On-line monitoring method, system and equipment for abrasion of cutting pick of cantilever type heading machine and medium

Abstract

The invention belongs to the technical field of cutting pick wear monitoring, and aims to solve the problems that the wear state of a cutting pick is difficult to detect and the periodic maintenance cost is increased. The invention provides an on-line monitoring method, system, equipment and medium for wear of a cutting pick of a cantilever type heading machine. The method comprises the following steps: acquiring a pressure signal of a cutting oil cylinder, a vibration signal of a cutting speed reducer and a temperature signal of a cutting head in real time; performing feature extraction on the sensor data, and determining a feature weight value based on correlation among the sensors; analyzing the abrasion condition of the cutting tooth based on the extracted features and the correlation among the sensors, and obtaining an initial cutting tooth abrasion state; correcting the features based on a double-layer feature extraction model, and updating to obtain a corrected cutting pick wear state; and monitoring the wear condition of the cutting pick based on the corrected wear state of the cutting pick. According to the cutting pick on-line monitoring technology based on multi-index information fusion, the abrasion condition of the cutting pick can be mastered in real time, and intelligent diagnosis based on the state and maintenance based on the condition are carried out on the cutting pick.

Description

Technical Field

[0001] The invention belongs to the technical field of pick wear monitoring, and in particular relates to a method, system, equipment and medium for online monitoring of pick wear of a cantilever type tunneling machine. Background Art

[0002] As the rock-breaking tool of the cantilever roadheader, the pick is in direct contact with the coal and rock. The working conditions are extremely harsh and complex. It is in a high-stress and strong impact state for a long time and bears complex alternating impact loads. The power consumption accounts for more than 80% of the total machine power. The cutting efficiency, cutting resistance, cutting specific energy consumption, cutting stability and other performance indicators of the roadheader are closely related to the pick. During the operation of the roadheader, the downtime caused by the failure of the pick accounts for 20% of the total operating time, and the economic losses caused account for 30% of the total economic losses. It can be seen that the reliability, safety and economy of the operation of the pick directly determine the cutting performance, service life and operating cost of the roadheader, which in turn affects the tunneling efficiency and economic benefits of coal enterprises.

[0003] The pick is in direct contact with the coal rock during the cutting process. Since it is wrapped by the coal rock, it is impossible to directly obtain the wear status of the pick by visual or other sensing means; at the same time, the size of the pick is small, and it is impossible to install a sensor device inside it. Therefore, there is no mature online monitoring technology for the pick at this stage. At present, the wear status of the pick is detected by manual visual inspection, and the maintenance method adopts post-failure maintenance or periodic maintenance. The post-failure maintenance method is a passive and unplanned maintenance activity, that is, when the pick system fails, the failed pick is replaced. Since the time of failure cannot be predicted and there is a lack of preparation before maintenance, the tunnel boring machine often has a long downtime and serious system failures. Periodic maintenance is an active and planned maintenance activity, that is, the pick is regularly inspected and the failed pick is replaced according to the wear condition of the pick. However, since the maintenance cycle is determined by manual experience, maintenance activities are often too frequent, resulting in increased maintenance costs and waste of maintenance resources. Summary of the invention

[0004] In order to solve at least one of the above-mentioned technical problems existing in the prior art, the present invention provides a method, system, equipment and medium for online monitoring of pick wear of a cantilevered tunnel boring machine.

[0005] The present invention is implemented by the following technical solution: a cantilever type tunnel boring machine pick wear online monitoring method, comprising the following steps:

[0006] S1: real-time collection of sensor data that affect the wear state of the cutting teeth, including but not limited to the pressure signal of the cutting cylinder, the vibration signal of the cutting reducer and the temperature signal of the cutting head;

[0007] S2: extracting features from the sensor data and determining feature weight values ​​based on the correlation between the sensors;

[0008] S3: Analyze the wear of the pick based on the extracted features and the correlation between the sensors to obtain the initial wear state of the pick;

[0009] S4: Based on the double-layer feature extraction model, the above features are corrected to obtain a corrected pick wear state; and based on the corrected pick wear state, the wear condition of the pick is monitored.

[0010] Preferably, in S2, the extracted features include:

[0011] The first feature f1, the second feature f2 and the third feature f3, and the expressions of the above features are respectively:

[0012]

[0013] Where P is the pressure sensor data, ε1 is the preset pressure value, and ε1 is greater than 0, k1 is the pressure adjustment parameter, and tanh is a hyperbolic tangent function;

[0014]

[0015] In the formula, A is the acceleration sensor data, ε2 is the acceleration preset value, and ε2 is greater than 0, k2 is the acceleration adjustment parameter, and sigmoid is an S-type function;

[0016]

[0017] Where T is the temperature sensor data, ε3 is the temperature preset value, and ε3 is greater than 0, k3 is the temperature adjustment parameter, and relu is a rectified linear unit function.

