CN115600654A - Convolution LSTM-based shield hob abrasion loss real-time prediction method and system - Google Patents
Convolution LSTM-based shield hob abrasion loss real-time prediction method and system Download PDFInfo
- Publication number
- CN115600654A CN115600654A CN202211189012.4A CN202211189012A CN115600654A CN 115600654 A CN115600654 A CN 115600654A CN 202211189012 A CN202211189012 A CN 202211189012A CN 115600654 A CN115600654 A CN 115600654A
- Authority
- CN
- China
- Prior art keywords
- hob
- abrasion
- shield
- time
- lstm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005299 abrasion Methods 0.000 title claims abstract description 175
- 238000000034 method Methods 0.000 title claims abstract description 73
- 238000013135 deep learning Methods 0.000 claims abstract description 66
- 238000011156 evaluation Methods 0.000 claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 27
- 230000005641 tunneling Effects 0.000 claims description 38
- 238000010276 construction Methods 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 19
- 238000011176 pooling Methods 0.000 claims description 12
- 239000002893 slag Substances 0.000 claims description 9
- 238000009412 basement excavation Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 230000003993 interaction Effects 0.000 claims description 6
- 239000002689 soil Substances 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 4
- 238000009434 installation Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 abstract description 6
- 210000002569 neuron Anatomy 0.000 description 12
- 238000013527 convolutional neural network Methods 0.000 description 10
- 239000011159 matrix material Substances 0.000 description 8
- 230000006870 function Effects 0.000 description 7
- 230000035515 penetration Effects 0.000 description 6
- 230000001186 cumulative effect Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 239000011435 rock Substances 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000006260 foam Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000000691 measurement method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000010720 hydraulic oil Substances 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/08—Construction
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Business, Economics & Management (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Development Economics (AREA)
- Primary Health Care (AREA)
- Excavating Of Shafts Or Tunnels (AREA)
Abstract
The invention provides a method and a system for predicting the abrasion loss of a shield hob in real time based on convolution LSTM, which comprises the following steps: s1, constructing a hob abrasion parameter set; s2, establishing a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter; s3, establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training; and S4, predicting by using the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss. According to the method, the CNN and LSTM methods are adopted to construct the deep learning network to predict the hob abrasion in real time, so that the prediction accuracy of the deep learning network is improved, and the problem of low working efficiency of a direct cabin opening inspection method is solved.
Description
Technical Field
The invention relates to the field of tunnel construction, in particular to a method and a system for predicting shield hob abrasion loss in real time based on convolution LSTM.
Background
In recent years, the shield method is widely applied to construction of underground tunnels in China. The shield machine cuts and excavates the stratum in front through the cutter head, thereby realizing the aim of shield tunnel excavation and tunneling. However, in the forward tunneling process of the shield tunneling machine, the hob inevitably wears, even if the hob is seriously worn in a hard rock stratum, the blocking risk of the shield tunneling machine is greatly increased, the shield tunneling efficiency is reduced, the construction period is further influenced, and the construction cost is increased. The research on the abrasion of the cutter of the earth pressure balance shield based on the torsional energy of the cutter head is shown in the text of the abrasion analysis of the earth pressure balance shield cutter based on the torsional energy of the cutter head published in the university of Shanghai communications journal of Shanghai in 2019, and the important significance of the prediction work of the service life of the hob in the construction process is emphasized. Therefore, it is necessary to predict the tool wear amount to improve the tool changing efficiency.
At present, the abrasion of the hob in the engineering field is mainly judged by adopting an open cabin inspection mode, namely, the hob abrasion loss is measured one by manually entering an excavation cabin, so that the hob needing to be replaced is determined. The method is low in working efficiency, and if the method is in unfavorable geology, a severe tool changing operation environment is faced, so that the tool changing operation risk is greatly increased. Through the search of the prior art documents, the Chinese invention patent with the patent publication number of CN106570275A discloses a TBM hob wear prediction method based on a CAI value, and the service life of a hob is predicted by applying a regression analysis method and taking a hob wear coefficient and a rock CAI value as core parameters. However, the method needs a rock abrasiveness test and a hob ring wear coefficient measuring test, has complicated steps, large workload and poor practicability, and cannot accurately determine the wear state of a hob.
Therefore, a simple, efficient and reliable method for determining the abrasion loss in real time is urgently needed to solve the problem of hob abrasion in the shield tunneling process.
Disclosure of Invention
Aiming at the defects of the existing method, the invention provides a method and a system for predicting the abrasion loss of a shield hob in real time based on convolution LSTM, wherein a CNN and LSTM method is adopted to construct a deep learning network to predict the abrasion loss of the hob in real time. Meanwhile, the influence of the diameter of the hob ring of the hob is further considered, and the accuracy of a legal person for determining the abrasion state of a certain hob is further improved.
