A surface integrity online detection method based on multi-source information fusion

Abstract

The application relates to the technical field of numerical control machining state monitoring, and particularly discloses a surface integrity online detection method based on multi-source information fusion, which comprises the following steps: synchronously collecting cutting force signals and mechanical vibration signals in the machining process of a target workpiece, reconstructing a phase space, obtaining a first phase space trajectory and a second phase space trajectory, calculating a phase space state transition correlation matrix of the first phase space trajectory and the second phase space trajectory, converting to generate a nonlinear dynamics state transition atlas, extracting a structure distribution characteristic parameter and inputting the structure distribution characteristic parameter into a first mapping model and a second mapping model which are constructed in advance, and respectively outputting a roughness prediction value and a residual stress prediction value of a current machining surface of the target workpiece; and when the structure distribution characteristic parameter has a decrease or an increase exceeding a warning threshold value within a preset time window, it is determined that the target workpiece has a surface integrity trend deterioration, and a warning signal is sent. The application can analyze multiple signal couplings and realize multi-index online monitoring.

Description

Technical Field

[0001] This invention relates to the field of CNC machining condition monitoring technology, and more specifically to an online surface integrity detection method based on multi-source information fusion. Background Technology

[0002] Surface roughness and residual stress are core indicators for evaluating the surface quality of machined surfaces. Traditional offline inspection methods, such as stylus profilometers, can disrupt production cycles. In machine measurement methods, analysis based on a single vibration signal is susceptible to interference from machine tool resonance and environmental noise, resulting in a high false alarm rate under varying depth of cut and rotational speed conditions. While analysis based on a single cutting force signal can reflect load changes, it is difficult to capture system damping changes caused by the microscopic cutting edge state of the tool.

[0003] Furthermore, existing signal processing methods, such as the Fast Fourier Transform, struggle to characterize the nonlinear deterministic relationship between cutting force and vibration. When the tool is in a chaotic edge state transitioning from stable cutting to micro-chatter, the surface roughness has already deteriorated, but conventional time-frequency domain characteristics often have not yet exceeded limits, resulting in detection lag.

[0004] Therefore, how to overcome the technical shortcomings of existing single-sensor monitoring methods, such as high false alarm rate under varying working conditions and inability to simultaneously characterize the coupling relationship between surface roughness and residual stress, in order to obtain accurate online characterization of surface integrity during processing, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention proposes an online surface integrity detection method based on multi-source information fusion, so as to overcome the above problems or at least partially solve the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An online surface integrity detection method based on multi-source information fusion includes the following steps: S1. Synchronously acquire cutting force signals and mechanical vibration signals during the machining process of the target workpiece; S2. Perform phase space reconstruction on the cutting force signal and the mechanical vibration signal respectively to obtain the first phase space trajectory and the second phase space trajectory; S3. Based on a preset adaptive threshold related to the background noise generated by the machine tool idling, calculate the phase space state transition correlation matrix of the first phase space trajectory and the second phase space trajectory, present the phase space state transition correlation matrix as a binary image, and generate a nonlinear dynamic state transition map characterizing the degree of nonlinear coupling between excitation and response. S4. Extract the structural distribution characteristic parameters from the nonlinear dynamic state transition map; S5. Input the structural distribution characteristic parameters into the pre-constructed first mapping model and second mapping model respectively, and output the roughness prediction value and residual stress prediction value of the current machined surface of the target workpiece respectively. S6. Within a preset time window, when the decrease or increase of the structural distribution characteristic parameter exceeds the warning threshold, it is determined that the target workpiece has a trend of surface integrity deterioration, and a warning signal is issued.

[0008] Furthermore, S2 includes: S21. Extract a signal sample in a stable cutting state from the synchronously acquired cutting force signal and mechanical vibration signal. S22. Determine the optimal delay time of the cutting force signal and the mechanical vibration signal under stable cutting conditions using mutual information functions respectively; S23. The optimal embedding dimension of the cutting force signal and the mechanical vibration signal are determined by the pseudo nearest neighbor method, respectively. S24. Based on the optimal delay time and optimal embedding dimension of the cutting force signal, construct the first phase space trajectory; based on the optimal delay time and optimal embedding dimension of the mechanical vibration signal, construct the second phase space trajectory.

[0009] Furthermore, S22 includes: S221. Divide the amplitude range of the cutting force signal under stable cutting conditions into b equal-width quantization intervals bin; S222, Delay time t Starting from 0, with sampling interval Δ t To increase the step size incrementally, different values ​​are calculated using the mutual information function. t The mutual information value corresponding to the value is used to obtain the mutual information function curve; S223, the value corresponding to the first minimum value of the mutual information function curve. t The value is taken as the optimal delay time of the cutting force signal. The delayed signal corresponding to the optimal delay time provides the maximum amount of new information to the original signal. S224. Divide the amplitude range of the mechanical vibration signal under stable cutting conditions into b equal-width quantization intervals bin, and execute S222-S223 to obtain the optimal delay time of the mechanical vibration signal.

[0010] Furthermore, in S222, the expression for the mutual information function is:

[0011] in, The probability that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval bin; Delay of the original cutting force signal or the original mechanical vibration signal t After stepping into the first j The probability of a quantization interval bin of equal width; This indicates that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval, while being delayed. t After stepping into the first j The joint probability of equal-width quantization intervals; bin represents the equal-width quantization interval of the amplitude range. b Indicates the total number of equal-width quantization intervals. t Represents the delay time variable. I ( t ) represents the mutual information value.

