Method and system for recognizing coating deposition process anomaly based on multi-source sensing data

Abstract

The present application relates to the technical field of vacuum thin film deposition process monitoring, and particularly relates to a coating deposition process anomaly identification method and system based on multi-source sensing data, comprising: combining the angle data of a planetary rotation mechanism with a workpiece three-dimensional point cloud model to calculate the three-dimensional coordinates, spatial distance data and dynamic polarization angle of each microelement on the surface of the workpiece; then using the inverse matrix of the transmission attenuation matrix constructed therefrom to perform reverse amplitude and phase modulation on the original microwave signal to output high-fidelity near-field microwave data; then performing mutual information and false near neighbor discrete calculation through a time delay embedding algorithm to reconstruct a multi-dimensional phase space trajectory matrix; then performing double-track parallel calculation to obtain the chaotic dynamics correlation dimension value and the fourth-order cumulative quantity value of high-order statistics; and finally determining that the workpiece is in a localized discharge abnormal state and triggering suppression interruption when the two values reach a linkage threshold value. The present application effectively eliminates dynamic multipath stray interference and improves weak anomaly recognition accuracy.

Description

Technical Field

[0001] This invention relates to the field of vacuum thin film deposition process monitoring technology, and in particular to a method and system for identifying anomalies in coating deposition processes based on multi-source sensor data. Background Technology

[0002] In modern vacuum coating deposition processes, high-power pulsed magnetron sputtering or multi-arc ion plating are widely used to prepare high-performance coatings on workpiece surfaces. To monitor the plasma state within the deposition chamber in real time and detect potential process anomalies, near-field microwave sensing technology, as a highly sensitive non-contact non-destructive testing method, is increasingly being applied. This technology utilizes an antenna matrix positioned within the deposition chamber to input raw microwave signals. By receiving changes in the propagation characteristics of the microwave signals as they pass through the vacuum deposition environment, it achieves time-series data acquisition and analysis of plasma discharge behavior and coating deposition state.

[0003] Because the planetary rotation mechanism continuously revolves and rotates, the spatial geometry between the workpiece surface and the receiving antenna exhibits time-varying distortion, introducing severe spatial propagation loss and dynamic polarization mismatch distortion into the signal. Existing demodulation methods cannot eliminate these dynamic spatial spurious interferences in complex spatial motion scenarios, resulting in low physical confidence of the microwave signal and making it difficult to accurately reflect the abnormal state of the deposition process. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a method and system for identifying anomalies in coating deposition processes based on multi-source sensor data. It aims to improve the problem that existing demodulation methods cannot eliminate the aforementioned dynamic spatial stray interference in complex spatial motion scenarios, resulting in low physical confidence of microwave signals and difficulty in truly reflecting the abnormal state of the deposition process.

[0005] In a first aspect, the present invention provides the following technical solution: a method for identifying anomalies in a coating deposition process based on multi-source sensor data, comprising: S1. Real-time acquisition of the revolution and rotation angle data of the planetary rotation mechanism, combined with the pre-stored three-dimensional point cloud model of the workpiece, calculate the three-dimensional coordinates of each micro-element on the workpiece surface, and output the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. S2. Calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and the dynamic polarization angle, and construct a transmission attenuation matrix as a matrix element. Use the inverse matrix of the transmission attenuation matrix to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. S3. Using a time delay embedding algorithm, mutual information calculation is performed on the near-field microwave data to determine the optimal time delay, and pseudo-nearest neighbor discretization calculation is performed to determine the minimum embedding dimension. Based on this, a multi-dimensional phase space trajectory matrix is ​​reconstructed. S4. Calculate the correlation dimension value based on the multidimensional phase space trajectory matrix, and simultaneously perform high-order statistical calculations on the near-field microwave data to obtain the fourth-order cumulant value. S5. When the value of the correlation dimension decreases to the preset chaotic interval threshold and the value of the fourth-order cumulative quantity exceeds the preset pulse peak value limit, it is determined that the coating deposition process is in a local discharge abnormal state and a suppression interruption command is triggered.

[0006] Preferably, in S1, calculating the three-dimensional coordinates of each micro-element on the workpiece surface includes: Extract the initial coordinates of each micro-element in the pre-stored 3D point cloud model of the workpiece; Based on the aforementioned revolution and rotation angle data, a first rotation matrix corresponding to the revolution and a second rotation matrix corresponding to the rotation are constructed respectively; A spatial transformation matrix is ​​constructed using the first rotation matrix, the second rotation matrix, and a preset translation vector; The spatial transformation matrix is ​​used to perform matrix transformation operations on each of the initial coordinates to output the three-dimensional coordinates of each micro-element on the workpiece surface.

[0007] Preferably, in S2, the calculation of the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel includes: The instantaneous spatial propagation loss is output by calculating the inverse square ratio between the spatial distance data and the preset working wavelength. Based on the dynamic polarization angle, cosine projection calculation is performed to output the polarization mismatch loss; The instantaneous spatial propagation loss and the polarization mismatch loss are multiplied to obtain matrix elements, and the transmission attenuation matrix is ​​constructed using the matrix elements.

[0008] Preferably, in S2, the step of using the inverse matrix of the transmission attenuation matrix to perform inverse amplitude and phase modulation on the original microwave signal input to the antenna matrix includes: Extract the amplitude modulation operator and phase modulation operator from the inverse matrix of the transmission attenuation matrix; The amplitude modulation operator is used to perform a product operation on the original microwave signal to complete the amplitude inverse modulation of the original microwave signal; The phase modulation operator is used to perform a superposition operation on the original microwave signal after amplitude inversion modulation to complete the phase inversion modulation of the original microwave signal; The original microwave signal, after amplitude inversion modulation and phase inversion modulation, is output as near-field microwave data.

[0009] Preferably, in S3, determining the optimal time delay by performing mutual information calculation on the near-field microwave data includes: Extract the joint probability density distribution and marginal probability density distribution of the near-field microwave data under the time delay variable; By performing integral calculations using the joint probability density distribution and the marginal probability density distribution, the dynamic mutual information function curve of the near-field microwave data as the time-delay variable increases is output. Find the first inflection point in the dynamic mutual information function curve where the slope changes from negative to positive, and determine the time delay variable corresponding to the first inflection point as the optimal time delay.