The invention provides a method for predicting the wear loss of a shield hob in real time based on convolution LSTM, which comprises the following steps:
s1, establishing a hob abrasion parameter set, wherein the hob abrasion parameter set comprises data samples of the time-varying tunneling parameters influencing hob abrasion in shield construction;
s2, establishing a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter;
s3, establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
and S4, predicting by using the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
Optionally, the evaluation criterion of the wear state of the hob is specifically:
CLI＝W a /(d·t w )
in the formula, CLI is the index of the abrasion state of the hob, and the unit is 1/min; w a The accumulated abrasion loss of the hob is in mm; t is t w The unit of the actual working time corresponding to the accumulated abrasion loss of the hob is min; d is the diameter of the hob ring and the unit is mm.
In a second aspect of the present invention, a system for predicting the wear loss of a shield hob based on a convolution LSTM in real time is provided, which includes:
the hob abrasion parameter set building module is used for building a hob abrasion parameter set, and the hob abrasion parameter set comprises data samples of the time-varying tunneling parameters influencing the abrasion of the hob in shield construction;
the evaluation standard establishing module establishes a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter;
the hob abrasion loss prediction model establishing module is used for establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
and the prediction module is used for predicting by adopting the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
The invention provides a shield hob abrasion loss real-time monitoring terminal, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor is used for executing the shield hob abrasion loss real-time prediction method based on the convolution LSTM when executing the program.
In a fourth aspect of the present invention, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor is configured to execute the method for predicting the wear of a shield hob based on convolution LSTM in real time.
The method for predicting the abrasion loss of the shield hobbing cutter based on the convolution LSTM is used for comprehensively utilizing the CNN and LSTM methods and constructing a deep learning network to predict the abrasion loss of the hobbing cutter in real time aiming at the problems that a tunnel construction direct cabin opening inspection method is low in working efficiency and only the health degree of the whole hobbing cutter of a cutter head is considered in the conventional method, so that the real-time prediction of the abrasion loss of a certain hobbing cutter in a shield tunneling process is realized, and the tunneling efficiency of the shield tunnel construction is effectively improved.
Compared with the prior art, the embodiment of the invention has at least one of the following beneficial effects:
according to the method and the system for predicting the abrasion loss of the shield hob based on the convolution LSTM, the abrasion loss of the hob in the shield tunneling construction process is predicted based on a CNN method and an LSTM method, and particularly, in a convolution LSTM combined deep learning network, the CNN can effectively reduce the correlation among different parameters in a hob abrasion parameter set and improve the model iterative computation efficiency; the LSTM can fully consider historical information contained in time series data in the wear parameter set, and the prediction accuracy of the deep learning network is improved. The two methods are combined, the influence of the tunneling parameter change on the abrasion of the hob in the shield construction is fully considered, the abrasion loss of the single hob can be predicted in real time along with the construction progress, the problem of low working efficiency of a direct cabin opening inspection method is solved, and the defect that the health degree of the whole hob of a cutterhead is only considered in the conventional method is overcome.
The invention establishes the evaluation standard of the abrasion state of the hob, can help field constructors to judge the cutter checking time, reduces the times and time for checking and maintaining the cutter in construction, effectively improves the construction efficiency and reduces the construction cost. The method is high in accuracy, and a simple, convenient, reasonable and efficient new method is provided for determining the abrasion loss of the shield hob.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a flow chart of hob wear prediction according to an embodiment of the present invention;
FIG. 2 is a graph of cumulative wear of a hob as a function of operating time in accordance with one embodiment of the present invention;
FIG. 3 is a diagram of a convolutional LSTM deep learning network structure according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating the predicted results of the amount of wear of the hob according to an embodiment of the present invention.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications can be made by persons skilled in the art without departing from the spirit of the invention. All falling within the scope of the present invention.
Referring to fig. 1, in the preferred embodiment of the present invention, data of an actual project are collected first, and an accumulated wear parameter of a hob and an actual working time corresponding to the accumulated wear parameter are determined; establishing a hob abrasion state evaluation standard, and building a model database; constructing a convolutional LSTM deep learning network hob abrasion real-time prediction model by comprehensively adopting CNN and LSTM methods; and finally, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard in the database into a convolution LSTM deep learning network model for training, thereby predicting the real-time abrasion value of each hob.