[0012] Furthermore, S23 includes: S231. For the cutting force signal, starting from the embedding dimension m=1, and increasing step size 1 successively, perform false nearest neighbor analysis on each embedding dimension m. S232. For each embedding dimension m, utilize the optimal latency time. t F The cutting force signal is reconstructed in phase space to generate N m-dimensional state vectors; S233. For each m-dimensional state vector Find its nearest neighbor vector Make the Euclidean distance between the two Minimum, construct the original nearest neighbor pair; S234. Increase the embedding dimension to m+1, reconstruct the m+1-dimensional extended vector of each state vector, and calculate the Euclidean distance between the original nearest neighbor pairs in the m+1-dimensional space. ; S235. Define the distance ratio criterion for the i-th state vector. R F ( i ),when R F ( i )> R th When the original nearest neighbor pair is determined to be a false nearest neighbor pair, then, R th To determine the threshold; S236. Count the total number of pairs of points identified as false nearest neighbors among all N state vectors under the current embedding dimension m. And calculate the total number of false nearest neighbor pairs. The proportion of false nearest neighbors is the percentage of the total number of points in the state vector. ; S237. Let m = m + 1, and repeat S232-S236 until the proportion of false nearest neighbors is reached. When the cutting force signal first drops below a preset threshold or stabilizes as the embedding dimension m increases, the corresponding embedding dimension is taken as the optimal embedding dimension. m F ; S238. For the mechanical vibration signal, repeat steps S231-S237 to obtain the optimal embedding dimension of the mechanical vibration signal. m A .

[0013] Furthermore, S24 includes: S241, Optimal delay time based on the cutting force signal t F and optimal embedding dimension m F For each sampling time i Construct a m F dimensional state vector , ,in, i The range of values ​​is i =1,2,…, N F ; N F This represents the total number of phase space trajectory points obtained after reconstructing the cutting force signal. N F ,satisfy: Where L is the total number of sampling points for the cutting force signal; Arrange the state vectors at all times in chronological order to form the first phase space trajectory, denoted as: ; S25. Based on the optimal delay time of the mechanical signal t A and optimal embedding dimension m A For each sampling time i Construct a m A dimensional state vector , ,in, i The range of values ​​is i =1,2,…, N A ; N A This represents the total number of phase space trajectory points obtained after reconstructing the mechanical signal, satisfying: ; The state vectors at all times are arranged in chronological order to form the second phase space trajectory. ; S26, Take N F and N A The smaller of the two is used as the uniform length for constructing the phase space state transition correlation matrix.

[0014] Furthermore, in S3, the formula for calculating the elements in the phase space state transition correlation matrix is ​​as follows:

[0015] in, The elements represent the phase space state transition correlation matrix. This indicates the phase space adjacency state value of the cutting force channel. This indicates the phase space adjacency state value of the vibration channel. This represents the logical AND operator; This represents the adaptive threshold for the neighborhood of the cutting force channel. This represents the adaptive threshold of the vibration channel neighborhood. and These represent the two phase space vectors of the cutting force signal, and These represent two phase space vectors of the mechanical vibration signal, with subscript i indicating the index of the first sampling time and j indicating the index of the second sampling time. It is the Heaviside step function.

[0016] Furthermore, adaptive threshold and The determination process includes: While the machine tool spindle is rotating but the tool has not yet contacted the workpiece, the output signals of the cutting force sensor and the vibration acceleration sensor are recorded simultaneously, and the standard deviations of the two output signals are calculated to obtain the standard deviation of the cutting force channel background noise. and the standard deviation of the background noise of the vibration channel ; The standard deviation of the background noise of the cutting force channel is respectively... and the standard deviation of the background noise of the vibration channel Multiply by a preset floating coefficient to obtain the adaptive threshold of the cutting force channel neighborhood. Adaptive threshold for vibration channel neighborhood ; At preset time intervals, the output signals of the cutting force sensor and vibration acceleration sensor under idling conditions are re-acquired within the last few seconds, and the standard deviation of the two output signals is recalculated. and And update the adaptive threshold in real time. and .

[0017] Furthermore, in S4, the structural distribution characteristic parameters include at least the state transition determinism index and the state dwell stability index; in S6, when the decrease of the state transition determinism index exceeds the warning threshold and the increase of the state dwell stability index exceeds the warning threshold within a preset time window, it is determined to be a trend of surface integrity deterioration caused by tool wear, built-up edge, or cutting chatter. The process of extracting the state transition determinism index includes: Traverse the line segments with continuous diagonal values ​​of 1 in the nonlinear dynamic state transition map. When the cutting force signal and the mechanical vibration signal are both state transition coincident points at the same coordinate position, the corresponding element takes a value of 1; otherwise, it takes a value of 0. When the value is 1, it is represented by a black pixel; when the value is 0, it is represented by a white pixel. Define the length as The line segment is a state transition diagonal. The number of state transition diagonals of various lengths appearing in the nonlinear dynamic state transition spectrum is counted to obtain a histogram of diagonal length frequency distribution. And exclude extremely short lines whose length is less than the preset length; The state transition determinism index is calculated according to the following formula:

[0018] in, l Indicates the length of the diagonal of the state transition. l min This represents the minimum statistical length of the diagonal of the state transition. L max This indicates the maximum length of the diagonal during the state transition. This indicates that the length of the diagonal is l The number of line segments, This represents the element in the i-th row and j-th column of the phase space state transition correlation matrix; The process of extracting the state dwell stability index includes: Traverse all line segments with a continuous value of 1 in all column directions of the nonlinear dynamic state transition diagram; Define a line segment of length v as a state-stationary perpendicular line, and count the frequency of occurrence of state-stationary perpendicular lines of various lengths to obtain the frequency distribution of perpendicular line lengths. P ( v ), and exclude extremely short lines whose length is less than the preset length; The state-dwelling stability index is calculated using the following formula:

[0019] in, Indicates the length of the vertical line where the state resides. v min The minimum statistical length of the state-resident vertical line, V max This indicates the maximum statistical length of the vertical line where the state resides. The length of the vertical line indicating the state is 1. v The number of line segments.