[0010] Preferably, in S3, the step of performing pseudo-nearest neighbor discrete calculus to determine the minimum embedding dimension includes: Combining the optimal time delay, the near-field microwave data is subjected to Euclidean distance calculation in the transient space of dimension 1 and dimension 2 plus one, respectively, and the distance ratio of corresponding trajectory points in the two spaces is extracted; The number of elements whose distance ratio is greater than a preset distance threshold is counted to obtain the proportion of false nearest neighbors in the corresponding dimension; The dimension is incremented for iterative calculation. The iteration is terminated when the proportion of false nearest neighbors is less than a preset proportion threshold, and the current dimension is determined as the minimum embedding dimension. A phase space reconstruction operator is constructed using the optimal time delay and the minimum embedding dimension to perform mapping transformation on the near-field microwave data and output a multi-dimensional phase space trajectory matrix.

[0011] Preferably, in S4, higher-order statistical calculations are performed simultaneously on the near-field microwave data to obtain fourth-order cumulant values, including: Calculate the mean of the near-field microwave data within the current sampling window, and use the mean to perform DC removal operation on the near-field microwave data to obtain DC-removed time series data; Calculate the second-order autocorrelation function matrix and the fourth-order instantaneous moment value of the DC-free time series data respectively; The fourth-order cumulant value is output by performing a weighted difference calculation using the sum of squares of the diagonal elements of the second-order autocorrelation function matrix and the fourth-order instantaneous moment value.

[0012] Preferably, in S5, when the value of the correlation dimension decreases to a preset chaotic interval threshold, it includes: Extract the Euclidean distance between each pair of spatial phase points in the multidimensional phase space trajectory matrix, and calculate the correlation integral value based on the Euclidean distance under a given spatial correlation radius; The spatial correlation radius is changed to perform multiple iterative calculations, and the resulting multiple correlation integral values ​​are linearly fitted in double logarithmic coordinates. The slope of the fitted line segment is determined as the correlation dimension value. Obtain the cumulative running time of the current coating deposition process, and use the cumulative running time to calculate the attenuation slope of the initial threshold, and output the preset chaotic interval threshold. The calculated correlation dimension value is compared with the output preset chaotic interval threshold to determine whether the correlation dimension value has decreased to the preset chaotic interval threshold.

[0013] Preferably, in S5, the determination that the coating deposition process is in a localized discharge abnormal state and the triggering of the suppression interruption instruction includes: Set a rolling time window, and perform time-stream synchronization alignment of the associated dimension values ​​and the fourth-order cumulative values ​​within the rolling time window; The cumulative duration during which the correlation dimension value is lower than the preset chaotic interval threshold within the rolling time window is counted, as well as the frequency of abrupt changes in the fourth-order cumulative quantity value exceeding the preset pulse peak value upper limit. When the cumulative duration exceeds a preset time threshold and the mutation frequency exceeds a preset frequency threshold, it is determined that the local discharge is in an abnormal state, and a suppression interruption command is issued.

[0014] Secondly, the present invention provides the following technical solution: an anomaly identification system for coating deposition process based on multi-source sensor data, the system comprising: The geometric space calculation module is used to acquire the revolution and rotation angle data of the planetary rotation mechanism in real time. Combined with the pre-stored three-dimensional point cloud model of the workpiece, it calculates the three-dimensional coordinates of each micro-element on the workpiece surface and outputs the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. The matrix modulation inverse operation module is used to calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and the dynamic polarization angle, and to construct a transmission attenuation matrix as a matrix element. The inverse matrix of the transmission attenuation matrix is ​​used to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. The phase space trajectory reconstruction module is used to perform mutual information calculation on the near-field microwave data using a time delay embedding algorithm to determine the optimal time delay, and to perform pseudo-nearest neighbor discretization calculation to determine the minimum embedding dimension, thereby reconstructing a multi-dimensional phase space trajectory matrix. The feature digital calculation module is used to calculate the correlation dimension value based on the multidimensional phase space trajectory matrix, and simultaneously perform high-order statistical calculations on the near-field microwave data to obtain the fourth-order cumulant value. The abnormal chain blocking module is used to determine that the coating deposition process is in a local discharge abnormal state and trigger a suppression interruption command when the value of the correlation dimension decreases to a preset chaotic interval threshold and the value of the fourth-order cumulative quantity exceeds a preset pulse peak value.

[0015] The present invention has the following beneficial effects: 1. In this invention, by combining the revolution and rotation angle data of the planetary rotation mechanism and the three-dimensional point cloud model of the workpiece to dynamically reconstruct the spatial transformation matrix, and by using the inverse matrix of the transmission attenuation matrix to perform reverse modulation of the amplitude and phase of the original microwave signal, physical-level reverse compensation for complex multipath propagation loss and polarization mismatch distortion is achieved, solving the problem of severe stray interference and low physical confidence of near-field microwave sensing data in complex spatial motion scenarios in vacuum cavities.

[0016] 2. In this invention, the correlation dimension values ​​are extracted by adaptively reconstructing the multidimensional phase space trajectory matrix through a time delay embedding algorithm, and the weighted difference calculation of the second-order autocorrelation function matrix and the fourth-order instantaneous moment is simultaneously used to output the fourth-order cumulative value. This realizes the deep joint mining of nonlinear chaotic dynamic degradation characteristics and high-order non-Gaussian pulse characteristics, and solves the problem that the weak micro-arc signals in the early stage of local discharge are easily submerged by process environment noise and difficult to be sensitively captured.

[0017] 3. In this invention, the initial threshold is dynamically adjusted by using the cumulative running time to perform adaptive attenuation slope calculation. The intersection of the cumulative duration and the frequency of sudden changes is introduced by combining the rolling time window to realize the dynamic tracking of the time-varying drift of the sensing baseline and the robust filtering of occasional noise. This solves the problem that single-index threshold monitoring is prone to false alarms and unnecessary shutdown control caused by hardware drift and single-point noise triggering the system during long-term process operation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the coating deposition process anomaly identification method based on multi-source sensor data proposed in this invention; Figure 2 This is a flowchart detailing the phase space trajectory reconstruction and adaptive threshold locking algorithm of the coating deposition process anomaly identification method based on multi-source sensor data proposed in this invention. Figure 3 This is an architecture diagram of the coating deposition process anomaly identification system based on multi-source sensor data proposed in this invention. Detailed Implementation

[0019] The technical solutions in 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.

