Intelligent machining method for injection hole of liquid rocket engine

Abstract

The invention belongs to the field of mechanical engineering, and discloses an intelligent machining method for injection holes of a liquid rocket engine. A system used in the intelligent machining method comprises a five-axis numerical control machining tool provided with a system capable of reading data information, a line laser sensor capable of measuring height information of an object, a force measuring cutter handle capable of reading force signals in real time in the machining process and the like. The injection disc of the rocket engine is fixed on a numerical control machine tool workbench through the clamp, so that the position change in the machining process is prevented; a high-precision tool is used for connecting a line laser sensor with a cutter handle, and a workpiece is scanned to obtain three-dimensional point cloud information of the injection hole; solving processing parameters through point cloud reconstruction; a force measuring cutter handle is adopted for machining, and the machining process is regulated and controlled according to real-time force signals. According to the intelligent machining method, the problem that workpiece machining parameters are difficult to determine in a traditional machining method is abandoned, the machining precision can be remarkably improved, and the labor intensity of workers is reduced.

Description

Technical Field

[0001] The invention belongs to the field of mechanical engineering and relates to an intelligent processing method for a liquid rocket engine injection hole. Background Art

[0002] As a key component of a liquid rocket engine, the injection plate has hundreds of injection holes distributed in a circular array. The machining accuracy of the injection holes can directly affect the fuel atomization combustion efficiency and the stability of the rocket's flight attitude. The injection holes are brazed by a copper substrate and a steel nozzle. Affected by multiple factors such as the low machining accuracy of the copper substrate array holes, the poor consistency of the manual assembly posture of the copper substrate and the steel nozzle, and the solder flow in the gap between the substrate and the steel pipe during the brazing process, the spatial posture of the injection holes shows a differentiated and irregular offset from the theoretical coordinates. The chamfer heights of the steel pipe parts of the injection holes are different, and the conventional CNC machining method based on the design model to generate machining codes cannot be used. At present, the chamfering of the injection holes is generally completed manually by a radial drilling machine, which requires a high level of experience for the workers, has a long machining cycle, and is difficult to ensure product consistency. Therefore, there is an urgent need for an intelligent machining method for rocket engine injection holes that can improve production efficiency and the overall machining quality of the injection plate.

[0003] Zhejiang University disclosed a method for finishing the intersection holes of outer wings in its invention patent "A finishing method for outer wing docking", CN113000881A. Although this method can provide higher measurement accuracy, it has complex operation links, more manual operation links, and low processing speed. China Construction Eighth Engineering Bureau Co., Ltd. disclosed a method for chamfering the connecting holes that can be straddled on the connecting plate in its invention patent "Chamfering processing equipment and chamfering processing method for connecting holes on connecting plates", CN118060632A. The spatial coordinates of the connecting holes are strictly distributed according to the array, and the chamfering processing amount of each hole is a fixed value. Xi'an Qunjian Aviation Precision Manufacturing Co., Ltd. disclosed an operation of obtaining the first coordinate value and the second coordinate value corresponding to each of a plurality of array holes in its invention patent "A method for processing array holes", CN113449393A. The method performs actual processing based on the theoretical position and does not consider the posture offset of the hole position.

[0004] None of the above studies mentioned an intelligent processing method for liquid rocket engine injection holes. Summary of the invention

[0005] The present invention mainly aims at the problems of strong manual dependence, long processing cycle and low precision caused by the personalized differences of the injection holes in the existing liquid rocket engine injection hole processing process, and proposes an intelligent processing method for the injection holes of liquid rocket engines. The method uses a line laser sensor to obtain the actual spatial posture of the injection hole of the injection disk, calculates the characteristic parameters of the injection hole to be processed based on the point cloud reconstruction method, and relies on the secondary development function module of the numerical control system to realize the automatic generation of the processing G code; the cutting force trend prediction model of the processing process is trained to assist in the realization of real-time feedback control of the cutting process.

