An antenna processing abnormality tracing system and method based on multi-source data analysis

Abstract

The application discloses an antenna processing abnormality tracing system and method based on multi-source data analysis, relates to the technical field of antenna processing monitoring, and comprises the following steps: firstly, multi-source sensing monitoring signals of each process in the whole process of antenna processing are acquired, time synchronization collection and preprocessing are performed, and standard sensing monitoring signal sequences of each process are obtained; then, a directed causal transfer matrix is constructed according to the processing sequence of each process, abnormal contributions transferred by previous processes are deducted, the abnormal contribution degrees of each process are obtained, and candidate time-domain abnormal processes are determined; meanwhile, adaptive characteristic frequency bands are set according to the processing types and physical characteristics of each process, frequency domain features are extracted, candidate frequency-domain abnormal processes are determined; finally, time-frequency fusion determination is performed on the candidate abnormal processes, the abnormal source is determined, and the abnormal propagation path is deduced; the application can accurately identify process self-abnormality and previous transfer abnormality, and completely depict the abnormal propagation path, thereby improving the accuracy and efficiency of antenna processing abnormality tracing.

Description

Technical Field

[0001] This invention relates to the field of antenna processing monitoring technology, specifically to an antenna processing anomaly tracing system and method based on multi-source data analysis. Background Technology

[0002] Antenna fabrication typically involves multiple steps, such as mounting, soldering, and assembly. The quality of each step directly affects the final electrical performance of the antenna. To ensure antenna fabrication quality, various quality monitoring methods have been proposed in the existing technology. One type of method is based on statistical process control (SPC). This method monitors the process status by collecting processing parameters for each step and establishing control charts. When parameters exceed control limits, an anomaly is identified in the process. However, this method can only determine whether a process is abnormal; it cannot trace the source process of the anomaly or its propagation path. Another type of method is fault diagnosis based on signal analysis. This method identifies the fabrication status by analyzing the time-domain or frequency-domain characteristics of sensor signals such as vibration, force, and current. However, this type of method usually only analyzes a single signal type independently and does not fully consider the physical characteristics of the causal transmission of anomalies between different steps in antenna fabrication.

[0003] In actual antenna manufacturing, defects in preceding processes often manifest as signal anomalies in subsequent processes, rather than anomalies generated by the subsequent processes themselves. Current technologies typically treat each process as an independent unit for anomaly assessment, failing to establish a model of anomaly propagation relationships between processes. When anomalies occur in a subsequent process, it's impossible to distinguish whether the anomaly originated from a preceding process or was generated by the process itself, leading to low accuracy in anomaly source localization. Furthermore, existing frequency domain analysis methods generally use a uniform characteristic frequency band to process all processes, neglecting the differences in signal frequency components under different processing types and lacking the ability to extrapolate the propagation path of anomalies from the source process to subsequent processes. The tracing results only output the anomalous process number, failing to provide on-site personnel with visual information on the anomaly propagation relationships between processes, resulting in low troubleshooting efficiency.

[0004] To address the shortcomings of the existing technologies, there is a need for an antenna manufacturing anomaly tracing scheme that can establish the anomaly transmission relationship between processes, distinguish between process-specific anomalies and transmitted anomalies, and deduce the anomaly propagation path. Summary of the Invention

[0005] The purpose of this invention is to provide an antenna manufacturing anomaly tracing system and method based on multi-source data analysis to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a system and method for tracing antenna manufacturing anomalies based on multi-source data analysis. The methods include: Step S1: Acquire sensor monitoring signals from multiple processes in the entire antenna manufacturing process. The sensor monitoring signals include vibration signals, current signals, force signals, and acoustic emission signals. Time-synchronized acquisition and preprocessing of the sensor monitoring signals from each process are performed to obtain the standard sensor monitoring signal sequence for each process. Step S2: Based on the processing sequence of each process in the entire antenna processing process, calculate the comprehensive signal anomaly degree of each process and the temporal correlation strength value between any two processes using the standard sensor monitoring signal sequence of each process. Construct a directed causal transfer matrix based on the processing sequence and the temporal correlation strength value. Use the directed causal transfer matrix to deduct the anomaly contribution part transmitted to the subsequent process from the comprehensive signal anomaly degree of the preceding process to obtain the anomaly contribution degree of each process. Based on the distribution of the anomaly contribution degree of each process, determine the process with the largest deviation of the anomaly contribution degree from the overall distribution. Select the process that exceeds the preset threshold as the temporal candidate anomaly source and record the anomaly trigger time of the temporal candidate anomaly source. Step S3: Based on the processing type of each process in the entire antenna processing process and the physical characteristics of each type of sensor monitoring signal, set the characteristic frequency bands corresponding to each type of sensor monitoring signal in each process. Utilize the frequency domain characteristics of the standard sensor monitoring signal sequence of each process within the characteristic frequency bands corresponding to each type of sensor monitoring signal in each process to calculate the frequency domain anomaly index of each process. Processes with frequency domain anomaly indexes exceeding a preset threshold are taken as candidate frequency domain anomaly sources. Step S4: Construct a candidate anomaly source set based on the time-domain candidate anomaly source and the frequency-domain candidate anomaly source. Enhance the scoring of processes that appear in both the time-domain and frequency-domain candidate anomaly sources. Calculate the dual-evidence collaborative anomaly score for all processes. Take the process with the highest score as the anomaly source process. Based on the directed causal transmission matrix, deduce the propagation path of the anomaly from the anomaly source process to the subsequent processes step by step along the processing sequence. Output the anomaly source process information, the anomaly propagation path, and the anomaly trigger time, and generate an anomaly tracing report.