[0020] Example 1: In a first embodiment of the present invention, the present invention provides a method for identifying anomalies in the coating deposition process based on multi-source sensor data, such as... Figures 1-2 As shown, it includes the following steps: S1. Real-time acquisition of the revolution and rotation angle data of the planetary rotation mechanism, combined with the pre-stored three-dimensional point cloud model of the workpiece, calculate the three-dimensional coordinates of each micro-element on the workpiece surface, and output the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. Furthermore, in S1, calculating the three-dimensional coordinates of each micro-element on the workpiece surface includes: Extract the initial coordinates of each micro-element in the pre-stored 3D point cloud model of the workpiece; Based on the revolution and rotation angle data, construct the first rotation matrix corresponding to the revolution and the second rotation matrix corresponding to the rotation. A spatial transformation matrix is ​​constructed using the first rotation matrix, the second rotation matrix, and a preset translation vector; The spatial transformation matrix is ​​used to perform matrix transformation operations on each initial coordinate to output the three-dimensional coordinates of each micro-element on the workpiece surface.

[0021] Specifically, during operation, sensors on the planetary rotary mechanism collect real-time data on its revolution and rotation angles. The workpiece is placed on the fixture of the planetary rotary mechanism and moves synchronously with the mechanism in both revolution and rotation. The system pre-stores a three-dimensional point cloud model of the workpiece, which consists of the initial coordinates of numerous discretely distributed micro-elements on the workpiece surface. The initial coordinates are defined in the local coordinate system of the workpiece and expressed as a three-dimensional coordinate vector in matrix format.

[0022] To calculate the real-time position of each micro-element on the workpiece surface in the absolute coordinate system of the deposition chamber, the initial coordinates of each micro-element are first read from the pre-stored 3D point cloud model of the workpiece. The initial coordinates are represented as column vectors. Next, using the real-time acquired revolution angle data... Construct the first rotation matrix for revolution Using real-time acquired rotation angle data Construct the second rotation matrix of the rotation Among them, the first rotation matrix The second rotation matrix is ​​used to describe the angular displacement of the workpiece about the central axis of the deposition chamber. Used to describe the rotational angular displacement of a workpiece about its own geometric center axis.

[0023] After constructing the first rotation matrix Second rotation matrix Then, a preset translation vector is introduced. Translation vector This represents a fixed spatial translation distance between the geometric center axis of the workpiece and the center axis of the deposition chamber. This is achieved by using the first rotation matrix... Second rotation matrix and translation vector Perform matrix composition operations to construct a spatial transformation matrix. Use the spatial transformation matrix to adjust the initial coordinates. Perform matrix transformation operations to output the three-dimensional coordinates of each infinitesimal element on the workpiece surface. The specific mathematical formula is as follows: ; in, This represents the real-time three-dimensional coordinate column vector of each micro-element in the absolute coordinate system of the sedimentation chamber. This is the second rotation matrix for rotation. To store the initial coordinate column vector of the infinitesimal element in the pre-stored 3D point cloud model of the workpiece, For the preset spatial translation column vector of the center of rotation, This is the first rotation matrix for the revolution. In a feasible implementation, the translation vector... The absolute spatial coordinate constants are set to 450 mm, 0 mm, and 150 mm.

[0024] Obtaining the three-dimensional coordinates of each micro-element on the workpiece surface Then, combining the known preset absolute position coordinates of each receiving antenna in the antenna matrix in the absolute coordinate system of the deposition chamber... The calculation involves performing a three-dimensional Euclidean distance calculation. The logic is to take the square root of the sum of the squared differences between the three-dimensional coordinates of each infinitesimal element and the position coordinates of each receiving antenna, thereby outputting the corresponding spatial distance data between the workpiece surface and each receiving antenna for each receiving channel. The mathematical formula for its calculation is as follows: ; in, For spatial distance data, Let be the three-dimensional coordinate column vector of each infinitesimal element on the workpiece surface. This is a column vector of known absolute position coordinates of each receiving antenna in the antenna matrix. This represents the 2-norm operation of vectors in three-dimensional space.

[0025] Simultaneously, the real-time normal vector of the workpiece surface is determined using the three-dimensional coordinates of each infinitesimal element at this time. And combined with the fixed polarization axis vector of the receiving antenna The angle change between these two vectors is calculated in three-dimensional absolute space to solve for the dynamic polarization angle. The calculation formula is as follows: ; in, For dynamic polarization angle, This represents the normal vector of the workpiece surface corresponding to the infinitesimal element in the absolute coordinate system. This corresponds to the fixed polarization axial vector of the receiving antenna. This represents the dot product operation between two vectors.

[0026] The spatial distance data calculated above and dynamic polarization angle As intermediate process parameters, they are involved in the subsequent calculation of instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel in real time, providing a precise data flow basis for the subsequent construction of transmission attenuation matrix and the reverse phase and amplitude modulation of microwave signal.

[0027] S2. Calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and dynamic polarization angle, and use them as matrix elements to construct a transmission attenuation matrix. Use the inverse matrix of the transmission attenuation matrix to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. Furthermore, in S2, the calculation of the instantaneous spatial propagation loss and polarization mismatch loss for each receiving channel includes: The instantaneous spatial propagation loss is output by calculating the inverse square ratio between the spatial distance data and the preset working wavelength. The polarization mismatch loss is output by calculating the cosine projection based on the dynamic polarization angle. The instantaneous spatial propagation loss and polarization mismatch loss are multiplied to obtain matrix elements, and the matrix elements are used to construct the transmission attenuation matrix.

[0028] Furthermore, in S2, the inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix using the inverse of the transmission attenuation matrix includes: Extract the amplitude modulation operator and phase modulation operator from the inverse matrix of the transmission attenuation matrix; The amplitude of the original microwave signal is reverse-modulated by multiplying the original microwave signal using an amplitude modulation operator. The phase inverse modulation of the original microwave signal is achieved by superimposing the original microwave signal after amplitude inverse modulation using a phase modulation operator. The original microwave signal, after amplitude inversion modulation and phase inversion modulation, is output as near-field microwave data.

[0029] Specifically, the spatial distance data and dynamic polarization angle obtained in the previous steps are input into the data calculation process of this step in real time. To quantitatively characterize the physical attenuation characteristics of microwave signals in the complex multipath environment of the deposition chamber, it is first necessary to calculate the attenuation of each receiving channel. During the calculation, the spatial distance data output from the previous steps and the preset operating wavelength constant are read. In a feasible implementation, the preset operating wavelength constant is set to 0.122 meters. The spatial distance data is inversely proportional to the preset operating wavelength by its square, i.e., the spatial distance data is divided by the square of the preset operating wavelength and the reciprocal is taken; the output is the instantaneous spatial propagation loss. The specific mathematical formula is as follows: ; in, For instantaneous spatial propagation loss, To preset the operating wavelength constant, For spatial distance data, Pi is the mathematical constant of a circle.