[0018] Preferably, in S2, the feature weight values ​​obtained based on the correlation between the sensors include:

[0019] The first feature weight value w1, the second feature weight value w2 and the third feature weight value w3, and the expressions of the above feature weight values ​​are respectively:

[0020]

[0021] Where corr(X,Y) represents the correlation between sensor X and sensor Y, P is the pressure sensor, T is the temperature sensor, and A is the acceleration sensor.

[0022] Preferably, in S3, the formula for the initial pick wear state is:

[0023] Pick wear state = (w1*f1+w2*f2+w3*f3) / (w1+w2+w3)

[0024] Preferably, in S4, the step of correcting the feature includes:

[0025] The first feature f1, the second feature f2 and the third feature f3 are set as an input sequence;

[0026] A two-layer feature extraction model is used to extract features from the input sequence and output the corrected pick wear state;

[0027] The corrected pick wear state formula is:

[0028]

[0029] Where y1, y2, and y3 represent the pick wear state sequence calculated by the double-layer feature extraction model.

[0030] The present invention also provides an online monitoring system for pick wear of a cantilever type tunnel boring machine, comprising a sensor module, a data acquisition module, a data analysis module and a monitoring module;

[0031] The sensor module includes a pressure sensor, an acceleration sensor and a temperature sensor; the pressure sensor is installed on the oil ports of the cutting lifting and cutting rotating joints of the multi-way reversing valve of the hydraulic system of the cantilever roadheader to indirectly monitor the force of the cutting teeth; the acceleration sensor is installed on the cutting reducer of the cantilever roadheader to monitor the vibration of the cutting teeth; the temperature sensor is installed inside the cutting head of the cantilever roadheader to monitor the temperature changes of the cutting teeth;

[0032] The data acquisition module is used to receive the data sent by the sensor module and transmit it to the data analysis module;

[0033] The data analysis module is used to extract features from sensor data, determine the correlation between sensors, and analyze the wear of the pick based on the sensor data and the correlation between sensors to obtain the initial wear state of the pick. At the same time, the extracted features are optimized through a double-layer feature extraction model to update the corrected wear state of the pick.

[0034] The monitoring module is used to obtain the pick status information sent by the data analysis module in real time and monitor the wear of the pick.

[0035] Preferably, the data analysis module includes a preprocessing module, an extraction module and a state feature determination module;

[0036] The preprocessing module is used to preprocess the sensor data to obtain intermediate data;

[0037] The extraction module is used for extracting state features for describing the wear state of the pick from the intermediate data;

[0038] The state characteristic determination module is used to determine the pressure sensor characteristics, the acceleration sensor characteristics and the temperature sensor characteristics, and based on the above characteristics, comprehensively evaluate the wear condition of the pick, and determine the state information of the pick.

[0039] Preferably, the input sequence X={x1,x2,x3} of the double-layer feature extraction model is x1, x2, x3 are the first feature f1, the second feature f2 and the third feature f3 respectively;

[0040] The calculation formula for feature extraction of the double-layer feature extraction model is:

[0041] h 1,t ,c 1,t =LSTM1(x t ,h 1,t-1 ,c 1,t-1 )

[0042] h 2,t ,c 2,t =LSTM2(h 1,t ,h 2,t-1 ,c 2,t-1 )

[0043] In the formula, h 1,t and c 1,t They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t, respectively. 2,t and c 2,t They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t; h 1,t-1 and c 1,t-1 They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t-1; h 2,t-1 and c 2,t-1 They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t-1; x t represents the input sequence X={x1,x2,x3} of the first layer of the two-layer feature extraction model in time step t; LSTM1 represents the first layer of the long short-term memory network model; LSTM2 represents the second layer of the long short-term memory network model;

[0044] The output of the two-layer feature extraction model is:

[0045] y t =g(W hy h 2,t +b y )

[0046] In the formula, g is the activation function, Why and b y is the weight matrix and bias term of the output layer, y t As corrected pick wear condition.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The pick online monitoring technology based on multi-index information fusion of the present invention can grasp the wear condition of the pick in real time, perform state-based intelligent diagnosis and condition-based maintenance on the pick, ensure cutting performance, and ensure safe, efficient and economical operation of the tunnel boring machine.

[0049] Through the sensor module and data acquisition module, the system monitors the stress, vibration and temperature changes of the pick in real time, realizing real-time monitoring of the wear status of the pick. The data analysis module accurately evaluates the wear of the pick by analyzing the sensor data, and provides precise information on the status of the pick, which helps to take maintenance measures in time and extend the service life of the cutting head of the tunnel boring machine. The monitoring module obtains the status information of the pick in real time, discovers the wear of the pick in time, avoids safety hazards caused by excessive wear of the pick, and improves the operation safety of the tunnel boring machine.

[0050] By using a two-layer feature extraction model, the complex relationship between the input features can be better captured, and more representative feature representations can be extracted to help improve the prediction accuracy of the pick wear state. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] 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 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 work.