[0006] The technical solution of the present invention:

[0007] Firstly, the workpiece, i.e., the liquid rocket engine injection plate, is fixed on the worktable of a five-axis CNC machine tool by an adaptive centering fixture, and the relative position relationship between the workpiece coordinate system and the machine tool coordinate system is determined;

[0008] Secondly, the injection disk is scanned at multiple angles using a line laser sensor to ensure the integrity of the three-dimensional point cloud data of the injection holes, without being restricted by the structure of the injection disk.

[0009] Then, the spatial geometric features of the processing part are accurately analyzed through point cloud reconstruction technology to obtain the actual center coordinates of the injection hole and the height information of the chamfer to be processed;

[0010] Secondly, based on the solved machining feature parameters, the machining G code is automatically generated with the help of the secondary development interface of the CNC system;

[0011] Finally, a force measuring tool holder is used to collect cutting force signals in real time during the machining process, and the cutting force data is dynamically analyzed in combination with a deep learning model. The machining parameters are adaptively adjusted to optimize the machining state, thus achieving intelligent monitoring and refined control of the entire process.

[0012] The specific steps are as follows:

[0013] Step 1: Use adaptive centering fixture to position and clamp the workpiece

[0014] The adaptive centering fixture is mainly composed of two semicircular structures, a fixed end and a movable end. During the workpiece clamping process, the outer wall of the liquid rocket engine injection disk is first pressed against the fixed end, and then the movable end is closed and tightened by screws.

[0015] Step 2: Line laser scanning measurement

[0016] The line laser sensor establishes communication with the machine tool CNC system through a network cable and is integrated at the end of the machine tool spindle using a tool holder. After completing the position and posture calibration of the line laser sensor, execute the "SPOS" command to lock the machine tool spindle. In order to ensure the integrity of the 3D point cloud data of each injection hole, a multi-directional line laser scanning measurement trajectory is designed.

[0017] Step 3: Solving the processing related quantities

[0018] Select the minimum bounding box Ω that can completely cover the geometric features of a single injection hole, extract the point cloud data in the bounding box to establish a sub-point cloud N; group the coordinate points in the sub-point cloud N according to the different laser scanning lines to which they belong, and number them as P1, P2…P i …P n , n is the total number of line laser scanning lines contained in the acquired workpiece point cloud, from P i Every 3 consecutive lines of laser scanning line data P i+1 , P i+2 , P i+3 Form a data matrix D i ; Use the convolution kernel K of size 3×3 to perform the convolution operation to obtain the data matrix D after convolution i ';

[0019] D i '=D i *K (1)

[0020] The upper surface data points of several injection holes of the steel nozzle are manually labeled as labels H, and end-to-end training is performed using a multi-layer perceptron. The input data is the data matrix D'=(D'1, D'2, ... D' i ,..D' n-2 ), the output data is the data point set on the upper surface of the injection hole steel nozzle Data point set The amount of data is consistent with the label H; the mean square error function value MSE is used as the loss function calculation and parameter training standard of the neural network;

[0021]

[0022] Among them, i is the data point set The number of data with the label H; Use the trained neural network to obtain the data point set P on the upper surface of a single injection hole steel nozzle = {P j (x j ,y j ,z j ), j∈[1,n]}, calculate the average height of all points as the upper surface height h of the injection hole top ;

[0023]

[0024] According to the data point set P on the upper surface of the injection hole steel nozzle, find the outermost edge point set S of the injection hole steel nozzle side , expand it outward to obtain the cone point set S of the injection hole copper matrix cone ;

[0025] S1: Solution of injection hole center coordinates

[0026] At the outermost edge point set S of the injection hole steel nozzle side Three points s1(x1,y1), s2(x2,y2), and s3(x3,y3) are randomly selected to form the initial sample, and the least squares fitting algorithm is used for circle fitting. The circle fitting equation is:

[0027]