[0030] Simultaneously, the dynamic polarization angle output from the previous steps is read, and this angle is input into a cosine projection function to calculate its cosine value, outputting the polarization mismatch loss. Subsequently, the calculated instantaneous spatial propagation loss and polarization mismatch loss are multiplied by a scalar to obtain the composite attenuation coefficient, which is then used as the corresponding matrix element in the transmission attenuation matrix. By arranging the matrix elements of each receiving channel in spatial topology, the complete transmission attenuation matrix H is finally constructed. The specific mathematical formula is as follows: ; in, Transmission attenuation matrix The Middle Line 1 Column matrix elements, For the first The launch position to the first Instantaneous spatial propagation loss of each receiving channel, For the first The launch position to the first The dynamic polarization angle of each receiving channel.

[0031] After constructing the transmission attenuation matrix Subsequently, in order to physically compensate for signal distortion caused by spatial propagation and polarization deflection, the transmission attenuation matrix was modified. Perform the inverse operation by dividing the adjoint matrix by the determinant to generate the corresponding inverse matrix. Next, from the inverse matrix... The argument and modulus are extracted from the complex elements respectively, and the amplitude modulation operator corresponding to the complex number is calculated. and phase modulation operator Specifically, amplitude modulation operator The modulus is obtained by summing and then taking the square root of the squares of the real and imaginary parts of the inverse matrix; phase modulation operator. The argument is obtained by performing an arctangent function operation on the ratio of the imaginary part to the real part of the inverse matrix.

[0032] Subsequently, the original microwave signal input from the antenna matrix is ​​used as the basic data stream, and amplitude modulation operator is applied. A scalar product operation is performed with the real amplitude of the original microwave signal, and the amplitude is proportionally scaled to achieve amplitude inverse modulation. Then, a phase modulation operator is used... By performing algebraic addition on the current phase angle of the original microwave signal after amplitude inversion modulation, the phase increments are linearly superimposed, thus completing phase inversion modulation. The mathematical formula for inversion modulation in the complex domain is as follows: ; in, This is the column vector of the final output complex matrix of near-field microwave data. It is the inverse of the transmission attenuation matrix. The input to the antenna matrix is ​​the column vector of the complex matrix of the original microwave signal. Through the baseband signal inverse demodulation at the complex matrix level, the original microwave signal after amplitude inverse modulation and phase inverse modulation is finally mapped and determined as near-field microwave data, and then transmitted as a continuous time-series data stream in real time to the subsequent time delay embedding reconstruction analysis process.

[0033] S3. Using the time delay embedding algorithm, mutual information calculation is performed on near-field microwave data to determine the optimal time delay, and pseudo-nearest neighbor discretization calculation is performed to determine the minimum embedding dimension. Based on this, a multi-dimensional phase space trajectory matrix is ​​reconstructed. Furthermore, in S3, the mutual information calculation of near-field microwave data to determine the optimal time delay includes: Extract the joint probability density distribution and marginal probability density distribution of near-field microwave data under time delay variables; By performing integral calculations using the joint probability density distribution and the marginal probability density distribution, the dynamic mutual information function curve of the near-field microwave data as the time delay variable increases is output. Find the first inflection point in the dynamic mutual information function curve where the slope changes from negative to positive, and determine the time delay variable corresponding to the first inflection point as the optimal time delay.

[0034] Furthermore, in S3, the process of determining the minimum embedding dimension through pseudo-nearest neighbor discretization includes: By combining the optimal time delay, Euclidean distance calculations are performed on near-field microwave data in two transient spaces of dimension 1 and dimension plus one, respectively, and the distance ratio of corresponding trajectory points in the two spaces is extracted. The number of elements whose distance ratio is greater than a preset distance threshold is counted to obtain the proportion of false nearest neighbors in the corresponding dimension; The incremental dimension is used for iterative calculation. The iteration is terminated when the proportion of false nearest neighbors is less than a preset proportion threshold, and the current dimension is determined as the minimum embedding dimension. A phase space reconstruction operator is constructed using the optimal time delay and minimum embedding dimension to perform mapping transformation on near-field microwave data and output a multi-dimensional phase space trajectory matrix.

[0035] Specifically, the near-field microwave data obtained in the previous steps is input in real time into the time-delay embedding data calculation process in this step. Near-field microwave data is represented as one-dimensional time-series data. To recover its evolution trajectory in a multi-dimensional nonlinear dynamic system, it is first necessary to determine the core parameters of the reconstruction space. This is achieved by setting time-delay variables with increasing discrete step sizes. The original near-field microwave data timing signal is time-shifted and aligned with itself. This is done for variables with different time delays. The two time-series data streams are subjected to gridded interval statistics to calculate the joint probability density distribution of the data streams in a multidimensional two-dimensional distribution. At the same time, the marginal probability density distributions of the original near-field microwave data and the near-field microwave data after time shift are calculated respectively.

[0036] Subsequently, the product of the joint probability density distribution and the corresponding marginal probability density distributions is calculated as a ratio, and the natural logarithm of this ratio is taken. Then, the logarithmized ratio is integrally calculated using a double summation formula to quantitatively output the time-delay variable of the near-field microwave data. The curve represents an increasing dynamic mutual information function. Its specific mathematical formula is as follows: ; in, This represents the mutual information value under the corresponding time delay variable. For near-field microwave data and its own delay Subsequent time series data within the numerical value range and The joint probability density distribution at [location] This represents the marginal probability density distribution of the original near-field microwave data. This represents the marginal probability density distribution of the delayed near-field microwave data. First-order difference calculations are performed on the obtained dynamic mutual information function curve to simulate partial differential differentiation. The first node where the slope changes from negative to positive and crosses zero along the increasing time delay direction is searched; that is, the first local minimum inflection point appears in the dynamic mutual information function curve. The time delay variable corresponding to this first inflection point is then... This was determined to be the optimal latency.