[0007] Furthermore, step S1 includes: Step S1-1: Let P = {P1, P2, ..., P} be the set of processes consisting of all the processes in the entire antenna manufacturing process. i ,...,P j ,...,P I}, where P1, P2, ..., P i ,...,P j ,...,P ILet represent the 1st, 2nd, ..., i-th, ..., j-th, ..., i-th steps in the entire antenna fabrication process; the sensor monitoring signals collected from each step are time-synchronized, and the sampling times of each type of signal are calibrated using a unified time base, so that the sampled values ​​of all sensor monitoring signals have the same sampling start time t0 and sampling frequency f. s Let the set of discrete sampling times after time synchronization be t={t1,t2,...,t...} n ,...,t N}, where t1, t2, ..., t n ,...,t N These represent the 1st, 2nd, ..., nth, ..., Nth discrete sampling times, respectively; where t n =t0+n / f s For the i-th process P i At the nth discrete time t n The sampled value is denoted as X. i (t n ); Step S1-2: Standardize the format of the sensor monitoring signals of each process after time synchronization, and set the format of the i-th process P... i At the nth discrete time t n The vibration signal, current signal, force signal, and acoustic emission signal are combined to form the i-th process P. i At the nth discrete time t n Sensing monitoring signal vector X i (t n )=[x i,1 (t n ),x i,2 (t n ),x i,3 (t n ),x i,4 (t n )] T ;where x i,1 (t n ),x i,2 (t n ),x i,3 (t n ),x i,4 (t n ) represent the i-th process P respectively i At the nth discrete time t n The amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal, with the superscript T indicating transpose; Step S1-3: For the i-th process P i At the nth discrete time t nSensing monitoring signal vector X i (t n The signal is filtered and denoised using a digital bandpass filter to remove environmental noise and interference components from various types of sensor monitoring signals. The filtered sensor monitoring signal vector is denoted as X. i,f (t n )=X i (t n )*h(t n ); where * represents convolution operation, h(t) n )=[h1(t n ),h2(t n ),h3(t n ),h4(t n )] T Let h1(t) be the unit impulse response vector of the digital filter corresponding to each sensor monitoring signal type. n ),h2(t n ),h3(t n ),h4(t n The units impulse responses of the digital filters corresponding to the vibration signal, current signal, force signal, and acoustic emission signal are respectively denoted as X. The filtered sensing signal vector is denoted as X. i,f (t n Amplitude normalization is performed to obtain the standard sensor monitoring signal vector Y. i (t n )=(X i,f (t n )-μ i ) / σ i ;where μ i and σ i P represents the i-th process. i The mean vector and standard deviation vector of the filtered sensor monitoring signal vector; the standard sensor monitoring signal vector Y at each discrete sampling time. i (t n The i-th process P is constructed according to the sampling time sequence. i Standard sensor monitoring signal sequence Y i ={Y i (t1),Y i (t2),...,Y i (t n ),...,Y i (t N )}, where Y i (t1),Y i (t2),...,Y i (t n ),...,Y i (t N) represent the i-th process P respectively i The standard sensor monitoring signal vector at the 1st, 2nd, ..., nth, ..., Nth discrete sampling times; where Y i (t n )=[y i,1 (t n ),y i,2 (t n ),y i,3 (t n ),y i,4 (t n )] T y i,1 (t n ),y i,2 (t n ),y i,3 (t n ),y i,4 (t n ) represent the i-th process P respectively i At the nth discrete sampling time t n The standard vibration signal amplitude, standard current signal amplitude, standard force signal amplitude, and standard acoustic emission signal amplitude.

[0008] Furthermore, step S2 includes: Step S2-1: According to the i-th process P i Standard sensor monitoring signal sequence Y i ={Y i (t1),Y i (t2),...,Y i (t n ),...,Y i (t N )}, calculate the i-th process P i At all sampling times t1 to t N Comprehensive signal anomaly within ; where y i,m (t n ) represents the i-th process P i The m-th type of standard sensor monitoring signal at the nth discrete sampling time t n Standard sensor monitoring signal value, y i,m,avg P represents the i-th process. i The average value of the m-th type of standard sensor monitoring signal over all sampling times, where m=1,2,3,4 correspond to standard vibration signal, standard current signal, standard force signal, and standard acoustic emission signal, respectively. Step S2-2: For the i-th process P i and the j-th process P j According to the i-th process Pi Standard sensor monitoring signal sequence Y i ={Y i (t1),Y i (t2),...,Y i (t n ),...,Y i (t N )} and the j-th process P j Standard sensor monitoring signal sequence Y j ={Y j (t1),Y j (t2),...,Y j (t n ),...,Y j (t N )}, calculate P for the i-th process respectively. i Standard sensor monitoring signal sequence and the j-th process P j The dot product of the standard sensor monitoring signal sequence at each discrete sampling time is calculated, and cross-correlation is performed to obtain the cross-correlation function between the i-th and j-th processes. Where τ is the number of lag sampling points, and its value range is τ=-(N-1),...,-1,0,1,...,N-1. When τ<0, it indicates that the j-th process is ahead of the i-th process, and when τ>0, it indicates that the j-th process is behind the i-th process. The superscript T indicates transpose. The cross-correlation function G is extracted. i,j The peak value of (τ) is taken as the temporal correlation strength value between the i-th process and the j-th process. Record the cross-correlation function G. i,j The number of hysteresis sampling points corresponding to the peak value of (τ) Where arg max represents the number of lag sampling points corresponding to the maximum value, and τ i,j The sign of indicates whether the j-th process is ahead or behind the i-th process; Step S2-3: Utilize the processing sequence of the entire antenna fabrication process and the timing correlation strength value C obtained in step S2-2. i,j Construct the directed causal transit matrix A∈R between processes. I×I , The expression is Where R represents the set of real numbers; and the elements A in the directed causal transit matrix A are... ij P represents the i-th process. i For the j-th process P j The anomaly propagation coefficient; depending on the processing sequence, an anomaly in a subsequent process may originate from the propagation of an anomaly in a preceding process, while an anomaly in an preceding process cannot originate from a subsequent process. Therefore, when i>j, A ij=0, when i<j, A ij =C i,j , when i=j, A ij =0; Step S2-4: According to the comprehensive signal anomaly degree a of each process i and the directed causal transfer matrix A, calculate the anomaly contribution degree of the i-th process P i under the constraint of processing sequence ; where represents the total anomaly contribution transferred from all preorder processes to the i-th process P i , the anomaly contribution degree of the i-th process P i under the constraint of processing sequence represents the additional anomaly degree of the i-th process P i exceeding the preorder anomaly transfer influence; taking the anomaly contribution degree of the i-th process P i under the constraint of processing sequence as the time-domain confidence weight of the i-th process P i ; Step S2-5: According to the anomaly contribution degree c of the i-th process P i i , calculate the mean value of the process anomaly contribution degrees and the standard deviation ; calculate the deviation multiple value d of the anomaly contribution degree of the i-th process P i i =(c i -c avg ) / s c ; if the anomaly contribution degree deviation multiple value of the i-th process P i satisfies d i >d th , where d th is the anomaly contribution degree deviation threshold; then mark the i-th process P i as a time-domain candidate anomaly source; if there is no process that satisfies d i >d th , it is determined that there is no time-domain anomaly in the process machining.

[0009] Step S2-6: For the i-th process P marked as a time-domain candidate anomaly source i , according to the signal time sequence relationship between the i-th process P i and the adjacent preorder process, the (i-1)-th process P i-1 , using the lag sampling point number τ corresponding to the peak value recorded in step S2-2 i-1,i , combined with the sampling frequency f s , calculate the propagation time Δt for the anomaly to propagate from the i-th process P i to the adjacent process P j i-1,i =τ i-1,i / f​​​s According to the propagation time Δt i-1,i The anomaly is traced back to the time when the candidate anomaly source in the time domain occurs; if there are multiple candidate anomaly source processes in the time domain, the anomaly trigger time corresponding to each candidate anomaly source process in the time domain is recorded respectively.