[0052] Figure 1 A schematic diagram of a module of a cantilever type tunnel boring machine pick wear monitoring system provided by an embodiment of the present invention;

[0053] Figure 2 A schematic flow chart of a method for monitoring pick wear of a cantilevered tunnel boring machine provided in an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of sensor installation positions in a cantilever type tunnel boring machine pick wear monitoring system provided by an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of the installation position of a pressure sensor in a cantilever tunneling machine pick wear monitoring system provided by an embodiment of the present invention;

[0056] Figure 5A schematic diagram of the installation position of the acceleration sensor in the pick wear monitoring system of the cantilever type tunnel boring machine provided by an embodiment of the present invention;

[0057] Figure 6 A schematic diagram of the installation position of the temperature sensor in the pick wear monitoring system for a cantilever tunneling machine provided in an embodiment of the present invention.

[0058] In the figure: 1-cutting lifting joint and cutting rotary joint; 2-cutting reducer; 3-cutting head; 4-pressure sensor; 5-acceleration sensor; 6-temperature sensor. DETAILED DESCRIPTION

[0059] In conjunction with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. 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 implementation methods obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size should fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention. It should be noted that in this specification, relational terms such as first and second are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0061] The present invention provides an embodiment:

[0062] like Figures 1 to 6 As shown, an online monitoring system for pick wear of a cantilever type tunnel boring machine includes a sensor module, a data acquisition module, a data analysis module and a monitoring module;

[0063] The sensor module includes a pressure sensor, an acceleration sensor and a temperature sensor; the pressure sensor is installed on the oil ports of the cutting lifting and cutting rotating joints of the multi-way reversing valve of the hydraulic system of the cantilever roadheader to indirectly monitor the force of the cutting teeth; the acceleration sensor is installed on the cutting reducer of the cantilever roadheader to monitor the vibration of the cutting teeth; the temperature sensor is installed inside the cutting head of the cantilever roadheader to monitor the temperature changes of the cutting teeth;

[0064] The data acquisition module is used to receive the data sent by the sensor module and transmit it to the data analysis module;

[0065] The data analysis module is used to extract features from sensor data, determine the correlation between sensors, and analyze the wear of the pick based on the sensor data and the correlation between sensors to obtain the initial wear state of the pick. At the same time, the extracted features are optimized through a double-layer feature extraction model to update the corrected wear state of the pick.

[0066] The monitoring module is used to obtain the pick status information sent by the data analysis module in real time and monitor the wear of the pick; the monitoring module can provide real-time pick status information to the equipment operator through display screens, alarms, remote communications, etc., so that maintenance measures can be taken in time.

[0067] In the past, manual visual inspection of the pick’s appearance could be used to directly determine whether the pick had failed; the failure of the pick was divided into soft failure due to wear and hard failure due to impact. When the height of the pick after wear reaches the soft failure threshold, it is a soft failure. At this time, the height of the pick can be measured with a vernier caliper to determine whether it has failed; when the pick is chipped or broken, it is a hard failure. At this time, the appearance of the pick can be observed to determine whether it has failed. By replacing severely worn picks in a timely manner, the normal operation of the tunnel boring machine can be guaranteed and the service life of the tunnel boring machine cutting head can be extended. Since the pick is wrapped in coal and rock during operation, it is impossible to directly monitor the wear amount by visual and other sensing means. Therefore, the physical quantity that indirectly reflects the wear state of the pick is used to fuse multi-index information, so as to comprehensively evaluate the wear state of the pick, which has become a key technology for online monitoring of the pick.

[0068] Among them, the pressure sensor can be installed on the oil port of the cutting lifting and cutting rotary joint of the multi-way reversing valve of the hydraulic system of the cantilever roadheader to indirectly monitor the force of the pick. During the cutting process, the wear degree of the pick can be inferred by monitoring the pressure changes of the cutting lifting cylinder and the cutting rotary cylinder. The greater the pressure of the cutting cylinder, the harder the coal and rock cut by the pick, and the faster the pick wears. As the wear degree of the pick increases, the pressure of the cutting cylinder will also change accordingly.

[0069] The acceleration sensor can be installed on the cantilever roadheader cutting reducer to monitor the vibration of the pick. During the cutting process, the wear degree of the pick can be inferred by monitoring the vibration changes of the cutting reducer. The larger the vibration amplitude, the harder the coal rock cut by the pick, and the faster the pick wears. As the wear degree of the pick increases, the vibration of the cutting reducer will also change accordingly.

[0070] Temperature sensors can be installed inside the cantilevered roadheader cutting head to monitor the temperature changes of the cutting teeth. During the cutting process, by monitoring the temperature changes of the cutting head, the degree of wear of the cutting teeth can be inferred. The higher the temperature, the harder the coal rock cut by the cutting teeth, and the faster the cutting teeth wear. As the degree of wear of the cutting teeth increases, the temperature of the cutting head will also change accordingly.

[0071] The pressure sensor, temperature sensor and acceleration sensor should be installed at a position on the tunnel boring machine where the feedback signal of the pick can be collected. The specific installation position can be determined according to the actual situation. The embodiment of the present invention does not make specific limitations on this. Ultimately, the wear state of the pick is indirectly comprehensively evaluated through the signals fed back by the above three sensors and the deep learning technology based on multi-indicator information fusion.