[0028] Among them, (x c ,y c ) is the coordinate of the center of the circle to be fitted, and R is the radius of the circle to be fitted;

[0029] Calculate the outermost edge point set S of the injection hole steel nozzle side The error between other points in the fitted circle and the circle is i (x i ,y i ), if the error is within the preset radius threshold δ of the fitted circle center, the point is considered to be an interior point; the judgment basis is as follows:

[0030]

[0031] e i <δ (6)

[0032] Repeat the above steps several times, select the parameters of the fitting circle with the largest number of inner points as the final result, and obtain the coordinates of the center of the circle o(x0, y0), which are the coordinates of the center of the injection hole;

[0033] S2: Solution for injection hole chamfer height

[0034] The first straight line is obtained by fitting the point cloud of the upper surface of the steel nozzle of the injection hole using the least squares method. The second straight line is obtained by fitting the point cloud of the side of the steel nozzle using the least squares method. The third straight line is obtained by solving the vertical horizontal plane through the intersection of the first and second straight lines. The points on the cone surface of the copper matrix of the injection hole are fitted using the least squares method to obtain the fourth straight line. The intersection of the fourth straight line and the third straight line is the lowest point, and the height of the lowest point is the height h of the lower surface. bottom , the injection hole nozzle height is expressed as:

[0035] h=h top -h bottom (7)

[0036] Assume that the data set of all injection hole center coordinates and chamfer height to be processed is D = {(x i ,y i ,h i)},i∈1,2,...,N, where N is the number of injection holes. For each data point, it is necessary to verify whether it satisfies the following conditions:

[0037]

[0038] Among them, X ci , Y ci , H ci is the theoretical coordinates of the injection hole and the height to be processed, σ x , σ y , σ h They are the difference between the theoretical coordinates and the actual coordinates of the injection hole and the standard deviation of the height distribution to be processed;

[0039] Step 4: Processing G code generation

[0040] The general processing G code template is written by using the secondary development interface of the CNC system, and the height coordinate value of the injection hole to be processed and the center coordinate of the injection hole are designed as dynamic variables; in the G code template, the initial parameter value is the theoretical coordinate of the workpiece design and the processing height; combined with the data information of the center coordinates of all injection holes and the height of the chamfer to be processed obtained in the third step, the G code template is imported through dynamic assignment and real-time adjustment to realize the dynamic and personalized parameters; the code structure of the processing process of the i-th injection hole is as follows:

[0041]

[0042] Among them, G90 is the instruction for executing the operation in the machine tool coordinate system, G01 is the instruction for linear motion, and κ is the increased safety distance to prevent tool collision;

[0043] After data processing, the parameters to be written into the template are stored. If a power outage or system abnormality occurs, the last stable state can be restored. Through the combination of dynamic adjustment of the template and fault-tolerant logic, intelligent generation and optimization of machining G code can be achieved.

[0044] Step 5: Intelligent processing based on force measuring tool holder

[0045] The cutting force signal during the machining process is collected using a force measuring tool holder. Due to the fast dynamic changes of the cutting force signal, the sampling frequency must satisfy the Nyquist sampling theorem to capture complete details; the sampling frequency should conform to the following formula:

[0046]

[0047] Among them, s is the spindle speed, z is the number of teeth of the tool, and f v is the natural frequency of the machining system, f sis the cutting force frequency; since the data acquisition contains noise or abnormal values, the signal needs to be filtered; since the cutting process belongs to a high-frequency force signal, a low-pass filter is used, and the formula is as follows:

[0048] f cutoff <0.5f s (11)

[0049] Among them, f cutoff is the cut-off frequency;

[0050] Since there is a time error in the system feedback, a deep learning framework based on long short-term memory network and one-dimensional convolutional neural network is designed to predict the cutting force signal and extract key features; the cutting force signal is collected as time series data: expressed as:

[0051] F(t)={F1,F2,...,F T} (12)