[0010] Furthermore, step S3 includes: Step S3-1: Based on the processing type of each step in the entire antenna manufacturing process and the physical characteristics of each type of sensing and monitoring signal, determine the process P for the i-th step. i The m-th type of standard sensor monitoring signal is set to the corresponding characteristic frequency band [f] L,i,m ,f H,i,m ], where f L,i,m and f H,i,m P represents the i-th process. i The lower and upper boundary frequencies of the characteristic frequency band of the m-th type standard sensor monitoring signal are given. Different processing steps and different signal types correspond to different characteristic frequency band intervals. Step S3-2: For the i-th process P i Standard sensor monitoring signal sequence Y i ={Y i (t1),Y i (t2),...,Y i (t n ),...,Y i (t N The standard sensor monitoring signal vector at each discrete sampling time is subjected to a discrete Fourier transform to obtain the i-th process P. i Frequency domain standard sensor monitoring signal vector ;where f q =q·f s / N represents the q-th discrete frequency point, q=0,1,...,N-1, and j is the imaginary unit; the frequency domain standard sensing monitoring signal vector Z i (f q (This is expressed in amplitude-phase form) ; where M i (f q P represents the i-th process. i At frequency point f q The amplitude spectrum vector at θ i (f q P represents the i-th process. i At frequency point f q Phase spectrum vector at; Step S3-3: Extract the i-th process P i The m-th type standard sensor monitoring signal in the characteristic frequency band [f L,i,m ,fH,i,m Amplitude spectrum M within ] i,m (f q ), calculate the i-th process P i Frequency domain energy concentration of the m-th standard sensor monitoring signal within the characteristic frequency band ; where q L,m =⌈f L,I,m ·N / f s ⌉,q H,m =⌊f H,i,m ·N / f s ⌋, representing the i-th process P respectively. i The discrete frequency point indices corresponding to the lower and upper boundaries of the characteristic frequency band of the m-th type of standard sensor monitoring signal are given by the i-th process P. i The total number of frequency points within the characteristic frequency band of the m-th type standard sensor monitoring signal. Indicates rounding up. Indicates rounding down; Step S3-4: Calculate the i-th process P i The total energy of the m-th type standard sensor monitoring signal within the characteristic frequency band ; Calculate the i-th process P i The energy percentage r of the m-th standard sensor monitoring signal within the characteristic frequency band i,m =(E i,m ·Q i,m ) / E i,m,total The energy percentage r i,m P represents the i-th process. i The concentration of the energy of the m-th type standard sensor monitoring signal within the characteristic frequency band. The larger the value, the more concentrated the signal energy of the process within the characteristic frequency band, and the more significant the characteristic signal under the corresponding processing type. Step S3-5: Based on the processing type of each process, assign P to the i-th process. i The expected energy percentage range corresponding to the m-th type standard sensor monitoring signal is set [r] i,m,low ,r i,m,high The expected energy percentage range [r] i,m,low ,r i,m,high According to the antenna manufacturing process standard for the i-th process P i The signal energy distribution requirements specified by the processing type are determined; the i-th process P i The energy percentage r of the m-th standard sensor monitoring signal i,m Compare the i-th process P with the expected energy percentage range to calculate the i-th process P. i Frequency domain anomaly index of the m-th type standard sensor monitoring signal within the characteristic frequency band The frequency domain anomaly index η i,m P represents the i-th process.i The relative extent to which the energy percentage of the m-th type standard sensor monitoring signal deviates from the expected range within the characteristic frequency band; Step S3-6: For the i-th process P i The frequency domain anomaly indicators of the four types of standard sensor monitoring signals are weighted and fused to obtain the i-th process P. i Comprehensive frequency domain anomaly index ;where β m The preset weighting coefficients for the m-th type of standard sensor monitoring signal satisfy... ; Step S3-7: Set the phase deviation threshold η based on the statistical distribution of the overall phase deviation of each process under historical normal processing conditions. th When the i-th process P i The overall phase deviation satisfies η i >η th When, then the i-th process P i Marked as a candidate anomaly source in the frequency domain; if no such anomaly source exists satisfying η i >η th If the process is not frequency domain abnormal, then it is determined that the process does not have frequency domain anomalies.

[0011] Furthermore, step S4 includes: Step S4-1: Define the set S of candidate anomaly sources in the time domain T ={P i ∣d i >d th} and the frequency domain candidate anomaly source set S F ={P i |η i >η th}; Step S4-2: If the set of candidate anomaly sources in the time domain is S T and the frequency domain candidate anomaly source set S F If all values ​​are empty, it is determined that there is no abnormality in the process, and the result of no processing abnormality is output and the subsequent steps are terminated. Step S4-3: For all operations, calculate P for the i-th operation. i Dual evidence co-existing abnormality score ;in As an indicator function, when P i ∈S T ∩S F The value is 1 if the condition is met, and 0 otherwise; α>0 is a preset enhancement coefficient used to enhance the scoring of processes that appear simultaneously in both the time-domain and frequency-domain candidate anomaly source sets; the process with the highest dual-evidence collaborative anomaly score is taken as the anomaly source process P. origin ; Step S4-4: Based on the directed causal transit matrix A, from the anomaly source process P... origin Initially, the anomaly propagation path is deduced step by step along the processing sequence: for the j-th process P j (j>origin), if the abnormal source process P origin For the j-th process P j The abnormal transmission coefficient A origin,j Greater than the preset propagation judgment threshold A th If the anomaly is determined, the source process P is identified as the anomaly. origin Passed to the j-th process P j The j-th process P j Mark the process as affected by the anomaly and include it in the anomaly propagation path; traverse all subsequent processes of the anomaly source process in the order of processing, and after judging all subsequent processes, obtain the complete anomaly propagation path; Step S4-5: Output the abnormal source process P origin The process number and its corresponding exception trigger time are output, and the exception propagation path contains information about the processes affected by the exception; if the exception source process P origin If the corresponding exception trigger time recorded in steps S2-6 exists, then the exception source process P will be... origin The abnormal trigger time is output as the abnormal trigger time of the final abnormal source process; if there is no corresponding abnormal trigger time, the number of hysteresis sampling points τ corresponding to the peak value recorded in step S2-2 is used. origin-1,origin Combined with sampling frequency f s Calculate the propagation time Δt from the anomaly to the anomaly source process. origin-1,origin =τ origin-1,origin / f s ; Taking the abnormal source process P origin Using the time point corresponding to the cross-correlation peak in the standard sensor monitoring signal sequence as a reference, the propagation time Δt is traced back. origin-1,origin It obtains the time when the anomaly occurred in the process at which the anomaly originated; and generates an anomaly tracing report containing information about the process at which the anomaly originated, the anomaly propagation path, and the time when the anomaly was triggered.

