Artillery barrel deep hole machining method and machining device based on magnetostriction actuator

Abstract

The invention relates to the field of deep machining of cutting blades, and provides an artillery barrel deep hole machining method and device based on a magnetostrictive actuator, and the machining method comprises the steps that multi-source data sensing network systems are installed on a boring rod, a cutter and an artillery barrel respectively; processing a multi-source signal acquired in real time, and extracting time-frequency information in an acoustic emission signal and a vibration signal; according to the characteristics of artillery barrel deep hole machining, a tool wear prediction model is established; inputting the time-frequency information into a tool wear prediction model to obtain a current signal of a control system; and a current signal is input to the coil to control the stretching of the magnetostrictive element, so that the cutter and the artillery barrel are kept in a contact state, and the fine adjustment of the displacement of the cutter is realized. The machining device comprises a spindle box, a guide rail, a boring rod assembly, a cutter and a floating support. The problem that dynamic characteristics of the large-length-diameter-ratio boring rod are affected by tool abrasion and time-varying excitation is solved, and the precision requirement of artillery barrel deep hole manufacturing is met.

Description

Technical Field

[0001] The present invention relates to the field of deep hole processing / cutting blade deep processing, and in particular to a gun barrel deep hole processing method and processing device based on a magnetostrictive actuator. Background Art

[0002] As an important component of artillery, the barrel seriously affects the firing accuracy of artillery. A gun barrel refers to a situation where the ratio of the length to the diameter of the barrel is greater than 20:1. In the manufacture of gun barrels, the inner diameter of the gun barrel is mostly 100-400mm, the total length is mostly 7-12m, and its aspect ratio can be as high as 120. As the front-end process of rifling processing, the dimensional accuracy, straightness and roundness of deep hole fine boring directly affect the rifling processing quality. However, with the application of high-strength and toughness artillery steel, its high hardness and high toughness cause rapid wear of the tool tip, and the nonlinear dynamics of the large aspect ratio boring bar causes the time-varying tool tip trajectory problem. It is very easy to cause barrel processing out of tolerance under a single long boring stroke.

[0003] To address these issues, boring tool radial micro-displacement control technology offers a new approach to effectively compensating for tool tip trajectory deviations. Eccentric boring micro-compensation devices have been developed based on hydraulically or electromechanically driven boring tool radial micro-displacement mechanisms. Hydraulic drive systems present complex system issues, while electromagnetically driven radial micro-displacement mechanisms are complex in design and difficult to achieve precise displacement control. Displacement converters are complex in structure, require extremely high rigidity and transmission accuracy, are prone to wear over time, and struggle to meet high-frequency response speeds.

[0004] Building on this foundation, the development of micro-displacement deformation actuation technology for components made of materials such as piezoelectric ceramics and giant magnetostrictive materials (GMMs) has enabled high-frequency response and high-precision microfeed control of cutting tools. Micro-displacement actuators driven by piezoelectric ceramics require high drive voltages, making energy transmission difficult. Furthermore, the energy density of individual piezoelectric ceramic elements is low, resulting in small output forces and displacements. GMMs offer advantages such as high energy density, large output forces and displacements, and fast response speeds, leading to the emergence of numerous high-performance giant magnetostrictive actuators (GMAs).

[0005] Therefore, the present invention introduces the giant magnetostrictive actuator GMA micro-displacement technology into the barrel deep hole boring processing system. Through the high-frequency and high-precision control of the boring tool micro-displacement, the trajectory deviation of the tool tip is compensated, the precision control of the gun barrel is achieved, and the processing accuracy is improved. Summary of the Invention

[0006] In response to the problems existing in the prior art, the present invention provides a method and device for deep hole machining of a gun barrel based on a magnetostrictive actuator. To address the problems of deep machining of cutting blades and out-of-tolerance machining of the gun barrel, a multi-source data sensing network system is first installed at the corresponding positions of the boring bar, tool, and gun barrel. The time-frequency information in the multi-source signals is then input into a tool wear prediction model. By embedding a high-frequency, high-precision boring cutter micro-displacement instantaneous stability dual-control strategy with a magnetostrictive actuator, online and accurate prediction of tool tip wear under complex and variable working conditions is carried out. The time-varying dynamic characteristics of the boring bar based on time-varying cutting force and time-varying overhang are studied, and the time-varying pattern of the boring cutter tip trajectory is analyzed. Finally, the magnetostrictive element is controlled to extend and retract, driving a push rod to perform micron-level displacement, keeping the tool and gun barrel in contact, thereby achieving fine-tuning of the tool displacement, effectively controlling machining errors, and realizing precise machining of deep holes in gun barrels.

[0007] The present invention provides a method for deep hole machining of a gun barrel based on a magnetostrictive actuator, and the specific implementation steps are as follows:

[0008] S1. Mount the gun barrel to be machined on a guide rail via a floating support, mount the tool on the working end of the boring bar, connect the driving end of the boring bar to the spindle box, and mount the fixed end of the boring bar on the guide rail via a floating support;

[0009] S2. Installing the multi-source data sensor network system at corresponding positions on the boring bar, the tool, and the gun barrel, and setting cutting parameters for the tool;

[0010] S3. Start the spindle box and the multi-source data sensing network system respectively, process the acoustic emission signals and vibration signals collected in real time by the multi-source data sensing network system, and extract the time-frequency information about the tool tip displacement from the acoustic emission signals and the vibration signals;