[0020] Furthermore, in S5, the construction process of the first mapping model and the second mapping model includes: A single-factor cutting experiment was conducted by changing only the target machining parameters and keeping the other machining parameters fixed. Cutting was performed under each set of single-factor cutting experimental parameters. Cutting force signals and mechanical vibration signals in the machining area were collected simultaneously, and structural distribution parameters were extracted according to S2-S4 as input features. The surface roughness and residual stress values ​​obtained from the actual measurements under the single-factor cutting experimental parameters for each group are used as output labels; Gaussian process regression algorithm is used to establish a first mapping model between input features and surface roughness, and a second mapping model between input features and residual stress.

[0021] As can be seen from the above technical solution, compared with the prior art, the present invention has the following beneficial effects: This invention effectively characterizes the nonlinear coupling relationship between excitation and response by simultaneously acquiring cutting force signals and mechanical vibration signals and constructing a nonlinear dynamic state transition spectrum of the two. This reflects the true dynamic behavior of the machining system better than single signal analysis, thus improving monitoring accuracy.

[0022] This invention extracts multiple structural distribution characteristic parameters as input features and establishes mapping models between the input features and surface roughness and residual stress, which can simultaneously output predicted values ​​of multiple indicators. This enables online monitoring and quantitative evaluation of multiple surface integrity indicators without interrupting the processing.

[0023] This invention achieves an optimal balance between sensitivity that preserves the true dynamic changes and robustness that suppresses noise interference by collecting the background noise during the machine tool's idle phase and setting an adaptive threshold based on the background noise. This avoids the subjective bias of manually setting parameters, enhances the robustness and versatility of the method, and makes the phase space state transition correlation matrix more accurately reflect the synchronous state transition mode of the phase space trajectories of cutting force signal and vibration signal, providing a key basis for subsequent feature extraction and surface integrity assessment. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] Figure 1 This is a flowchart of the online surface integrity detection method based on multi-source information fusion provided in this embodiment of the invention; Figure 2 This is a flowchart illustrating the generation process of the first phase spatial trajectory and the second phase spatial trajectory provided in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figure 1 As shown in the figure, this invention discloses an online surface integrity detection method based on multi-source information fusion, including the following steps: S1. Synchronously acquire cutting force signals and mechanical vibration signals during the machining process of the target workpiece; S2. Perform phase space reconstruction on the cutting force signal and the mechanical vibration signal respectively to obtain the first phase space trajectory and the second phase space trajectory; S3. Based on a preset adaptive threshold related to the background noise generated by the machine tool idling, calculate the phase space state transition correlation matrix of the first phase space trajectory and the second phase space trajectory, present the phase space state transition correlation matrix as a binary image, and generate a nonlinear dynamic state transition map characterizing the degree of nonlinear coupling between excitation and response. S4. Extract structural distribution characteristic parameters from the nonlinear dynamic state transition map; S5. Input the structural distribution characteristic parameters into the pre-built first mapping model and second mapping model respectively, and output the roughness prediction value and residual stress prediction value of the current machined surface of the target workpiece respectively. S6. Within a preset time window, when the decrease or increase of the structural distribution characteristic parameters exceeds the warning threshold, it is determined that the target workpiece has a trend of surface integrity deterioration, and a warning signal is issued.

[0028] The specific implementation methods of the above steps will be further explained below.

[0029] S1. Simultaneously acquire cutting force and mechanical vibration signals during the machining process of the target workpiece. The cutting force signal is acquired using a piezoelectric force sensor mounted on the tool, with a sampling frequency of no less than 2 kHz; the mechanical vibration signal is acquired using a triaxial accelerometer mounted radially on the spindle box, with a sampling frequency of no less than 10 kHz. Then, the original cutting force and mechanical vibration signals are preprocessed, specifically using a Butterworth filter for noise reduction. After filtering and noise reduction, the signals are Z-score standardized. After standardization, both the cutting force and vibration signals are converted into standardized sequences with a mean of 0 and a standard deviation of 1.

[0030] S2. Perform phase space reconstruction on the cutting force signal and the mechanical vibration signal respectively to obtain the first phase space trajectory and the second phase space trajectory. For example... Figure 2 As shown, the specific steps include: S21. From the synchronously acquired cutting force signal and mechanical vibration signal, extract a signal sample in a stable cutting state. The sample length is denoted as... L The purpose of selecting the stable segment is to eliminate the influence of transient processes such as tool feed and retraction on subsequent statistical analysis, and to ensure that the extracted time delay parameters reflect the inherent dynamic characteristics of the cutting system under steady-state conditions.

[0031] S22. Determine the optimal delay times for the cutting force signal and mechanical vibration signal under stable cutting conditions using mutual information functions, including: S221. Divide the amplitude range of the cutting force signal under stable cutting conditions into b equal-width quantization intervals bin.

[0032] S222, Delay time t Starting from 0, with sampling interval Δ t To increase the step size incrementally, different values ​​are calculated using the mutual information function. t The mutual information value corresponding to the value is used to obtain the mutual information function curve; the expression of the mutual information function is:

[0033] in, The probability that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval bin; Delay of the original cutting force signal or the original mechanical vibration signal t After stepping into the first j The probability of a quantization interval bin of equal width; This indicates that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval, while being delayed. t After stepping into the first jThe joint probability of equal-width quantization intervals; bin represents the equal-width quantization interval of the amplitude range. b Indicates the total number of equal-width quantization intervals. t Represents the delay time variable. I ( t ) represents the mutual information value.