[0012] The system includes: a signal acquisition and preprocessing module, a time-domain feature extraction and time-domain candidate determination module, a frequency-domain feature extraction and frequency-domain candidate determination module, and a time-frequency fusion and source tracing output module; The signal acquisition and preprocessing module is used to acquire and preprocess the sensor monitoring signals of multiple processes in the entire antenna manufacturing process, and to obtain the standard sensor monitoring signal sequence of each process through time synchronization acquisition and preprocessing. The time-domain feature extraction and time-domain candidate determination module is used to calculate time-domain features and determine time-domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process, and at the same time record the anomaly trigger time of the time-domain candidate anomaly source; The frequency domain feature extraction and frequency domain candidate determination module is used to calculate frequency domain features and determine frequency domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process. The time-frequency fusion and source tracing output module is used to calculate the dual-evidence collaborative anomaly score based on the time-domain candidate anomaly source and the frequency-domain candidate anomaly source, determine the anomaly source process, deduce the anomaly propagation path, and generate an anomaly source tracing report.

[0013] The signal acquisition and preprocessing module includes: The process set definition unit is used to define the process set, perform time synchronization processing on the sensing and monitoring signals of each process, and define the set of discrete sampling times; The sensing and monitoring signal vector combination unit is used to combine vibration signals, current signals, force signals and acoustic emission signals of the same process at the same time into a sensing and monitoring signal vector; The filtering, denoising, and normalization processing unit is used to filter, denoise, and normalize the amplitude of the sensing and monitoring signal vectors to obtain the standard sensing and monitoring signal sequences for each process.

[0014] The temporal feature extraction and temporal candidate determination module includes: The integrated signal anomaly calculation unit is used to calculate the integrated signal anomaly of each process based on the standard sensor monitoring signal sequence of each process. The time-series correlation strength calculation unit is used to perform cross-correlation calculations on the standard sensor monitoring signal sequences of any two processes to obtain the time-series correlation strength value and the number of hysteresis sampling points corresponding to the peak value. A directed causal transitivity matrix construction unit is used to construct a directed causal transitivity matrix based on the processing order and temporal correlation strength value; The anomaly contribution calculation unit is used to calculate the anomaly contribution of each process after deducting the effects of the preceding transmission based on the comprehensive signal anomaly degree and the directed causal transfer matrix. The time-domain candidate anomaly source marking unit is used to mark time-domain candidate anomaly sources according to the deviation multiple of the anomaly contribution of each process; The abnormal triggering time recording unit is used to infer and record the abnormal triggering time of the time-domain candidate abnormal source process based on the number of lag sampling points and the sampling frequency.

[0015] The frequency domain feature extraction and frequency domain candidate determination module includes: The characteristic frequency band setting unit is used to set the characteristic frequency band range corresponding to each type of standard sensor monitoring signal according to the processing type of each process and the physical characteristics of each type of sensor monitoring signal; The Discrete Fourier Transform (DFT) unit is used to perform DFT on the standard sensing and monitoring signal sequences of each process to obtain the frequency domain standard sensing and monitoring signal vector. The frequency domain energy concentration calculation unit is used to extract the amplitude spectrum of various types of standard sensor monitoring signals in the characteristic frequency band and calculate the frequency domain energy concentration. The energy percentage calculation unit is used to calculate the total energy of various types of standard sensor monitoring signals in the entire frequency band and the energy percentage in the characteristic frequency band. The frequency domain anomaly index calculation unit is used to calculate the frequency domain anomaly index of various types of standard sensor monitoring signals based on the expected energy ratio range and the actual energy ratio of each process. The comprehensive frequency domain anomaly index calculation unit is used to perform weighted fusion of the frequency domain anomaly indices of the four types of standard sensor monitoring signals in each process to obtain the comprehensive frequency domain anomaly index. The frequency domain candidate anomaly source marking unit is used to mark processes whose comprehensive frequency domain anomaly index exceeds the threshold as frequency domain candidate anomaly sources according to a preset threshold.

[0016] The time-frequency fusion and source tracing output module includes: The candidate anomaly source set definition unit is used to define the time-domain candidate anomaly source set and the frequency-domain candidate anomaly source set; The no-anomaly determination unit is used to determine whether there are candidate anomaly sources. If all are empty, the no-anomaly result is output. The dual-evidence collaborative anomaly scoring calculation unit is used to calculate the dual-evidence collaborative anomaly score for all processes, enhance the score of processes that appear in both candidate sets, and take the process with the highest score as the source of the anomaly. The anomaly propagation path deduction unit is used to deduce the anomaly propagation path step by step from the anomaly source process along the processing sequence based on the directed causal transmission matrix. The source tracing result output unit is used to output information on the source process of the anomaly, the propagation path of the anomaly, and the trigger time of the anomaly, and generate an anomaly source tracing report.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a directed causal transmission matrix between processes and uses process processing sequence constraints to transform the undirected temporal correlation strength into a directed anomaly transmission coefficient. It accurately deducts the anomaly contribution from the preceding process from the comprehensive signal anomaly degree of each process, thus obtaining the true anomaly contribution degree that reflects the anomaly generated by the process itself. This overcomes the defect of the prior art that cannot distinguish between the anomaly of the process itself and the anomaly transmitted from upstream, and misjudges the transmitted anomaly as the anomaly of the current process itself, thus significantly improving the accuracy of locating the anomaly source process.

[0018] 2. Based on the processing type of different processes and the physical characteristics of each type of sensor monitoring signal, this invention sets independent characteristic frequency band intervals for each type of sensor monitoring signal in each process, calculates the energy ratio of each type of signal in the characteristic frequency band, and determines frequency domain anomalies by comparing it with the expected energy ratio range specified in the antenna processing technology standard. This overcomes the shortcomings of existing technologies that use a unified characteristic frequency band for frequency domain analysis, which cannot adapt to different process types and have inaccurate detection results, making frequency domain anomaly detection more in line with the actual processing characteristics of different processes.

[0019] 3. This invention jointly determines candidate anomaly sources in the time domain and candidate anomaly sources in the frequency domain, enhances the scoring of processes that are marked in both the time and frequency domains, and then uses a directed causal transmission matrix to deduce the propagation path of the anomaly from the source process to the subsequent process step by step along the processing sequence. It also outputs the source process of the anomaly, the affected process of the anomaly, and the trigger time of the anomaly. This overcomes the shortcomings of the prior art, which only outputs the abnormal process number and cannot provide information on the anomaly transmission relationship between processes. It provides on-site personnel with a complete anomaly propagation link, which facilitates the rapid location of the root cause of the anomaly and the assessment of the scope of its impact. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an antenna fabrication anomaly tracing method based on multi-source data analysis according to the present invention. Figure 2 This is a schematic diagram of the structure of an antenna processing anomaly tracing system based on multi-source data analysis according to the present invention. Detailed Implementation

[0021] 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.