[0037] After determining the optimal time delay, the near-field microwave data is input into the pseudo-nearest neighbor discrete calculus process to determine the minimum embedding dimension. First, the embedding dimension is initialized. Combining the determined optimal time delay, the near-field microwave data are analyzed in different dimensions. sum dimension Reconstruction is performed in the transient space. In dimension... In the reconstructed space, find the nearest neighbor of each trajectory point and calculate the spatial geometric Euclidean distance between these two points; then in dimension... The Euclidean distance between these two points is recalculated in the reconstructed space. By calculating the algebraic ratio of the new Euclidean distance to the original Euclidean distance, the distance ratio of corresponding trajectory points in the two spaces is extracted. The mathematical formula is as follows: ; in, It is the distance ratio. For in dimension The reconstructed space contains phase point trajectory vectors with multiple time delay elements. For phase point The nearest neighbor trajectory vector, and The two points mentioned above are respectively increased in dimension to The corresponding phase point trajectory vector in the reconstructed space, This represents the L2 norm distance operation for spatial vectors.

[0038] Set a preset distance threshold, iterate through and count the distance ratios. The number of elements greater than a preset distance threshold is divided by the total number of phase points in the reconstructed space to obtain the proportion of false nearest neighbors representing false spatial proximity relationships at the corresponding dimension. Subsequently, the dimension is increased sequentially... The algorithm performs iterative calculations and sets a preset ratio threshold. When the calculated proportion of false nearest neighbors is less than the preset ratio threshold, the iteration is terminated, and the current dimension is determined as the minimum embedding dimension.

[0039] A phase space reconstruction operator is constructed using the determined optimal time delay and minimum embedding dimension. This phase space reconstruction operator is then used to perform time-delay coordinate mapping transformation on one-dimensional near-field microwave data. Specifically, the time-series data is sliced ​​into multidimensional segments and concatenated with matrix row vectors according to the optimal time delay interval. This outputs a multidimensional phase space trajectory matrix with high-dimensional dynamic characterization properties, which is then used as a standard matrix data stream and fed into the subsequent chaotic correlation dimension and high-order statistical calculation process in real time.

[0040] S4. The correlation dimension value is calculated based on the multidimensional phase space trajectory matrix, and the high-order statistics of the near-field microwave data are calculated simultaneously to obtain the fourth-order cumulant value. Furthermore, in S4, higher-order statistical calculations are performed simultaneously on the near-field microwave data to obtain fourth-order cumulant values, including: Calculate the mean of the near-field microwave data within the current sampling window, and use the mean to perform DC removal operation on the near-field microwave data to obtain DC-removed time series data; Calculate the second-order autocorrelation function matrix and the fourth-order instantaneous moment value of the DC-deferred time series data respectively; The fourth-order cumulant value is output by performing a weighted difference calculation using the sum of squares of the diagonal elements of the second-order autocorrelation function matrix and the fourth-order instantaneous moment value.

[0041] Specifically, the multidimensional phase space trajectory matrix reconstructed in the previous steps and the calculated near-field microwave data are input in real time into the dual-track parallel calculation process of this step. To extract the microscopic changes in the process state in terms of chaotic dynamics and non-Gaussian statistical characteristics, the correlation dimension values ​​are first calculated using the multidimensional phase space trajectory matrix. Simultaneously, a parallel data processing channel is activated to perform high-order statistical calculations on the one-dimensional time-series near-field microwave data to extract fourth-order cumulant values.

[0042] In the calculation of the correlation dimension, since the multidimensional phase space trajectory matrix contains the phase point trajectories in the reconstructed space, the system first extracts the Euclidean distances between each pair of spatial phase points in the multidimensional phase space trajectory matrix and sets a given spatial correlation radius. By statistically analyzing the proportion of phase point pairs whose Euclidean distance is less than the given spatial correlation radius to the total number of phase point pairs, the spatial correlation integral value is calculated. Subsequently, multiple iterative calculations are performed by continuously changing the given spatial correlation radius to obtain a set of spatial correlation radius values ​​and a set of corresponding spatial correlation integral values. The obtained sets of data are then subjected to least-squares linear fitting in a double logarithmic coordinate system, and the slope of the fitted line segment is determined as the correlation dimension value.

[0043] In the high-order statistical calculation channel, a sliding current sampling window is first set. A segment of near-field microwave data within the current sampling window is extracted, and the mean of the near-field microwave data within the current sampling window is calculated by summing the arithmetic mean of all sampled values ​​within this segment. Then, for each sampled value in the near-field microwave data within the current sampling window, the mean is subtracted to perform deDC operation, removing the non-stationary DC component from the signal and outputting deDC-determined time-series data with zero mean characteristics. Next, spatial transformation operations of low-order and high-order statistical moments are performed using the obtained deDC-determined time-series data. The deDC-determined time-series data is input into the autocorrelation function operator, and the second-order autocorrelation function matrix of the deDC-determined time-series data is calculated by multiplying the time-series data with itself at different time delay steps and calculating the expected value. Simultaneously, each sampled value in the deDC-determined time-series data is raised to the fourth power, and the arithmetic mean of all fourth-power values ​​within the sampling window is calculated to obtain the corresponding fourth-order instantaneous moment value.

[0044] Extract the diagonal elements from the second-order autocorrelation function matrix. These diagonal elements are essentially the autocorrelation coefficients at zero time delay and at a specific time delay. Squaring these diagonal elements and summing them yields the sum of squares. This sum of squares is then compared with the previously calculated fourth-order instantaneous moment values ​​using a weighted difference calculation. The specific mathematical formula is as follows: ; in, This is the final output value of the fourth-order cumulant. The fourth-order instantaneous moment value is obtained by removing DC timing data. The second-order autocorrelation function matrix is ​​the first... The diagonal elements corresponding to each delay position The number 3 represents the index label of the discrete time-delay sequence, and the number 3 is a fixed weighting coefficient constant for the Gaussian probability distribution feature elimination term. Through the above difference balancing calculation, the stationary Gaussian colored noise introduced by the environment is completely filtered out, and the fourth-order cumulant value is finally output. This fourth-order cumulant value, together with the correlation dimension value calculated by the multidimensional phase space trajectory matrix, serves as the core quantitative indicator of the process status and is transmitted in real time to the subsequent dual-indicator joint anomaly monitoring and judgment process.

[0045] S5. When the correlation dimension value decreases to the preset chaotic interval threshold and the fourth-order cumulative value exceeds the preset pulse peak value limit, the coating deposition process is determined to be in a local discharge abnormal state, and a suppression interruption command is triggered.

[0046] Furthermore, in S5, when the correlation dimension value decreases to a preset chaotic interval threshold, including: Extract the Euclidean distance between each pair of spatial phase points in the multidimensional phase space trajectory matrix, and calculate the correlation integral value based on the Euclidean distance under a given spatial correlation radius; Multiple iterative calculations are performed by changing the spatial correlation radius, and the obtained multiple correlation integral values ​​are linearly fitted in double logarithmic coordinates. The slope of the fitted line segment is determined as the correlation dimension value. Obtain the cumulative running time of the current coating deposition process, and use the cumulative running time to calculate the decay slope of the initial threshold, and output the preset chaotic interval threshold. The calculated correlation dimension value is compared with the output preset chaotic interval threshold to determine whether the correlation dimension value has decreased to the preset chaotic interval threshold.

[0047] Furthermore, in S5, determining that the coating deposition process is in a localized discharge abnormal state and triggering a suppression interruption instruction includes: Set a rolling time window and synchronize the time flow of the associated dimension values ​​and the fourth-order cumulant values ​​within the rolling time window; The statistics show the cumulative duration during which the correlation dimension value is below the preset chaotic interval threshold within the rolling time window, and the frequency of abrupt changes where the fourth-order cumulative value exceeds the preset pulse peak value upper limit. When the cumulative duration exceeds a preset time threshold and the frequency of sudden changes exceeds a preset frequency threshold, it is determined that the local discharge is in an abnormal state, and a suppression interrupt command is issued.

[0048] Specifically, the correlation dimension value and fourth-order cumulant value obtained from parallel calculations in the preceding steps are input in real time into the joint threshold chain judgment process in this step. To accurately capture the microscopic microwave signal abrupt changes caused by localized non-uniform discharge during vacuum coating deposition, the judgment process first needs to dynamically construct a comparison boundary. Since the slopes of multiple sets of double logarithmic linear fittings of spatial phase points have been determined as correlation dimension values ​​in the preceding steps, the system then obtains the cumulative running time of the current coating deposition process through timer hardware. Considering the time-varying influence of deposits on the vacuum chamber wall and the gradual thickening of the antenna surface coating on the nonlinear microwave transmission characteristics, the initial threshold is adaptively multiplied by the attenuation slope using the cumulative running time, thereby calculating the dynamic criterion reflecting the current process stage, and outputting a preset chaotic interval threshold. The specific threshold calculation formula is as follows: ; in, The preset chaotic interval threshold is the output. This is the initial threshold constant preset when the process starts. The decay slope constant is the rate at which a preset threshold evolves over time. This is used to obtain the cumulative runtime of the current coating deposition process. Subsequently, the correlation dimension value obtained in real time is compared with the preset chaotic interval threshold output by the numerical comparator to determine whether the correlation dimension value has decreased to the preset chaotic interval threshold.

[0049] To avoid system malfunctions caused by sporadic noise at single points, the decision-making process introduces a transient dynamic time stream alignment and joint counting mechanism based on a rolling time window. The system sets a rolling time window and, within the rolling time window, timestamps are aligned between the correlation dimension values ​​and the fourth-order cumulant values ​​of the discrete time series to maintain data stream synchronization.

[0050] The system activates the bidirectional counting analyzer. On one hand, it continuously monitors each sampling moment within the rolling time window, accumulating and statistically analyzing the cumulative duration when the correlation dimension value is below the preset chaotic interval threshold. On the other hand, it captures the instantaneous mutation peaks of the fourth-order cumulative quantity value in real time, and statistically analyzes the mutation frequency when the fourth-order cumulative quantity value exceeds the preset pulse peak value upper limit. The aforementioned cumulative duration and mutation frequency, as intermediate state indicators reflecting the abnormal energy accumulation and high-frequency micro-arc pulse characteristics, are directly fed as independent variables into the dual-indicator interlocking logic decision unit.

[0051] The dual-index interlocking logic controller compares the calculated cumulative duration with a preset time threshold, and simultaneously compares the statistically calculated mutation frequency with a preset frequency threshold. When the intersection control condition of both the cumulative duration and mutation frequency exceeding the preset frequency threshold is met, a high-risk state lockout signal is triggered, determining that the coating deposition process is in a localized discharge abnormal state. Once this abnormal state is established, the system immediately suspends the current timing data processing and sends a high-level trip trigger pulse to the circuit breaker actuator of the vacuum power supply, i.e., issues a suppression interruption command, thereby achieving dynamic interlocking protection against micro-arc discharge phenomena on the workpiece surface.

[0052] Example 2: In industrial applications involving high-vacuum magnetron sputtering coating deposition on large, irregularly shaped workpieces, the confined space within the deposition chamber and the accompanying high-power plasma discharge prevent sensors from being directly attached to the rotating workpiece surface for close-range monitoring. Typically, a microwave receiving antenna matrix is ​​fixedly arranged around the inner wall of the vacuum chamber. However, during actual deposition, the planetary rotation mechanism drives the workpiece in continuous, high-speed combined revolution and rotation motion to ensure coating uniformity. This high-speed mechanical motion causes the spatial geometric distance between every tiny area on the workpiece surface and the fixed receiving antenna, as well as the electromagnetic field polarization axis angle between them, to be in a high-frequency, time-varying overlapping state. In this specific scenario, the raw microwave signal directly captured by the antenna matrix is ​​mixed with extremely strong free-space propagation divergence loss caused by changes in spatial distance, and dynamic polarization mismatch noise caused by angular distortion. These severe dynamic spatial stray interferences directly mask the microscopic discharge characteristics of the workpiece surface. To solve these problems, this invention provides a coating deposition process anomaly identification system based on multi-source sensor data, the structure of which is as follows: Figure 3 As shown. The specific implementation process of this system is as follows: The geometric space calculation module is used to acquire the revolution and rotation angle data of the planetary rotation mechanism in real time. Combined with the pre-stored three-dimensional point cloud model of the workpiece, it calculates the three-dimensional coordinates of each micro-element on the workpiece surface, and outputs the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. The matrix modulation inverse operation module is used to calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and dynamic polarization angle, and to construct the transmission attenuation matrix as a matrix element. The inverse matrix of the transmission attenuation matrix is ​​used to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. The phase space trajectory reconstruction module is used to determine the optimal time delay by performing mutual information calculation on near-field microwave data using a time delay embedding algorithm, and to determine the minimum embedding dimension by performing pseudo-nearest neighbor discretization calculation, thereby reconstructing a multi-dimensional phase space trajectory matrix. The feature digital calculation module is used to calculate the correlation dimension value based on the multidimensional phase space trajectory matrix, and simultaneously perform high-order statistical calculations on the near-field microwave data to obtain the fourth-order cumulant value. The abnormal chain blocking module is used to determine that the coating deposition process is in a local discharge abnormal state when the correlation dimension value drops to the preset chaotic interval threshold and the fourth-order cumulative value exceeds the preset pulse peak value limit, and to trigger a suppression interruption command.

[0053] Specifically, the geometric space calculation module receives real-time angle data from the planetary rotation mechanism via sensors and calls a pre-stored 3D point cloud model of the workpiece. This module constructs a first rotation matrix and a second rotation matrix using the revolution and rotation angles respectively, and combines this with a fixed translation vector to complete a spatial homogeneous matrix transformation, calculating the 3D coordinates of each micro-element on the workpiece surface in the deposition absolute coordinate system point by point. Subsequently, the geometric space calculation module performs a difference L2 norm operation between the coordinates of each micro-element and the preset absolute coordinates of the antenna array to generate spatial distance data, and simultaneously calculates the 3D angle between the real-time normal vector of the workpiece surface and the antenna polarization axis vector, outputting the dynamic polarization angle.

[0054] The matrix modulation inverse operation module seamlessly receives the aforementioned spatial distance data and dynamic polarization angle. This module calculates the instantaneous spatial propagation loss of each channel using the inverse square law and the polarization mismatch loss using the cosine projection relationship. Multiplying these two values ​​creates a two-dimensional transmission attenuation matrix. Subsequently, the matrix modulation inverse operation module solves for the inverse matrix using a determinant algorithm, extracting the corresponding magnitude from the inverse matrix elements as the amplitude modulation operator and the argument as the phase modulation operator. These two operators are then used to perform a complex product and linear superposition of the complex amplitude of the original microwave signal, inversely canceling out spatial multipath interference and outputting high-fidelity near-field microwave data.

[0055] The phase space trajectory reconstruction module performs time delay embedding reconstruction on the aforementioned near-field microwave data. This module first incrementally increments the time delay variable, calculates the joint probability density and marginal probability density of the time-shifted sequence and the original sequence, and locks in the optimal time delay by constructing a mutual information function curve and optimizing its first first-order partial differential inflection point. Next, the module performs pseudo-nearest neighbor discretization in the adjacent transient high-dimensional space. By statistically analyzing the proportion of elements whose phase point distance ratio exceeds a preset distance threshold, the iteration terminates when the proportion converges to below a preset proportion threshold, thus determining the minimum embedding dimension. Finally, using the optimal time delay and the minimum embedding dimension, a Hankel matrix mapping is performed on the near-field microwave data to generate a multi-dimensional phase space trajectory matrix.

[0056] The feature numerical calculation module employs a dual-track parallel logic. The first track utilizes the multidimensional phase space trajectory matrix to perform iterative calculations of correlation integrals under different spatial correlation radii, and outputs the slope of the fitted line segment as a correlation dimension value characterizing the chaotic dynamic state of the system through least-squares linear fitting in double logarithmic coordinates. The second track synchronously captures near-field microwave data within the current sampling window, performs DC-free calculations to obtain DC-free time-series data, calculates its second-order autocorrelation function matrix and fourth-order instantaneous moment values, and finally outputs the fourth-order cumulant value through weighted difference calculations to eliminate Gaussian components.

[0057] The abnormal chain interruption module retrieves the correlation dimension value and the fourth-order cumulant value in real time. This module first obtains the cumulative running time of the current process, performs adaptive decay slope calculation on the initial threshold, and outputs a preset chaotic interval threshold that dynamically decreases over time, thereby eliminating baseline time-varying drift caused by the thickening of the hardware surface coating. Subsequently, the module timestamps the correlation dimension value and the fourth-order cumulant value within a rolling system time window, and simultaneously counts the cumulative duration of the correlation dimension value below the preset chaotic interval threshold, as well as the frequency of abrupt changes in the fourth-order cumulant value exceeding the preset pulse peak value upper limit. When the bidirectional counting results simultaneously satisfy the condition that the cumulative duration exceeds the preset time threshold and the abrupt change frequency exceeds the preset frequency threshold, the module issues a suppression interrupt command, driving the main circuit breaker to trip and interrupt, cutting off the local discharge energy source, thus achieving automated safety assurance for the thin film deposition process of large irregularly shaped workpieces.

[0058] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying anomalies in coating deposition processes based on multi-source sensor data, characterized in that, include: S1. Real-time acquisition of the revolution and rotation angle data of the planetary rotation mechanism, combined with the pre-stored three-dimensional point cloud model of the workpiece, calculate the three-dimensional coordinates of each micro-element on the workpiece surface, and output the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. S2. Calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and the dynamic polarization angle, and construct a transmission attenuation matrix as a matrix element. Use the inverse matrix of the transmission attenuation matrix to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. S3. Using a time delay embedding algorithm, mutual information calculation is performed on the near-field microwave data to determine the optimal time delay, and pseudo-nearest neighbor discretization calculation is performed to determine the minimum embedding dimension. Based on this, a multi-dimensional phase space trajectory matrix is ​​reconstructed. S4. Calculate the correlation dimension value based on the multidimensional phase space trajectory matrix, and simultaneously perform high-order statistical calculations on the near-field microwave data to obtain the fourth-order cumulant value. S5. When the value of the correlation dimension decreases to the preset chaotic interval threshold and the value of the fourth-order cumulative quantity exceeds the preset pulse peak value limit, it is determined that the coating deposition process is in a local discharge abnormal state and a suppression interruption command is triggered.

2. The method for identifying anomalies in the coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S1, calculating the three-dimensional coordinates of each micro-element on the workpiece surface includes: Extract the initial coordinates of each micro-element in the pre-stored 3D point cloud model of the workpiece; Based on the aforementioned revolution and rotation angle data, a first rotation matrix corresponding to the revolution and a second rotation matrix corresponding to the rotation are constructed respectively; A spatial transformation matrix is ​​constructed using the first rotation matrix, the second rotation matrix, and a preset translation vector; The spatial transformation matrix is ​​used to perform matrix transformation operations on each of the initial coordinates to output the three-dimensional coordinates of each micro-element on the workpiece surface.