Specifically, the embodiment of the invention provides a method for predicting the wear loss of a shield hob based on convolution LSTM in real time, which comprises the following steps:
s1, establishing a hob abrasion parameter set, wherein the hob abrasion parameter set comprises data samples of the time-varying tunneling parameters influencing hob abrasion in shield construction;
in this step, the hob abrasion parameter set includes the following specific parameters: shield machine power system parameters, cutter head system parameters, slag discharge system parameters, slag soil adjusting system parameters and shield tunnel geometric parameters.
S2, establishing a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the diameter of a hob ring;
s3, establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
in a preferred embodiment, after normalization processing is carried out on the hob wear parameter set and the hob wear state evaluation standard, a convolution LSTM deep learning network model is input.
In this step, the convolution LSTM deep learning network model is composed of an input layer, a convolution layer, a pooling layer, an LSTM network layer, a full connection layer and an output layer which are connected in sequence, wherein: the LSTM network layer is composed of LSTM units including input gates, forgetting gates, output gates and memory cells, and learns information of different long periods and short periods in a time sequence.
And S4, predicting by using the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
According to the embodiment, the CNN and LSTM methods are adopted to construct the deep learning network to predict the abrasion of the hobbing cutter in real time, so that the abrasion loss of a certain hobbing cutter in a shield tunneling process is predicted in real time, the accuracy and the reliability of tool changing opportunity prediction are improved, the shield tunneling efficiency is effectively improved, and the risk of tool changing operation is reduced.
In some embodiments, the hob abrasion parameter refers to an accumulated abrasion loss of the hob at a certain installation position, and is obtained through a manual measurement method. The cumulative wear amount is the sum of the wear amounts at the time of replacement of the hob at a certain mounting position. The abrasion loss can be obtained through manual measurement or image processing and other modes, for example, the manual measurement refers to measuring the radial abrasion loss of the hob when the shield tunneling machine is opened for inspection by using a measuring tool caliper matched with the size of a hob ring of the hob.
In some embodiments, the actual working time refers to the total interaction time of the hob and the stratum in the shield tunneling process at a certain installation position, and the sum of the interaction time of the hob and the stratum in the section is recorded for multiple abrasion losses. Specifically, the actual working time is as follows: and filtering out the non-working time of the shield machine in the hob abrasion loss recording section, filtering out the data of the initial section and the final section of the shield machine in each ring of the tunneling process in the recording section, and keeping the data of the stable section.
The non-working time of the shield tunneling machine is determined by the following formula:
F＝min{f(AR),f(PE),f(CRS)}
in the formula, if the value of F is 0, the shield machine is in a non-working state, AR is a tunneling speed, PE is a penetration degree, CRS is a cutter head rotating speed, F (x) is a function for judging whether data contain a zero value, if yes, F (x) is 0, and if not, F (x) is 1.
The stable stage data meets the condition that five parameter values of the tunneling speed AR, the penetration degree PE, the shield thrust TF, the cutter head rotating speed CRS and the cutter head torque TO fluctuate within the range of an average value and a standard deviation in the rings before and after tunneling.
In the above embodiment, the hob wear parameter set refers to a data sample of a tunneling parameter that affects hob wear in shield construction and changes with time. Further, in a preferred embodiment, the wear parameter set and the evaluation criterion of the wear state of the hob are normalized, where the normalization refers to performing non-dimensionalization on the parameter set, and mapping the parameters to the value range interval [0,1], so that errors caused by different data units and magnitude in prediction can be avoided, and the convergence rate can be improved.
In the above embodiment, the convolution LSTM deep learning network model is composed of an input layer, a convolution layer, a pooling layer, an LSTM network layer, a full connection layer, and an output layer.
In the above embodiment, in the convolutional LSTM deep learning network model, the input layer neurons are determined by parameters in the hob wear parameter set. The convolution layer is determined by the depth of a convolution kernel, the size of the convolution kernel and the convolution step length; the pooling layer is determined by the pooling area size and the pooling step size.
In the above embodiment, the LSTM network layer is composed of LSTM units including input gates, forgetting gates, output gates and memory cells, and can learn information of different periods in a time sequence, and better process and predict intervals and delay events.
The embodiment of the invention trains the convolution LSTM deep learning network model and carries out iterative computation for many times until the requirement is met. And the iteration is to substitute the hob wear parameter set and the hob life evaluation standard into the convolution LSTM deep learning network model for calculation. And measuring the error of the predicted value and the measured value of the convolution LSTM deep learning network model by adopting a cost function. Meanwhile, an optimizer is adopted to optimize parameters, namely, an algorithm of trainable parameters in the convolution LSTM deep learning network model is optimized in each iteration process. The training parameters refer to the weight matrix and bias matrix in the convolved LSTM model.