[0022] Example: Figures 1-2 As shown, the present invention provides a technical solution: a system and method for tracing antenna manufacturing anomalies based on multi-source data analysis. The entire process of manufacturing a certain type of antenna includes 7 steps, numbered in the following order: P1 (positioning ceramic block to fixture), P2 (attaching ceramic block to shielding cover), P3 (positioning PCB to shielding cover), P4 (soldering ceramic pins), P5 (soldering and sealing the shielding cover), P6 (electrical testing), and P7 (full inspection and packaging).

[0023] The methods include: Step S1: Acquire sensor monitoring signals from multiple processes in the entire antenna manufacturing process. The sensor monitoring signals include vibration signals, current signals, force signals, and acoustic emission signals. Time-synchronized acquisition and preprocessing of the sensor monitoring signals from each process are performed to obtain the standard sensor monitoring signal sequence for each process. Step S1-1: Define the process set P={P1,P2,P3,P4,P5,P6,P7}; perform time synchronization processing on the sensor monitoring signals collected by each process, ensuring that all signals have the same sampling start time t0=0s and sampling frequency f. s =1000Hz; the discrete sampling times after time synchronization are: t1=0.001s, t2=0.002s, t3=0.003s, t4=0.004s, t5=0.005s; Step S1-2: Taking process P2 (attaching the ceramic block to the shielding cover) as an example, the original sensor monitoring signal vectors at the 5 sampling times are as follows: At time t1 = 0.001s, the amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal are 0.82, 1.2, 2.1, and 0.45, respectively, forming the vector X2(t1) = [0.82, 1.2, 2.1, 0.45]. T ; At time t2 = 0.002s, the amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal are 0.95, 1.35, 2.4, and 0.52, respectively, forming the vector X2(t2) = [0.95, 1.35, 2.4, 0.52]. T ; At time t3 = 0.003s, the amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal are 1.56, 1.8, 3.2, and 0.78, respectively, forming the vector X2(t3) = [1.56, 1.8, 3.2, 0.78]. T ; At time t4 = 0.004s, the amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal are 1.2, 1.5, 2.7, and 0.63, respectively, which combine to form the vector X2(t4) = [1.2, 1.5, 2.7, 0.63]. T ; At time t5 = 0.005s, the amplitudes of the vibration signal, current signal, force signal, and acoustic emission signal are 0.88, 1.25, 2.2, and 0.48, respectively, which combine to form the vector X2(t5) = [0.88, 1.25, 2.2, 0.48]. T ; Steps S1-3: The sensor monitoring signal vector of process P2 is filtered and denoised. The filter passband is set to 200Hz~500Hz based on the effective frequency range of the ceramic block bonding process. After filtering, amplitude normalization is performed to obtain the standard sensor monitoring signal vector Y2(t). n ): At time t1 = 0.001s, the amplitudes of the standard vibration signal (-0.65), standard current signal (-0.70), standard force signal (-0.68), and standard acoustic emission signal (-0.62) are combined into a vector Y2(t1) = [-0.65, -0.7, -0.68, -0.62]. T ; At time t2 = 0.002s, the amplitudes of the standard vibration signal (-0.35), standard current signal (-0.4), standard force signal (-0.38), and standard acoustic emission signal (-0.38) are combined into a vector Y2(t2) = [-0.35, -0.4, -0.38, -0.38]. T ; At time t3 = 0.003s, the amplitudes of the standard vibration signal (1.85), standard current signal (1.9), standard force signal (1.92), and standard acoustic emission signal (1.8) are combined into the vector Y2(t3) = [1.85, 1.9, 1.92, 1.8]. T ; At time t4 = 0.004s, the amplitudes of the standard vibration signal (0.25), standard current signal (0.2), standard force signal (0.18), and standard acoustic emission signal (0.22) are combined into the vector Y2(t4) = [0.25, 0.2, 0.18, 0.22]. T ; At time t5 = 0.005s, the amplitudes of the standard vibration signal (-0.48), standard current signal (-0.5), standard force signal (-0.52), and standard acoustic emission signal (-0.45) are combined into the vector Y2(t5) = [-0.48, -0.5, -0.52, -0.45]. T ; The standard sensor monitoring signal sequence Y2={Y2(t1),Y2(t2),Y2(t3),Y2(t4),Y2(t5)} of process P2 is formed by the standard sensor monitoring signal vectors at the above 5 time points in chronological order.

[0024] Similarly, the above operations are performed on all seven processes to obtain the standard sensor monitoring signal sequence for each process.

