Method and apparatus for determining geotechnical parameters of excavated material

Abstract

The invention relates to a method for determining at least one geotechnical parameter of tunnel boring machine excavation material, wherein: the excavation material is fed onto a conveyor device (2) for conveying the excavation material; an impact body (5) is arranged in the conveying flow (3) of the excavation material such that the impact body is at least partially immersed in the excavation material, the excavation material in the conveying flow (3) having a preset or measured flow rate; a measurement parameter corresponding to the force exerted by the excavation material of the conveying flow (3) onto the impact body (5) is measured, a time sequence of measurement values of the measurement parameter being acquired; and the at least one geotechnical parameter is determined from a section of the sequence of measurement values in such a way that the section of the time sequence of measurement values is input into a data processing device and the section is processed by the data processing device in such a way that at least one output value corresponding to the at least one geotechnical parameter is output, the processing of the measurement values being based on a preset flow rate or a value of the measured flow rate being input and taken into account in the processing of the measurement values.

Description

Technical Field

[0001] The present invention relates to a method for determining at least one geotechnical parameter of the excavation material for a tunnel boring machine. Background Technology

[0002] The type and composition of the excavation material for a tunnel boring machine (TBM) depends, of course, on the characteristics of the rock mass (rock or soil) present at the excavation face, as well as on the type of excavation and the substances added to the excavation material at that time.

[0003] In soft, cohesive soils, tunnel boring machines (TBMs) with earth pressure support, also known as earth pressure balance (EPB) shield tunneling machines, are preferred. In an EPB shield, a slurry composed of excavated material and water is used as the plastic support medium. The TBM's rotating cutting wheel, equipped with blades, presses against the excavation face and strips away the existing soil. The excavated material enters the excavation chamber through an opening, where it is mixed with the existing slurry. A conveyor auger transports the excavated material from the bottom of the excavation chamber onto a conveyor belt. Here, the support pressure of the slurry can be precisely controlled by coordinating the conveyor auger's rate with the tunneling speed. Pressure conditions are continuously monitored using earth pressure sensors within the excavation chamber. Thus, even under varying geological conditions, all tunneling parameters can be optimally coordinated by the machine operator. This achieves necessary compensation for the pressure conditions at the excavation face, preventing uncontrolled soil intrusion into the atmospheric pressure portion of the TBM, and thereby creating conditions for rapid, low-settlement tunneling.

[0004] Not all soils possess ideal properties for earth pressure balance (EPB) tunneling under natural conditions. However, the application of this method can be expanded through soil amendment. This involves adding different amendments, such as water, clay, or foam, to alter the plasticity, deformability, consistency, and permeability of the geological mass. In this way, EPB tunneling can achieve good tunneling performance even in heterogeneous soils with varying gravel, sand, or water content, or in inherently unstable rock masses.

[0005] However, if excessive water, clay, and / or foam are added to the excavated material, for example, excavated material conveyed by a conveyor screw to a conveyor belt becomes difficult to handle due to its flowable consistency; such excavated material must, for example, be temporarily stored in a pit or settling pond or undergo further treatment. This can lead to increased costs. Furthermore, there are generally regulations for the use of disposal sites stipulating that the minimum shear strength of the excavated material to be disposed of must not be less than 10 kN / m. 2 .

[0006] Therefore, in these and other applications, it is desirable to automatically determine the geotechnical parameters, particularly rheological parameters, such as viscosity and yield strength, (vane) shear strength, or slump (“slump”; according to DIN EN 12350-2) of the excavated material being transported from the excavation chamber during ongoing operations. For example, determining the slump (according to DIN EN 12350-2) requires extracting a sample from the transport stream and, within the scope of a slump test, filling a truncated cone-shaped metal mold of predetermined dimensions, lifting the mold in a defined manner, and measuring the difference between the height of the mold and the height of the collapsed truncated cone of excavated material after lifting. This is time-consuming and cannot be fully automated; in particular, continuous measurement acquisition is not possible. A similar situation applies to the determination of alternative geotechnical parameters, such as spread, vane shear strength, or dynamic viscosity and yield strength. Summary of the Invention

[0007] Therefore, the object of the present invention is to automatically determine the geotechnical parameters of the excavation material delivered from the excavation chamber during operation.

[0008] According to the present invention, the objective is achieved by a method having the features of claim 1 for determining at least one geotechnical parameter of excavation material of a tunnel boring machine, and an apparatus having the features of claim 14 for determining at least one geotechnical parameter of excavation material generated in a tunnel boring machine.

[0009] The determination of at least one geotechnical parameter is preferably performed automatically.