In some embodiments, training the convolved LSTM deep learning network model further comprises determining a prediction mode of the convolved LSTM deep learning network model, i.e., the amount of historical information and the amount of output information considered for predicting hob wear. The historical information amount represents the length of time series data input by each batch in the training and prediction process of the convolution LSTM deep learning network model. The output information quantity represents the time length of the convolution LSTM deep learning network model for predicting the hob abrasion in advance relative to the current time. The prediction mode uses the amount of history information as an input for a certain batch, and uses the amount of output information as an output for the batch. The batch input and output is determined by:
I T ＝{t 1 ,t 2 ,…,t m }
O T ＝{t m+1 ,t m+2 ,…,t m+n }
I T+1 ＝{t m-n+2 ,t m-n+3 ,…,t m+n }
O T+1 ＝{t m+n+1 ,t m+n+2 ,…,t m+2n }
wherein m is the amount of history information, n is the amount of output information, I T Inputting a time series, O, for an arbitrary batch of convolved LSTM deep learning network models T Is a reaction of T The corresponding convolved LSTM deep learning network model outputs a time series. I is T+1 Is I T The next batch of convolution LSTM deep learning network model inputs the time series, O T+1 Is a reaction of T+1 The corresponding convolved LSTM deep learning network model outputs a time series. And by analogy, the time sequence data are sequentially used as the input and the output of the convolution LSTM deep learning network model.
In the embodiment of the invention, a deep learning network (convolution LSTM deep learning network model) is constructed by adopting CNN and LSTM methods, wherein CNN can effectively reduce the correlation among different parameters in a hob abrasion parameter set, and improve the iterative computation efficiency of the model. The hob abrasion is closely related to historical shield operation parameters and historical stratum properties, and the LSTM can fully consider historical information contained in time sequence data in an abrasion parameter set and continuously update and output the data based on new input data, so that the prediction accuracy of a deep learning network is improved.
The parts of the present invention not specifically described in the above embodiments can be implemented by using the prior art, and are not described herein again.
Based on the same technical concept, another embodiment of the present invention further provides a system for predicting the wear loss of a shield hob based on a convolution LSTM, including: the hob abrasion parameter set building module, the evaluation standard building module, the hob abrasion loss prediction model building module and the prediction module are provided, wherein: the hob abrasion parameter set construction module is used for constructing a hob abrasion parameter set, and the hob abrasion parameter set comprises data samples of the time change of the excavation parameters influencing the hob abrasion in the shield construction; the evaluation standard establishing module establishes a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter of the hob; the hob abrasion loss prediction model establishing module is used for establishing a convolution LSTM deep learning network model, inputting a hob abrasion parameter set and a hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training; and the prediction module predicts by adopting a hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
In this embodiment, the evaluation criterion establishing module establishes the evaluation criterion of the wear state of the hob, which can be expressed by the following notations:
CLI＝W a /(d·t w )
in the formula, CLI is the index of the abrasion state of the hob, and the unit is 1/min; w a The accumulated abrasion loss of the hob is in mm; t is t w The unit of the actual working time corresponding to the accumulated abrasion loss of the hob is min; d is the diameter of the hob ring and the unit is mm.
In the above real-time shield hob wear extent prediction system based on the convolution LSTM, the technologies adopted by the modules may refer to the technologies implemented in the corresponding steps of the real-time shield hob wear extent prediction method based on the convolution LSTM in the above embodiments, and are not described herein again.
In another embodiment of the present invention, a shield hob abrasion loss real-time monitoring terminal is further provided, which includes a memory, a processor, and a computer program stored on the memory and capable of running on the processor, and when the processor executes the program, the processor is configured to execute the method for predicting the abrasion loss of the shield hob based on the convolution LSTM in the foregoing embodiments in real time.
In another embodiment of the present invention, a computer readable storage medium is further provided, on which a computer program is stored, which when executed by a processor is used to execute the method for real-time prediction of the wear amount of the shield hob based on the convolution LSTM in the above embodiments.
For a better understanding of the above embodiments of the present invention, the following examples are given for further illustration and are not intended to limit the present invention.
Taking the construction of a newly-built railway shield tunnel in a certain city as an example, the excavation diameter of the earth pressure balance shield machine is 9.15m, the inner diameter and the outer diameter of each duct piece are 8m and 8.8m respectively, and the width is 1.8m. The opening ratio of the cutter head is about 35%, and the cutter head is provided with 6 17-inch double-edge central hobs, 48-inch single-edge front hobs, 88 cutters, 24 pairs of edge scrapers, 24 gauge cutters, 30 shell cutters and 12 single-edge hob cutters. Wherein the diameter of the 17-inch double-edged hob is 432mm, the distance between knives is 90mm, and the height of the knives is 160mm; the diameter of the 19-inch single-edge hob is 483mm, the distance between knives is 85mm, and the height of the knife is 160mm. In this example, the wear of a front 33 single-blade hob is predicted as an example and is explained in detail.