[0034] S223, with t The increase, I ( t ) Usually from the maximum value ( t Mutual information is at its maximum when the mutual information function curve reaches zero (i.e., the signal is perfectly correlated with itself), and then gradually decays. The point at which the mutual information function curve reaches its first minimum value is the point where the mutual information function curve corresponds to this maximum value. t The value is used as the optimal delay time for the cutting force signal. The delay signal corresponding to the optimal delay time provides the maximum amount of new information to the original signal. This ensures that the coordinate components of the reconstructed phase space are not highly redundant due to excessive delay, and also avoids the loss of dynamic continuity due to excessive delay.

[0035] S224. Divide the amplitude range of the mechanical vibration signal under stable cutting conditions into b equal-width quantization intervals bin, and execute S222-S223 to obtain the optimal delay time of the mechanical vibration signal.

[0036] S23. The optimal embedding dimension for the cutting force signal and mechanical vibration signal is determined using the pseudo-nearest neighbor method. The pseudo-nearest neighbor ratio threshold is set to 0.1, meaning that the embedding dimension is the selected value when the percentage of pseudo-nearest neighbors first drops below 10%. In phase space reconstruction theory, when the embedding dimension is too low, two state points that are far apart in the high-dimensional phase space and belong to different dynamic trajectories will be incorrectly compressed and projected into the low-dimensional phase space, becoming adjacent to each other, forming so-called pseudo-nearest neighbors. As the embedding dimension gradually increases, these pseudo-nearest neighbors will gradually separate due to the full expansion of the phase space. When the embedding dimension increases to a certain critical value, the proportion of pseudo-nearest neighbors drops to zero or approximately zero, indicating that the phase space has been fully expanded. The corresponding embedding dimension at this point is the minimum fully embedded dimension. m Specifically, this includes: S231. For the cutting force signal, starting from the embedding dimension m=1, and increasing by a step size of 1, perform spurious nearest neighbor analysis for each embedding dimension m.

[0037] S232. For each embedding dimension m, utilize the optimal latency time. t F The cutting force signal is reconstructed in phase space to generate N m-dimensional state vectors, represented as follows:

[0038] in, This represents the m-dimensional cutting force phase space state vector.

[0039] S233. For each m-dimensional state vector Find its nearest neighbor vector Make the Euclidean distance between the two The minimum, the Euclidean distance between the two is denoted as:

[0040] in, express m The nearest neighbor vector in 3D space, This represents the Euclidean distance.

[0041] and This is a pair of original nearest neighbor points.

[0042] S234. Increase the embedding dimension to m+1 and reconstruct the m+1-dimensional extended vectors of each state vector. Calculate the Euclidean distance between the original nearest neighbor pairs in the (m+1)-dimensional space. :

[0043] in, express m +1-dimensional cutting force phase space state vector express m +1-dimensional space nearest neighbor vector.

[0044] S235. Define the distance ratio criterion for the i-th state vector. R F ( i ):

[0045] in, Let represent the Euclidean distance in m-dimensional space. This represents the Euclidean distance in m+1 dimensional space. A threshold value is set. R th Typically, an empirical value of 10 to 15 is used. When R F ( i )> R th When the original nearest neighbor pair is determined to be a false nearest neighbor pair, the pair is identified as such.

[0046] S236. Count the total number of pairs of points identified as false nearest neighbors among all N state vectors under the current embedding dimension m. And calculate the total number of false nearest neighbor pairs. The proportion of false nearest neighbors is the percentage of the total number of points in the state vector. Its expression is:

[0047] In the formula, N This represents the total number of points in the state vector. This represents the number of false nearest neighbors in N state vectors. The first drop below the preset threshold, or Follow m When the cutting force signal increases and then stabilizes, no longer decreasing significantly, the minimum embedding dimension is the optimal embedding dimension of the cutting force signal. m F ; S237. Let m = m + 1, and repeat S232-S236. m The gradual increase yields a sequence of false nearest neighbor proportions. , , ... When the proportion of fake nearest neighbors... When the cutting force signal first drops below a preset threshold or stabilizes and no longer decreases significantly with increasing embedding dimension m, the corresponding embedding dimension is taken as the optimal embedding dimension. m F ; S238. For the mechanical vibration signal, repeat steps S231-S237 to obtain the optimal embedding dimension of the mechanical vibration signal. m A .

[0048] S24. Based on the optimal delay time and optimal embedding dimension of the cutting force signal, construct the first phase spatial trajectory; based on the optimal delay time and optimal embedding dimension of the mechanical vibration signal, construct the second phase spatial trajectory, specifically including: S241. Reconstruct the phase space of the cutting force signal: Suppose the original dynamic cutting force signal is a one-dimensional time series sampled at equal intervals:

[0049] in, L This represents the total number of sampling points for the cutting force signal.

[0050] Optimal delay time based on cutting force signal t F and optimal embedding dimension m F For each sampling time i Construct a m F dimensional state vector :

[0051] in, i The range of values ​​is i =1,2,…, N F ; N F This represents the total number of points in the phase space trajectory obtained after reconstructing the cutting force signal. N F ,satisfy: ; Arrange the state vectors at all times in chronological order to form the first phase space trajectory, denoted as: .

[0052] S25. Reconstruct the phase space of the mechanical vibration signal: Let the original mechanical vibration signal be:

[0053] in, L This represents the total number of sampling points for the mechanical vibration signal.

[0054] Optimal delay time based on mechanical signal t A and optimal embedding dimension m A For each sampling time i Construct a m A dimensional state vector :

[0055] in, i The range of values ​​is i =1,2,…, N A ; N A This represents the total number of points in the phase space trajectory obtained after reconstructing the mechanical signal, satisfying: ; Arranging the state vectors at all times in chronological order forms the second phase space trajectory. ; S26, due to t F and t A , m F and m A They are usually different due to differences in signal characteristics. N F and NA They may not be equal. In practice, take... N F and N A The smaller of N =min( N F , N A This serves as a uniform length for constructing the phase space state transition correlation matrix, ensuring a one-to-one correspondence between the two phase space trajectories in the time index.