[0010] In a method according to the invention for determining at least one geotechnical parameter of excavation material for a tunnel boring machine, the excavation material is supplied to a conveying device for conveying the excavation material. An impact body is placed in the conveying flow of the excavation material such that the impact body is at least partially submerged in the excavation material, the excavation material in the conveying flow having a preset or measured flow velocity. A measurement parameter corresponding to the force exerted on the impact body by the excavation material in the conveying flow is then measured. At this time, a time series of the measured values ​​of the measurement parameter is acquired, and the at least one geotechnical parameter is determined from a segment of the sequence of measured values ​​by inputting the segment of the time series of measured values ​​into a data processing device and processing the segment to output at least one output value corresponding to the at least one geotechnical parameter. The processing of the measured values ​​is based on the preset flow velocity, or the measured flow velocity value is input and taken into account when processing the measured values.

[0011] The data processing apparatus preferably implements a trained neural network having multiple layers of nodes, assigning weights to all connections between the nodes and assigning predetermined activation functions to the nodes, the weights having been pre-set through a training method. An input data matrix is ​​obtained from the segments of the sequence of measured values ​​and is input to the input layer of the nodes of the neural network. The at least one output value corresponding to the at least one geotechnical parameter is output by at least one node of the output layer of the neural network.

[0012] Advantageously, measurements can be taken directly on the existing transport flow, and the desired geotechnical parameters, such as slump, can be automatically provided after a very short data processing time. This, for example, enables the setting and / or adjustment of the amount of water and / or foam and / or other additives that affect the rheological properties of the excavation material based on the determined at least one geotechnical parameter, or the classification of the excavation material itself based on the geotechnical parameter.

[0013] In one implementation, the predetermined activation function is, for example, the reLU activation function (reLU(x) = max(0,x)), or alternatively, a differentiable approximation function as the reLU activation function is the function f(x) = ln(1 + ex). Initial values ​​are assigned to the weights before the training method begins, preferably small, randomly chosen initial values, such as uniformly distributed numbers in the interval (-0.05, 0.05) or random numbers normally distributed around zero. The training method, for example, uses the squared error as the loss function and employs gradient descent to adjust the weights.

[0014] In one embodiment of the method for determining at least one geotechnical parameter, the transport flow of excavated material may be guided through a pipe in which the impactor, such as a rod extending into the pipe, is installed. However, in a preferred embodiment of the method, the impactor is positioned within an open transport flow of the excavated material. This is preferably an upper open transport flow on a conveyor belt.

[0015] In a preferred embodiment of the method for determining at least one geotechnical parameter, a predetermined flow rate is set for the transport flow and the predetermined flow rate is kept constant during the time series of measurements of the measured parameter. This simplifies data processing because it eliminates the need to process measurement-time pairs, requiring only the measurement sequence. The prerequisite is that the set flow rate is equal to the flow rate set when obtaining the pairings of the measurement sequence and the measured geotechnical parameter used in the training method.

[0016] The time series of the measurements are preferably acquired at a constant sampling frequency. This also simplifies data processing, as only one value of the sampling frequency or the time interval between measurement points needs to be processed, or only processed at that time.

[0017] In a preferred embodiment of the method for determining at least one geotechnical parameter, the time series segments of the measured values ​​correspond to measurement durations of 0.25 s to 60 s, preferably 0.5 s to 1 s. Practical tests using common compositions and flow velocities of excavated materials have shown that the important motion processes of the impactor, related to the determination of geotechnical parameters, particularly slump, and consequently the important force curves to be acquired, occur within said time intervals.

[0018] In one embodiment of the method for determining at least one geotechnical parameter, a predetermined number of characterizing features are calculated from the time series segments of the measured values ​​and used as input data matrices. For example, the mean, median, maximum, minimum, and / or standard deviation are calculated from the time series segments of the measured values—also called measurement sequences—as characterizing features, or the Fourier coefficients of the discrete Fourier transform are also calculated. These features are arranged in a multidimensional matrix and, after normalization, are passed to the input layer.

[0019] In another simple embodiment of the method for determining at least one geotechnical parameter, the time series segment of the measured values, i.e., the measurement sequence, is normalized and used as a one-dimensional input data matrix.

[0020] It is conceivable that the data processing device implements the neural network using dedicated hardware with circuit elements having a large number of simulated neurons. However, a preferred embodiment of the method for determining at least one geotechnical parameter is characterized by the data processing device simulating the neural network (a so-called artificial neural network) in a program-controlled manner. This means that the data processing device performs mathematical operations simulating the functions of each neuron layer of the neural network.

[0021] The data processing device can simulate different types of neural networks, such as feedforward networks (including multilayer perceptrons), convolutional networks, or recurrent neural networks. In a preferred improvement of the method for determining at least one geotechnical parameter, the data processing device simulates a feedforward neural network with a one-dimensional convolutional network. Compared to fully connected networks, the amount of computation to be performed is significantly reduced by using convolutional neural networks (CNNs). Simultaneously, locally neighboring neurons, and thus temporally neighboring measurements in the current case (i.e., the measurement sequence), have a greater influence on neurons in the next layer.

[0022] In a currently preferred embodiment of the method for determining at least one geotechnical parameter, the data processing device simulates a convolutional neural network having a first convolutional layer with subsequent pooling layers and a subsequent second convolutional layer with multiple subsequent fully connected layers. It has been demonstrated that this configuration of the neural network avoids overfitting on the one hand and achieves good predictions on the other.

[0023] A preferred embodiment of the method for determining at least one geotechnical parameter is characterized in that, prior to determining the at least one geotechnical parameter from the time series segment of the input measurements, the weights are set in the training method in the following manner:

[0024] a) For multiple excavation material samples with different compositions and different geotechnical parameters selected in advance, respectively:

[0025] a1) Measure or determine and store as target parameter values ​​the at least one geotechnical parameter, such as slump, of the excavated material sample.

[0026] a2) The impactor is placed in the conveying flow of the excavated material sample such that the impactor is at least partially submerged in the excavated material, the excavated material in the conveying flow having a predetermined flow velocity, and measurement parameters corresponding to the force exerted on the impactor by the excavated material in the conveying flow of the excavated material sample are measured, and time series of the measured values ​​of the measurement parameters are acquired.

[0027] a3) Store pairs consisting of a segment of the sequence of measured values ​​and a corresponding target parameter value.

[0028] b) Assigning initial values ​​to the weights in the neural network implemented by the data processing device.

[0029] c) For pairings consisting of segments of a sequence of measurements and corresponding target parameter values:

[0030] c1) Determining the at least one geotechnical parameter from the segment of the sequence of measured values ​​by processing the segment of the time series of measured values ​​by a data processing device implementing the neural network, such that at least one output value corresponding to the at least one geotechnical parameter is output.

[0031] c2) Then determine the difference between the output value and the corresponding target parameter value.

[0032] c3) Change the weights based on the difference.

[0033] In this process, for the next pairing consisting of a segment of the sequence of measured values ​​and the corresponding target parameter value, steps c1) to c3) are repeated, in which case the pairings are selected in a random order.

[0034] For the execution of steps c1) to c3) of the predetermined portion of the pair stored in step a), the predetermined portion is used as a training set.

[0035] An application of the method for determining at least one geotechnical parameter according to the present invention is a method for adjusting the consistency of modified excavation material in an earth pressure balance shield tunneling machine, wherein, in order to modify the excavation material in the excavation chamber, a modifier, such as a surfactant foam, is added to the excavation material, and the modified excavation material is conveyed from the excavation chamber using a conveying device. Here, in an open conveying flow formed on one of the conveying devices, at least one geotechnical parameter, preferably slump, is automatically determined according to one of the methods described above, and the amount of modifier to be added is adjusted according to the geotechnical parameter thus determined, preferably by increasing or decreasing the amount of modifier to be added based on the difference between the geotechnical parameter thus determined and the expected value of the geotechnical parameter.

[0036] An application of the method for determining at least one geotechnical parameter according to the present invention is a method for classifying excavation material of a tunnel boring machine, wherein modified excavation material is conveyed from an excavation chamber using at least one conveying device, and in an open conveying flow formed on one of the conveying devices, at least one geotechnical parameter, preferably slump, is automatically determined according to one of the methods described above, and the excavation material is classified according to the determined geotechnical parameter, preferably by classifying the excavation material according to the difference between the thus determined geotechnical parameter and an expected value of the geotechnical parameter.

[0037] An apparatus according to the invention for determining at least one geotechnical parameter of excavated material generated in a tunnel boring machine comprises: (a) a conveying device for conveying the excavated material such that a conveying flow of the excavated material with a certain flow velocity is formed in an open conveying channel, the conveying device having means for setting a predetermined flow velocity and / or means for measuring the flow velocity; (b) an impactor disposed in the open conveying channel, the impactor being at least partially submerged in the conveying flow of the excavated material; (c) a measuring device connected to the impactor for measuring a measurement parameter corresponding to a force exerted on the impactor by the excavated material in the conveying flow; (d) means coupled to the measuring device for acquiring sampled values ​​of the measurement parameter, converting the sampled values ​​of the measurement parameter into digital measurement values, and outputting a time series of the digital measurement values; (e) a storage device for temporarily storing the sequence of digital measurement values; and (f) a data processing device coupled to the storage device, the data processing device being configured such that the data processing device processes segments of the sequence of digital measurement values ​​and outputs at least one output value corresponding to the at least one geotechnical parameter.

[0038] Advantageously, the data processing device is configured such that it generates and processes an input data matrix from the segments of the sequence of digital measurements, and outputs at least one output value corresponding to the at least one geotechnical parameter, taking into account a predetermined flow rate or the measured value of the flow rate during processing.

[0039] The data processing device implements a trained neural network having multiple layers of nodes, assigns weights to all connections between the nodes and assigns predetermined activation functions to the nodes, the weights being pre-set by a training method, inputs the input data matrix into the input layer of the neural network consisting of nodes, and outputs at least one output value corresponding to the at least one geotechnical parameter at at least one node of the output layer of the neural network.

[0040] The impactor can be a structure of any shape, introduced into the transport flow of excavated material such that forces related to the rheological properties of the excavated material act on the impactor. The impactor can, for example, be a rod with a circular cross-section extending into the transport flow. In a preferred embodiment of the device for determining at least one geotechnical parameter, the impactor is constructed as a sphere. An advantage of this type of impactor is that it does not need to be positioned in a defined orientation within the transport flow. A plowshare-shaped body is also an advantageous embodiment.

[0041] In a preferred device for determining at least one geotechnical parameter, the impactor is a steel ball weighing 500g to 2000g, preferably 800g to 1000g. This steel ball, for example, has a diameter of about 6cm and a weight of about 900g. Under typical flow velocities (in the range of 1m / s to 3m / s) and compositions of the transport flow, a steel ball of this size is heavy enough that it does not excessively "jump" across the surface of the excavated material transport flow, but not so heavy that it remains in a stable position almost unaffected by the transport flow.

[0042] For example, the impactor in the form of a sphere can be mounted on a fixed support, such as a rod, and extend into the delivery stream. For example, the strain gauges of the measuring device can record the force-related bending of such a rod.

[0043] However, in a preferred improvement of the device used to determine at least one geotechnical parameter, the spherical impactor is fixed to a swingably suspended rod or cable, and a sensor for acquiring the measurement parameter, corresponding to the tension borne by the rod or cable, is coupled to the rod or cable. Suspending the impactor to a swingably suspended rod or cable is advantageous because this suspension allows the impactor to avoid larger objects (e.g., rocks) carried in the excavated material flow.

[0044] In a preferred apparatus for determining at least one geotechnical parameter, the data processing device is configured to simulate a feedforward neural network in the form of a one-dimensional convolutional network. Preferably, the data processing device is configured to simulate a one-dimensional convolutional neural network having a first convolutional layer with subsequent pooling layers and a subsequent second convolutional layer with multiple subsequent fully connected layers. This has the advantages mentioned above in conjunction with the corresponding preferred embodiments of the method.

[0045] Advantageous and / or preferred improvements of the invention are described in the dependent claims. Attached Figure Description

[0046] The present invention will now be described in detail with reference to the preferred embodiments shown in the accompanying drawings. In the drawings:

[0047] Figure 1 A schematic diagram of the principle of a measuring device for determining at least one geotechnical parameter is shown.

[0048] Figure 2 An exemplary illustration of a measurement series and a fragment of the measurement series constituting a measurement sequence are shown;

[0049] Figure 3A schematic diagram illustrating the principle of a neural network simulated by a data processing device is shown.

[0050] Specific implementation form

[0051] Figure 1 The diagram illustrates a rheological measuring device 1 used in an apparatus according to the invention for determining at least one geotechnical parameter, particularly an index parameter, such as slump.

[0052] The measuring device 1 is used to determine soil and rock parameters, such as, in particular, slump, in the conveying flow 3, which moves at a predetermined speed in the conveying channel 2. The movement of the conveying flow 3 is indicated by arrow 4.

[0053] For example, the conveying channel 2 is located on the conveyor belt of the tunnel boring machine. The speed of the conveying flow 3 is in the range of 1 m / s to 3 m / s, preferably in the range of 1.8 m / s to 2.5 m / s.

[0054] An impactor, preferably in the form of a steel ball 5, suspended on cable 6, is at least partially submerged in the conveyor flow 3. A force sensor 7 is installed at the suspension portion of cable 6 on a support with a crossbeam 8, the force sensor measuring the force acting on cable 6. For common excavated material moving in the conveyor channel 2 at a speed of 1.8 m / s to 2.5 m / s, a steel ball with a diameter of 6 cm and a weight of approximately 900 g is preferably used.

[0055] Force sensor 7 and evaluation device for evaluating sensor output signal ( Figure 1 (Not shown in the diagram) The evaluation device samples the output signal of the force sensor 7 at a predetermined sampling frequency and outputs and / or temporarily stores the sampled values. In the test apparatus, the sampling frequency is, for example, 200 Hz, so that force sampled values ​​are acquired and stored at intervals of 5 ms.

[0056] Figure 2 The diagram above shows a series of measurements collected in the test apparatus, in which sampled values ​​were acquired over a duration of, for example, approximately 60 seconds. Here, for sampled values ​​acquired during the start-up of the test apparatus, sampled values ​​acquired in the first 20 seconds are discarded, for example.

[0057] The measurement series of force measurements acquired at 5ms intervals were divided into 61 measurement sequences, each of which had, for example, 128 sample values ​​and thus lasted for approximately 0.64 seconds.

[0058] Figure 2The chart below exemplifies the sensor signal curve over a time interval of 0.64 s, during which 128 sample values ​​were acquired. Choosing 128 sample values ​​per sequence is also advantageous because, for example, in the Discrete Fourier Transform used to obtain characterizing features from the measurement sequence, integer powers of 2 facilitate computational processing.

[0059] After training the neural network in the manner detailed below, 128 sampled values ​​corresponding to a measurement sequence are input into the nodes of the neural network's input layer. The correctly trained neural network then outputs geotechnical parameter values ​​corresponding to the input measurement sequence including the 128 sampled values; in a preferred embodiment, the output is the slump value ("Slump value"). In alternative embodiments, multiple other geotechnical parameter values ​​may also be output.

[0060] For training, i.e. supervised learning of the neural network, it is necessary to provide training data, which consists of pairs consisting of a measurement sequence of 128 force values ​​and the corresponding values ​​of the measured soil and rock parameters, i.e., the slump under the current conditions.

[0061] In a preferred embodiment, to generate training data for the neural network, a process described below is performed. First, excavated material samples with different compositions and consistencies are provided, for example, 13 samples, characterized by having a range of different slumps within a relevant range.

[0062] For each sample, geotechnical parameters, such as slump in the current case, are collected using measurement techniques. The samples are then placed sequentially into the test apparatus, where the transport flow of each sample is adjusted to a predetermined transport speed of 1.8 m / s or 2.5 m / s.

[0063] An impactor in the form of a steel ball fixed to a cable is introduced into the conveyor flow, and after the system stabilizes at a predetermined speed, a series of force sampling values ​​are acquired and stored, as shown in... Figure 2 As illustrated in the diagram above.

[0064] For each of the 13 samples, for example, a measurement series of the type shown is generated, preferably three measurement series. Each of these three measurement series is further divided into 61 measurement sequences, each with 128 sample values, thus generating 183 measurement sequences for each of the 13 excavated material samples, that is, a total of 13 × 183 = 2379 measurement sequences are generated and stored. The measured geotechnical parameters of the sample, i.e., the slump, are associated with the measurement sequences collected for that sample and stored accordingly. Thus, 2379 pairs of measurement sequences with 128 force values ​​and corresponding slump values ​​are stored.

[0065] Approximately 2000 pairs were randomly selected from a total of 2379 measurement sequence-collapse pairings as training pairs for training the neural network. About 300 randomly selected pairs from the remaining pairs were used as a validation dataset to examine the quality and suitability of the trained neural network.

[0066] The neural network was trained by using squared error as the loss function and gradient descent.

[0067] First, the weights of the neural network are preset to arbitrarily small values. Then, 128 sampled values ​​of the force curve measured within a 0.64-second time interval are input into the input layer of the neural network, and the output values ​​corresponding to the soil and rock parameters are obtained at the output layer based on the preset weights. The difference between the soil and rock parameters determined by the neural network and the actual values ​​of the soil and rock parameters measured for the sample is then determined. Based on the difference between the parameter values ​​output by the neural network and the actual values ​​measured for the sample, the weights are adjusted in a known manner. This process is repeated for all pairs consisting of force sampled values ​​from the measurement sequence and soil and rock parameters (slump) measured for the corresponding sample. The training pairs are input in a random order.

[0068] The details of supervised learning performed using the squared error loss function and gradient descent are well known to those skilled in the field of neural networks and need not be described in detail here.

[0069] The following describes the structure of an exemplary neural network, which is also used in the experimental setup. Figure 3 A schematic diagram of a neural network simulated by a data processing device is shown, in which the values ​​of geotechnical parameters (here, slump) are determined by a measurement sequence of 128 force samples. This artificial neural network first includes... Figure 3 The top row shows the 128 nodes of the input layer. Below that is shown the computational simulation of the convolution layer.

[0070] In convolution, Figure 3 The convolutional window (also called a filter, kernel, or convolutional kernel), with a predetermined width and nine weight values ​​(represented by the shaded line), is associated with the values ​​of the nine input nodes. Each weight value is multiplied by the value of a node, and the resulting nine products are summed (vector scalar product). The result is evaluated using an activation function and stored in a location known as a feature map.

[0071] Next, move the convolution window one node from the input layer. Figure 3 (From center to right), so that the nine weight values ​​of the convolution window are now associated with the values ​​of nodes 2 to 10 of the input layer (multiplied respectively and summed).

[0072] exist Figure 3 In the image below the input layer, a first convolutional window with a width of 9 is first shown below input layer nodes 1 to 9. Here, the result of the scalar product is stored in the first field of the first row of the feature map shown below the convolutional window. This is indicated by an arrow connecting the first convolutional window to the result field.

[0073] The first convolution window, moved eleven steps, is shown in dashed lines next to it on the right. This is intended to show that at this position, the nine weights of the first convolution window are associated with the values ​​of nodes 11 to 19 of the input layer, and the results are stored in the eleventh field of the first row of the feature map.

[0074] Finally, the first convolution window is shown on the far right with another dashed line. The first convolution window is positioned such that its weight values ​​are associated with nodes 120 to 128 of the input layer. At this point, the result is stored in the 120th field of the first row of the feature map, which is also shown by the corresponding arrow.

[0075] exist Figure 3 In the preferred embodiment shown, a total of 64 different convolutional windows or filters are used in the first convolutional layer, indicated by numbers 1, 2 to 64 to the left of the filtering / convolutional modules. Accordingly, results are generated through the entire convolutional operation and stored in a feature map, which has 120 fields per row in 64 rows. These operations simulate the first one-dimensional convolutional layer of a neural network. Since the convolutional layer is one-dimensional, the one-dimensional convolutional window moves across the one-dimensional input layer.

[0076] Before further processing the values ​​stored in the feature map, it is preferable to also evaluate these values ​​using a so-called activation function. This evaluation can be performed either during the computation of the values ​​(scalar product) before they are stored in the feature map, or afterward on the entire computed feature map. Both approaches are equivalent when simulating a neural network.

[0077] In the preferred embodiment shown here, the so-called ReLU activation function is used, i.e., reLU(x) = Max(0, x). This means that all negative values ​​are replaced with zero, while positive values ​​remain unchanged.

[0078] To reduce computation time and workload, it is preferable to reduce the amount of data (120×64 elements) stored in the feature map within the so-called pooling layer by obtaining a new element from a predetermined number of elements in the feature map. Figure 3 In the preferred embodiment shown, each is composed of features Figure 1The average of the 10 elements in a row is calculated and temporarily stored (in another node level). This averaging of 10 elements is exemplified by curly braces and arrows for the first 10 elements of rows 1 through 64.

[0079] A total of 12 × 64 = 768 averages are obtained. This averaging operation is also called average pooling. It is also conceivable to use a neural network that replaces average pooling with so-called max pooling in its pooling layers, where only the maximum value is selected from a predetermined number of output values ​​(also known as downsampling).

[0080] Following this pooling is another convolutional layer, such as... Figure 3 As shown in the diagram. In the other convolution, 64 different convolutional windows, each with a width of 9, are also moved sequentially over rows of 12 elements each in the pooling results. In this case, four result elements are formed in each movement of the 9-width convolutional window over the 12-element rows. Here, the 64 different convolutional windows or filters move sequentially (or possibly in parallel) over the 12-element rows, and the first, second, third, and fourth result elements of the 64 filters are summed, respectively.

[0081] This process is repeated for each of the 64 rows in the pooling result, thus forming a matrix with a total of 4 × 64 elements, such as... Figure 3 As shown in the diagram. Activation functions, preferably reLU activation functions, are also used to evaluate the results of these convolution operations. Then, in a so-called flattening process, the 4×64 evaluated convolution results are rearranged into a vector with 256 elements, where this vector is... Figure 3 It is shown as a node layer with 256 nodes.

[0082] Following this are three more layers of the neural network, with 64, 32, and 16 nodes respectively. Figure 3 The four hidden layers shown below, with 256, 64, 32, or 16 nodes, are fully connected layers, meaning that each node is connected to every node in the next node layer, where these connections are... Figure 3 It is not fully shown in the image.

[0083] Assign a weight to each connection between two nodes. Furthermore, Figure 3 Each node in the diagram also implements an activation function, preferably the reLU activation function. Following the last layer with 16 nodes is the output layer. Figure 3In the example shown, the output layer includes only one output node for outputting, for example, a value for collapse. No activation function is assigned to this node.

[0084] In the supervised learning process, initial values ​​are assigned not only to the nine weights of each of the 64 convolutional windows in the first convolutional layer and the nine weights of each of the 64 windows in the second convolutional layer, but also to the 256×64+64×32+32×16+16=18960 weights of the fully connected layer and the weights of the additional bias neurons, and then these values ​​are adjusted by appropriate methods.

[0085] Of course, various alternative configurations can be envisioned for suitable neural networks. For example, instead of 64 filters in a convolutional layer, only 16 filters can be used, omitting the pooling after the first convolutional layer. A trade-off must always be struck in network configuration: to achieve good prediction results, the number of learnable parameters (weights) should not be too small, but also not too high, otherwise overfitting may occur. [Neural network reference here] Figure 3 The structure described is the result of numerous experiments using different neural networks.

Claims

1. A method for determining at least one geotechnical parameter of the material being excavated by a tunnel boring machine, wherein... The excavated material is supplied to the conveying device (2) for conveying the excavated material. An impactor (5) is placed in the excavation material conveying flow (3) such that the impactor is at least partially submerged in the excavation material, the excavation material in the conveying flow (3) having a preset or measured flow velocity. Measure the parameters corresponding to the force applied to the impact body (5) by the excavation material from the conveying flow (3), collect the time series of the measured values ​​of the parameters, and The at least one geotechnical parameter is determined from a segment of the sequence of measured values ​​by inputting the segment of the time series of the measured values ​​into a data processing device and processing the segment by the data processing device to output at least one output value corresponding to the at least one geotechnical parameter. The processing of the measured values ​​is based on a preset flow rate, or the measured flow rate value is input and taken into account when processing the measured values.

2. The method for determining at least one geotechnical parameter according to claim 1, characterized in that, The impactor is placed in the open conveying flow of the excavation material.

3. The method for determining at least one geotechnical parameter according to claim 1 or 2, characterized in that, A predetermined flow rate is set for the transport flow and the predetermined flow rate is kept constant during the time series of measurements of the measured parameters.

4. The method for determining at least one geotechnical parameter according to any one of claims 1 to 3, characterized in that, A time series of measured values ​​is acquired at a constant sampling frequency.

5. The method for determining at least one geotechnical parameter according to any one of claims 1 to 4, characterized in that, The time series segments of the measured values ​​correspond to measurement durations of 0.25 s to 60 s, preferably 0.5 s to 1 s.

6. The method for determining at least one geotechnical parameter according to any one of claims 1 to 5, characterized in that, The data processing device implements a trained neural network with multiple layers of nodes, assigning weights to all connections between the nodes and assigning predetermined activation functions to the nodes, the weights having been pre-adjusted through a training method. An input data matrix is ​​obtained from the segments of the sequence of measured values, and the input data matrix is ​​input into the input layer of the nodes of the neural network. The at least one output value corresponding to the at least one geotechnical parameter is output by at least one node of the output layer of the neural network.

7. The method for determining at least one geotechnical parameter according to claim 6, characterized in that, A predetermined number of characterizing features are calculated from the segments of the time series of the measured values, and these features are used as an input data matrix.

8. The method for determining at least one geotechnical parameter according to claim 6, characterized in that, The time series segments of the measured values ​​are normalized into a one-dimensional input data matrix.

9. The method for determining at least one geotechnical parameter according to any one of claims 6 to 8, characterized in that, The data processing device simulates a neural network in a program-controlled manner, or the data processing device simulates a feedforward neural network with a convolutional network, or the data processing device simulates a convolutional neural network having a first convolutional layer with subsequent pooling layers and a subsequent second convolutional layer with multiple subsequent fully connected layers.

10. The method for determining at least one geotechnical parameter according to any one of claims 6 to 9, characterized in that, Before determining the at least one geotechnical parameter from the time series of input measurements, the weights are set in the training method in the following manner: a) For multiple excavation material samples with different compositions and different geotechnical parameters selected in advance, respectively: a1) Measure or determine and store the at least one geotechnical parameter of the excavated material sample as a target parameter value. a2) The impactor is placed in the conveying flow of the excavated material sample such that the impactor is at least partially submerged in the excavated material, the excavated material in the conveying flow having a predetermined flow velocity, and measurement parameters corresponding to the force exerted on the impactor by the excavated material in the conveying flow of the excavated material sample are measured, and time series of the measured values ​​of the measurement parameters are acquired. a3) Store pairs consisting of a segment of the sequence of measured values ​​and a corresponding target parameter value. b) Assigning initial values ​​to the weights in the neural network implemented by the data processing device. c) For pairings consisting of segments of a sequence of measurements and corresponding target parameter values: c1) Determining the at least one geotechnical parameter from the segment of the sequence of measured values ​​by processing the segment of the time series of measured values ​​by a data processing device implementing the neural network, such that at least one output value corresponding to the at least one geotechnical parameter is output. c2) Then determine the difference between the output value and the corresponding target parameter value. c3) Change the weights based on the difference. Specifically, for the next pairing consisting of a segment of the sequence of measured values ​​and the corresponding target parameter value, steps c1) to c3) are repeated, in which case the pairings are selected in a random order. Execution steps c1) to c3) are performed on the predetermined portion of the pair stored in step a), which is used as a training set.

11. The method for determining at least one geotechnical parameter according to any one of claims 1 to 10, characterized in that, The slump of the excavation material is determined as a geotechnical parameter.

12. A method for adjusting the consistency of a modified excavation material in an earth pressure balance shield tunneling machine, wherein, To improve the excavation material, a modifier is added to the excavation material in the excavation chamber. Modified excavation material is conveyed from the excavation chamber using a conveying device. In an open conveying flow formed on one of the conveying devices, at least one geotechnical parameter, preferably the slump, is automatically determined according to the method described in any one of claims 1 to 11. The amount of soil amendment to be added is adjusted based on the determined soil and rock parameters.

13. A method for classifying excavation materials from a tunnel boring machine, wherein, Modified excavation material is conveyed from an excavation chamber using at least one conveying device, and in an open conveying flow formed on one of the conveying devices, at least one geotechnical parameter, preferably slump, is automatically determined according to the method of any one of claims 1 to 11, and the excavation material is classified according to the determined geotechnical parameter.

14. An apparatus for determining at least one geotechnical parameter of excavated material produced in a tunnel boring machine, said apparatus being particularly for use with the method according to any one of claims 1 to 11, said apparatus comprising: A conveying device for conveying the excavated material, such that a conveying flow (3) of the excavated material with a flow rate is formed in an open conveying channel (2), the conveying device having means for setting a predetermined flow rate and / or means for measuring the flow rate. An impactor (5) is disposed in the open conveying channel (2), the impactor being at least partially submerged in the conveying flow (3) of the excavated material. A measuring device (7) connected to the impactor (5) is used to measure a parameter corresponding to the force applied to the impactor (5) by the excavation material from the conveying flow (3). A device coupled to the measuring device (7) for acquiring sampled values ​​of the measuring parameters, for converting the sampled values ​​of the measuring parameters into digital measuring values ​​and outputting a time series of the digital measuring values. Storage device for temporarily storing the sequence of digital measurements, and A data processing device coupled to the storage device, the data processing device being configured such that the data processing device processes a segment of the sequence of digital measurements and outputs at least one output value corresponding to the at least one geotechnical parameter.

15. The apparatus for determining at least one geotechnical parameter according to claim 14, characterized in that, The data processing device is configured such that it generates and processes an input data matrix from the segments of the sequence of digital measurements, and outputs at least one output value corresponding to the at least one geotechnical parameter, taking into account a predetermined flow rate or a measured flow rate value during processing. The data processing device implements a trained neural network having multiple layers of nodes, assigning weights to all connections between the nodes and assigning predetermined activation functions to the nodes, the weights being pre-set through a training method. The input data matrix is ​​input into the input layer of the nodes of the neural network, and at least one output value corresponding to the at least one geotechnical parameter is output at at least one node of the output layer of the neural network.

16. The apparatus for determining at least one geotechnical parameter according to claim 14 or 15, characterized in that, The impactor (5) is constructed as a sphere, preferably as a steel ball with a weight of 500g to 2000g, more preferably 800g to 1000g.

17. The apparatus for determining at least one geotechnical parameter according to claim 16, characterized in that, The sphere (5) is fixed to a swing-suspended rod or to a cable (6), and a sensor (7) for acquiring the measurement parameters is coupled to the rod or cable (6), the measurement parameters corresponding to the tension borne by the rod or cable (6).

18. The apparatus for determining at least one geotechnical parameter according to claim 15, characterized in that, The data processing device is configured to simulate a feedforward neural network with a one-dimensional convolutional network. Preferably, the data processing device is configured to simulate a convolutional neural network having a first convolutional layer with subsequent pooling layers and a subsequent second convolutional layer with multiple subsequent fully connected layers.

Images