[0025] Step S2: Based on the processing sequence of each process in the entire antenna processing process, calculate the comprehensive signal anomaly degree of each process and the temporal correlation strength value between any two processes using the standard sensor monitoring signal sequence of each process. Construct a directed causal transfer matrix based on the processing sequence and the temporal correlation strength value. Use the directed causal transfer matrix to deduct the anomaly contribution part transmitted to the subsequent process from the comprehensive signal anomaly degree of the preceding process to obtain the anomaly contribution degree of each process. Based on the distribution of the anomaly contribution degree of each process, determine the process with the largest deviation of the anomaly contribution degree from the overall distribution. Select the process that exceeds the preset threshold as the temporal candidate anomaly source and record the anomaly trigger time of the temporal candidate anomaly source. Step S2-1: Taking process P2, the standard vibration signal values ​​at its 5 sampling times are [-0.65, -0.35, 1.85, 0.25-0.48]; first, calculate the average value of this type of signal over all sampling times: y 2,1,avg =(-0.65-0.35+1.85+0.25-0.48) / 5=0.124; Similarly, the average value of the standard current signal over all sampling times can be obtained: y 2,2,avg =0.1; The average value of the standard force signal over all sampling times: y 2,3,avg =0.104; The average value of the standard acoustic emission signal over all sampling times: y 2,4,avg =0.114; According to the formula for calculating the comprehensive signal anomaly degree: Substituting the four types of standard sensor monitoring signal data from process P2 into the above formula, we get: ; Similarly, the overall signal anomaly degree for other processes can be obtained: The overall signal anomaly degree of process P1 is: a1=2.1; The overall signal anomaly degree of process P3 is: a3=1.1; The overall signal anomaly rate for process P4 is: a4 = 2.4; The overall signal anomaly rate for process P5 is: a5 = 1.3; The overall signal anomaly rate for process P6 is: a6 = 0.95; The overall signal anomaly rate for process P7 is: a7 = 0.7; Step S2-2: Using the vibration signals from processes P1 (positioning the ceramic block to the fixture) and P2 (attaching the ceramic block to the shielding cover), the five standard vibration signal values ​​for process P1 are known to be [0.5, 0.6, 1.2, 0.8, 0.55], and the five standard vibration signal values ​​for process P2 are known to be [-0.65, -0.35, 1.85, 0.25, -0.48]. Calculate using the cross-correlation function formula: Calculate the cross-correlation value when τ=-2 using the vibration signal: ; Similarly, calculating the cross-correlation values ​​for τ = -1, 0, 1, 2 yields 0.213, 0.324, 0.202, and -0.014 respectively. Comparing the cross-correlation values ​​for each τ value, the maximum value of 0.412 is obtained when τ = -2. Therefore, the time series correlation strength value is: The number of hysteresis sampling points corresponding to the peak value: ;τ 1,2 =-2 indicates that process P1 lags behind process P2 by 2 sampling points, meaning the signal propagates from process P2 to process P1; Similarly, the temporal correlation strength values ​​and the number of lag sampling points between each pair of other processes can be obtained; Similarly, the temporal correlation strength values ​​and the number of lag sampling points between other process pairs can be obtained: C 1,3 =0.12, τ 1,3 =1 indicates that P1 is one sampling point ahead of P3; C 1,4 =0.56, τ 1,4 =-1 indicates that P1 lags behind P4 by 1 sampling point; C 1,5 =0.08, τ 1,5 =2, indicating that P1 is 2 sampling points ahead of P5; C 1,6 =0.05, τ 1,6 =0 indicates that process P1 and process P6 are synchronized; C 1,7 =0.02, τ 1,7 =-1 indicates that process P1 lags behind process P7 by 1 sampling point; C 2,3 =0.15, τ 2,3 =0 indicates that process P2 and process P3 are synchronized; C 2,4 =0.62, τ 2,4 =0 indicates that process P2 and process P4 are synchronized; C 2,5 =0.1, τ 2,5 =0 indicates that process P2 and process P5 are synchronized; C 2,6 =0.06, τ 2,6 =1, which indicates that process P2 leads process P6 by 1 sampling point; C 2,7 =0.03, τ 2,7 =0, which indicates that process P2 is synchronized with process P7; C 3,4 =0.18, τ 3,4 =1, which indicates that process P3 leads process P4 by 1 sampling point; C 3,5 =0.21, τ 3,5 =-1, which indicates that process P3 lags behind process P5 by 1 sampling point; C 3,6 =0.14, τ 3,6 =0, which indicates that process P3 is synchronized with process P6; C 3,7 =0.09, τ 3,7 =0, which indicates that process P3 is synchronized with process P7; C 4,5 =0.85, τ 4,5 =-1, which indicates that process P4 lags behind process P5 by 1 sampling point; C 4,6 =0.22, τ 4,6 =1, which indicates that process P4 leads process P6 by 1 sampling point; C 4,7 =0.11, τ 4,7 =0, which indicates that process P4 is synchronized with process P7; C 5,6 =0.88, τ 5,6 =-1, which indicates that process P5 lags behind process P6 by 1 sampling point; C 5,7 =0.35, τ 5,7 =-1, which indicates that process P5 lags behind process P7 by 1 sampling point; C 6,7 =0.79, τ 6,7 =0, which indicates that process P6 is synchronized with process P7; Step S2-3: Using the processing sequence constraint, construct a directed causal transfer matrix A∈R 7×7 , whose element A ij satisfies: when i>j, A ij =0; when i<j, A ij =C i,j ; when i=j, A ij =0; the expanded form of A is: ; taking A1,2 =C 1,2 =0.412, indicating that the anomaly transmission coefficient from process P1 to process P2 is 0.412; with A 2,4 =C 2,4 =0.62, indicating that the anomaly transmission coefficient from process P2 to process P4 is 0.62; with A 4,5 =C 4,5 =0.85, indicating that the anomaly transmission coefficient from process P4 to process P5 is 0.85; with A 5,6 =C 5,6 =0.88, indicating that the anomaly transmission coefficient from process P5 to process P6 is 0.88; with A 6,7 =C 6,7 =0.79, indicating that the anomaly transmission coefficient from process P6 to process P7 is 0.79; Step S2-4: Calculate the abnormal contribution rate using the following formula: For operation P2, the preceding operation is only P1: c2 = a2 - A 1,2 a1 = 4.145 - 0.412 × 2.1 = 3.28; Similarly, the abnormal contribution of other processes can be obtained as follows: The abnormal contribution of process P1 is c1=2.1; The abnormal contribution of process P3 is c3=0.225; The abnormal contribution of process P4 is c4=-1.547; The abnormal contribution of process P5 is c5=-1.554; The abnormal contribution of process P6 is c6=-1.23; The abnormal contribution of process P7 is c7=-1.036; Step S2-5: Calculate the mean and standard deviation based on the abnormal contribution of each process: ; ; For process P2, calculate the deviation multiple of the abnormal contribution of process P2: d2 = (c2 - c avg ) / s c =1.671; Similarly, the deviation multiples of abnormal contribution rates for other processes are: The deviation multiple of the abnormal contribution of process P1 is: d1=1.063; The deviation multiple of the abnormal contribution of process P3 is: d3=0.098; The deviation multiple of the abnormal contribution of process P4 is: d4=-0.814; The deviation multiple of the abnormal contribution of process P5 is: d5=-0.817; The deviation multiple of the abnormal contribution of process P6 is: d6=-0.651; The deviation multiple of the abnormal contribution of process P7 is: d7 = -0.551; Let the abnormal contribution deviation from the threshold be d. th =1; satisfies d i >d th The process consists of process P1 and process P2. Process P1 and process P2 are marked as candidate anomaly sources in the time domain. Step S2-6: For process P2, based on the number of lag sampling points τ between the adjacent preceding processes P1 and P2 recorded in step S2-2... 1,2 =-2, combined with the sampling frequency f s =1000Hz, calculate the propagation time Δt of the anomaly from process P1 to process P2. 1,2 =τ 1,2 / f s =(-2) / 1000=-0.002s; The negative number indicates that process P1 lags behind process P2 by 2 sampling points, meaning that the anomaly in process P2 occurs before process P1; The signal peak of process P2 occurs at t3=0.003s. Going back 0.002s, the anomaly trigger time of process P2 is obtained as follows: For process P1, which is the first process without any preceding processes, the abnormal trigger time is the time point corresponding to the signal peak value at 0.003s; record the abnormal trigger times of processes P1 and P2 respectively.