[0056] S3. Construction of nonlinear dynamic state transition maps, specifically including: S31. Based on the first phase space trajectory and the second phase space trajectory, calculate the phase space state transition correlation matrix of the two phase space trajectories, where the formula for calculating a certain element in the phase space state transition correlation matrix is:

[0057] in, The elements represent the phase space state transition correlation matrix. This indicates the phase space adjacency state value of the cutting force channel. This indicates the phase space adjacency state value of the vibration channel. This represents the logical AND operator; This represents the adaptive threshold for the neighborhood of the cutting force channel. This represents the adaptive threshold of the vibration channel neighborhood. and These represent the two phase space vectors of the cutting force signal, and These represent two phase space vectors of the mechanical vibration signal, with subscript i indicating the index of the first sampling time and j indicating the index of the second sampling time. It is the Heaviside step function.

[0058] Among them, adaptive threshold and The determination process includes: 1) During the idle state where the machine tool spindle is rotating but the tool has not yet contacted the workpiece, the output signals of the cutting force sensor and the vibration acceleration sensor are recorded simultaneously. The signals during this time period do not contain cutting information and only reflect background random fluctuations caused by the machine tool transmission system, hydraulic system, and electromagnetic interference. Then, the standard deviations of the two output signals are calculated to obtain the standard deviation of the cutting force channel background noise. and the standard deviation of the background noise of the vibration channel .

[0059] 2) The standard deviation of the background noise of the cutting force channel is calculated separately. and the standard deviation of the background noise of the vibration channel Multiply by a preset floating coefficient to obtain the adaptive threshold of the cutting force channel neighborhood. Adaptive threshold for vibration channel neighborhood :

[0060]

[0061] In the formula, This represents the standard deviation of the background noise in the cutting force channel. This represents the standard deviation of the background noise in the vibration channel. The 1.5 factor is based on the statistical law of normal distribution; approximately 86.6% of random noise fluctuations will fall within ±1.5 standard deviations. Using this factor achieves an optimal balance between sensitivity in preserving true dynamic changes and robustness in suppressing noise interference. If the distance between the spatial vectors of two cutting forces is ≤1.5 times the noise fluctuation amplitude, they are considered to be essentially in the same dynamic state (the difference is only caused by noise); if the distance is >1.5 times the noise fluctuation amplitude, they are considered to have shifted to different dynamic states (the difference is caused by the actual changes in the cutting process).

[0062] 3) Incorporate the adaptive threshold into the construction of the phase space state transition correlation matrix.

[0063] 4) At preset time intervals, re-acquire the output signals of the cutting force sensor and vibration acceleration sensor during the most recent few seconds of idling, and recalculate the standard deviation of the two output signals. and And update the adaptive threshold in real time. and This ensures that the neighbor determination circle always matches the noise level of the current environment.

[0064] S42. In the phase space state transition correlation matrix, the corresponding element is set to 1 only when the cutting force signal and the vibration signal are both state transition coincident points at the same coordinate position; otherwise, it is set to 0. The phase space state transition correlation matrix is ​​visualized as a binary image, generating a force-vibration coupled nonlinear dynamic state transition map. In the map, black pixels represent state transition coincident points, and white pixels represent non-state transition coincident points. This map intuitively demonstrates the synchronous state transition mode of the phase space trajectories of the cutting force signal and the vibration signal. The continuity and degree of breakage of its diagonal structure directly reflect the strength of the excitation-response nonlinear coupling relationship, providing a crucial basis for subsequent feature extraction and surface integrity assessment.

[0065] S4. Extract the structural distribution characteristic parameters from the nonlinear dynamic state transition spectrum, perform quantitative recursive analysis on the nonlinear dynamic state transition spectrum, and extract the state transition deterministic index and the state residence stability index.

[0066] 1) The process of extracting the deterministic index of state transition includes: Traverse the line segments in the nonlinear dynamic state transition diagram where the diagonal direction is continuously 1.

[0067] Define the length as The line segment is a state transition diagonal. By counting the number of state transition diagonals of various lengths appearing in the state transition spectrum of nonlinear dynamics, a histogram of the frequency distribution of diagonal lengths is obtained. ,in l =1,2,…, L max , L max This is the length of the longest diagonal in the graph. Excludes graphs with lengths less than the preset length. l min Extremely short-term, l min The usual value is 2.

[0068] The state transition determinism index is calculated using the following formula:

[0069] in, l Indicates the length of the diagonal of the state transition. l min This represents the minimum statistical length of the diagonal of the state transition. L max This indicates the maximum length of the diagonal during the state transition. This indicates that the length of the diagonal is l The number of line segments, Let represent the element in the i-th row and j-th column of the phase space state transition correlation matrix; in the above expression, the molecule This characterizes the weighted contribution of the total diagonal length of the matrix (i.e., the continuous step size of the system state repetition). When the system state is dominated by deterministic laws, the phase space trajectory will form a long and continuous diagonal structure, i.e. l Larger values P(l) Dominantly, these long diagonals contribute significantly to the molecule, making JDET The value approaches 1. Conversely, if the system is dominated by random noise, the phase space trajectory will exhibit isolated and scattered state points, lacking a long diagonal structure. JDET The value decreased significantly. During the processing, JDETA value close to 1 indicates that the relative motion between the tool and the workpiece exhibits a high degree of regularity (such as controlled feed motion and periodic chatter). At this time, the system is dominated by deterministic physical laws, the machining process is stable, and thus good surface roughness is maintained. JDET A decrease in the value indicates that the deterministic structure is disrupted, random disturbances increase, and processing quality deteriorates.

[0070] 2) The process of extracting the state dwell stability index includes: Traverse all line segments with a continuous value of 1 in all column directions of the nonlinear dynamic state transition diagram.