[0072] The data analysis module includes a preprocessing module, an extraction module and a state feature determination module;

[0073] The preprocessing module is used to preprocess the sensor data to obtain intermediate data; the preprocessing module is used to preprocess the sensor data to improve the data quality and accuracy. For example, filtering, denoising, data correction and other operations can be performed to remove noise and eliminate interference, making subsequent data analysis more reliable and accurate.

[0074] The extraction module is used to extract state features for describing the wear state of the pick from the preprocessed intermediate data. These features can be statistical features, frequency domain features, time domain features, etc., and can be extracted through signal processing methods, time series analysis, spectrum analysis and other technologies. For example, the average value, maximum value, minimum value of the pick force, peak value of the vibration frequency and other features can be extracted.

[0075] Take the pressure sensor data processing as an example: when the pick is severely worn, the pressure sensor data may show a larger fluctuation amplitude or change rate. By processing the data, this feature can be better captured, so as to timely detect and predict the wear of the pick.

[0076] Assuming that the common linear normalization method is used, that is, the pressure sensor data P is scaled to the range of 0 to 1, the calculation formula is:

[0077]

[0078] Substitute the above data into the formula for calculation:

[0079]

[0080] Through ordinary linear normalization processing, the normalized pressure sensor data obtained is 0.6.

[0081] Now assume that no normalization is performed and the original pressure sensor data is used directly for analysis. In this case, if the range of the data is large, it may make it difficult to compare and analyze the data because the numerical differences between different sensor data are large, making it difficult to intuitively compare the changing trends and characteristics of the data.

[0082] Therefore, by performing nonlinear normalization processing on the pressure sensor data, such as logarithmic function transformation, the data can be mapped to a more appropriate range, better reflecting the changing characteristics and trends of the data, and improving the accuracy and reliability of data processing. By combining nonlinear transformation and ordinary normalization processing, the following improvements are achieved in the field of cantilever roadheader pick wear monitoring technology:

[0083] Better capture of data features: Nonlinear transformation can better capture the changing trends and fluctuation characteristics of data, and ordinary normalization processing can scale data to an appropriate range, and combined to more comprehensively analyze data features. Improve data processing accuracy: By combining the two processing methods, it can more accurately reflect the changes in data and improve the accuracy and reliability of data processing. More comprehensive monitoring: Combining nonlinear and linear normalization processing can more comprehensively monitor the wear status of the cantilever roadheader pick, improving the comprehensiveness and effectiveness of wear monitoring.

[0084] The pressure sensor data is extracted from the intermediate data, and the pressure sensor data is nonlinearly transformed to capture the change trend and fluctuation trend of the pressure sensor data, and the pressure sensor feature is obtained, wherein the calculation expression of the pressure sensor feature is as follows:

[0085]

[0086] Where P represents the pressure sensor data, P normalized represents the normalized pressure sensor characteristics, P min Indicates the minimum value of the pressure sensor data, P max Indicates the maximum value of the pressure sensor data.

[0087] The pressure sensor data is extracted from the intermediate data, and a nonlinear transformation is performed on the pressure sensor data to capture the change trend and fluctuation trend of the pressure sensor data, and the pressure sensor feature is obtained. The calculation expression of the pressure sensor feature is as follows:

[0088]

[0089] Where P represents the pressure sensor data, P normalized represents the normalized pressure sensor characteristics, P min Indicates the minimum value of the pressure sensor data, P maxIndicates the maximum value of the pressure sensor data;

[0090] Assume the following data:

[0091] P=80、P min =50, P max =100;

[0092] Substituting these values ​​into the formula for calculation:

[0093] Therefore, after calculation, the normalized pressure sensor data P is obtained. normalized ≈1.69.

[0094] Such processing of pressure sensor data has the following benefits:

[0095] Nonlinear transformation: The nonlinear transformation of logarithmic functions can better capture the changing trends and fluctuation characteristics of data, improve the expressiveness of data processing and the accuracy of feature extraction. Strong adaptability: By introducing data processing formulas, it is applicable to different data distributions and characteristics, improving the flexibility and adaptability of data processing. Improved accuracy: It can more accurately capture the data characteristics and changes of pressure sensors in the wear monitoring of cantilevered roadheaders, and improve the monitoring effect and prediction accuracy.

[0096] The state feature determination module is used to determine the pressure sensor features, acceleration sensor features, and temperature sensor features, and comprehensively evaluate the wear of the pick based on the above features to determine the state information of the pick. For example, the force condition of the pick is determined by analyzing the change trend, amplitude and other characteristics of the pressure sensor data, the vibration condition of the pick is evaluated by analyzing the vibration frequency, peak value and other characteristics of the acceleration sensor data, and the temperature change of the pick is monitored by analyzing the change trend, peak value and other characteristics of the temperature sensor data. The data analysis module can use machine learning algorithms, statistical analysis methods, etc. to perform data processing and wear state prediction.