[0026] Step S3: Based on the processing type of each process in the entire antenna processing process and the physical characteristics of each type of sensor monitoring signal, set the characteristic frequency bands corresponding to each type of sensor monitoring signal in each process. Utilize the frequency domain characteristics of the standard sensor monitoring signal sequence of each process within the characteristic frequency bands corresponding to each type of sensor monitoring signal in each process to calculate the frequency domain anomaly index of each process. Processes with frequency domain anomaly indexes exceeding a preset threshold are taken as candidate frequency domain anomaly sources. Step S3-1: Based on the antenna manufacturing process standards, define the characteristic frequency bands for different process types: The standard vibration signal characteristic frequency band for process P1 (positioning ceramic block to fixture) is [200, 400] Hz; The standard vibration signal characteristic frequency band for process P2 (attaching ceramic blocks to the shielding cover) is [200, 500] Hz; The standard vibration signal characteristic frequency band for the P3 process (positioning the PCB to the shielding cover) is [50, 150] Hz; The standard vibration signal characteristic frequency band for the P4 process (welding ceramic pins) is [100, 300] Hz; The standard vibration signal characteristic frequency band for the P5 process (shielding cap tin sealing) is [10, 50] Hz; The standard vibration signal characteristic frequency band for the P6 process (electrical testing) is [5, 20] Hz; The P7 process (full appearance inspection and packaging) does not have a designated frequency band. For current, force, and acoustic emission signals, the corresponding characteristic frequency bands are set according to their respective physical characteristics, referring to the above rules. Step S3-2: Perform a Discrete Fourier Transform on the standard vibration signal sequence [-0.65, -0.35, 1.85, 0.25, -0.48] of process P2: ;where f q =q·1000 / 5=q·200Hz, q=0,1,2,3,4, the corresponding frequency points are 0,200,400,600,800Hz respectively; The calculated amplitude spectrum values ​​at each frequency point are: 0.5 at 0Hz, 2.1 at 200Hz, 1.8 at 400Hz, 0.4 at 600Hz, and 0.2 at 800Hz. Step S3-3: The characteristic frequency band of the standard vibration signal in process P2 is [200, 500] Hz, and the corresponding discrete frequency points are 200 Hz and 400 Hz, with a total number of frequency points Q. 2,1 =2; Calculate the frequency domain energy concentration of the standard vibration signal for process P2 within the characteristic frequency band: ; Step S3-4: Calculate the total energy of the standard vibration signal of process P2 within its characteristic frequency band: ; Calculate the energy percentage of the standard vibration signal in the characteristic frequency band of process P2: r 2,1 =(E 2,1 ×Q 2,1 ) / E 2,1,total =0.944; This value indicates the degree of concentration of signal energy within the characteristic frequency band; Step S3-5: Referring to the antenna manufacturing process standards, set the expected energy percentage range [r] for the vibration signal during the process of attaching the ceramic block to the shielding cover. i,m,low ,r i,m,high = [0.6, 0.8]; Since the actual energy percentage of 0.944 is greater than the expected upper limit of 0.8, the frequency domain anomaly index is calculated: η 2,1 =(r 2,1 -r 2,1,high ) / r 2,1,high =(0.944-0.8) / 0.8=0.18; Similarly, frequency domain anomaly indicators for other types of signals can be obtained: Frequency domain anomaly index of the standard current signal in process P2 within its characteristic frequency band: η2,2 =0.035; Frequency domain anomaly index of the standard force signal in process P2 within the characteristic frequency band: η 2,3 =0.12; Frequency domain anomaly index of the standard acoustic emission signal in the characteristic frequency band of process P2: η 2,4 =0.01; Step S3-6: Assume the weighting coefficients for the standard vibration signal, standard current signal, standard force signal, and standard acoustic emission signal are 0.4, 0.2, 0.3, and 0.1, respectively. The comprehensive frequency domain anomaly index for process P2 is: ; Similarly, the comprehensive frequency domain anomaly indexes for other processes can be obtained: The comprehensive frequency domain anomaly index for process P1 is: η1=0.072; The comprehensive frequency domain anomaly index for process P3 is: η3=0.045; The comprehensive frequency domain anomaly index for process P4 is: η4 = 0.032; The comprehensive frequency domain anomaly index for process P5 is: η5 = 0.085; The comprehensive frequency domain anomaly index for process P6 is: η6 = 0.062; The comprehensive frequency domain anomaly index for process P7 is: η7=0.02; Step S3-7: Set the phase deviation threshold η th =0.08; satisfies η i >η th The processes are P2 and P5, and P2 and P5 are marked as candidate anomaly sources in the frequency domain.

[0027] Step S4: Construct a candidate anomaly source set based on the time-domain candidate anomaly source and the frequency-domain candidate anomaly source. Enhance the scoring of processes that appear in both the time-domain and frequency-domain candidate anomaly sources. Calculate the dual-evidence collaborative anomaly score for all processes. Take the process with the highest score as the anomaly source process. Based on the directed causal transmission matrix, deduce the propagation path of the anomaly from the anomaly source process to the subsequent processes step by step along the processing sequence. Output the anomaly source process information, the anomaly propagation path, and the anomaly trigger time, and generate an anomaly tracing report. Step S4-1: Define the set S of candidate anomaly sources in the time domain T ={P1,P2}, the set of candidate anomaly sources in the frequency domain S F ={P2,P5}; Step S4-2: Set of Temporal Candidate Anomaly Sources S T and the frequency domain candidate anomaly source set S F If none of the values ​​are empty, continue with the next steps. Step S4-3: Set the enhancement coefficient α = 1.5; according to the dual-evidence co-existing anomaly scoring formula: Calculate the dual-evidence co-existence anomaly score for each process: The dual-evidence co-existence anomaly score for process P1 is: γ1 = 1.063 × 0.072 = 0.077; The dual-evidence co-existence anomaly score for process P2 is: γ2 = 1.671 × 0.116 × (1 + 1.5) = 0.485; The dual-evidence co-existence anomaly score for process P3 is: γ3 = 0.098 × 0.045 = 0.004; The dual-evidence co-existence anomaly score for process P4 is: γ4 = -0.814 × 0.032 = -0.026; The dual-evidence co-existence anomaly score for process P5 is: γ5 = -0.817 × 0.085 = -0.069; The dual-evidence co-existence anomaly score for process P6 is: γ6 = -0.651 × 0.062 = -0.04; The dual-evidence co-existence anomaly score for process P7 is: γ7 = -0.551 × 0.02 = -0.011; γ²=0.485 is the highest score, therefore the abnormal source process P origin =P2; Step S4-4: Set the propagation determination threshold A th =0.3, starting from the anomaly source P2 process, the anomaly propagation path is deduced step by step along the processing sequence, and the anomaly propagation coefficient in the directed causal propagation matrix A is used for determination: A 2,3 =0.15<0.3, the anomaly does not propagate to process P3; A 2,4 =0.62>0.3, the error is passed to process P4; A 2,5 =0.10<0.3, the anomaly does not propagate to process P5; A 2,6 =0.06<0.3, the anomaly does not propagate to process P6; A 2,7 =0.03<0.3, the anomaly does not propagate to process P7; Continue to check whether the P4 process continues to propagate forward: A 4,5 =0.85>0.3, the anomaly continues to propagate from process P4 to process P5; A 4,6 =0.22<0.3, the anomaly does not propagate to process P6; A 4,7 =0.11<0.3, the anomaly does not propagate to process P7; Continue to check whether the P5 process continues to propagate forward: A 5,6 =0.88>0.3, the anomaly continues to propagate from process P5 to process P6; A 5,7 =0.35>0.3, the anomaly continues to propagate from process P5 to process P7; Continue to check whether the P6 process continues to propagate forward: A 6,7 =0.79>0.3, the anomaly continues to propagate from process P6 to process P7; The final anomaly propagation path is: from process P2 to process P4 to process P5 to process P6 and finally to process P7; Steps S4-5: The source process of the anomaly is process P2 (attaching the ceramic block to the shielding cover), and the anomaly trigger time is 0.001s. The affected processes include process P4 (soldering the ceramic pins), process P5 (sealant sealing the shielding cover), process P6 (electrical testing), and process P7 (appearance inspection and packaging). The anomaly propagation path is: from process P2 to process P4 to process P5 to process P6 and finally to process P7.