[0011] S4. Based on the characteristics of deep hole machining for artillery barrels, a tool wear prediction model is established, which specifically includes the following sub-steps:

[0012] S41. Using the state update expression of the bidirectional recurrent neural network, with the acoustic emission signal and vibration signal collected in step S3 as the core input and the tool tip displacement as the output, a tool tip wear prediction model under a single working condition is established, and the nonlinear mapping relationship between the characteristics of the multi-source signals and the tool tip wear sensitive characteristics of the tool is mined, wherein the state update expression of the bidirectional recurrent neural network is:

[0013] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0014] i t =σ(Wi ·[h t-1 ,x t ]+b i )

[0015]

[0016] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0017] h t =o t *tanh(C t )

[0018] Among them, f t Indicates the forgetting of historical signal data during processing, i t Indicates the acceptance of new signal data during the processing, C t Represents the long-term storage memory of signal data during processing and is updated at each time step. Indicates the update of C generated based on the current input data t Status information, h t Indicates the adjustment of the current prediction during the processing, o t is the output of the final data, x t represents the signal collected from the multi-source data sensor network system at the current moment, σ represents the Sigmoid activation function, ensuring that the signal output value is between (0,1), and W f Represents the weight matrix of the forget gate, W i Represents the weight matrix of the input gate, W C Represents the weight matrix for calculating the processing state update amount, W o represents the weight matrix of the output gate, b f represents the bias term of the forget gate, b i represents the bias term of the input gate, b C Represents the bias term used to update the current memory unit state, b o Represents the bias term h of the output gate t-1 represents the hidden state at the previous moment, C t-1 Indicates the storage state of the tool displacement compensation at the previous moment, tanh represents the hyperbolic tangent activation function, and normalizes the signal data to between (-1,1);

[0019] S42. Based on the tool tip wear prediction model under the single working condition obtained in step S41, the single working condition source domain data sample set is used as input, and the target domain to be learned under the variable feed working condition is used as output. A feature extractor is constructed using a convolutional neural network and a bidirectional long short-term memory network to extract multidimensional spatiotemporal features of the single working condition source domain data sample set and the target domain to be learned under the variable feed working condition, and a domain adaptive migration model is constructed.

[0020] S5. Input the time-frequency information of the acoustic emission signal and the vibration signal obtained in step S3 into the tool wear prediction model constructed in step S4, calculate the tool wear state, and generate a current signal for the control system;

[0021] S6. Input the current signal obtained in step S5 to the coil wound around the permanent magnet to change the magnetic field strength, thereby controlling the expansion and contraction of the magnetostrictive element, driving the push rod to perform micron-level displacement, so that the tip of the tool and the gun barrel maintain contact, thereby achieving fine-tuning of the tool displacement. Continuously collect and monitor acoustic emission signals and vibration signals to form a closed-loop feedback control.

[0022] Preferably, the position distribution of the acceleration sensor on the boring bar is obtained by analyzing the sensitivity between the signals of the acceleration sensor at different installation positions on the boring bar and the tool tip wear using the Pearson correlation coefficient method. The expression of the Pearson correlation coefficient is:

[0023]

[0024] Among them, ρ X,Y is the signal of the acceleration sensor at different installation positions on the boring bar, X and Y are the correlation coefficients in the Pearson correlation coefficient method, cov(X,Y) is the covariance of X and Y, σ X and σ Y are the standard deviations of X and Y respectively.

[0025] Preferably, in step S2, the cutting parameters include cutting speed, feed rate and current intensity; in step S3, the multi-source signal includes tool position deviation value, cutting force abnormality value, vibration abnormality value and acoustic signal characteristic value.

[0026] Preferably, the specific implementation steps of step S3 of extracting the time-frequency information from the acoustic emission signal and the vibration signal are:

[0027] S31. Based on the characteristics of tool wear and machining conditions, a sliding frame sampling method is used to intercept signal data of a specified length, and the discreteness of the signal data is analyzed. Abnormal data points exceeding or falling below three times the standard deviation of the mean are removed to obtain an initial data sample.

[0028] S32. Based on the initial data samples, a single working condition source domain data sample set is constructed using the tool tip wear value as a label;

[0029] S33. Based on the single-operating-condition source domain data sample set obtained in step S32, a wavelet transform method is used to simultaneously decompose the low-frequency and high-frequency parts of the single-operating-condition source domain data sample set to obtain frequency domain information. The expression of the wavelet transform method is:

[0030]

[0031] Among them, W a,b (t) represents the time-frequency local characteristics of machining vibration, acceleration, and acoustic emission signals at different scales and positions obtained after wavelet transform, and x(t) represents the original signals acquired by multi-source sensors during the actual machining process. represents the complex conjugate of the mother wavelet function used in wavelet transform, a and b represent the different levels and time characteristics of tool wear respectively;

[0032] S34. The frequency domain information obtained in S33 is processed using the Hilbert-Huang method to extract the instantaneous frequency components of the vibration frequency change and cutting force fluctuation caused by tool wear, thereby obtaining the time-frequency information in the multi-source signal. The expression of the Hilbert-Huang method is:

[0033]

[0034] Among them, ω(t) represents the instantaneous frequency characteristics of the signal caused by the change in the contact characteristics between the tool and the gun barrel, and θ(t) represents the phase characteristics caused by the change in the tool vibration and machining force with the degree of wear.