[0097] The sensor data is processed by the preprocessing module, the extraction module extracts the characteristics of the pick wear state from the intermediate data, and the state characteristic determination module determines the wear state information of the pick based on these characteristics. This can improve the accuracy and reliability of the system's assessment of the pick wear state, help users better understand the use of the pick, and make the most reasonable maintenance decisions.

[0098] Determine the state information of the pick according to the characteristics of the pressure sensor, the acceleration sensor, and the temperature sensor: Based on the determined sensor characteristics, comprehensively evaluate the wear state of the pick. By establishing a comprehensive evaluation model, the characteristics of different sensors are weighted and combined to obtain information such as the overall wear degree and remaining life of the pick. For example, machine learning algorithms, fuzzy logic, and other methods can be used to achieve comprehensive evaluation.

[0099] Through the above steps, the state feature determination module can determine the pressure sensor features, acceleration sensor features and temperature sensor features based on the state features, and comprehensively evaluate the wear state information of the pick according to these features. Such a state feature determination module can provide accurate and reliable pick state information, helping users to understand the use of the pick in a timely manner and make maintenance and replacement decisions.

[0100] A second aspect of an embodiment of the present invention provides an online monitoring method for pick wear of a cantilevered tunnel boring machine, comprising the following steps:

[0101] S1: real-time collection of sensor data that affect the wear state of the cutting teeth, including but not limited to the pressure signal of the cutting cylinder, the vibration signal of the cutting reducer and the temperature signal of the cutting head;

[0102] S2: extracting features from the sensor data and determining feature weight values ​​based on the correlation between the sensors;

[0103] S3: Analyze the wear of the pick based on the extracted features and the correlation between the sensors to obtain the initial wear state of the pick;

[0104] S4: Based on the double-layer feature extraction model, the above features are corrected to obtain a corrected pick wear state; and based on the corrected pick wear state, the wear condition of the pick is monitored.

[0105] In S2, the extracted features include:

[0106] The first feature f1, the second feature f2 and the third feature f3, and the expressions of the above features are respectively:

[0107]

[0108] Where P is the pressure sensor data, ε1 is the preset pressure value, and ε1 is greater than 0, k1 is the pressure adjustment parameter, and tanh is a hyperbolic tangent function;

[0109]

[0110] In the formula, A is the acceleration sensor data, ε2 is the acceleration preset value, and ε2 is greater than 0, k2 is the acceleration adjustment parameter, and sigmoid is an S-type function;

[0111]

[0112] Where T is the temperature sensor data, ε3 is the temperature preset value, and ε3 is greater than 0, k3 is the temperature adjustment parameter, and relu is a rectified linear unit function.

[0113] Design principle:

[0114] For the first feature f1, a logarithmic operation is used to convert the variation range of the pressure sensor data into a more appropriate scale, and the data with a large variation range is compressed by the tanh function to prevent an excessively large variation range from having an excessive impact on the calculation of the wear state.

[0115] For the second feature f2, the square root operation is used to convert the variation range of the acceleration sensor data into a more appropriate scale, and the sigmoid function is used to compress the data with a larger variation range, so that the features with a larger variation range do not occupy too much weight in the calculation.

[0116] For the third feature f3, an exponential operation is used to convert the variation range of the temperature sensor data into a more appropriate scale, and the data with a larger variation range is compressed through the relu function, so that the features with a larger variation range do not occupy too much weight in the calculation.

[0117] Through the above design principle, the different characteristics of the sensor data can be more accurately considered and given corresponding weights according to the importance and contribution of the characteristics, so as to calculate a more accurate wear state of the pick.

[0118] The expressions of the first feature weight value w1, the second feature weight value w2, and the third feature weight value w3 and the association between the sensors may be selected and determined based on the following design principles and strategies:

[0119] Based on feature contribution: The weight value is determined according to the contribution of different sensors to the wear state. If the data of a certain sensor plays a more important role in predicting the wear state, then the corresponding feature weight value can be increased accordingly.

[0120] Based on the correlation between features: Consider the inherent correlation between sensors and reflect the correlation between different sensors by setting weight values. If the data of different sensors have a high correlation in the wear state prediction, then the corresponding feature weight values ​​can be increased accordingly.

[0121] The specific expressions of the first characteristic weight value w1, the second characteristic weight value w2 and the third characteristic weight value w3 and the association between the sensors can be determined according to the above design principles and strategies. These expressions can be adjusted and optimized according to the actual application scenarios and the results of data analysis to achieve more accurate and effective wear state prediction.

[0122] And the expressions of the above feature weight values ​​are:

[0123]

[0124] Wherein, corr(X,Y) represents the correlation between sensor X and sensor Y, which reflects the intrinsic association between sensors by calculating the correlation between each sensor; P is a pressure sensor, T is a temperature sensor, and A is an acceleration sensor.