[0052] The actual processing data of multiple injection holes in the injection disk are used as training samples, and the long short-term memory neural network is used to capture the time dependence of the cutting force signal. The one-dimensional convolutional neural network is used to extract the processing features in the predicted cutting force signal. The model expressions of the long short-term memory neural network and the one-dimensional convolutional neural network are:

[0053] h t ,C t =LSTM(x t ,h t-1 ,C t-1 ) (13)

[0054]

[0055] Among them, h t is the hidden state of the current time step, C t is the cell state at the current time step, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, C t-1 is the cell state at the previous time step; z t,k represents the output of the kth convolution kernel at time step t, K represents the size of the convolution kernel, that is, the number of input sequence elements involved in the convolution operation, and w k,j represents the weight of the jth position in the kth convolution kernel, x t+j-1 represents the value of the input sequence at time step t+j-1, b k Represents the bias term of the kth convolution kernel;

[0056] When the height value obtained by the prediction result of the completed feature is lower than the calculated result of the geometric feature, the CNC system skips the current line instruction and interrupts the current injection hole processing in advance and continues the subsequent operations. The processing geometric parameters are backed up in the document. After the overall processing task is completed, the data and processing quality are reviewed to determine the degree of processing completion.

[0057] The beneficial effects of the invention are as follows: the invention can significantly improve the processing accuracy, reduce the labor intensity of workers, and provide an advanced solution for the efficient and intelligent processing of the injection holes of liquid rocket engines. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The overall schematic diagram of the processing system.

[0059] Figure 2 Schematic diagram of the distribution of injection holes on the injection disk.

[0060] Figure 3 Schematic diagram of the injection hole structure.

[0061] Figure 4 This is the edge fitting effect of the scanned point cloud of the injection hole.

[0062] Figure 5 Schematic diagram for solving the injection hole height.

[0063] Figure 6 This is a flow chart of the intelligent processing method for rocket engine injection holes.

[0064] In the figure, 1 is the machine tool spindle connecting flange, 2 is the injection plate, 3 is the self-centering fixture, 4 is the machine tool worktable, 5 is the line laser sensor tooling, and 6 is the line laser sensor. DETAILED DESCRIPTION

[0065] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.

[0066] The processing object of the embodiment of the present invention is a certain type of rocket engine injection disk with a diameter of 430mm and a height of 300mm. The line laser sensor used is the Keyence LJX-8200 sensor, the machine tool is a five-axis CNC machining center, the CNC system is the SINUMERIKONE system, and the force measuring tool holder is a Spike mobile tool holder type force measuring sensor. Figure 6 The overall flow chart of the large-scale point cloud simplification method based on the improved feature descriptor is as follows:

[0067] Step 1: Use a self-centering fixture to position and clamp the workpiece

[0068] The adaptive centering fixture consists of two semicircular parts, a fixed end and a movable end. To ensure the coaxiality between the workpiece and the rotation center of the machine tool table, the arc surface of the fixed end is aligned by dialing during the first installation. During the workpiece clamping process, first place the outer wall of the injection plate close to the fixed end, and then close and tighten the movable end with screws.

[0069] Step 2: Line laser scanning measurement

[0070] Considering the high sensitivity of the line laser sensor to posture, the high-precision contour surface is aligned after the sensor is clamped and fixed. After completing the posture calibration of the sensor, execute the "SPOS" command to lock the machine tool spindle to avoid the introduction of systematic errors due to changes in the sensor posture in subsequent operations. Design an S-shaped scanning trajectory to cover the area to be processed to scan the workpiece.

[0071] Step 3: Solving the processing related quantities

[0072] Select the minimum bounding box Ω that can completely cover the geometric features of a single injection hole, extract the point cloud data in the bounding box to establish a sub-point cloud N. Group the coordinate points in the sub-point cloud N according to the different laser scanning lines to which they belong, and number them as P1, P2…P i …P n , starting from P1, every 3 consecutive scan lines of data P i+1 , P i+2 , P i+3 Form a data matrix D i . Use the convolution kernel K of size 3×3 to perform the convolution operation to obtain the data matrix D after the convolution i '.