[0035] Preferably, in step S31, the judgment expression of the abnormal data point is:

[0036]

[0037] Where z is the normalized value, X is the signal data, μ is the mean of the signal data set, and σ is the standard deviation of the signal data set. If |z|>3, the data is considered an outlier.

[0038] Preferably, in step S41, the expressions of the correlation analysis and kernel principal component analysis (KPCA) are:

[0039]

[0040] Among them, φ(X) is the feature representation mapped to the high-dimensional space by the kernel function K, α i is the projection coefficient.

[0041] The second aspect of the present invention provides a deep hole processing device for a gun barrel based on a magnetostrictive actuator, which includes a spindle box, a guide rail, a boring bar assembly and a floating support, wherein the fixed end of the spindle box is connected to the first mounting end of the guide rail, the output end of the spindle box is connected to the driving end of the boring bar in the boring bar assembly, the fixed end of the boring bar in the boring bar assembly is connected to the second mounting end of the guide rail through a floating support, and the working end of the slide rail in the boring bar assembly is connected to the fixed end of the tool; the boring bar assembly includes a tool, A slide rail, a push rod, a boring bar, a magnetostrictive element, a permanent magnet, a coil, and a mounting plate. The mounting plate is connected to the mounting end inside the boring bar, the fixed end of the mounting plate is connected to the fixed end of the permanent magnet, the permanent magnet is provided with a coil, and the permanent magnets are distributed in an annular manner along the axis of the magnetostrictive element. The fixed end of the magnetostrictive element is connected to the first mounting end of the push rod, the second mounting end of the push rod is connected to the fixed end of the slide rail, and the working end of the slide rail is connected to the fixed end of the tool.

[0042] Furthermore, it also includes a multi-source data sensing network system, which includes a current sensor, a vibration sensor, an acceleration sensor and an acoustic emission sensor. The current sensor is installed on the spindle box, the vibration sensor is installed at the connection position of the boring bar and the tool, the acoustic emission sensor is installed in the working area of ​​the tool, the acoustic emission sensors are evenly spaced along the length direction of the gun barrel, and the acceleration sensors are respectively installed at the driving end, middle and fixed end of the boring bar and the position of the gun barrel near the processing hole.

[0043] Compared with the prior art, the present invention has the following advantages:

[0044] 1. The boring bar and tool of the present invention are equipped with a multi-source data sensing network system that monitors tool wear, cutting force, and boring bar dynamics in real time. The control system then adjusts the current to extend or retract the magnetostrictive actuator to change the tool feed and cutting depth. This micro-displacement, high-frequency, and high-precision control compensates for tool tip trajectory deviations, ensuring the desired tool tip trajectory. This prevents tool tip wear and time-varying excitation, which can severely affect the dynamics of high-aspect-ratio boring bars and cause time-varying tool tip trajectory, effectively controlling the tool tip trajectory and achieving precise control of the artillery barrel, thereby ensuring machining accuracy.

[0045] 2. The tool in the present invention is made of highly wear-resistant material and is designed with a special cutting edge shape to improve cutting efficiency and extend service life; the boring bar has a large aspect ratio, which can provide good rigidity and ensure the stability of the tool in deep hole processing.

[0046] 3. During the boring process, the multi-source data sensing network system of the present invention monitors the wear status and cutting force of the tool in real time. The controller dynamically adjusts the feed rate of the boring bar and the position of the tool based on this information to ensure processing accuracy. By adopting a method that combines feedforward predictive control and feedback closed-loop control, the processing error is effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of a method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to the present invention;

[0048] Figure 2 This is an overall structural diagram of the gun barrel deep hole processing device based on the magnetostrictive actuator of the present invention;

[0049] Figure 3 This is a schematic diagram of the magnetostrictive actuator in the gun barrel deep hole processing device based on the magnetostrictive actuator of the present invention;

[0050] Figure 4 This is a diagram showing the ideal tool tip motion trajectory in the gun barrel deep hole machining method based on a magnetostrictive actuator according to the present invention;

[0051] Figure 5 This is a deflection diagram of a boring bar in a method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to the present invention;

[0052] Figure 6 This is a diagram showing the barrel machining error caused by the deflection of the boring bar in the gun barrel deep hole machining method based on the magnetostrictive actuator of the present invention;

[0053] Figure 7 This is a diagram of tool tip wear in the gun barrel deep hole machining method based on the magnetostrictive actuator of the present invention;

[0054] Figure 8 This is a diagram showing the barrel machining error caused by tool tip wear in the gun barrel deep hole machining method based on a magnetostrictive actuator of the present invention.

[0055] Main reference numerals:

[0056] Spindle box 1, guide rail 2, boring bar assembly 3, tool 301, slide rail 302, push rod 303, boring bar 304, magnetostrictive element 305, permanent magnet 306, coil 307, mounting plate 308, floating support 4, gun barrel 5, upper deviation a of tool tip motion trajectory, lower deviation b of tool tip motion trajectory, actual barrel inner surface shape c after machining, undeflected boring bar under ideal machining state d, deflected boring bar under actual machining state e, actual barrel shape f after machining, ideal barrel shape g after machining, ideal tool state h, worn tool state i. DETAILED DESCRIPTION

[0057] To fully describe the technical content, objectives and effects of the present invention, the following will be described in detail with reference to the accompanying drawings.

[0058] A method for deep hole machining of gun barrel based on magnetostrictive actuator, such as Figure 1 As shown, the specific implementation steps are:

[0059] S1. Install the gun barrel 5 to be processed on the guide rail 2 through the floating support 4, install the tool 301 on the working end of the boring bar 304, connect the driving end of the boring bar 304 to the spindle box 1, and install the fixed end of the boring bar 304 on the guide rail 2 through the floating support 4.