As shown in fig. 1, the method for predicting the wear loss of the shield hob based on the convolution LSTM in real time in the embodiment specifically includes the following steps:
the method comprises the following steps: and collecting engineering data including tunnel geometric parameters, shield tunneling parameters, hob positions and cutter ring size data.
In the embodiment, the excavation diameter of the tunnel can be determined to be 9.15m and the buried depth of the top of the tunnel in the research section can be determined to be between 5 and 25m according to the actual engineering situation and the design report. The shield tunneling parameters such as tunneling speed, penetration degree, cutter head rotating speed, shield machine power system parameters, cutter head system parameters, slag discharge system parameters, slag soil adjusting system parameters and the like are obtained from a self-contained acquisition system of the shield tunneling machine. The diameter of the No. 33 single-edge hob is 483mm, and the height of the hob is 160mm.
Step two: and determining the wear parameters of the No. 33 hob and the corresponding actual working time.
In this embodiment, the abrasion parameter of the No. 33 hob is the accumulated abrasion loss, and when the No. 33 hob is replaced in the shield tunneling process, the sum of the radial abrasion losses of the hobs is measured by a manual measurement method by using a special measuring tool caliper for the abrasion loss of the 19-inch hob.
In this embodiment, the actual working time refers to the total interaction time of the 33 # hob and the stratum in the shield tunneling process, and is the sum of the interaction time of the hob and the stratum in the 28-time wear loss recording section. The actual working time needs to be filtered out, the abrasion loss of the No. 33 hob needs to be filtered, the non-working time of the shield machine in the section is recorded, the data of the initial section and the final section of the shield machine in each ring of the tunneling process in the section is recorded, and the data of the stable section is reserved.
In this embodiment, the non-operating time of the shield machine is determined by the following formula:
F＝min{f(AR),f(PE),f(CRS)}
in the formula, if the value of F is 0, the shield machine is in a non-working state, AR is a tunneling speed, PE is a penetration degree, CRS is a cutter head rotating speed, F (x) is a function for judging whether data contain a zero value, if yes, F (x) is 0, and if not, F (x) is 1.
In this embodiment, the data in the stable stage needs TO satisfy the fluctuation of five parameter values, i.e., the tunneling speed AR, the penetration PE, the shield thrust TF, the cutter rotation speed CRS, and the cutter torque TO, in the ranges of the average value and the standard deviation before and after tunneling.
Step three: and establishing a 33 # hob abrasion state evaluation standard and building a model database.
In this embodiment, the change of the cumulative wear loss of the # 33 hob with the working time is shown in fig. 2, the wear state of the # 33 hob is determined by the evaluation index of the wear state of the hob in the shield tunneling process, and the evaluation index is determined by the following formula:
CLI＝W a /(d·t w )
in the formula, CLI is the index of the abrasion state of a No. 33 hob, and the unit is 1/min; w is a group of a The accumulated abrasion loss of the No. 33 hob is in mm; t is t w The actual working time corresponding to the accumulated abrasion loss of the No. 33 hob is min; d is the diameter of a 33 # hob ring and is 483mm.
In this embodiment, the database is built, and the method includes two steps of establishing a 33 # hob wear parameter set and data preprocessing. The number 33 hob abrasion parameter set refers to a data sample of the tunneling parameters influencing hob abrasion in shield construction along with time change, and comprises the shield machine power system parameters, the cutter head system parameters, the slag discharge system parameters, the slag soil adjusting system parameters and the shield tunnel geometric parameters collected in the step one. The power system parameters comprise shield thrust, propulsion speed, penetration degree, soil cabin pressure and six groups of shield hydraulic oil cylinder jacking force; the cutter parameters comprise the rotating speed of the cutter and the torque of the cutter; the slag discharge system parameters comprise the rotating speed of the screw machine, the torque of the screw machine and the soil pressure of the screw machine; the parameters of the muck adjusting system comprise nine groups of foam pipe pressure, nine groups of foam flow rate and nine groups of air flow rate; the geometric parameters of the shield tunnel are taken as the distance from the tunnel top plate to the ground.
In this embodiment, the data preprocessing is to perform normalization processing on the 33 # hob wear parameter set and the hob wear state evaluation standard, and map the parameters to the value domain interval [0,1], so as to avoid errors in prediction caused by different data units and magnitude levels, and improve the convergence rate.