[0073] D i '=D i *K (15)

[0074] The upper surface data points of several injection holes of the steel nozzle are manually labeled as labels H, and end-to-end training is performed using a multi-layer perceptron. The data matrix D'=(D'1, D'2, ... D' i ,..D' n-2 ), output the data point set of the upper surface of the injection hole steel nozzle Point Set The amount of internal data is consistent with the label H. The mean square error function value (MSE) is used as the loss function calculation and parameter training standard of the neural network.

[0075]

[0076] Among them, i is the point set The number of data with the label H. The trained neural network is used to obtain the data point set P on the upper surface of a single injection hole steel nozzle = {P j(x j ,y j ,z j ), j∈[1,n]}, the average value of all points z is 0.2232mm as the height h of the upper surface of the injection hole top .

[0077] According to the upper surface data point set P, find the outermost edge point set S of the steel pipe side , expand it outward to obtain the cone point set S of the injection hole copper matrix cone .

[0078] S1: Solution of injection hole center coordinates

[0079] In S side Three points s1(x1,y1), s2(x2,y2), and s3(x3,y3) are randomly selected to form the initial sample, and the least squares fitting algorithm is used for circle fitting. The circle fitting equation is:

[0080]

[0081] Among them, (x c ,y c ) is the coordinate of the center of the fitted circle. Calculate S side The error between other points in the fitted circle and the fitted circle, if the error is within the preset radius threshold of 0.1 of the fitted circle center, then the point is considered to be an inner point. The judgment basis is as follows:

[0082]

[0083] Repeat the above steps several times, select the parameters of the fitting circle with the largest number of inner points as the final result, and obtain the center coordinates o(x0, y0), which are the center coordinates of the injection hole. The center coordinates of the first injection hole are calculated to be (-0.494039, 0.2888). Repeat this process to obtain all the center coordinates.

[0084] S2: Solution for injection hole chamfer height

[0085] The point cloud of the upper surface of the steel pipe of the injection hole is fitted using the least squares method to obtain a straight line z1=0.023y+0.03506. The point cloud of the side of the steel pipe is fitted using the least squares method to obtain a straight line z2=-0.71y+5.82. The vertical horizontal plane through the intersection of the straight line z1 and the straight line z2 is solved to obtain a straight line y=8.18. The points on the conical surface of the copper substrate of the injection hole are fitted using the least squares method to obtain a straight line z4=-0.71y+5.82. The intersection 13 of the straight line 12 and the straight line 11 is the lowest point, that is, the height of the lower surface is 0.0122mm.

[0086] Take the points on a generatrix of the cone surface of the copper matrix of the first injection hole to form a point set L, and use the least squares method to linearly fit to get the straight line equation z. The straight line equation of the inclined plane is: z2 = 0.64y-5.22, and the intersection point of the straight line z and the outer wall straight line z1 = -0.71y + 5.82 of the steel pipe is the lowest point, and the lower surface height value h can be obtained. bottom Then the nozzle height of the first injection hole is 0.211 mm. Repeat this process to obtain the nozzle heights of all injection holes.

[0087] In order to eliminate the influence of errors and noise in the process of point cloud data analysis, improve the stability and reliability of the processing process, trigger an alarm and stop when the parameters exceed the safe processing range, and reduce the risk of accidents caused by data abnormalities. Assume that the data set of the center coordinates of all injection holes and the height of the chamfer to be processed is D = {(x i ,y i ,h i )},i∈1,2,...,N, then for each data point, it is necessary to verify whether it satisfies the following conditions:

[0088]

[0089] Among them, X ci , Y ci , H ci is the theoretical coordinates of the injection hole and the height to be processed, σ x , σ y , σ h They are the difference between the theoretical coordinates and the actual coordinates of the injection hole and the standard deviation of the height distribution to be processed. After the threshold is judged to meet the requirements, combined with the line laser baseline distance, the processing height of the first injection hole in the machine tool coordinate system is -311.025mm.