[0060] S2. Install the multi-source data sensing network system at the corresponding positions of the boring bar 304, the tool 4 and the gun barrel 5, and set the cutting parameters of the tool 301; the cutting parameters include cutting speed, feed rate and current intensity.

[0061] Specifically, the multi-source data sensor network system can monitor tool wear and cutting force changes in real time, providing important data support for the controller. By receiving data feedback from the multi-source data sensor network system and combining it with the tool wear prediction model, the controller can dynamically adjust the feed speed of the boring bar 304 and the micro-displacement of the tool 301. Specifically, it includes a current sensor, a vibration sensor, an acceleration sensor, and an acoustic emission sensor. The current sensor is installed on the spindle box 1, the vibration sensor is installed at the connection between the boring bar 304 and the tool 301, and the acoustic emission sensor is installed in the working area of ​​the tool 301. The acoustic emission sensors are evenly spaced along the length of the gun barrel 5. Acceleration sensors are installed at the driving end, middle part, and fixed end of the boring bar 304, as well as at a position on the gun barrel 5 near the processing hole.

[0062] The position distribution of the acceleration sensor on the boring bar 304 is obtained by analyzing the sensitivity of the signals of the acceleration sensor at different installation positions on the boring bar 304 to the wear of the tool tip 301 using the Pearson correlation coefficient method. The expression of the Pearson correlation coefficient is:

[0063]

[0064] Among them, ρ X,Y is the signal of the acceleration sensor at different installation positions on the boring bar 304, X and Y are the correlation coefficients in the Pearson correlation coefficient method, cov(X,Y) is the covariance of X and Y, σ X and σ Y is the standard deviation of X and Y.

[0065] S3. Start the spindle box 1 and the multi-source data sensor network system separately. Process and analyze the multi-source signals collected in real time by the multi-source data sensor network system to construct a sample set based on the multi-source signals. Time-frequency information from the multi-source signals is obtained from the sample set. The obtained time-frequency information is used to construct a tool wear prediction model based on deep learning using a neural network. Specifically, the multi-source signals consist of acoustic emission signals and vibration signals, including tool position deviation values, cutting force anomalies, vibration anomalies, and acoustic signal characteristic values.

[0066] S4. Ideally, the tip of the tool 301 should move along the set trajectory. Figure 4 As shown in the figure, the inner surface shape c of the barrel after actual processing is between the upper deviation a of the tool tip motion trajectory and the lower deviation b of the tool tip motion trajectory, as shown in the figure. Figure 5 As shown, the deflection of the boring bar 304 is formed by the difference between the undeflected boring bar d in the ideal processing state and the deflected boring bar e in the actual processing state, as shown in FIG. Figure 7 As shown, the wear of the tool 301 is formed by the difference between the ideal tool state h and the worn tool state i. Figure 7 Wear of tool 301 and Figure 5 The deflection of the boring bar 304, the tool tip trajectory in actual processing will be Figure 6 and Figure 8 The offset shown in the figure causes a deviation between the actual barrel shape f after machining and the ideal barrel shape g after machining. According to the unidirectional long stroke characteristics of the deep hole boring machining system on the gun barrel 5, a tool wear prediction model is established.

[0067] S5. Input the acoustic emission signal and vibration signal obtained in step S3 into the tool wear prediction model constructed in step S4, calculate the wear state of the tool 301, and generate a current signal for the control system.

[0068] S6. Input the current signal obtained in step S5 to the coil 307 wound around the permanent magnet 306. The current directly determines the magnetic field strength, controls the expansion and contraction of the magnetostrictive element 305, and drives the push rod 303 to move in micrometers, so that the tool 301 and the gun barrel 5 maintain contact, thereby achieving fine adjustment of the tool 301 displacement. Continuously collect and monitor acoustic emission signals and vibration signals to form a closed-loop feedback control system.

[0069] Preferably, the specific implementation steps of step S3 of extracting time-frequency information from multi-source signals are:

[0070] S31. Analyze the spatiotemporal characteristics and singularities of multi-source signals. Based on the characteristics of tool wear and machining conditions, use sliding frame sampling to intercept signal data of a specified length to ensure that the signal data intercepted each time is sufficient to capture the key timing characteristics of tool wear. Analyze the discreteness of the signal data, remove abnormal data points that exceed or fall below three times the standard deviation of the mean, and obtain the initial data sample.

[0071] Specifically, for each data, its normalized value z is calculated to determine whether it is abnormal data. The formula is:

[0072]

[0073] Where z is the normalized value, X is the signal data, μ is the mean of the signal data set, and σ is the standard deviation of the signal data set. If |z|>3, the data is considered an outlier. This method can remove outliers that exceed or fall below three times the standard deviation of the mean, ensuring high accuracy of the signal data.

[0074] S32. Based on the initial data samples, a single working condition source domain data sample set is constructed with the tool tip wear value as a label.

[0075] S33. Based on the single-operating-condition source domain data sample set obtained in step S32, its time domain information is obtained. The low-frequency and high-frequency parts of the single-operating-condition source domain data sample set are simultaneously decomposed using a wavelet transform method to obtain frequency domain information. The expression of the wavelet transform method is:

[0076]

[0077] Among them, W a,b (t) represents the time-frequency local characteristics of machining vibration, acceleration, and acoustic emission signals at different scales and positions obtained after wavelet transform, and x(t) represents the original signals acquired by multi-source sensors during the actual machining process. It represents the complex conjugate of the mother wavelet function used in wavelet transform, and a and b represent the different levels and time characteristics of tool wear, respectively.

[0078] Wavelet transform can accurately extract the time domain and frequency domain information of the signal, and can identify the instantaneous features related to the wear of the tool 301 in the processing method of the present invention.

[0079] S34. The frequency domain information obtained in S33 is processed using the Hilbert-Huang method to extract the instantaneous frequency components of the vibration frequency changes and cutting force fluctuations caused by tool wear, achieve high-resolution time-frequency analysis, and obtain the time-frequency information in the multi-source signal. The expression of the Hilbert-Huang method is:

[0080]

[0081] Wherein, ω(t) represents the instantaneous frequency characteristics of the signal caused by the change in the contact characteristics between the tool 301 and the gun barrel 5, and θ(t) represents the phase characteristics caused by the vibration and machining force of the tool 301 as the degree of wear changes.

[0082] Preferably, the specific process of establishing the tool wear prediction model in step S4 is:

[0083] S41. First, the multi-scale spatiotemporal features of the time-frequency information obtained in step S3 are extracted using deep autoencoding technology, and the dimensionality reduction of the multi-scale spatiotemporal features is achieved using correlation analysis and kernel principal component analysis (KPCA) methods. Then, a deep bidirectional recurrent neural network (Bi-LSTM) method is used to introduce an attention mechanism, adaptively calculate and adjust input weights, and explore the nonlinear mapping relationship between the characteristics of multi-source signals and the tool tip wear sensitive characteristics of tool 301. The tool tip wear prediction model under a single working condition is established using the time series signal characteristics.

[0084] The expressions of correlation analysis and KPCA method are:

[0085]

[0086] Among them, φ(X) is the feature representation mapped to the high-dimensional space by the kernel function K, α i is the projection coefficient. The state update expression of the bidirectional recurrent neural network is:

[0087] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0088] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0089]

[0090] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0091] h t =o t *tanh(C t )

[0092] Among them, ft Indicates the forgetting of historical signal data during processing, i t Indicates the acceptance of new signal data during the processing, C t Represents the long-term storage memory of signal data during processing and is updated at each time step. Indicates the update of C generated based on the current input data t Status information, h t Indicates the adjustment of the current prediction during the processing, o t is the output of the final data, x t represents the signal collected from the multi-source data sensor network system at the current moment, σ represents the Sigmoid activation function, ensuring that the signal output value is between (0,1), and W f Represents the weight matrix of the forget gate, W i Represents the weight matrix of the input gate, W C Represents the weight matrix for calculating the processing state update amount, W o represents the weight matrix of the output gate, b f represents the bias term of the forget gate, b i represents the bias term of the input gate, b C Represents the bias term used to update the current memory unit state, b o Represents the bias term h of the output gate t-1 represents the hidden state at the previous moment, C t-1 It represents the storage state of the tool displacement compensation at the previous moment, and tanh represents the hyperbolic tangent activation function, which normalizes the signal data to between (-1, 1).

[0093] S42. To improve the generalization capability of the model, based on the tool tip wear prediction model for a single working condition obtained in step S41, a feature extractor is constructed using a convolutional neural network (CNN) and a bidirectional long short-term memory network (BLSTM) with a single-working condition source domain data sample set as input and a target domain to be learned for a variable feed condition as output. This extracts multidimensional spatiotemporal features from the single-working condition source domain data sample set and the target domain to be learned for the new working condition, and constructs a domain-adaptive transfer model. The domain-adaptive transfer model primarily includes a feature extractor composed of a CNN and a BLSTM, a tool wear classifier composed of an FC layer, and a metric function, the Maximum Mean Discrepancy (MMD). Labeled samples from the source domain and unlabeled samples from the target domain are first input into the feature extractor. Through multi-source data fusion and nonlinear mapping feature modeling of sequence feature modeling, high-dimensional feature vectors are simultaneously extracted from the source and target domain samples. Multiple-kernel MMD (MK-MMD) is used to quantify and minimize the distribution differences of transferable features between domains to obfuscate the spatial distribution differences of source and target domain features, thereby realizing the transfer and sharing of knowledge and improving the generalization ability of the prediction model.

[0094] In a preferred embodiment of the present invention, a gun barrel deep hole processing device based on a magnetostrictive actuator is provided. Figure 2 As shown, it includes a spindle box 1, a guide rail 2, a boring bar assembly 3 and a floating support 4, the fixed end of the spindle box 1 is connected to the first mounting end of the guide rail 2, the spindle hole of the spindle box 1 is connected to the positioning component of the boring bar 304 in the boring bar assembly 3, the fixed end of the boring bar 304 in the boring bar assembly 3 is connected to the second mounting end of the guide rail 2 through the floating support 4, and the working end of the slide rail 302 in the boring bar assembly 3 is connected to the fixed end of the tool 301; the floating support 4 can move along the guide rail 2, and the boring bar 304 can move axially along the floating support 4, thereby driving the tool 301 to move axially along the boring bar 304, and the tool 301 can be driven by the magnetostrictive element 305 to telescopically move in a direction perpendicular to the axial direction of the boring bar 304, thereby realizing variable diameter boring.

[0095] Boring bar assembly 3, such as Figure 3As shown, the tool 301 includes a slide rail 302, a push rod 303, a boring bar 304, a magnetostrictive element 305, a permanent magnet 306, a coil 307, and a mounting plate 308. The mounting plate 308 is connected to the inner mounting end of the boring bar 304, and the fixed end of the mounting plate 308 is connected to the fixed end of the permanent magnet 306. The permanent magnet 306 is provided with a coil 307. The permanent magnet 306 is distributed in a ring along the axis of the magnetostrictive element 305. The fixed end of the magnetostrictive element 305 is connected to the first mounting end of the push rod 303, and the second mounting end of the push rod 303 is connected to the fixed end of the slide rail 302. The working end of the slide rail 302 is connected to the fixed end of the tool 301. The slide rail 302 can convert the axial movement of the push rod 303 into the longitudinal movement of the tool 301, and realize high-frequency micro-displacement control through current regulation. The magnetostrictive element 305 ensures precise adjustment of the tool 301 during the machining process.

[0096] The magnetostrictive element 305, located inside the boring bar 304, achieves high-frequency micro-displacement control through current regulation, ensuring precise adjustment of the tool 301 during machining. The control system is connected to various sensors in the multi-source data sensing network system, including a data acquisition module and a control module, responsible for real-time monitoring of machining status and adjusting control strategies. The magnetostrictive element 305 and the control system are connected by a cable, transmitting current signals to adjust the output of the magnetostrictive element 305.

[0097] A magnetic field is generated around the magnetostrictive element 305 by the coil 307 and the permanent magnet 306, causing the magnetostrictive element 305 to shorten or lengthen under the influence of the magnetic field. The coil 307 is controlled by a current control module. The control system of the machining device of the present invention adjusts the current intensity in the coil 307 through the current control module, thereby controlling the intensity of the magnetic field. The magnetostrictive element 305 expands and contracts under the influence of the magnetic field, thereby adjusting the axial movement of the push rod 303 inside the boring bar 304 in real time, promoting micron-level displacement of the push rod 303, and precisely controlling the feed and cutting depth of the tool 301.

[0098] The current sensor is located in the power supply path of the spindle system, monitoring the current changes of the spindle in real time, evaluating the cutting force and the load status of the tool, and providing early warning of tool 301 wear or cutting abnormalities; the vibration sensor is located in the clamping area of ​​the boring bar 304 and the tool 301, monitoring the vibration of the boring bar 304 and the tool 301 in real time, identifying potential vibration problems and the wear status of the tool 301, and optimizing processing stability; the acoustic emission sensor is located in the clamping area of ​​the tool 301, capturing the acoustic emission signal during the cutting process, and analyzing the wear status of the tool 301 and the cutting quality through the acoustic signal; each sensor in the multi-source data sensing network system monitors the data in real time and converts it into electrical signals, which are transmitted to the data acquisition module of the control system through wireless communication based on the OPC UA protocol for data acquisition. The control system is responsible for real-time monitoring of the processing status and adjusting the control strategy.

[0099] The following is a further description of a method and apparatus for deep hole machining of a gun barrel based on a magnetostrictive actuator in conjunction with an embodiment of the present invention:

[0100] First, the gun barrel 5 is installed on the guide rail 2 through the floating support 4, and the position of the gun barrel 5 is adjusted to ensure that the cutting trajectory of the tool 301 is aligned with the gun barrel 5 during the processing; the multi-source signals collected by the multi-source data sensing network system are input into the data acquisition module to ensure that the processing status can be monitored in real time; the control system is started and the initial parameters are set, including cutting speed, feed rate and current intensity; the current of the coil 307 located on the permanent magnet 306 is adjusted by the control system to change the magnetic field intensity, causing the magnetostrictive element 305 to extend or contract, and the push rod 303 to move forward and backward, thereby adjusting the tool 301 to extend and retract along the position perpendicular to the axis of the boring bar 304. During the boring process, the tool tip wear, vibration and current data of the tool 301 are collected in real time.

[0101] Then, a tool wear prediction model is established. Considering the unidirectional long stroke characteristics of the deep hole boring processing system of the gun barrel 5, the CNC system of the boring machine is used to read the signal of the current sensor located on the spindle box 1 in the multi-source data sensing network system. Acceleration sensors are installed on different parts of the boring bar 304 and the gun barrel 5, and acoustic emission sensors are evenly spaced along the length of the gun barrel 5. Multi-source signals are collected using the multi-source data sensing network system. The spatiotemporal characteristics and singularity of the multi-source signals are analyzed, and sliding frame sampling is used to intercept signal data of a specified length. The degree of dispersion is analyzed, and outliers that exceed or fall below three times the standard deviation of the mean are removed to obtain the initial data sample.

[0102] Then, using multi-source signals as data samples and the tool tip wear value of tool 301 as a label, a single-condition source domain data sample set is constructed. Based on the above source domain data samples, their time domain information is obtained, and the wavelet transform method is used to simultaneously decompose the denoised low-frequency and high-frequency parts to obtain frequency domain information. The Hilbert-Huang method is then used to obtain the instantaneous frequency components with actual physical significance in the signal, thereby achieving high-resolution time-frequency analysis. The multi-scale spatiotemporal features in the time-frequency information of multi-source signals are extracted through deep autoencoding technology, and feature dimensionality reduction is achieved using correlation analysis and KPCA methods. A tool wear prediction model based on a deep bidirectional recurrent neural network is established under a single working condition. The attention mechanism is introduced to adaptively calculate and adjust the input weights, and the nonlinear mapping relationship between the fusion of multi-source signal features and the tool tip wear sensitive features in tool 301 is explored. The domain adaptive migration model is constructed using the source domain samples of the original working condition as input and the target domain to be learned in the new working condition as output.

[0103] Finally, a pre-set tool wear model is used for real-time analysis. Based on the analysis results, the control strategy is dynamically adjusted to optimize the current output, achieving high-frequency micro-displacement control of tool 301 and ensuring tool tip stability throughout the entire machining process. After boring is completed, the equipment is shut down and the machining data is extracted for analysis and improvement. The machining quality is evaluated, and the straightness, roundness, and internal dimensions of the deep hole are checked to ensure that they meet design requirements.

[0104] like Figure 4 As shown, in an ideal machining process, the tip of the tool 301 moves stably along a predetermined trajectory, and within the allowable tolerance range, the contact surface between the tool 301 and the gun barrel 5 remains stable, ensuring machining accuracy; Figure 5 As shown in FIG, during high-speed cutting, the boring bar 304 will bend to a certain extent due to the cutting force and changes in processing conditions. The deflection of the boring bar 304 will directly affect the motion trajectory of the tool 301, especially in deep hole processing, where the deflection effect is particularly obvious due to the long length of the boring bar 304. Figure 6 As shown in FIG, due to the accumulation of the boring bar deflection effect, the motion trajectory of the tool 301 deviates from the ideal trajectory, resulting in an irregular machined surface, which in turn reduces the machining accuracy of the deep hole of the gun barrel 5, and the machining error is Δ1; Figure 7 As shown, during the contact process between the tool 301 and the gun barrel 5, the tip of the tool 301 will be worn due to the cutting of the material and the accumulation of heat. The wear of the tip of the tool 301 will gradually increase with the extension of the processing time. As the wear increases, the shape and position of the tip of the tool 301 change, which directly leads to the deviation of the trajectory of the tool 301. Figure 8As shown, due to wear on the tip of tool 301, the tool's trajectory has significantly shifted, resulting in errors in deep hole machining. This error primarily manifests itself in a change in the tip position of tool 301, preventing it from cutting along the ideal trajectory during machining, leading to a dimensional deviation Δ1 in the deep hole machining of the gun barrel 5. To compensate for the errors caused by the deflection of the boring bar 304 and the wear of the tool's tip, a multi-source data sensing network system is used to monitor dynamic signals during machining, particularly the status of tool 301 and the vibration signals of boring bar 304. By acquiring acoustic emission and vibration signals in real time, the control system adjusts the current in coil 307 on permanent magnet 306 based on the transmitted signals, thereby changing the magnetic field strength, causing the magnetostrictive element 305 to expand and contract, which in turn drives the axial movement of push rod 303, causing tool 301 to adjust its expansion and contraction displacement L along slide rail 302. This compensates for the deviation in the tool's tip trajectory, ensuring that the tool's tip trajectory conforms to the ideal trajectory and thus ensuring machining accuracy.

[0105] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A method for deep hole machining of a gun barrel based on a magnetostrictive actuator, characterized in that: It includes the following steps: S1. Mount the gun barrel with a large aspect ratio to be machined on a guide rail via a floating support. Mount the tool on the working end of a boring bar. Connect the driving end of the boring bar to the spindle box. Mount the fixed end of the boring bar on the guide rail via a floating support. S2. Installing the multi-source data sensor network system at corresponding positions on the boring bar, the tool, and the gun barrel, and setting cutting parameters for the tool; S3. Start the spindle box and the multi-source data sensing network system respectively, process the acoustic emission signals and vibration signals collected in real time by the multi-source data sensing network system, and extract the time-frequency information about the tool tip displacement from the acoustic emission signals and the vibration signals; S4. Based on the characteristics of deep hole machining for artillery barrels, a tool wear prediction model is established, which specifically includes the following sub-steps: S41. Using the state update expression of the bidirectional recurrent neural network, with the acoustic emission signal and vibration signal collected in step S3 as the core input and the tool tip displacement as the output, a tool tip wear prediction model under a single working condition is established, and the nonlinear mapping relationship between the characteristics of the multi-source signals and the tool tip wear sensitive characteristics of the tool is mined, wherein the state update expression of the bidirectional recurrent neural network is: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(C t ) Among them, f t Indicates the forgetting of historical signal data during processing, i t Indicates the acceptance of new signal data during the processing, C t Represents the long-term storage memory of signal data during processing and is updated at each time step. Indicates the update of C generated based on the current input data t Status information, h t Indicates the adjustment of the current prediction during the processing, o t is the output of the final data, x t represents the signal collected from the multi-source data sensor network system at the current moment, σ represents the Sigmoid activation function, ensuring that the signal output value is between (0,1), and W f Represents the weight matrix of the forget gate, W i Represents the weight matrix of the input gate, W C Represents the weight matrix for calculating the processing state update amount, W o represents the weight matrix of the output gate, b f represents the bias term of the forget gate, b i represents the bias term of the input gate, b C Represents the bias term used to update the current memory unit state, b o Represents the bias term h of the output gate t-1 represents the hidden state at the previous moment, C t-1 Indicates the storage state of the tool displacement compensation at the previous moment, tanh represents the hyperbolic tangent activation function, and normalizes the signal data to between (-1,1); S42. Based on the tool tip wear prediction model under the single working condition obtained in step S41, the single working condition source domain data sample set is used as input, and the target domain to be learned under the variable feed working condition is used as output. A feature extractor is constructed using a convolutional neural network and a bidirectional long short-term memory network to extract multidimensional spatiotemporal features of the single working condition source domain data sample set and the target domain to be learned under the variable feed working condition, and a domain adaptive migration model is constructed. S5. Input the time-frequency information of the acoustic emission signal and the vibration signal obtained in step S3 into the tool wear prediction model constructed in step S4, calculate the tool wear state, and generate a current signal for the control system; S6. Input the current signal obtained in step S5 to the coil wound around the permanent magnet to change the magnetic field strength, thereby controlling the expansion and contraction of the magnetostrictive element, driving the push rod to perform micron-level displacement, so that the tip of the tool and the gun barrel maintain contact, thereby achieving fine-tuning of the tool displacement. Continuously collect and monitor acoustic emission signals and vibration signals to form a closed-loop feedback control.

2. The method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to claim 1, characterized in that: The position distribution of the acceleration sensor on the boring bar is obtained by analyzing the sensitivity of the acceleration sensor signals at different installation positions on the boring bar to the tool tip wear using the Pearson correlation coefficient method. The expression of the Pearson correlation coefficient is: Among them, ρ X,Y is the signal of the acceleration sensor at different installation positions on the boring bar, X and Y are the correlation coefficients in the Pearson correlation coefficient method, cov(X,Y) is the covariance of X and Y, σ X and σ Y are the standard deviations of X and Y respectively.

3. The method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to claim 1, characterized in that: In step S2, the cutting parameters include cutting speed, feed rate and current intensity; in step S3, the multi-source signals include tool position deviation value, cutting force abnormal value, vibration abnormal value and acoustic signal characteristic value.

4. The method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to claim 3, characterized in that: The specific implementation steps of step S3 for extracting time-frequency information from multi-source signals are as follows: S31. Based on the characteristics of tool wear and machining conditions, a sliding frame sampling method is used to intercept signal data of a specified length, and the discreteness of the signal data is analyzed. Abnormal data points exceeding or falling below three times the standard deviation of the mean are removed to obtain an initial data sample. S32. Based on the initial data samples, a single working condition source domain data sample set is constructed using the tool tip wear value as a label; S33. Based on the single-operating-condition source domain data sample set obtained in step S32, a wavelet transform method is used to simultaneously decompose the low-frequency and high-frequency parts of the single-operating-condition source domain data sample set to obtain frequency domain information. The expression of the wavelet transform method is: Among them, W a,b (t) represents the time-frequency local characteristics of machining vibration, acceleration and acoustic emission signals at different scales and positions obtained after wavelet transform, and x(t) represents the original signals obtained by multi-source sensors during the actual machining process. represents the complex conjugate of the mother wavelet function used in wavelet transform, a and b represent the different levels and time characteristics of tool wear respectively; S34. The frequency domain information obtained in S33 is processed using the Hilbert-Huang method to extract the instantaneous frequency components of the vibration frequency change and cutting force fluctuation caused by tool wear, thereby obtaining the time-frequency information in the multi-source signal. The expression of the Hilbert-Huang method is: Among them, ω(t) represents the instantaneous frequency characteristics of the signal caused by the change in the contact characteristics between the tool and the gun barrel, and θ(t) represents the phase characteristics caused by the change in the tool vibration and machining force with the degree of wear.

5. The method for deep hole machining of a gun barrel based on a magnetostrictive actuator according to claim 4, characterized in that: In step S31, the judgment expression of the abnormal data point is: Where z is the normalized value, X is the signal data, μ is the mean of the signal data set, and σ is the standard deviation of the signal data set. If |z|>3, the data is considered an outlier.

6. A gun barrel deep hole processing device based on a magnetostrictive actuator, characterized in that: It includes the spindle box, guide rails, boring bar assembly and floating support. The fixed end of the spindle box is connected to the first mounting end of the guide rail, the output end of the spindle box is connected to the driving end of the boring bar in the boring bar assembly, the fixed end of the boring bar in the boring bar assembly is connected to the second mounting end of the guide rail through a floating support, and the working end of the slide rail in the boring bar assembly is connected to the fixed end of the tool; the boring bar assembly includes a tool, a slide rail, a push rod, a boring bar, a magnetostrictive element, a permanent magnet, a coil and a mounting plate, the mounting plate is connected to the mounting end inside the boring bar, the fixed end of the mounting plate is connected to the fixed end of the permanent magnet, the permanent magnet is provided with a coil, and the permanent magnets are distributed in an annular manner along the axis of the magnetostrictive element, the fixed end of the magnetostrictive element is connected to the first mounting end of the push rod, the second mounting end of the push rod is connected to the fixed end of the slide rail, and the working end of the slide rail is connected to the fixed end of the tool.

7. The gun barrel deep hole processing device based on magnetostrictive actuator according to claim 6 is characterized in that: It also includes a multi-source data sensing network system, which includes a current sensor, a vibration sensor, an acceleration sensor and an acoustic emission sensor. The current sensor is installed on the spindle box, the vibration sensor is installed at the connection position between the boring bar and the tool, the acoustic emission sensor is installed in the working area of ​​the tool, the acoustic emission sensors are evenly spaced along the length direction of the gun barrel, and the acceleration sensors are respectively installed at the driving end, middle and fixed end of the boring bar and at the position of the gun barrel near the processing hole.

Images