization measures are to increase the number of samples, adjust the model structure, introduce regularization methods, and finally obtain the optimized model.

2. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In the above 3), the steps of the median filter algorithm are: a) defining the window size, b) taking the median of the surrounding samples of each sample point as the new sample value, and repeating this process until the entire signal is processed.

3. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In the above 5), the steps of extracting the MFCC coefficients are: a) dividing the sound sample into several frames, each frame containing several sample points, b) applying Fourier transform to each frame to obtain a spectrum, c) mapping the spectrum energy to a Mel-scale filter bank, d) calculating the logarithmic energy of each Mel filter output, and e) applying discrete cosine transform to the logarithmic energy to extract the first several coefficients as MFCC feature parameters.

4. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In the above 5), the input value Pij of the characteristic parameter is standardized before being input into the recognition model, so that the value ranges of different characteristic parameters are normalized to improve the accuracy of the recognition model.

5. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In 6), the kernel function of the support vector machine recognition model is a radial basis function, and the cross-validation method is used to optimize and adjust the model parameters.

6. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In 6), when training the recognition model, the proportion of abnormal sound samples is weighted, and the weight of the weighted processing is set as follows: the weight of the normal working sample of the instrument is set to 1, and the weight of the abnormal working sample of the instrument is calculated according to formula (1). Where N 总 is the total number of samples, N 异 is the number of abnormal samples.

7. The equipment monitoring method for tunnel excavation of a three-arm drilling rig according to claim 1 is characterized in that: In 6), the loss function adopts a weighted loss function, and the weighted loss function is formula (2), In the formula, ω i is the weight of the i-th sample, y i The actual output result is: is the model predicted probability.

Images