[0071] Define a line segment of length v as a state-stationary perpendicular line, and count the frequency of occurrence of state-stationary perpendicular lines of various lengths to obtain the frequency distribution of perpendicular line lengths. P ( v ),in v =1,2,…, V max Exclude lengths smaller than the preset length. v min Extremely short-term, v min The usual value is 2.

[0072] The state dwell stability index is calculated using the following formula:

[0073] in, Indicates the length of the vertical line where the state resides. v min The minimum statistical length of the state-resident vertical line, V max This indicates the maximum statistical length of the vertical line where the state resides. The length of the vertical line indicating the state is 1. v The number of line segments. JLAM An increase in the value often indicates that the system frequently enters the laminar flow state. In nonlinear dynamics, the laminar flow state specifically refers to the state in which the phase space trajectory of the system remains in a local region for a long time or undergoes slow evolution. In cutting, this corresponds to the periodic growth stage of built-up edge. The stable growth process of the built-up edge is slow, the change amplitude of the phase space trajectory is small, and it shows a slow evolution trend. Subsequently, the instability and shedding of the built-up edge leads to the instability of the system. This growth-shedding process causes an intermittent stable-instability alternation phenomenon. The continuation of this state will lead to an increase in surface roughness fluctuations.

[0074] In other embodiments, the structural distribution feature parameters extracted by the present invention also include the state transition periodic complexity entropy, and the specific extraction process of the state transition periodic complexity entropy includes: The time interval of state transition events is observed along the main diagonal of the nonlinear dynamic state transition map. In practice, this is achieved by statistically analyzing the length of the continuous white lines (zero-value lines) in the map. For the diagonal profile, the length of a line segment consisting of continuous zeros is defined as... w This interval is called a state transition period, which represents the time the system experiences after one state transition. w The state transition occurs again only after a certain time step. Traverse the entire graph diagonally and count the length of each gap. w Frequency of occurrence P w ( w ), w =1,2,…, W max Normalize the gap length distribution to a probability distribution, and then calculate its Shannon information entropy, i.e., the state transition cycle complexity entropy:

[0075]

[0076] In the formula, This represents the normalized probability distribution of the periodic interval. Represents the frequency distribution of the periodic gap. Indicates the length of the state transition cycle interval. Indicates the minimum gap length. This represents the maximum gap length. A higher state transition cycle complexity entropy indicates a greater degree of variability in the system's state transition cycle, and more complex and unpredictable dynamic behavior. This complexity increases significantly during severe tool wear or chatter development, and can serve as a sensitive indicator for monitoring surface roughness degradation.

[0077] In addition, state transition density can be introduced as a global index, which measures the total probability of state transition correlation events between cutting force signals and vibration signals in phase space, and serves as the numerical basis for all subsequent structural analyses.

[0078] The state transition determinism index and the state dwelling stability index decompose state transition events from two dimensions: diagonal structure and vertical structure, respectively. The former measures the proportion dominated by deterministic rules, while the latter measures the proportion dominated by laminar dwelling behavior. Together, they distinguish between the regular and random components of the system's dynamic behavior. The entropy of state transition cycle complexity further measures the degree of regularity and disorder of state transition events in the time dimension by starting from the uniformity of the distribution of state transition cycle gaps, thus making up for the shortcomings of deterministic index and dwell stability index in characterizing the complexity of time structure.

[0079] The aforementioned characteristic parameters work together. When the cutting process gradually deteriorates from stable to chatter or built-up edge instability, the parameter combination exhibits a trend of linked changes, such as a decrease in state transition density, a decrease in deterministic index, an increase followed by a decrease in dwell stability index, and a continuous increase in periodic complexity entropy. This enables a multidimensional dynamic characterization of the surface roughness degradation process and effectively improves the robustness and early warning sensitivity of the roughness mapping model.

[0080] S5. Input the structural distribution characteristic parameters into the pre-built first mapping model and second mapping model respectively, and output the roughness prediction value and residual stress prediction value of the current machined surface of the target workpiece respectively.

[0081] The construction process of the first mapping model and the second mapping model includes: 1) Single-factor cutting experiments were conducted by changing only the target machining parameters and keeping the other machining parameters fixed. In this embodiment, cutting speed, maximum depth of cut, and number of tool teeth were selected as key process parameters. For example, the depth of cut and number of tools were fixed, and only the cutting speed was changed; the cutting speed and depth of cut were fixed, and only the number of tools was changed; the cutting speed and number of tools were fixed, and only the depth of cut was changed; the cutting speed and depth of cut were fixed, and only the number of tools was changed. The signal acquisition time for each set of parameters was no less than 5 seconds, and the acquired raw signals were preprocessed by detrending and low-pass filtering to remove power frequency interference and high-frequency noise components. Among them, the cutting speed was selected at four levels: 60, 100, 140, and 180 m / min. At each level, the depth of cut or number of tool teeth was kept constant, and another parameter was systematically adjusted to construct a controlled comparative experimental group. A total of 12 sets of process parameter combinations were designed. The specific parameters are shown in Table 1. Table 1 Single-factor experimental parameters

[0082] 2) Cutting was performed under each set of single-factor cutting experimental parameters. Cutting force signals and mechanical vibration signals of the machining area were collected simultaneously, and structural distribution parameters were extracted according to S2-S4 as input features. The input features can be two variables: the state transition determinism index and the state dwell stability index. Alternatively, state transition density or state transition periodic complexity entropy can be added to make the input features a combination of three or four variables.

[0083] 3) The surface roughness and residual stress values ​​obtained from the actual measurements under the single-factor cutting experimental parameters for each group were used as output labels. Surface roughness values ​​were obtained using an Rtec MFT-5000 multi-functional friction and wear testing machine. Measurements were taken at three equidistant locations along the circumference of the workpiece, and the average of the three measurements was taken as the final representative value. Residual stress values ​​were obtained using a μ-X360s X-ray residual stress analyzer. A non-contact measurement method based on single-incident X-ray exposure (cosα method) was employed. Four measurement points were selected at equal intervals along the bottom of the raceway, and the average of the four points was taken as the final residual stress measurement result.

[0084] 4) The Gaussian process regression algorithm is used to establish the first mapping model between input features and surface roughness, and the second mapping model between input features and residual stress.

[0085] S6. Surface integrity trend analysis and early warning judgment.

[0086] Trend analysis is performed on the predicted surface roughness value sequence and the residual stress state assessment value sequence continuously monitored during the machining process. At the beginning of machining, the tool is in a brand new state and the cutting parameters are the optimal combination verified by the process, which is taken as the initial stable cutting state. Dynamic cutting force signals and vibration signals with a length of 1 to 3 seconds are automatically extracted in this state, and a nonlinear dynamic state transition map is constructed according to the aforementioned steps. In this embodiment, the state transition determinism index and the state dwell stability index of multiple consecutive analysis windows are calculated, and their arithmetic mean is taken as the initial stable cutting reference mean. JDET 0 and JLAM 0.

[0087] The warning threshold for the state transition determinism index is set to 85% of the average value under the initial stable cutting state, and the warning threshold for the state dwell stability index is set to 120% of the average value under the initial stable cutting state. When the decrease of the state transition determinism index within a preset time window exceeds the warning threshold, and the increase of the state dwell stability index exceeds the warning threshold, it is determined that the surface integrity is trending downward due to tool wear, built-up edge, or cutting chatter, and a warning signal is sent to the machine tool control system.

[0088] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0089] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for online surface integrity detection based on multi-source information fusion, characterized in that, Includes the following steps: S1. Synchronously acquire cutting force signals and mechanical vibration signals during the machining process of the target workpiece; S2. Perform phase space reconstruction on the cutting force signal and the mechanical vibration signal respectively to obtain the first phase space trajectory and the second phase space trajectory; S3. Based on a preset adaptive threshold related to the background noise generated by the machine tool idling, calculate the phase space state transition correlation matrix of the first phase space trajectory and the second phase space trajectory, present the phase space state transition correlation matrix as a binary image, and generate a nonlinear dynamic state transition map characterizing the degree of nonlinear coupling between excitation and response. S4. Extract the structural distribution characteristic parameters from the nonlinear dynamic state transition map; S5. Input the structural distribution characteristic parameters into the pre-constructed first mapping model and second mapping model respectively, and output the roughness prediction value and residual stress prediction value of the current machined surface of the target workpiece respectively. S6. Within a preset time window, when the decrease or increase of the structural distribution characteristic parameter exceeds the warning threshold, it is determined that the target workpiece has a trend of surface integrity deterioration, and a warning signal is issued.

2. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, S2 include: S21. Extract a signal sample in a stable cutting state from the synchronously acquired cutting force signal and mechanical vibration signal. S22. Determine the optimal delay time of the cutting force signal and the mechanical vibration signal under stable cutting conditions using mutual information functions respectively; S23. The optimal embedding dimension of the cutting force signal and the mechanical vibration signal are determined by the pseudo nearest neighbor method, respectively. S24. Based on the optimal delay time and optimal embedding dimension of the cutting force signal, construct the first phase space trajectory; based on the optimal delay time and optimal embedding dimension of the mechanical vibration signal, construct the second phase space trajectory.

3. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, S22 includes: S221. Divide the amplitude range of the cutting force signal under stable cutting conditions into b equal-width quantization intervals bin; S222, Delay time τ Starting from 0, with sampling interval Δ t To increase the step size incrementally, different values ​​are calculated using the mutual information function. τ The mutual information value corresponding to the value is used to obtain the mutual information function curve; S223, the value corresponding to the first minimum value of the mutual information function curve. τ The value is taken as the optimal delay time of the cutting force signal. The delayed signal corresponding to the optimal delay time provides the maximum amount of new information to the original signal. S224. Divide the amplitude range of the mechanical vibration signal under stable cutting conditions into b equal-width quantization intervals bin, and execute S222-S223 to obtain the optimal delay time of the mechanical vibration signal.

4. The online surface integrity detection method based on multi-source information fusion as described in claim 3, characterized in that, In S222, the expression for the mutual information function is: in, The probability that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval bin; Delay of the original cutting force signal or the original mechanical vibration signal τ After stepping into the first j The probability of a quantization interval bin of equal width; This indicates that the original cutting force signal or the original mechanical vibration signal falls into the i-th equal-width quantization interval, while being delayed. τ After stepping into the first j The joint probability of equal-width quantization intervals; bin represents the equal-width quantization interval of the amplitude range. b Indicates the total number of equal-width quantization intervals. τ Represents the delay time variable. I ( τ ) represents the mutual information value.

5. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, S23 includes: S231. For the cutting force signal, starting from the embedding dimension m=1, and increasing step size 1 successively, perform false nearest neighbor analysis on each embedding dimension m. S232. For each embedding dimension m, utilize the optimal latency time. τ F The cutting force signal is reconstructed in phase space to generate N m-dimensional state vectors; S233. For each m-dimensional state vector Find its nearest neighbor vector Make the Euclidean distance between the two Minimum, construct the original nearest neighbor pair; S234. Increase the embedding dimension to m+1, reconstruct the m+1-dimensional extended vector of each state vector, and calculate the Euclidean distance between the original nearest neighbor pairs in the m+1-dimensional space. ; S235. Define the distance ratio criterion for the i-th state vector. R F ( i ),when R F ( i )> R th When the original nearest neighbor pair is determined to be a false nearest neighbor pair, then, R th To determine the threshold; S236. Count the total number of pairs of points identified as false nearest neighbors among all N state vectors under the current embedding dimension m. And calculate the total number of false nearest neighbor pairs. The proportion of false nearest neighbors is the percentage of the total number of points in the state vector. ; S237. Let m = m + 1, and repeat S232-S236 until the proportion of false nearest neighbors is reached. When the cutting force signal first drops below a preset threshold or stabilizes as the embedding dimension m increases, the corresponding embedding dimension is taken as the optimal embedding dimension. m F ; S238. For the mechanical vibration signal, repeat steps S231-S237 to obtain the optimal embedding dimension of the mechanical vibration signal. m A .

6. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, S24 includes: S241, Optimal delay time based on the cutting force signal τ F and optimal embedding dimension m F For each sampling time i Construct a m F dimensional state vector , ,in, i The range of values ​​is i =1,2,…, N F ; N F This represents the total number of phase space trajectory points obtained after reconstructing the cutting force signal. N F ,satisfy: Where L is the total number of sampling points for the cutting force signal; Arrange the state vectors at all times in chronological order to form the first phase space trajectory, denoted as: ; S25. Based on the optimal delay time of the mechanical signal τ A and optimal embedding dimension m A For each sampling time i Construct a m A dimensional state vector , ,in, i The range of values ​​is i =1,2,…, N A ; N A This represents the total number of phase space trajectory points obtained after reconstructing the mechanical signal, satisfying: ; The state vectors at all times are arranged in chronological order to form the second phase space trajectory. ; S26, Take N F and N A The smaller of the two is used as the uniform length for constructing the phase space state transition correlation matrix.

7. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, In S3, the formula for calculating the elements in the phase space state transition correlation matrix is ​​as follows: in, The elements represent the phase space state transition correlation matrix. This indicates the phase space adjacency state value of the cutting force channel. This indicates the phase space adjacency state value of the vibration channel. This represents the logical AND operator; This represents the adaptive threshold for the neighborhood of the cutting force channel. This represents the adaptive threshold of the vibration channel neighborhood. and These represent the two phase space vectors of the cutting force signal, and These represent two phase space vectors of the mechanical vibration signal, with subscript i indicating the index of the first sampling time and j indicating the index of the second sampling time. It is the Heaviside step function.

8. The online surface integrity detection method based on multi-source information fusion as described in claim 7, characterized in that, Adaptive threshold and The determination process includes: While the machine tool spindle is rotating but the tool has not yet contacted the workpiece, the output signals of the cutting force sensor and the vibration acceleration sensor are recorded simultaneously, and the standard deviations of the two output signals are calculated to obtain the standard deviation of the cutting force channel background noise. and the standard deviation of the background noise of the vibration channel ; The standard deviation of the background noise of the cutting force channel is respectively... and the standard deviation of the background noise of the vibration channel Multiply by a preset floating coefficient to obtain the adaptive threshold of the cutting force channel neighborhood. Adaptive threshold for vibration channel neighborhood ; At preset time intervals, the output signals of the cutting force sensor and vibration acceleration sensor under idling conditions are re-acquired within the last few seconds, and the standard deviation of the two output signals is recalculated. and And update the adaptive threshold in real time. and .

9. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, In S4, the structural distribution characteristic parameters include at least the state transition determinism index and the state dwell stability index; in S6, when the decrease of the state transition determinism index exceeds the warning threshold and the increase of the state dwell stability index exceeds the warning threshold within a preset time window, it is determined to be a trend of surface integrity deterioration caused by tool wear, built-up edge or cutting chatter. The process of extracting the state transition determinism index includes: Traverse the line segments with continuous diagonal values ​​of 1 in the nonlinear dynamic state transition map. When the cutting force signal and the mechanical vibration signal are both state transition coincident points at the same coordinate position, the corresponding element takes a value of 1; otherwise, it takes a value of 0. When the value is 1, it is represented by a black pixel; when the value is 0, it is represented by a white pixel. Define the length as The line segment is a state transition diagonal. The number of state transition diagonals of various lengths appearing in the nonlinear dynamic state transition spectrum is counted to obtain a histogram of diagonal length frequency distribution. And exclude extremely short lines whose length is less than the preset length; The state transition determinism index is calculated according to the following formula: in, l Indicates the length of the diagonal of the state transition. l min This represents the minimum statistical length of the diagonal of the state transition. L max This indicates the maximum length of the diagonal during the state transition. This indicates that the length of the diagonal is l The number of line segments, This represents the element in the i-th row and j-th column of the phase space state transition correlation matrix; The process of extracting the state dwell stability index includes: Traverse all line segments with a continuous value of 1 in all column directions of the nonlinear dynamic state transition diagram; Define a line segment of length v as a state-stationary perpendicular line, and count the frequency of occurrence of state-stationary perpendicular lines of various lengths to obtain the frequency distribution of perpendicular line lengths. P ( v ), and exclude extremely short lines whose length is less than the preset length; The state-dwelling stability index is calculated using the following formula: in, Indicates the length of the vertical line where the state resides. v min The minimum statistical length of the state-resident vertical line, V max This indicates the maximum statistical length of the vertical line where the state resides. The length of the vertical line indicating the state is 1. v The number of line segments.

10. The online surface integrity detection method based on multi-source information fusion as described in claim 1, characterized in that, In S5, the construction process of the first mapping model and the second mapping model includes: A single-factor cutting experiment was conducted by changing only the target machining parameters and keeping the other machining parameters fixed. Cutting was performed under each set of single-factor cutting experimental parameters. Cutting force signals and mechanical vibration signals in the machining area were collected simultaneously, and structural distribution parameters were extracted according to S2-S4 as input features. The surface roughness and residual stress values ​​obtained from the actual measurements under the single-factor cutting experimental parameters for each group are used as output labels; Gaussian process regression algorithm is used to establish a first mapping model between input features and surface roughness, and a second mapping model between input features and residual stress.

Images