Step four: establishing a convolution LSTM deep learning network model, inputting a hob abrasion parameter set and a hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
in some embodiments, a hob wear parameter set is substituted into an input layer of a convolution LSTM deep learning network model, a hob life evaluation standard is substituted into an output layer of the convolution LSTM deep learning network model, iterative training is performed on the convolution LSTM deep learning network model, a weight matrix and a bias matrix in the model are updated, and a final prediction model is obtained after the training is completed. The hob life evaluation standard is output of a convolution LSTM deep learning network model, and the difference between the output value of the convolution LSTM deep learning network model and the output value of the convolution LSTM deep learning network model in the iterative process is used as model error back propagation, so that a weight matrix and a bias matrix are updated to obtain an optimized model, and a final prediction result is obtained by adopting the optimized model.
Specifically, the structure of the convolution LSTM deep learning network model is shown in fig. 3, and the model is composed of an input layer, a convolution layer, a pooling layer, an LSTM network layer, a full-link layer, and an output layer, and includes model parameters such as the number of neurons in each layer, the number of iterations, a cost function, and an optimization algorithm.
And (4) determining the neuron of the input layer by the parameters in the hob abrasion parameter set in the step three. In this example, the input layer neuron number is 43. The full-connectivity layer and output layer neuron numbers are determined in step five.
The convolutional layer is determined by the convolutional kernel depth, convolutional kernel size, and convolutional step size, and the pooling layer is determined by the pooling region size and pooling step size. In this embodiment, the convolution kernel depth in the convolution layer is 24, the convolution kernel size is 6, and the step size is 2. The size of the pooling area in the pooling layer is 2 and the step size is 1.
The LSTM network layer is composed of LSTM units including input gates, forgetting gates, output gates and memory cells, and can learn information of different periods in time sequence, and better process and predict intervals and delay events.
In this embodiment, the number of iterations is set to 200, and a mean square error function (MSE) is used as the cost function. And adopting an Adam algorithm as an optimization algorithm of the weight matrix and the bias matrix.
Training the convolution LSTM deep learning network model according to the parameter setting, which specifically comprises the following steps:
1) And inputting the hob abrasion parameter set and the hob abrasion state evaluation standard obtained in the third step into the quadruple convolution LSTM deep learning network model for training.
2) And determining the historical information amount and the output information amount considered by predicting the abrasion of the hob.
The historical information amount represents the length of time series data input by each batch in the training and prediction process of the convolution LSTM deep learning network model. The output information quantity represents the time length of the convolution LSTM deep learning network model for predicting the hob abrasion in advance relative to the current time. The amount of history information m is determined by the following equation:
in the formula, r is the average time required for the shield to tunnel a ring, in this embodiment, r =225min, and the calculated historical information amount m is 5min.
Quantity of output information t o Is determined by the following formula:
in this embodiment, the calculated output information amount n is 3min.
3) The prediction mode of the convolved LSTM model is determined. The prediction mode uses the historical information amount as an input of a certain batch, and uses the output information amount as an output of the batch. The batch input and output is determined by:
I 1 ＝{t 1 ,t 2 ,…,t m }＝{t 1 ,t 2 ,t 3 ,t 4 ,t 5 }
O 1 ＝{t m+1 ,t m+2 ,…,t m+n }＝{t 6 ,t 7 ,t 8 }
I 2 ＝{t m-n+2 ,t m-n+3 ,…,t m+n }＝{t 4 ,t 5 ,t 6 ,t 7 ,t 8 }
O 2 ＝{t m+n+1 ,t m+n+2 ,…,t m+2n }＝{t 9 ,t 10 ,t 11 }
wherein m is the amount of history information, n is the amount of output information, I 1 Input the time series, O, for the first batch of convolved LSTM deep learning network models 1 And outputting a time sequence for the first batch of convolution LSTM deep learning network models. I.C. A 2 Second batch convolution LSTM deep learning network model input time series, O 2 And outputting a time sequence for the second batch of convolution LSTM deep learning network models. And by analogy, the time sequence data are sequentially used as the input and the output of the convolution LSTM deep learning network model.
4) Determining input layer, full connectivity layer and output layer neuron numbers.
And the number of neurons in the input layer is equal to the number of parameters in the hob abrasion parameter set in the step three. In this example 43.
The number of output layer neurons equals the amount of output information. In this example, the number of output layer neurons is 3.
The number of full-connectivity layer neurons q is determined by:
q＝3p-1
in the formula, p is the number of neurons in the output layer, and the number q of neurons in the full connection layer in this embodiment is calculated to be 8.