[0090] Step 4: Processing G code generation

[0091] The secondary development interface of the CNC system is used to write a universal processing G code template, and the height coordinate value and the center coordinate of the injection hole are designed as dynamic variables. In the G code template, the initial parameter value is the theoretical coordinate and processing height of the workpiece design. Combined with the workpiece geometry information solved in the third step, the actual calculated height and center coordinates and other processing parameters are imported into the G code template through dynamic assignment and real-time adjustment to realize the dynamic and personalized parameters. The code structure of the first injection hole processing process is as follows:

[0092] G90 G01 X-0.494039Y0.2888Z-301.025F500

[0093] G90 G01 X-0.494039Y0.2888Z-311.025F10

[0094] G90 G01X-0.494039Y0.2888Z-301.025F500

[0095] Step 5: Intelligent processing based on force measuring tool holder

[0096] The cutting force signal during the machining process is collected using a force measuring tool holder. Due to the fast dynamic change of the cutting force signal, the sampling frequency satisfies the Nyquist sampling theorem. In this embodiment, the spindle speed s = 100, the number of teeth of the tool z = 4, and the natural frequency of the machining system f v =1kHz, to satisfy:

[0097]

[0098] Set the cutting force sampling frequency f s =2.5kHz. Since the data acquisition contains noise or abnormal values, low-pass filtering is used to process the signal, and the cutoff frequency f cutoff Let f v =1.25kHz

[0099] Due to the time error of system feedback, a deep learning framework based on long short-term memory network and one-dimensional convolutional neural network is designed to realize the prediction of cutting force signal and extraction of key features. The cutting force signal is collected as time series data: represented as

[0100] F(t)={F1,F2,...,F T} (twenty one)

[0101] The actual processing data of multiple injection holes in the injection plate are used as training samples to train the long short-term memory network to capture the time dependence of the cutting force signal. The model expression is:

[0102] h t ,C t =LSTM(x t ,h t-1 ,C t-1 ) (twenty two)

[0103] Among them, h t is the hidden state of the current time step, C t is the cell state at the current time step, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, C t-1 is the cell state at the previous time step.

[0104] A one-dimensional convolutional network is used to extract machining features from cutting force signals to prevent overcutting. The specific formula is:

[0105]

[0106] When the height value obtained by the prediction result of the completed feature is lower than the calculated result of the geometric feature, the CNC system skips the current line instruction to interrupt the current injection hole processing in advance and continue the subsequent operation. The processing geometric parameters are backed up in a special document. After the overall processing task is completed, the data and processing quality are rechecked to determine the processing completion degree. Other prediction results represent that the processing process is safe and can be processed normally according to the code.

[0107] The specific implementation cases described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation case of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent processing method for a liquid rocket engine injection hole, characterized in that: Here are the steps: Firstly, the workpiece, i.e., the liquid rocket engine injection plate, is fixed on the worktable of a five-axis CNC machine tool by an adaptive centering fixture, and the relative position relationship between the workpiece coordinate system and the machine tool coordinate system is determined; Secondly, the injection disk is scanned at multiple angles using a line laser sensor to ensure the integrity of the three-dimensional point cloud data of the injection holes, without being restricted by the structure of the injection disk. Then, the spatial geometric features of the processing part are accurately analyzed through point cloud reconstruction technology to obtain the actual center coordinates of the injection hole and the height information of the chamfer to be processed; Secondly, based on the solved processing feature parameters, the automatic generation of processing G code is realized with the help of the secondary development interface of the CNC system; Finally, a force measuring tool holder is used to collect cutting force signals in real time during the machining process, and the cutting force data is dynamically analyzed in combination with a deep learning model. The machining parameters are adaptively adjusted to optimize the machining state, thus achieving intelligent monitoring and refined control of the entire process.