3. The method for identifying anomalies in the coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S2, the calculation of the instantaneous spatial propagation loss and polarization mismatch loss for each receiving channel includes: The instantaneous spatial propagation loss is output by calculating the inverse square ratio between the spatial distance data and the preset working wavelength. Based on the dynamic polarization angle, cosine projection calculation is performed to output the polarization mismatch loss; The instantaneous spatial propagation loss and the polarization mismatch loss are multiplied to obtain matrix elements, and the transmission attenuation matrix is ​​constructed using the matrix elements.

4. The method for identifying anomalies in the coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S2, the step of using the inverse matrix of the transmission attenuation matrix to perform inverse amplitude and phase modulation on the original microwave signal input to the antenna matrix includes: Extract the amplitude modulation operator and phase modulation operator from the inverse matrix of the transmission attenuation matrix; The amplitude modulation operator is used to perform a product operation on the original microwave signal to complete the amplitude inverse modulation of the original microwave signal; The phase modulation operator is used to perform a superposition operation on the original microwave signal after amplitude inversion modulation to complete the phase inversion modulation of the original microwave signal; The original microwave signal, after amplitude inversion modulation and phase inversion modulation, is output as near-field microwave data.

5. The method for anomaly identification in coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S3, determining the optimal time delay by performing mutual information calculation on the near-field microwave data includes: Extract the joint probability density distribution and marginal probability density distribution of the near-field microwave data under the time delay variable; By performing integral calculations using the joint probability density distribution and the marginal probability density distribution, the dynamic mutual information function curve of the near-field microwave data as the time-delay variable increases is output. Find the first inflection point in the dynamic mutual information function curve where the slope changes from negative to positive, and determine the time delay variable corresponding to the first inflection point as the optimal time delay.

6. The method for anomaly identification in coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S3, the step of performing pseudo-nearest neighbor discrete calculus to determine the minimum embedding dimension includes: Combining the optimal time delay, the near-field microwave data is subjected to Euclidean distance calculation in the transient space of dimension 1 and dimension 2 plus one, respectively, and the distance ratio of corresponding trajectory points in the two spaces is extracted; The number of elements whose distance ratio is greater than a preset distance threshold is counted to obtain the proportion of false nearest neighbors in the corresponding dimension; The dimension is incremented for iterative calculation. The iteration is terminated when the proportion of false nearest neighbors is less than a preset proportion threshold, and the current dimension is determined as the minimum embedding dimension. A phase space reconstruction operator is constructed using the optimal time delay and the minimum embedding dimension to perform mapping transformation on the near-field microwave data and output a multi-dimensional phase space trajectory matrix.

7. The method for anomaly identification in coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S4, higher-order statistical calculations are performed synchronously on the near-field microwave data to obtain fourth-order cumulant values, including: Calculate the mean of the near-field microwave data within the current sampling window, and use the mean to perform DC removal operation on the near-field microwave data to obtain DC-removed time series data; Calculate the second-order autocorrelation function matrix and the fourth-order instantaneous moment value of the DC-free time series data respectively; The fourth-order cumulant value is output by performing a weighted difference calculation using the sum of squares of the diagonal elements of the second-order autocorrelation function matrix and the fourth-order instantaneous moment value.

8. The method for anomaly identification in coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S5, when the correlation dimension value decreases to a preset chaotic interval threshold, it includes: Extract the Euclidean distance between each pair of spatial phase points in the multidimensional phase space trajectory matrix, and calculate the correlation integral value based on the Euclidean distance under a given spatial correlation radius; The spatial correlation radius is changed to perform multiple iterative calculations, and the resulting multiple correlation integral values ​​are linearly fitted in double logarithmic coordinates. The slope of the fitted line segment is determined as the correlation dimension value. Obtain the cumulative running time of the current coating deposition process, and use the cumulative running time to calculate the attenuation slope of the initial threshold, and output the preset chaotic interval threshold. The calculated correlation dimension value is compared with the output preset chaotic interval threshold to determine whether the correlation dimension value has decreased to the preset chaotic interval threshold.

9. The method for anomaly identification in coating deposition process based on multi-source sensor data according to claim 1, characterized in that, In S5, determining that the coating deposition process is in a localized discharge abnormal state and triggering a suppression interruption command includes: Set a rolling time window, and perform time-stream synchronization alignment of the associated dimension values ​​and the fourth-order cumulative values ​​within the rolling time window; The cumulative duration during which the correlation dimension value is lower than the preset chaotic interval threshold within the rolling time window is counted, as well as the frequency of abrupt changes in the fourth-order cumulative quantity value exceeding the preset pulse peak value upper limit. When the cumulative duration exceeds a preset time threshold and the mutation frequency exceeds a preset frequency threshold, it is determined that the local discharge is in an abnormal state, and a suppression interruption command is issued.

10. A coating deposition process anomaly identification system based on multi-source sensor data, characterized in that, The system for the coating deposition process anomaly identification method based on multi-source sensor data according to any one of claims 1-9, the system comprising: The geometric space calculation module is used to acquire the revolution and rotation angle data of the planetary rotation mechanism in real time. Combined with the pre-stored three-dimensional point cloud model of the workpiece, it calculates the three-dimensional coordinates of each micro-element on the workpiece surface and outputs the spatial distance data and dynamic polarization angle between the workpiece surface and each receiving antenna in the antenna matrix based on the three-dimensional coordinates. The matrix modulation inverse operation module is used to calculate the instantaneous spatial propagation loss and polarization mismatch loss of each receiving channel based on the spatial distance data and the dynamic polarization angle, and to construct a transmission attenuation matrix as a matrix element. The inverse matrix of the transmission attenuation matrix is ​​used to perform inverse modulation of the amplitude and phase of the original microwave signal input to the antenna matrix, and output near-field microwave data. The phase space trajectory reconstruction module is used to perform mutual information calculation on the near-field microwave data using a time delay embedding algorithm to determine the optimal time delay, and to perform pseudo-nearest neighbor discretization calculation to determine the minimum embedding dimension, thereby reconstructing a multi-dimensional phase space trajectory matrix. The feature digital calculation module is used to calculate the correlation dimension value based on the multidimensional phase space trajectory matrix, and simultaneously perform high-order statistical calculations on the near-field microwave data to obtain the fourth-order cumulant value. The abnormal chain blocking module is used to determine that the coating deposition process is in a local discharge abnormal state and trigger a suppression interruption command when the value of the correlation dimension decreases to a preset chaotic interval threshold and the value of the fourth-order cumulative quantity exceeds a preset pulse peak value.

Images