5) The model training parameters are updated using the optimizer.
6) Repeating the steps 1) to 5) until the iteration number of the convolution LSTM deep learning model reaches 200 times.
Step five: and D, adopting the hob abrasion loss prediction model obtained after the training in the step four to predict the abrasion loss of the shield No. 33 hob and output the predicted value of the abrasion loss of the hob. Fig. 4 is a graph showing the predicted wear amount of the No. 33 hob of this example.
For example, as can be seen from fig. 4, when the shield advance time is 3750min, the predicted value of the cumulative wear loss of the 33 th hob is 95mm, the measured value of the cumulative wear of the last hob at that time is 92mm, the difference between the two is 3mm, and is less than the limit value of the wear of the 33 th hob by 15mm, so that the 33 th hob does not need to be replaced when the shield machine operating time reaches 3750min, and if the difference between the two is greater than or equal to 15mm, the 33 th hob needs to be replaced. If the difference between the two approaches 15mm (for example, reaches 14 mm), the tool change can be prepared in advance according to the field situation.
According to the shield hob abrasion loss real-time prediction method and system based on the convolution LSTM, a CNN (convolutional neural network) method and an LSTM method are comprehensively adopted, a deep learning network is built to predict hob abrasion in real time, and the prediction accuracy of the deep learning network is improved; meanwhile, a hob service life evaluation standard is established to judge the hob service life, the defect that the health degree of the whole hob of the hob head is only considered in the existing method is overcome, field construction personnel can be helped to judge the time of cutter inspection, and a more simple, convenient, reasonable and efficient new method is provided for determining the abrasion loss of the shield hob.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The parts of the present invention not specifically described above can be implemented by using the prior art, and are not described herein again.
Claims (10)
1. A shield hob abrasion loss real-time prediction method based on convolution LSTM is characterized by comprising the following steps:
s1, building a hob abrasion parameter set, wherein the hob abrasion parameter set comprises data samples of the time variation of excavation parameters influencing the abrasion of a hob in shield construction;
s2, establishing a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter;
s3, establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
and S4, predicting by using the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
2. The method for predicting the wear of a shield hob according to claim 1, wherein the parameters of the hob wear parameter set include the following specific parameters: shield machine power system parameters, cutter head system parameters, slag discharge system parameters, slag soil adjusting system parameters and shield tunnel geometric parameters.
3. The method for predicting the abrasion loss of the shield hob based on the convolution LSTM according to claim 1, wherein the hob abrasion parameter set and the hob abrasion state evaluation standard are normalized and then input into the convolution LSTM deep learning network model.
4. The method for predicting the abrasion loss of a shield hob based on the convolution LSTM according to claim 1, wherein the evaluation criteria of the abrasion state of the hob are specifically as follows:
CLI＝W a /(d·t w )
in the formula, CLI is the index of the abrasion state of the hob, and the unit is 1/min; w is a group of a The unit is the accumulated abrasion loss of the hob and is mm; t is t w The unit of the actual working time corresponding to the accumulated abrasion loss of the hob is min; d is the diameter of the hob ring and the unit is mm.
5. The real-time shield hob wear loss prediction method based on the convolution LSTM according to claim 1, wherein the convolution LSTM deep learning network model is composed of an input layer, a convolution layer, a pooling layer, an LSTM network layer, a full connection layer and an output layer which are connected in sequence, wherein:
the LSTM network layer is composed of LSTM units including input gates, forgetting gates, output gates and memory cells, and learns information of different long periods and short periods in a time sequence.
6. The method for predicting the abrasion loss of a shield hob based on convolution LSTM according to claim 1, wherein the actual working time is the total interaction time of the hob with the stratum in the shield tunneling process at a certain installation position, and the sum of the interaction time of the hob and the stratum in a section is recorded for multiple abrasion losses.
7. The convolution LSTM-based shield hob abrasion loss real-time prediction method of claim 5, wherein the actual working time is: and filtering out the non-working time of the shield machine in the hob abrasion loss recording section, filtering out the data of the initial section and the final section of the shield machine in each ring of the tunneling process in the recording section, and keeping the data of the stable section.