[0125] Design reason:

[0126] The design reason of the first characteristic weight value w1 is: considering the correlation of the pressure sensor P to the temperature sensor T and the acceleration sensor A, as well as the correlation between the temperature sensor T and the acceleration sensor A. By adding and normalizing the two correlations, the weight value corresponding to the sensor with higher correlation is increased to emphasize its contribution to the wear state.

[0127] The design reason of the second characteristic weight value w2 is: considering the correlation of the acceleration sensor A to the pressure sensor P and the temperature sensor T, and the correlation between the temperature sensor T and the pressure sensor P. By adding and normalizing the two correlations, the weight value corresponding to the sensor with higher correlation is increased to emphasize its contribution to the wear state.

[0128] The third characteristic weight value w3 is designed for the following reasons: considering the correlation of the temperature sensor T to the pressure sensor P and the acceleration sensor A, as well as the correlation between the pressure sensor P and the acceleration sensor A. By adding and normalizing the two correlations, the weight value corresponding to the sensor with higher correlation is increased to emphasize its contribution to the wear state.

[0129] The specific expressions of the first feature weight value w1, the second feature weight value w2 and the third feature weight value w3 can be adjusted and determined according to the actual application scenario and the results of data analysis.

[0130] In S3, the formula for the initial pick wear state is:

[0131] Pick wear state = (w1*f1+w2*f2+w3*f3) / (w1+w2+w3)

[0132] In order to improve the accuracy of the pick wear state, the first feature f1, the second feature f2 and the third feature f3 are corrected based on the double-layer feature extraction model to update the pick wear state formula. The step of correcting the features in S4 includes:

[0133] The first feature f1, the second feature f2 and the third feature f3 are set as the input sequence X = {x1, x2, x3};

[0134] A two-layer feature extraction model is used to extract features from the input sequence. The calculation formula for feature extraction by the two-layer feature extraction model is:

[0135] h 1,t ,c 1,t =LSTM1(x t ,h 1,t-1 ,c 1,t-1 )

[0136] h 2,t ,c 2,t =LSTM2(h 1,t ,h 2,t-1 ,c 2,t-1 )

[0137] In the formula, h 1,t and c 1,t They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t, respectively. 2,t and c 2,t They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t; h 1,t-1 and c 1,t-1 They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t-1; h 2,t-1 and c 2,t-1 They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t-1; x t represents the input sequence X={x1,x2,x3} of the first layer of the two-layer feature extraction model in time step t; LSTM1 represents the first layer of the long short-term memory network model; LSTM2 represents the second layer of the long short-term memory network model;

[0138] t-1 represents the previous time step, and LSTM represents the long short-term memory network model, which is used to process sequence data, such as time series analysis or natural language processing tasks. LSTM is a special recurrent neural network (RNN) designed to avoid long-term dependency problems and control the flow of information through a gating mechanism.

[0139] And output the corrected pick wear status; y t =g(Why h 2,t +b y )

[0140] In the formula, g is the activation function, W hy and b y is the weight matrix and bias term of the output layer, y t As corrected pick wear condition.

[0141] The corrected pick wear state formula is:

[0142]

[0143] Where y1, y2, and y3 represent the pick wear state sequence calculated by the double-layer feature extraction model.

[0144] Through the above steps, the preset feature extraction model and the formula of the pick wear state can be used to calculate the wear state of the pick, so as to improve the prediction accuracy and reliability of the wear state of the pick. The above method can effectively combine the feature extraction model and the wear state calculation formula to better understand and predict the wear state of the pick.

[0145] Combining the preset feature extraction model and the pick wear state formula brings the following benefits: More accurate prediction: By using a two-layer feature extraction model, the complex relationship between the input features can be better captured, and a more representative feature representation can be extracted. This can help improve the accuracy of the prediction of the pick wear state. More flexible model: Compared with simple linear formulas, the two-layer feature extraction model has stronger expressive power and flexibility. By performing multi-level abstraction and extraction of features, it can better adapt to different types of data and feature changes. Better generalization ability: The two-layer feature extraction model can learn more abstract and advanced representations of input features, thereby improving the generalization ability of the model, allowing the model to perform better on unseen data.

[0146] A specific example is to predict the wear state of the pick of a roadheader. Using the acceleration sensor data and the first feature f1, the second feature f2 and the third feature f3 as the input sequence, the hidden state h representing the wear state of the pick is extracted through the double-layer feature extraction model. 2,t Then, the hidden state h 2,t Substitute into the calculation formula to obtain the final wear state of the pick.

[0147] In practical applications, a large amount of acceleration sensor data and feature data can be collected to train a double-layer feature extraction model, and the model can be used to predict the wear state of the roadheader pick. By combining the feature extraction model and the wear state formula, the wear state of the roadheader pick can be predicted more accurately, improving the efficiency and accuracy of equipment maintenance, for example:

[0148] In this formula, a two-layer LSTM (Long Short-Term Memory Network) can be used as a feature extraction model to extract the feature representation of the input sequence. The following is the specific expression of each parameter and the value method in the example:

[0149] h 1,t and c 1,t : represents the hidden state and cell state of the first layer of the two-layer feature extraction model at time step t. In LSTM, the hidden state h and cell state c are obtained through operations inside the LSTM unit, and are memory units used to transmit information and control information flow, respectively.

[0150] h 2,t and c 2,t : represents the hidden state and cell state of the second layer of the two-layer feature extraction model at time step t. The hidden state and cell state of the second layer are calculated based on the hidden state and cell state of the first layer to learn more advanced feature representations.

[0151] y t : Represents the output result at time step t, through the weight matrix W of the output layer hy and the bias term b y For the second hidden state h 2,t It is obtained by performing linear transformation and activation function g.

[0152] In an embodiment of the present invention, hidden state and cell state are important concepts in the LSTM (Long Short-Term Memory) model, which are used to capture long-term dependencies and memory characteristics in time series data. In sensor data feature extraction, hidden state and cell state can be understood as the state of encoding and memorizing features and relationships in sensor data. Specifically, hidden state h can be regarded as the extracted features of sensor data at the current time step, which contains an abstract representation of the current data and reflects the important information contained in the current data; while cell state c is the memory of long-term dependencies between data, which controls the input and forgetting of information through the gating structure of the LSTM model, thereby capturing long-term dependencies between data. Therefore, in sensor data feature extraction, the use of hidden state and cell state can better capture and encode features and relationships in sensor data, helping the model to better understand and predict the state and changes of sensor data.

[0153] In the specific example, you can select the following parameters and value methods:

[0154] The hidden state size of LSTM1 is d1, and the cell state size is d1, that is, h 1,t and c 1,t The dimension of is d1.

[0155] The hidden state size of LSTM2 is d2, and the cell state size is d2, that is, h 2,t and c 2,t The dimension of is d2.

[0156] The activation function of the output layer can be selected from ReLU function or Sigmoid function.

[0157] The weight matrix W of the output layer hy The dimension is d2×m, where m is the dimension of the output and can be set according to the specific task.

[0158] The bias term b of the output layer y The dimension is m.

[0159] In the field of pick condition monitoring technology, it is possible to select a suitable activation function g and the weight matrix W of the output layer. hy and the bias term b y To build a model suitable for predicting the state of the pick. The specific expression is as follows: Activation function g: In the monitoring of the state of the pick, common activation functions include Sigmoid function and Softmax function. Both functions can map the output value to a range and are suitable for binary or multi-classification tasks. Sigmoid function: Map the output value to (0,1), which is suitable for binary classification tasks. Softmax function: The output values ​​are normalized and suitable for multi-classification tasks. The weight matrix W of the output layer hy and the bias term b y : In the condition monitoring of pick teeth, the weight matrix and bias term of the output layer can be defined as follows: Weight matrix W hy The dimension of is d2×1, where d2 is the dimension of the second layer hidden state, and the dimension of the output is 1, because it may be a binary classification task (such as predicting whether the pick is worn). y The dimension is 1, which is used to adjust the offset of the output result.

[0160] Based on the above, an output layer formula suitable for cutting tooth condition monitoring can be constructed as follows:

[0161] y t =g(W hy h 2,t +by )

[0162] Among them, g can choose Sigmoid function or Softmax function, W hy is a d2×1 weight matrix, b y is a bias term. Through this output layer formula, the second layer hidden state h 2,t It is mapped into a range and used to predict the wear state of the pick.

[0163] In practical applications, the parameters in the model can be learned through training data, including the weights and bias terms in LSTM, as well as the weight matrix and bias terms in the output layer. The model parameters are updated through the back propagation algorithm and loss function, so that the model can better fit the data and make predictions. Through the above design and parameter selection, a two-layer feature extraction model can be constructed, and combined with the output layer to predict the wear state of the pick.

[0164] An embodiment of the present invention further provides a device, comprising: a processor; a memory for storing instructions executable by the processor;

[0165] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0166] The embodiment of the present invention also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the aforementioned method when executed by a processor. The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0167] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A cantilever tunnel boring machine pick wear online monitoring method, characterized in that: The following steps are involved: S1: real-time collection of sensor data that affect the wear state of the cutting teeth, including but not limited to the pressure signal of the cutting cylinder, the vibration signal of the cutting reducer and the temperature signal of the cutting head; S2: extracting features from the sensor data and determining feature weight values ​​based on the correlation between the sensors; S3: Analyze the wear of the pick based on the extracted features and the correlation between the sensors to obtain the initial wear state of the pick; S4: Based on the double-layer feature extraction model, the above features are corrected to obtain a corrected pick wear state; and based on the corrected pick wear state, the wear condition of the pick is monitored.

2. The method for online monitoring of pick wear of a cantilever type tunnel boring machine according to claim 1, characterized in that: In S2, the extracted features include: The first feature f1, the second feature f2 and the third feature f3, and the expressions of the above features are respectively: Where P is the pressure sensor data, ε1 is the preset pressure value, and ε1 is greater than 0, k1 is the pressure adjustment parameter, and tanh is a hyperbolic tangent function; In the formula, A is the acceleration sensor data, ε2 is the acceleration preset value, and ε2 is greater than 0, k2 is the acceleration adjustment parameter, Sigmoid is an S-shaped function; Where T is the temperature sensor data, ε3 is the temperature preset value, and ε3 is greater than 0, k3 is the temperature adjustment parameter, and relu is a rectified linear unit function.

3. The on-line monitoring method for pick wear of a cantilever type tunnel boring machine according to claim 2, characterized in that: In S2, the feature weight values ​​obtained based on the correlation between sensors include: The first feature weight value w1, the second feature weight value w2 and the third feature weight value w3, and the expressions of the above feature weight values ​​are respectively: Where corr(X,Y) represents the correlation between sensor X and sensor Y, P is the pressure sensor, T is the temperature sensor, and A is the acceleration sensor.

4. The method for online monitoring of pick wear of a cantilever type tunnel boring machine according to claim 3, characterized in that: In S3, the formula for the initial pick wear state is: Pick wear state = (w1*f1+w2*f2+w3*f3) / (w1+w2+w3).

5. The on-line monitoring method for pick wear of a cantilever type tunnel boring machine according to claim 4, characterized in that: In S4, the step of modifying the feature includes: The first feature f1, the second feature f2 and the third feature f3 are set as an input sequence; A two-layer feature extraction model is used to extract features from the input sequence and output the corrected pick wear state; The corrected pick wear state formula is: Where y1, y2, and y3 represent the pick wear state sequence calculated by the double-layer feature extraction model.

6. An online monitoring system for pick wear of a cantilevered roadheader, used to implement the online monitoring method for pick wear of a cantilevered roadheader as claimed in any one of claims 1 to 5, characterized in that: It includes sensor module, data acquisition module, data analysis module and monitoring module; The sensor module includes a pressure sensor, an acceleration sensor and a temperature sensor; the pressure sensor is installed on the oil ports of the cutting lifting and cutting rotating joints of the multi-way reversing valve of the hydraulic system of the cantilever roadheader to indirectly monitor the force of the cutting teeth; the acceleration sensor is installed on the cutting reducer of the cantilever roadheader to monitor the vibration of the cutting teeth; the temperature sensor is installed inside the cutting head of the cantilever roadheader to monitor the temperature changes of the cutting teeth; The data acquisition module is used to receive the data sent by the sensor module and transmit it to the data analysis module; The data analysis module is used to extract features from sensor data, determine the correlation between sensors, and analyze the wear of the pick based on the sensor data and the correlation between sensors to obtain the initial wear state of the pick. At the same time, the extracted features are optimized through a double-layer feature extraction model to update the corrected wear state of the pick. The monitoring module is used to obtain the pick status information sent by the data analysis module in real time and monitor the wear of the pick.

7. The cantilever type roadheader pick wear online monitoring system according to claim 6, characterized in that: The data analysis module includes a preprocessing module, an extraction module and a state feature determination module; The preprocessing module is used to preprocess the sensor data to obtain intermediate data; The extraction module is used for extracting state features for describing the wear state of the pick from the intermediate data; The state characteristic determination module is used to determine the pressure sensor characteristics, the acceleration sensor characteristics and the temperature sensor characteristics, and based on the above characteristics, comprehensively evaluate the wear condition of the pick, and determine the state information of the pick.

8. The cantilever type roadheader pick wear online monitoring system according to claim 6, characterized in that: The input sequence X = {x1, x2, x3} of the two-layer feature extraction model is x1, x2, x3, respectively, the first feature f1, the second feature f2, and the third feature f3; The calculation formula for feature extraction of the double-layer feature extraction model is: h 1,t ,c 1,t =LSTM1(x t ,h 1,t-1 ,c 1,t-1 ) h 2,t ,c 2,t =LSTM2(h 1,t ,h 2,t-1 ,c 2,t-1 ) In the formula, h 1,t and c 1,t They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t, respectively. 2,t and c 2,t They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t; h 1,t-1 and c 1,t-1 They represent the hidden state and cell state of the first layer of the two-layer feature extraction model in time step t-1; h 2,t-1 and c 2,t-1 They represent the hidden state and cell state of the second layer of the two-layer feature extraction model in time step t-1; x t represents the input sequence X={x1,x2,x3} of the first layer of the two-layer feature extraction model in time step t; LSTM1 represents the first layer of the long short-term memory network model; LSTM2 represents the second layer of the long short-term memory network model; The output of the two-layer feature extraction model is: y t =g(W hy h 2,t +b y ) In the formula, g is the activation function, W hy and b y is the weight matrix and bias term of the output layer, y t As corrected pick wear condition.

9. A device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the cantilever tunneling machine pick wear monitoring method as described in any one of claims 1 to 5.

10. A readable storage medium, characterized in that: A computer program is stored in the readable storage medium, and the computer program includes a program code for controlling a process to execute a process, and the process includes the method for monitoring pick wear of a cantilever tunneling machine as described in any one of claims 1 to 5.

Images