2. An intelligent processing method for a liquid rocket engine injection hole, characterized in that: The specific steps are as follows: Step 1: Use adaptive centering fixture to position and clamp the workpiece The adaptive centering fixture is mainly composed of two semicircular structures, a fixed end and a movable end. During the workpiece clamping process, the outer wall of the liquid rocket engine injection disk is first pressed against the fixed end, and then the movable end is closed and tightened by screws. Step 2: Line laser scanning measurement The line laser sensor establishes communication with the machine tool CNC system through a network cable and is integrated at the end of the machine tool spindle using a tool holder. After completing the position and posture calibration of the line laser sensor, execute the "SPOS" command to lock the machine tool spindle. In order to ensure the integrity of the 3D point cloud data of each injection hole, a multi-directional line laser scanning measurement trajectory is designed. Step 3: Solving processing related quantities Select the minimum bounding box Ω that can completely cover the geometric features of a single injection hole, extract the point cloud data in the bounding box to establish a sub-point cloud N; group the coordinate points in the sub-point cloud N according to the different laser scanning lines to which they belong, and number them as P1, P2…P i …P n , n is the total number of line laser scanning lines contained in the acquired workpiece point cloud, from P i Every 3 consecutive lines of laser scanning line data P i+1 , P i+2 , P i+3 Form a data matrix D i ; Use the convolution kernel K of size 3×3 to perform the convolution operation to obtain the data matrix D after convolution i '; D′ i =D i *K (1) The upper surface data points of several injection holes of the steel nozzle are manually labeled as labels H, and end-to-end training is performed using a multi-layer perceptron. The input data is the data matrix D'=(D'1, D'2, ... D') obtained by processing different injection holes. i ,..D' n-2 ), the output data is the data point set on the upper surface of the injection hole steel nozzle Data point set The amount of data is consistent with the label H; the mean square error function value MSE is used as the loss function calculation and parameter training standard of the neural network; Among them, i is the data point set The number of data with the label H; Use the trained neural network to obtain the data point set P on the upper surface of a single injection hole steel nozzle = {P j (x j ,y j ,z j ), j∈[1,n]}, calculate the average height of all points as the upper surface height h of the injection hole top ; According to the data point set P on the upper surface of the injection hole steel nozzle, find the outermost edge point set S of the injection hole steel nozzle side , expand it outward to obtain the cone point set S of the injection hole copper matrix cone ; S1: Solution of injection hole center coordinates At the outermost edge point set S of the injection hole steel nozzle side Three points s1(x1,y1), s2(x2,y2), and s3(x3,y3) are randomly selected to form the initial sample, and the least squares fitting algorithm is used for circle fitting. The circle fitting equation is: Among them, (x c ,y c ) is the coordinate of the center of the circle to be fitted, and R is the radius of the circle to be fitted; Calculate the outermost edge point set S of the injection hole steel nozzle side The error between other points in the fitted circle and the circle is i (x i ,y i ), if the error is within the preset radius threshold δ of the fitted circle center, the point is considered to be an interior point; the judgment basis is as follows: e i <d (6) Repeat the above steps several times, select the parameters of the fitting circle with the largest number of inner points as the final result, and obtain the coordinates of the center of the circle o(x0, y0), which are the coordinates of the center of the injection hole; S2: Solution for injection hole chamfer height The first straight line is obtained by fitting the point cloud of the upper surface of the steel nozzle of the injection hole using the least squares method. The second straight line is obtained by fitting the point cloud of the side of the steel nozzle using the least squares method. The third straight line is obtained by solving the vertical horizontal plane through the intersection of the first and second straight lines. The points on the cone surface of the copper matrix of the injection hole are fitted using the least squares method to obtain the fourth straight line. The intersection of the fourth straight line and the third straight line is the lowest point, and the height of the lowest point is the height h of the lower surface. bottom , the injection hole nozzle height is expressed as: h=h top -h bottom (7) Assume that the data set of all injection hole center coordinates and chamfer height to be processed is D = {(x i ,y i ,h i )},i∈1,2,...,N, where N is the number of injection holes. For each data point, it is necessary to verify whether it satisfies the following conditions: Among them, X ci , Y ci , H ci is the theoretical coordinates of the injection hole and the height to be processed, σ x , σ y , σ h They are the difference between the theoretical coordinates and the actual coordinates of the injection hole and the standard deviation of the height distribution to be processed; Step 4: Processing G code generation The general processing G code template is written by using the secondary development interface of the CNC system, and the height coordinate value of the injection hole to be processed and the center coordinate of the injection hole are designed as dynamic variables; in the G code template, the initial parameter value is the theoretical coordinate of the workpiece design and the processing height; combined with the data information of the center coordinates of all injection holes and the height of the chamfer to be processed obtained in the third step, the G code template is imported through dynamic assignment and real-time adjustment to realize the dynamic and personalized parameters; the code structure of the processing process of the i-th injection hole is as follows: Among them, G90 is the instruction for executing the operation in the machine tool coordinate system, G01 is the instruction for linear motion, and κ is the increased safety distance to prevent tool collision; After data processing, the parameters to be written into the template are stored. If a power outage or system abnormality occurs, the last stable state can be restored. Through the combination of dynamic adjustment of the template and fault-tolerant logic, intelligent generation and optimization of machining G code can be achieved. Step 5: Intelligent processing based on force measuring tool holder The cutting force signal during the machining process is collected using a force measuring tool holder. Due to the fast dynamic changes of the cutting force signal, the sampling frequency must satisfy the Nyquist sampling theorem to capture complete details; the sampling frequency should conform to the following formula: Among them, s is the spindle speed, z is the number of teeth of the tool, and f v is the natural frequency of the machining system, f s is the cutting force frequency; since the data acquisition contains noise or abnormal values, the signal needs to be filtered; since the cutting process belongs to a high-frequency force signal, a low-pass filter is used, and the formula is as follows: f cutoff <0.5f s (11) Among them, f cutoff is the cut-off frequency; Since there is a time error in the system feedback, a deep learning framework based on long short-term memory network and one-dimensional convolutional neural network is designed to predict the cutting force signal and extract key features; the cutting force signal is collected as time series data: expressed as: F(t)={F1,F2,...,F T } (12) The actual processing data of multiple injection holes in the injection disk are used as training samples, and the long short-term memory neural network is used to capture the time dependence of the cutting force signal. The one-dimensional convolutional neural network is used to extract the processing features in the predicted cutting force signal. The model expressions of the long short-term memory neural network and the one-dimensional convolutional neural network are: h t ,C t =LSTM(x t ,h t-1 ,C t-1 ) (13) Among them, h t is the hidden state of the current time step, C t is the cell state at the current time step, x t is the input of the current time step, h t-1 is the hidden state of the previous time step, C t-1 is the cell state at the previous time step; z t,k represents the output of the kth convolution kernel at time step t, K represents the size of the convolution kernel, that is, the number of input sequence elements involved in the convolution operation, and w k,j represents the weight of the jth position in the kth convolution kernel, x t+j-1 represents the value of the input sequence at time step t+j-1, b k Represents the bias term of the kth convolution kernel; When the height value obtained by the prediction result of the completed feature is lower than the calculated result of the geometric feature, the CNC system skips the current line instruction and interrupts the current injection hole processing in advance and continues the subsequent operations. The processing geometric parameters are backed up in the document. After the overall processing task is completed, the data and processing quality are reviewed to determine the degree of processing completion.

3. The intelligent processing method for the injection hole of a liquid rocket engine according to claim 2, characterized in that: In order to ensure the coaxiality between the workpiece and the rotation center of the machine tool table, the arc surface of the fixed end needs to be checked by dialing when the adaptive centering fixture is installed for the first time.

Images