8. A shield hobbing cutter abrasion loss real-time prediction system based on convolution LSTM is characterized by comprising:
the hob abrasion parameter set building module is used for building a hob abrasion parameter set, and the hob abrasion parameter set comprises data samples of the time change of the tunneling parameters influencing the hob abrasion in the shield construction;
the evaluation standard establishing module establishes a hob abrasion state evaluation standard according to the hob abrasion parameters, the actual working time corresponding to the hob abrasion parameters and the hob ring diameter;
the hob abrasion loss prediction model establishing module is used for establishing a convolution LSTM deep learning network model, inputting the hob abrasion parameter set and the hob abrasion state evaluation standard into the convolution LSTM deep learning network model for training, and obtaining a hob abrasion loss prediction model after training;
and the prediction module is used for predicting by adopting the hob abrasion loss prediction model to obtain a real-time predicted value of the shield hob abrasion loss.
9. A real-time shield hob wear amount monitoring terminal, comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor is used for executing the real-time shield hob wear amount prediction method based on the convolution LSTM according to any one of claims 1 to 8 when executing the program.
10. A computer readable storage medium having stored thereon a computer program, when being executed by a processor, for executing the method for real-time prediction of shield hob wear according to any one of claims 1 to 8 based on convolution LSTM.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211189012.4A CN115600654A (en) | 2022-09-27 | 2022-09-27 | Convolution LSTM-based shield hob abrasion loss real-time prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211189012.4A CN115600654A (en) | 2022-09-27 | 2022-09-27 | Convolution LSTM-based shield hob abrasion loss real-time prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115600654A true CN115600654A (en) | 2023-01-13 |
Family
ID=84845182
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211189012.4A Pending CN115600654A (en) | 2022-09-27 | 2022-09-27 | Convolution LSTM-based shield hob abrasion loss real-time prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115600654A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118228056A (en) * | 2024-05-27 | 2024-06-21 | 中铁十四局集团有限公司 | Method and system for predicting cutter changing time of shield hob, shield monitoring platform and application |
-
2022
- 2022-09-27 CN CN202211189012.4A patent/CN115600654A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118228056A (en) * | 2024-05-27 | 2024-06-21 | 中铁十四局集团有限公司 | Method and system for predicting cutter changing time of shield hob, shield monitoring platform and application |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778010B (en) | TBM cutter life prediction method based on data-driven support vector regression machine | |
| CN111160571B (en) | Method for driving inversion of geology based on earth pressure balance shield machine parameter data | |
| CN110018670B (en) | Industrial process abnormal working condition prediction method based on dynamic association rule mining | |
| Song et al. | A new fuzzy c-means clustering-based time series segmentation approach and its application on tunnel boring machine analysis | |
| Leng et al. | A hybrid data mining method for tunnel engineering based on real-time monitoring data from tunnel boring machines | |
| CN113404502B (en) | Shield hob abrasion monitoring device and method based on ballast piece morphology | |
| CN115563863A (en) | Method and terminal for determining abrasion loss of slurry shield hobbing cutter in real time based on deep learning | |
| CN107368463B (en) | Roadway nonlinear deformation prediction method based on fiber bragg grating sensor network data | |
| CN114417697A (en) | Neural network-based TBM hob abrasion real-time prediction method and system | |
| CN112396320A (en) | Tunnel collapse risk assessment method based on ISM and fuzzy Bayesian network | |
| KR20180116922A (en) | Apparatus for predicting net penetration rate of shield tunnel boring machine and method thereof | |
| CN115600654A (en) | Convolution LSTM-based shield hob abrasion loss real-time prediction method and system | |
| CN115544753A (en) | Real-time assessment method, system, terminal and medium for shield hob abrasion state | |
| CN113063573A (en) | Shield cutter abrasion detection method and system | |
| CN114091188A (en) | Shield tunneling machine hydraulic drive cutter head system modeling and fault diagnosis method and device | |
| CN116245020A (en) | Cutter head abrasion and energy consumption optimization method and system based on shield tunneling machine | |
| Qiu et al. | TBM tunnel surrounding rock classification method and real-time identification model based on tunneling performance | |
| CN112966355B (en) | Method for predicting residual service life of shield machine cutter based on deep learning | |
| CN113762360B (en) | Surrounding rock grade prediction method in TBM tunneling process based on SMOTE+ADACCOST algorithm | |
| CN111340275B (en) | Tunnel support mode selection real-time prediction method based on detection while drilling technology | |
| CN115563872B (en) | TBM hob abrasion loss prediction method based on DE-SVR algorithm | |
| Simoes et al. | Fuzzy modeling approaches for the prediction of machine utilization in hard rock tunnel boring machines | |
| CN117786794A (en) | Shield tunneling existing tunnel deformation optimization control method and system | |
| CN115544874A (en) | Real-time monitoring system, method, terminal and medium for shield hob abrasion loss | |
| CN117172360A (en) | Drilling mechanical drilling speed optimization method, system, equipment and medium based on MLP and high-efficiency PSO |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |