A TBM ballast identification system and vibration compensation method thereof

Abstract

The present invention relates to a TBM ballast identification system and a vibration compensation method thereof, and belongs to the field of equipment control technology. The TBM ballast identification system includes: a static platform, the static platform is fixed on the top of the TBM ballast conveying device through a fixing device, and a posture detection device is fixed on the static platform; a dynamic platform, the dynamic platform is arranged below the static platform and above the TBM ballast conveying device, and a camera device is fixed on the dynamic platform; the dynamic platform and the static platform are fixedly connected through a number of actuators, and the actuators are retractable devices; a controller, the input end of the controller is connected to the posture detection device, and the output end of the controller controls and connects to each actuator; it is used to calculate the extension amount of each actuator according to the vibration data of the static platform, and then control the actuator to extend and retract. The present invention calculates the extension amount of the actuator according to the data collected by the posture detection device and then controls the movement of the actuator to adjust the posture of the dynamic platform, reduce the impact of vibration, and ensure the stability of the dynamic platform.

Description

Technical Field

[0001] The invention relates to a TBM ballast piece identification system and a vibration compensation method thereof, belonging to the technical field of equipment control. Background Art

[0002] Tunnel Boring Machine (TBM) is a tunnel construction equipment composed of multiple systems such as mechanical, electrical and hydraulic systems. During the tunnel construction process, the good operation of TBM is of great significance to the safety and efficiency of the construction. A priori understanding of the operating parameters of TBM during the excavation process can evaluate the operating status of TBM. Rock slag is a variety of slag with different morphologies such as block, flaky, and powdery produced by the rolling action of the cutter during the TBM excavation process. Its shape, size, surface roughness and other physical properties can directly reflect the mechanical state of TBM and the current geological conditions of construction. The particle size distribution statistics of rock slag produced during the excavation process have certain practical significance for the excavation status of TBM and the evaluation of surrounding rock categories.

[0003] Generally, cameras are installed on TBMs to collect and identify ballast rocks to determine the operating status of the TBM. However, due to the high vibration intensity at the construction site during the excavation of the TBM, the camera data collection is greatly disturbed and the quality of the collected images cannot be ensured, which brings great challenges to the ballast rock identification task. Therefore, it is necessary to compensate for the vibration of the camera to effectively reduce the impact of vibration on the image quality.

[0004] In the prior art, vibration compensation technology already exists in certain fields, for example: the Chinese invention patent application document with application publication number CN106444862 A, which discloses a six-degree-of-freedom wave compensation platform control method and system, in which six servo cylinders are used to form a six-degree-of-freedom wave compensation platform, and multiple sensors are installed to measure and collect the motion information of the ship, and the motion signal is converted into a drive signal to drive the compensation mechanism to reduce the impact of wave impact. This patent processes the current ship motion data and outputs a control signal, and the system has a certain lag; the ship's motion information is converted into wave impact information. When the motion information changes dramatically, the compensation system has poor robustness and untimely response, and the patent does not provide a system structure framework and specific implementation technology. The vibration frequency in TBM is high, and more direct collection and faster prediction are required. Therefore, this method is not suitable for vibration compensation of TBM ballast identification.

[0005] Another example: the Chinese invention patent application document with application publication number CN111994201A discloses a compensation method based on a wave compensation device. This compensation method estimates and updates the prediction models of the fixed platform and the moving platform in combination with the fixed platform posture prediction for the influence of waves to achieve dual feedforward control. This patent requires the dynamic platform and the static platform posture to be predicted separately, and the prediction algorithm given is a recursive calculation, which results in a slow system response due to the large amount of calculation. At the same time, from the application scenario, the impact of waves on ships is low-frequency, large in amplitude, and the allowable time for signal acquisition and processing is long. In addition, the wave impact signal cannot be directly obtained, and the motion information of the ship needs to be measured for compensation. In TBM vibration compensation, the vibration signal has the characteristics of high frequency, small amplitude, and obvious nonlinearity. At the same time, in the TBM ballast recognition system, the vibration signal can be directly obtained through the vibration sensor. Therefore, the requirements and applicability of the signal to the algorithm under the two working conditions are essentially different, and this method is not suitable for vibration compensation of TBM ballast recognition.

[0006] Another example: the Chinese invention patent document with the authorization announcement number CN 106791417 B discloses a dual-channel camera anti-shake system for the nacelle of a wind turbine. It uses a dual-slide sliding method and a gear rack transmission adjustment method to monitor the shaking phenomenon of the nacelle of the wind turbine, uses multi-sensor fusion to collect information, and uses multi-channel servo motors for control. The dual-slide structure used in this patent needs to rely on motors and gear transmissions in the yaw and pitch directions to adjust its position in space, which is not flexible. The vibration sensor only focuses on the shaking data in the X-axis and Y-axis directions, ignoring the Z-axis data, and uses interpolation planning to generate the motion trajectory when performing compensation motion. Not only is the calculation complicated, but the compensation structure is also inflexible. In the TBM ballast recognition task, the vibration signal is relatively dense, and the overall structure needs to be flexible. Therefore, this method is not suitable for vibration compensation of TBM ballast recognition.

[0007] Another example: the Chinese invention patent application document with application publication number CN112287552A discloses an aviation engine trend analysis method, which uses the engine's vibration fundamental frequency and speed data to screen the quality factor as the steady-state trend of vibration data analysis. The vibration data of the engine during normal operation is stable. This method focuses on the steady-state characteristics of the system and uses non-stationary data and nonlinear data as fault signals. However, the vibration signals normally generated by the TBM during the excavation process are nonlinear and time-varying, and do not have steady-state characteristics. Therefore, this method is not suitable for TBM ballast identification.

[0008] Another example: the journal article in the journal "Flow Transmission and Control" with the journal number of 2009(1):16-18, which discloses a PID control method based on repeated control compensation for a hydraulic servo vibration table, but the steady-state accuracy of the system is not high. Another example: the journal article in the journal "Manufacturing Technology and Machine Tools" with the journal number of 2019,000(005):72-75, which discloses a study on a PID control method for displacement compensation of a hydraulic vibration table, in which a double closed-loop control hydraulic vibration table is established, the displacement and pressure value signals are detected, and the PID-based control improves the displacement control accuracy to a certain extent through PID online compensation, but it shows performance defects such as low adaptive force and response lag when the TBM vibration signal is obviously nonlinear.

[0009] In summary, various existing compensation technologies are not suitable for TBM ballast identification. Therefore, it is necessary to propose a technical solution for vibration compensation suitable for TBM ballast identification. Summary of the invention

[0010] The purpose of this application is to provide a TBM ballast identification system and a vibration compensation method thereof, and to propose an effective technical solution for improving the identification quality of TBM ballast.

[0011] To achieve the above objectives, the present application proposes a technical solution for a TBM ballast identification system, including:

[0012] Static platform: The static platform is fixed on the top of the TBM ballast conveying device through a fixing device. A posture detection device is fixed on the static platform to detect the posture change of the static platform to obtain the vibration data of the static platform;

[0013] The moving platform is arranged below the static platform and above the TBM ballast conveying device. A camera is fixed on the moving platform to photograph the TBM ballast.

[0014] The moving platform and the static platform are fixedly connected via a plurality of actuators, wherein the actuators are retractable devices;

[0015] The controller has an input end connected to the posture detection device, and an output end of the controller controls and connects to each actuator; the controller is used to calculate the extension and retraction amount of each actuator according to the vibration data of the static platform, and then control the actuator to extend and retract to adjust the posture of the dynamic platform and realize vibration compensation action.

[0016] The beneficial effect of the technical solution of the TBM ballast identification system of the present invention is as follows: the present invention installs a static platform above the TBM ballast conveying device, fixes a posture detection device on the static platform, and connects a moving platform provided with a camera device through a retractable actuator. The controller calculates the telescopic amount of the actuator based on the data collected by the posture detection device and then controls the movement of the actuator to adjust the posture of the moving platform, reduce the impact of vibration, and ensure the stability of the moving platform.

[0017] Furthermore, the posture detection device includes a three-axis vibration sensor and a gyroscope.

[0018] Furthermore, the actuator is fixed to the static platform and the dynamic platform through a connecting device, and the connecting device is fixed to each platform in the form of a polygonal vertex. The actuator includes a telescopic part and a fixed part, the telescopic part is connected to the connecting device of the dynamic platform, and the fixed part is connected to the connecting device of the static platform.

[0019] Furthermore, the fixing device includes connecting frames arranged on both sides of the TBM ballast conveying device and hanging frames connected to the connecting frames, and the static platform is fixed on the hanging frames.

[0020] In addition, the present application also proposes a technical solution for a vibration compensation method applicable to the above-mentioned TBM ballast identification system, comprising the following steps:

[0021] 1) Obtaining vibration data of the static platform at the current time and in the previous period; the vibration data is N-degree-of-freedom data including position information and posture information;

[0022] 2) inputting the acquired vibration data into a trained recurrent neural network model to predict the vibration data in the next period; the trained recurrent neural network model is trained by historical static platform vibration data;

[0023] 3) Calculate the compensation posture of the moving platform based on the predicted vibration data;

[0024] 4) According to the compensation posture of the moving platform and the pre-established kinematic model of the TBM ballast piece identification system, the coordinates of the connection points between each actuator and the moving platform at the compensation posture at the next moment are obtained; the kinematic model is established by parameterizing a parallel structure including the moving platform, the static platform and several actuators;

[0025] 5) The coordinates of the connection points between each actuator and the moving platform at the compensation posture at the next moment are combined with the coordinates of the connection points between each actuator and the static platform at the current moment to obtain the length of each actuator corresponding to the compensation posture at the next moment, and the difference between the length of each actuator corresponding to the compensation posture at the next moment and the length of each actuator at the current moment is the extension amount of each actuator;

[0026] 6) Control each actuator to move with the extension amount at the next moment to achieve vibration compensation of the moving platform.

[0027] The technical solution of the vibration compensation method of the TBM ballast identification system of the present invention has the beneficial effect that: the present invention establishes a kinematic model and a recurrent neural network model of the identification system, inputs the vibration data obtained at the current moment and in the previous period into the recurrent neural network model to obtain the vibration data in the next period, and then calculates the compensation posture of the moving platform according to the vibration data combined with the kinematic model, and obtains the extension and contraction amount of each actuator through the compensation posture, controls the movement of each actuator, and realizes the vibration compensation of the moving platform. The present invention can solve the hysteresis problem of system compensation, and the calculation is simple, which improves the system efficiency.

[0028] Furthermore, the step 1) also includes a process of preprocessing and standardizing the vibration data, and the preprocessing includes the steps of screening and cleaning the vibration data.

[0029] Furthermore, the recurrent neural network model in step 2) is a GRU neural network model.

[0030] Furthermore, the calculation process of the compensation posture of the moving platform in step 3) is:

[0031] C qT =C cur -Tδ[y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTβ ,y ΔTγ ] T ;

[0032] Among them, C qT is the compensation posture of the moving platform within T time after the current moment; T is the prediction time interval; δ is the vibration coefficient; C cur is the position of the moving platform at the current moment; [y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTβ ,y ΔTγ ] T The predicted filtered six-degree-of-freedom vibration data.

[0033] Furthermore, the length of each actuator corresponding to the compensation posture at the next moment is:

[0034] l Ti =B TAi -A i , i=1,2,…,6;

[0035] Among them, l Ti B is the length of the i-th actuator corresponding to the compensation posture at the next moment; TAi A is the coordinate of the connection point between the ith actuator and the moving platform in the compensation posture in the static coordinate system at the next moment; i is the coordinate of the connection point between the ith actuator and the static platform in the static coordinate system at the current moment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a structural diagram of the TBM ballast piece identification system of the present invention;

[0037] Figure 2 is a compensation flow chart of the vibration compensation method of the present invention;

[0038] Figure 3 It is a schematic diagram of establishing coordinates of the TBM ballast piece identification system of the present invention;

[0039] Figure 4 It is a training block diagram of the GRU neural network model of the present invention;

[0040] In the figure: 1 is the ballast conveyor belt, 2 is the ballast to be identified, 3 is the connecting frame, 4 is the suspension frame, 5 is the camera stabilization device, 6 is the static platform, 7 is the three-axis vibration sensor, 8 is the gyroscope, 9 is the operation processor, 10 is the actuator, 11 is the dynamic platform, 12 is the six-degree-of-freedom platform controller, 13 is the industrial camera, 14 is the industrial light source, and 15 is the industrial lens. DETAILED DESCRIPTION

[0041] TBM ballast identification system implementation example:

[0042] The main idea of ​​the present invention is that, in order to solve the problem that the vibration intensity is high and the camera stability is poor during TBM construction, which affects the quality of pictures collected by the camera and further affects the recognition effect of slag, the present invention provides a camera stabilization device that can adjust the camera position according to the vibration signal. The device includes a vibration sensor, a gyroscope and a plurality of actuators. The vibration data of the camera for a period of time before the current moment is collected by the vibration sensor and the gyroscope, and then the vibration data of the camera for a period of time after the current moment is predicted. The extension and contraction amount of the actuator in the stabilization device is adjusted according to the predicted vibration data, and then the stability of the camera is adjusted, thereby providing a technical basis for further research on TBM slag recognition.

[0043] Specifically, the TBM ballast identification system is as follows: Figure 1 As shown, it includes a camera device and a camera stabilization device 5, and the camera device includes an industrial camera 13, an industrial light source 14, and an industrial lens 15. The camera stabilization device 5 includes a static platform 6, a three-axis vibration sensor 7, a gyroscope 8, a computing processor 9, a plurality of actuators 10, a dynamic platform 11, and a six-degree-of-freedom platform controller 12.

[0044] The camera stabilization device 5 is fixed above the ballast conveyor belt 1 through the connecting frame 3 and the hanging frame 4. The connecting frame 3 is fixed on both sides of the ballast conveyor belt 1, and the hanging frame 4 is connected to the two connecting frames 3. The static platform 6 (fixed by bolts) of the camera stabilization device 5 is fixed on the hanging frame 4. The static platform 6 is fixed with a three-axis vibration sensor 7, a gyroscope 8 and an operation processor 9. The static platform 6 and the dynamic platform 11 are fixedly connected through an actuator 10. The six-degree-of-freedom platform controller 12, the industrial camera 13, the industrial light source 14, and the industrial lens 15 are arranged on the dynamic platform 11 to collect the image information of the ballast 2 to be identified at a top-down angle. The connecting frame 3 and the hanging frame 4 serve as fixing devices for fixing the static platform 6. Its structure is not limited to the structure proposed in this embodiment, and can be improved as needed. The present invention does not limit this.

[0045] The ballast conveyor belt 1 is a device for transporting ballast inside the TBM. The ballast 2 to be identified is ballast of different shapes produced by the cutterhead during TBM excavation, and is the visual detection target in the ballast identification task.

[0046] The three-axis vibration sensor 7 and the gyroscope 8 on the static platform 6 are used as posture detection devices to collect vibration information at the TBM site and the posture information of the static platform 6. The three-axis vibration sensor 7 and the gyroscope 8 are connected to the input end of the operation processor 9, and the collected information is transmitted to the operation processor 9 as vibration data. The operation processor 9 is used to predict the vibration trend.

[0047] The actuator 10 can be an electric cylinder, a hydraulic cylinder or other device that can adjust the telescopic amount to realize the movement of the moving platform 11 in all directions, with simple control and flexible operation. Since both the static platform 6 and the moving platform 11 are hexagonal in shape, and each hexagonal platform is provided with a ball joint at its six vertices, six actuators 10 are provided, and the cylinder body of each electric cylinder (the cylinder body is the fixed part of the actuator 10) is connected to the ball joint of the static platform 6, and the piston rod of the electric cylinder (the piston rod is the telescopic part of the actuator 10) is connected to the ball joint of the moving platform 11. The industrial camera 13 and the industrial lens 15 cooperate to shoot the ballast 2 to be identified on the ballast conveyor belt 1, and the industrial light source 14 is used to increase the light intensity and improve the shooting brightness. Of course, the ball joint as a connecting device between the actuator 10 and each platform can also be replaced by other universal couplings, such as: cross hinge, tripod type, etc. The present invention does not limit the structure of the connecting device. Meanwhile, the specific shapes of the static platform 6 and the dynamic platform 11 may also be circular or square, etc., and the present invention does not impose any limitation on this.

[0048] The six-degree-of-freedom platform controller 12 is connected to the computing processor 9 in communication, and the six-degree-of-freedom platform controller 12 controls the connected actuators 10. The six-degree-of-freedom platform controller 12 receives the predicted vibration data of the computing processor 9, and solves the six actuators 10 according to the predicted vibration data, and then realizes the movement of the dynamic platform 11 in six degrees of freedom (X, Y, Z, α, β, γ) in three-dimensional space by controlling the telescopic movement of the six actuators 10, thereby realizing vibration compensation and improving the quality of the captured pictures. Of course, the six-degree-of-freedom platform controller 12 and the computing processor 9 can also be combined into one controller, and the present invention does not limit this.

[0049] The number of the actuators 10 and the shapes of the static platform 6 and the dynamic platform 11 can be set as required, and the present invention does not impose any limitation thereto.

[0050] The six-degree-of-freedom platform is the type of vibration table selected for the present invention (the dynamic platform 11 is a vibration table). When applied, different types of vibration tables such as three-degree-of-freedom, four-degree-of-freedom, and five-degree-of-freedom can be selected according to specific working conditions. The signal transmission between the operation processor 9 and the six-degree-of-freedom platform controller 12 can be through serial ports, network cables, wireless transmission, etc.

[0051] The calculation processor 9 and the six-degree-of-freedom platform controller 12 implement the vibration compensation method of the TBM ballast piece identification system through calculation. Specifically, the vibration compensation method is as follows: Figure 2 As shown, the following steps are included:

[0052] 1) Establish a kinematic model of the camera stabilization device 5.

[0053] The model structure is as follows Figure 3 As shown, it consists of a static platform 6 and two upper and lower platforms of a dynamic platform 11, and six actuators 10. Sensors and processors are mounted on the two platforms respectively. Through the telescopic movement of the six actuators 10, the dynamic platform 11 can move in six degrees of freedom (X, Y, Z, α, β, γ) within the space, and then the following kinematic model is established. The kinematic model is established by parameterizing the parallel structure including the dynamic platform 11, the static platform 6 and several actuators 10.

[0054] The static platform 6 and the dynamic platform 11 respectively use six ball joints and other connecting devices as vertices to form a convex polygon with inner vertices in the circle with radius r and R. For the upper and lower platforms, the plane midpoint passes through the midpoint of the plane, with the midpoint of the plane as the origin, the plane midpoint of the static platform 6 as the origin o, and the plane midpoint of the dynamic platform 11 as the origin O, then:

[0055] The coordinate vector A of the vertex in the static platform 6 in the static coordinate system (x, y, z) i =[A ix ,A iy ,A iz ] T;

[0056] Coordinate vectors B of each hinge point in the moving platform 11 in the moving coordinate system (X, Y, Z) i =[B ix ,B iy ,B iz ] T ,i=1,2,…6.

[0057] Assume that the posture transformation from the static coordinate system to the moving coordinate system is that the coordinate point rotates around the x-axis, y-axis, and z-axis in the coordinate system o-xyz at angles α, β, and γ respectively, then the coordinate transformation matrix R from the moving coordinate system to the static coordinate system is obtained as:

[0058]

[0059] The coordinate transformation matrix R is simplified to:

[0060]

[0061] Assume that the position of the origin of the moving coordinate system relative to the static coordinate system is P o =[X,Y,Z] T , then any point P in the moving coordinate system B Point P in static coordinates A For: P A =R·P B +P o , transform the hinge coordinates of the moving platform 11 in the moving coordinate system to the static coordinate system, and obtain the length vector l of each actuator 10 in the static coordinate system i for:

[0062] l i =B Ai -A i =(R·B i +P 0 )-A i =[m i1 ,m i2 ,m i3 ] T , i=1,2,…6;

[0063] Among them, B Ai is the coordinate of the hinge point on the moving platform 11 in the static coordinate system, and then the length of each actuator is obtained. i |For:

[0064] The six degrees of freedom have the advantages of simple structure, high rigidity, high precision and high load capacity. By controlling the six actuators 10, the camera can move in six degrees of freedom (X, Y, Z, α, β, γ) in three-dimensional space. The kinematic inverse solution is used to compensate the movement and is easy to program.

[0065] 2) Establish a recurrent neural network model, train the recurrent neural network model, and obtain a trained vibration prediction model.

[0066] According to the time series and nonlinearity of TBM vibration data, combined with relevant theories and technologies, a GRU neural network model based on GRU is given. The entire GRU neural network model framework and training process are as follows Figure 4 As shown, there are input layer, output layer and hidden layer. Of course, long short-term memory neural network can also be used.

[0067] The GRU neural network model is trained to obtain the optimal trained GRU neural network model, which is stored in the operation processor 9, and the trained parameters can be directly used to predict the vibration trend of the input vibration data. The output value of the hidden neuron is: y=W·x+b, where y represents the output value, x represents the input value, W represents the weight value, and b represents the bias value.

[0068] The loss function during training is:

[0069]

[0070] Among them, loss represents the loss value, y i is the output value of each neuron, y reat Represents the real original data.

[0071] Finally, the Adam algorithm is used to dynamically optimize the learning rate. The Adam algorithm can handle sparse gradients and non-stationary targets and is suitable for multi-dimensional high-frequency vibration signals. The loss function is used to evaluate the gap between the neural network target and the actual output. The smaller the function value, the smaller the difference between the actual output and the target output, that is, the more appropriate the weight. When the loss value is ≤e, stop training.

[0072] 3) The vibration data of the current time and the previous time period are obtained through the three-axis vibration sensor 7 and the gyroscope 8.

[0073] The vibration signal generated by the TBM is measured in real time by the three-axis vibration sensor 7, which is three-dimensional position data, that is, the magnitude of the vibration signal in the directions of the static platform 6 (x, y, z). At the same time, the posture change of the static platform 6 (α, β, γ) is measured by the gyroscope 8, which is also three-dimensional data. Therefore, the signals measured by the three-axis vibration sensor 7 and the gyroscope 8 together constitute a six-dimensional vibration data (x, y, z, α, β, γ).

[0074] 4) The vibration data collected in step 3) is input into the operation processor 9, and the vibration data in the next time period is predicted according to the trained GRU neural network model.

[0075] The processing of the operation processor 9 includes collecting raw data, data preprocessing, data standardization, time series data prediction, prediction data processing, and prediction result output.

[0076] In this step, the processor 9 performs data preprocessing on the collected multi-dimensional vibration data in the form of screening and cleaning, so that the vibration data is arranged in a certain time series, there is no wrong time point or time series confusion, and the integrity and correctness of the data are guaranteed. The preprocessed vibration data is in the form of a sequence:

[0077] X=(X 1 ,X 2 ,…X n ),X n =[X nx ,X ny ,X nz ,X nα ,X nβ ,X nγ ] T ;

[0078] Among them, each vibration data X n All are six-dimensional data, (X nx ,X ny ,X nz ,X nα ,X nβ ,X nγ ) is the amount of vibration data acting on the platform in the (x, y, z) direction and (α, β, γ) angle.

[0079] Then the preprocessed vibration data X is segmented into pieces of length L, and the segmented vibration data is standardized using the Z-score method. After data scaling, the data value will fall within a certain range. At this time, using it as input data for the network model can improve the efficiency and accuracy of the neural network.

[0080] The Z-score standardization formula is:

[0081]

[0082]

[0083]

[0084] Among them, S X is the standardized data series, is the average value of the sequence, S is the sequence X i The standard deviation of the data after Z-score processing is 0 and the variance is 1.

[0085] Input the standardized vibration data into the trained GRU neural network model In the above example, the standardized vibration data S is X The length is L, S X ={S n-L+1 ,S n-L+2 ,…,S n}; The corresponding output result is:

[0086]

[0087] It can be seen that the predicted value at the n+1th moment is Y n+1 , and then input data S X The last (L-1) data and Y n+1 Merge into a new row of data: S X+1 ={S n-L+2 ,S n-L+3 ,…,S n ,Y n+1}, then the new sequence data S X+1 Input to The predicted value at the n+2th moment is Y n+2 , and so on, the predicted sequence is Y nT = {Y n+1 ,Y n+2 ,…,Y n+T}, T is the period for predicting the future vibration trend, which is determined based on how long the vibration trend needs to be predicted in the future.

[0088] The predicted values ​​in the prediction sequence are the data types after data standardization, so it is necessary to perform an inverse operation of standardization to restore the predicted data to the same type as the original data. The formula for inverse standardization is as follows:

[0089]

[0090] Among them, y nT For the predicted data that conforms to the original data type, Y nT For S X go through The predicted data calculated by the model, S is the sequence X i The standard deviation of is the average value of the series.

[0091] The GRU neural network structure has good environmental adaptability and nonlinear approximation capabilities. Using the GRU structure to predict the vibration trend of TBM can solve the defects of low system adaptability and delayed response when using traditional methods such as PID.

[0092] 5) According to the vibration data predicted in step 4), the predicted position and predicted posture (collectively referred to as predicted posture) and the compensated posture of the moving platform 11 are obtained.

[0093] Predict the vibration data y within time T nT ={y n+1 ,y n+2 ,…,y n+T},y n =[y nx ,y ny ,y nz ,y nα ,y nβ ,y nγ ] T At this time, the predicted vibration data is the predicted trend of the static platform 6 within the time T, based on the fixed relationship between the dynamic platform 11 and the static platform 6, that is, the predicted vibration data is the predicted trend of the dynamic platform 11. Therefore, the predicted posture of the dynamic platform 11 can be obtained by combining the current posture of the dynamic platform 11 and the predicted vibration data.

[0094] Regarding the value of T, when the value of T is small, the accuracy of the system compensation displacement can be guaranteed, and the predicted vibration data can be filtered. The filtering process is to take the average value of the vibration data within the time T:

[0095]

[0096] Among them, y ΔT Predicted filtered vibration data, y ΔT =[y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTβ ,y ΔTγ ] T , the result is transmitted to the six-freedom platform controller, and combined with the current position of the moving plane, the position of the moving plane 11 after T in the future under the influence of vibration can be predicted as:

[0097]

[0098] Among them, CT is the predicted position of the moving platform 11 within T time after the current moment; δ is the vibration coefficient, which is the corresponding motion parameter between the vibration data and the displacement of the moving platform 11 under the influence of vibration; C cur is the position and posture of the moving platform 11 at the current moment; [y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTβ ,y ΔTγ ] T It is the mean value of the vibration trend within time T predicted by the neural network based on the vibration data of the current and past periods.

[0099] In order to achieve vibration compensation, the compensation posture of the moving platform is:

[0100] C qT =C cur -Tδ[y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTβ ,y ΔTγ ] T ;

[0101] Among them, C qT is the compensation posture of the moving platform within T time after the current moment, which is the same as C T Corresponding and opposite direction postures.

[0102] The reasons for the above filtering process are: 1. The present invention focuses on the vibration compensation within the unit time T as an action result, rather than a continuous action result, so it is necessary to convert y nT Consider it as a whole to find out the impact on the moving platform 11; 2. The vibration signal is nonlinear, and the mean filter can be used to extract the useful signal value that best represents this period of time from the vibration signal within the unit T.

[0103] 6) Based on the compensated posture of the moving platform 11 obtained in step 5), the kinematic inverse solution is performed in combination with the motion model to obtain the length of each actuator 10 at the next moment. The difference between the length of each actuator 10 at the next moment and the length of each actuator 10 at the current moment is the extension and retraction amount of each actuator 10.

[0104] In this step, the kinematic inverse solution is obtained, that is, the coordinate transformation matrix from the point in the moving platform coordinate system to the upper platform coordinate system is derived according to the moving and static coordinate systems and the Euler angle rotation transformation formula, and then the coordinates of the lower fixed points of the six actuators 10 in the static coordinate system at any time are obtained, and the lengths of the six actuators 10 are solved.

[0105] Specifically, first, under the effect of the predicted vibration data, the position and posture of the moving platform 11 to be compensated are:

[0106] C qT =[C qTx ,C qTy ,C qTz ,C qTα ,C qTβ ,C qTγ ] T ;

[0107] Next, combined with the kinematic model established in step 1), the coordinate vector B of the connection point between each actuator 10 and the moving platform 11 in the moving coordinate system corresponding to the compensation posture at the next moment can be obtained: Ti =[B Tix ,B Tiy ,B Tiz ] T , i = 1, 2, ... 6, and the coordinate origin of the moving platform P To =[X T ,Y T ,Z T ] T , B in the initial kinematic model i and P o According to the coordinate transformation matrix R, the coordinates B of the connection point between each actuator 10 and the moving platform 11 in the compensation posture at the next moment in the static coordinate system are obtained. TAi For B TAi =R·B Ti +P To , the length vector l of each actuator 10 corresponding to the compensation posture at the next moment Ti for:

[0108] l Ti =B TAi -A i =(R·B Ti +P To )-A i ,(i=1,2,…,6);

[0109] Among them, l Ti B is the length of the i-th actuator 10 corresponding to the compensation posture at the next moment; TAi A is the coordinate of the connection point between the i-th actuator 10 and the moving platform 11 in the compensation posture in the static coordinate system at the next moment; i is the coordinate of the connection point between the ith actuator 10 and the static platform 6 in the static coordinate system at the current moment.

[0110] The length of each actuator 10 at the next moment is calculated according to the space vector two-norm formula. Ti |For: Then, under the influence of the predicted vibration trend, the expansion and contraction amount of each actuator 10 to achieve the compensation effect is |Δl i |=|l Ti |-|l oi |, (i=1,2,3,…,6), (|l oi | is the length of each actuator 10 at the current moment).

[0111] 7) According to the obtained extension and contraction amount of each actuator 10 , each actuator 10 is controlled to move at the next moment to achieve vibration compensation of the moving platform 11 .

[0112] In the TBM ballast identification, the present invention installs a camera stabilization device 5, and establishes a kinematic model for the camera stabilization device 5 to analyze its kinematics and inverse kinematics. At the same time, a GRU neural network model for predicting vibration trends is established, and the vibration compensation of the moving platform 11 is realized by combining the GRU neural network model and the kinematic model. The camera stabilization device 5 of the present invention is installed inside the TBM, and the vibration signal can be directly obtained through the vibration sensor, which is more accurate. Compensation is performed through vibration trend prediction, which can solve the lag problem of the system. At the same time, the six-freedom platform adopted by the present invention has a simple structure and high rigidity, and can easily realize the movement of the camera in six degrees of freedom in three-dimensional space, and the compensation movement can be completed by only controlling the extension and contraction amount of the six actuators 10. The number of sensors and drivers required is small, and the calculation process of the compensation amount is simple and efficient, which improves the system efficiency.

[0113] Embodiment of the vibration compensation method of the TBM ballast identification system:

[0114] The specific implementation process of the vibration compensation method has been introduced in the above-mentioned TBM ballast identification system embodiment and will not be repeated here.

Claims

1. A TBM ballast identification system, characterized in that: It includes a static platform, a dynamic platform and a controller; the dynamic platform and the static platform are fixedly connected through a plurality of actuators, and the actuators are retractable devices; the input end of the controller is connected to a posture detection device, and the output end of the controller controls and connects to each actuator; the static platform is fixed above the TBM ballast conveying device through a fixing device, and a posture detection device is fixed on the static platform to detect the posture change of the static platform to obtain vibration data of the static platform; the dynamic platform is arranged below the static platform and above the TBM ballast conveying device, and a camera is fixed on the dynamic platform to photograph the TBM ballast; the controller is used to calculate predicted vibration data according to the vibration data of the static platform, calculate predicted posture according to the predicted vibration data and the current posture, calculate the compensation posture in the opposite direction to the predicted posture in combination with the current posture of the dynamic platform, and then perform kinematic inverse solution to obtain the extension and retraction amount of each actuator, and then control the actuator to extend and retract to adjust the posture of the dynamic platform to achieve vibration compensation action.

2. The TBM ballast identification system according to claim 1, characterized in that: The posture detection device includes a three-axis vibration sensor and a gyroscope.

3. The TBM ballast identification system according to claim 1, characterized in that: The actuator is fixed to the static platform and the dynamic platform through a connecting device, and the connecting device is fixed to each platform in the form of a polygonal vertex. The actuator includes a telescopic part and a fixed part. The telescopic part is connected to the connecting device of the dynamic platform, and the fixed part is connected to the connecting device of the static platform.

4. The TBM ballast identification system according to claim 1, characterized in that: The fixing device includes a device arranged on the TBM The ballast conveying device has connecting frames on both sides and hanging frames connected with the connecting frames, and the static platform is fixed on the hanging frames.

5. A vibration compensation method for a TBM ballast identification system according to claim 1, characterized in that: The following steps are involved: 1) Obtaining vibration data of the static platform at the current time and in the previous period; the vibration data is N-degree-of-freedom data including position information and posture information; 2) inputting the acquired vibration data into a trained recurrent neural network model to predict the vibration data in the next period; the trained recurrent neural network model is trained by historical static platform vibration data; 3) Calculate the compensation posture of the moving platform based on the predicted vibration data; 4) According to the compensation posture of the moving platform and the pre-established kinematic model of the TBM ballast piece identification system, the coordinates of the connection points between each actuator and the moving platform at the compensation posture at the next moment are obtained; the kinematic model is established by parameterizing a parallel structure including the moving platform, the static platform and several actuators; 5) The coordinates of the connection points between each actuator and the moving platform at the compensation posture at the next moment are combined with the coordinates of the connection points between each actuator and the static platform at the current moment to obtain the length of each actuator corresponding to the compensation posture at the next moment, and the difference between the length of each actuator corresponding to the compensation posture at the next moment and the length of each actuator at the current moment is the extension amount of each actuator; 6) Control each actuator to move with the extension amount at the next moment to achieve vibration compensation of the moving platform.

6. The vibration compensation method of the TBM ballast identification system according to claim 5, characterized in that: The step 1) also includes a process of preprocessing and standardizing the vibration data, wherein the preprocessing includes the steps of screening and cleaning the vibration data.

7. The vibration compensation method of the TBM ballast identification system according to claim 5, characterized in that: The recurrent neural network model in step 2) is a GRU neural network model.

8. The vibration compensation method of the TBM ballast identification system according to claim 6, characterized in that: The calculation process of the compensation posture of the moving platform in step 3) is: C qT =C cur -Tδ[and ΔTx ,and ΔTy ,and ΔTz ,and ΔTα ,and ΔTβ ,and ΔTγ ] T ; Among them, C qT is the compensation posture of the moving platform within T time after the current moment; T is the prediction time interval; δ is the vibration coefficient; C cur is the position of the moving platform at the current moment; [y ΔTx ,y ΔTy ,y ΔTz ,y ΔTα ,y ΔTB ,y ΔTγ′ ] T The predicted filtered six-degree-of-freedom vibration data.

9. The vibration compensation method of the TBM ballast identification system according to claim 5, characterized in that: The length of each actuator corresponding to the compensation posture at the next moment is: l Ti =B TAi -A i ,i=1,2,...,6; Among them, l Ti B is the length of the i-th actuator corresponding to the compensation posture at the next moment; TAi A is the coordinate of the connection point between the ith actuator and the moving platform in the compensation posture in the static coordinate system at the next moment; i is the coordinate of the connection point between the ith actuator and the static platform in the static coordinate system at the current moment.

Images