[0028] The system includes: a signal acquisition and preprocessing module, a time-domain feature extraction and time-domain candidate determination module, a frequency-domain feature extraction and frequency-domain candidate determination module, and a time-frequency fusion and source tracing output module; The signal acquisition and preprocessing module is used to acquire and preprocess the sensor monitoring signals of multiple processes in the entire antenna manufacturing process, and to obtain the standard sensor monitoring signal sequence of each process through time synchronization acquisition and preprocessing. The time-domain feature extraction and time-domain candidate determination module is used to calculate time-domain features and determine time-domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process, and at the same time record the anomaly trigger time of the time-domain candidate anomaly source; The frequency domain feature extraction and frequency domain candidate determination module is used to calculate frequency domain features and determine frequency domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process. The time-frequency fusion and source tracing output module is used to calculate the dual-evidence collaborative anomaly score based on the time-domain candidate anomaly source and the frequency-domain candidate anomaly source, determine the anomaly source process, deduce the anomaly propagation path, and generate an anomaly source tracing report.

[0029] The signal acquisition and preprocessing module includes: The process set definition unit is used to define the process set, perform time synchronization processing on the sensing and monitoring signals of each process, and define the set of discrete sampling times; The sensing and monitoring signal vector combination unit is used to combine vibration signals, current signals, force signals and acoustic emission signals of the same process at the same time into a sensing and monitoring signal vector; The filtering, denoising, and normalization processing unit is used to filter, denoise, and normalize the amplitude of the sensing and monitoring signal vectors to obtain the standard sensing and monitoring signal sequences for each process.

[0030] The temporal feature extraction and temporal candidate determination module includes: The integrated signal anomaly calculation unit is used to calculate the integrated signal anomaly of each process based on the standard sensor monitoring signal sequence of each process. The time-series correlation strength calculation unit is used to perform cross-correlation calculations on the standard sensor monitoring signal sequences of any two processes to obtain the time-series correlation strength value and the number of hysteresis sampling points corresponding to the peak value. A directed causal transitivity matrix construction unit is used to construct a directed causal transitivity matrix based on the processing order and temporal correlation strength value; The anomaly contribution calculation unit is used to calculate the anomaly contribution of each process after deducting the effects of the preceding transmission based on the comprehensive signal anomaly degree and the directed causal transfer matrix. The time-domain candidate anomaly source marking unit is used to mark time-domain candidate anomaly sources according to the deviation multiple of the anomaly contribution of each process; The abnormal triggering time recording unit is used to infer and record the abnormal triggering time of the time-domain candidate abnormal source process based on the number of lag sampling points and the sampling frequency.

[0031] The frequency domain feature extraction and frequency domain candidate determination module includes: The characteristic frequency band setting unit is used to set the characteristic frequency band range corresponding to each type of standard sensor monitoring signal according to the processing type of each process and the physical characteristics of each type of sensor monitoring signal; The Discrete Fourier Transform (DFT) unit is used to perform DFT on the standard sensing and monitoring signal sequences of each process to obtain the frequency domain standard sensing and monitoring signal vector. The frequency domain energy concentration calculation unit is used to extract the amplitude spectrum of various types of standard sensor monitoring signals in the characteristic frequency band and calculate the frequency domain energy concentration. The energy percentage calculation unit is used to calculate the total energy of various types of standard sensor monitoring signals in the entire frequency band and the energy percentage in the characteristic frequency band. The frequency domain anomaly index calculation unit is used to calculate the frequency domain anomaly index of various types of standard sensor monitoring signals based on the expected energy ratio range and the actual energy ratio of each process. The comprehensive frequency domain anomaly index calculation unit is used to perform weighted fusion of the frequency domain anomaly indices of the four types of standard sensor monitoring signals in each process to obtain the comprehensive frequency domain anomaly index. The frequency domain candidate anomaly source marking unit is used to mark processes whose comprehensive frequency domain anomaly index exceeds the threshold as frequency domain candidate anomaly sources according to a preset threshold.

[0032] The time-frequency fusion and source tracing output module includes: The candidate anomaly source set definition unit is used to define the time-domain candidate anomaly source set and the frequency-domain candidate anomaly source set; The no-anomaly determination unit is used to determine whether there are candidate anomaly sources. If all are empty, the no-anomaly result is output. The dual-evidence collaborative anomaly scoring calculation unit is used to calculate the dual-evidence collaborative anomaly score for all processes, enhance the score of processes that appear in both candidate sets, and take the process with the highest score as the source of the anomaly. The anomaly propagation path deduction unit is used to deduce the anomaly propagation path step by step from the anomaly source process along the processing sequence based on the directed causal transmission matrix. The source tracing result output unit is used to output information on the source process of the anomaly, the propagation path of the anomaly, and the trigger time of the anomaly, and generate an anomaly source tracing report.

[0033] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for tracing antenna manufacturing anomalies based on multi-source data analysis, characterized in that: Includes the following steps: The sensor monitoring signals of each process in the entire antenna manufacturing process are acquired, and the sensor monitoring signals are time-synchronized and preprocessed to obtain the standard sensor monitoring signal sequence of each process. Based on the processing sequence of each process, the signal anomaly degree of each process and the temporal correlation strength between any two processes are calculated using the standard sensor monitoring signal sequence of each process. A directed causal transfer matrix is ​​constructed based on the processing sequence and the temporal correlation strength. The anomaly contribution part transmitted to the subsequent process is deducted from the signal anomaly degree of the preceding process using the directed causal transfer matrix to obtain the anomaly contribution degree of each process. Candidate time-domain anomalous processes are determined based on the distribution of the anomaly contribution degree of each process, and the anomaly trigger time of the candidate time-domain anomalous processes is recorded. Based on the processing type of each process and the physical characteristics of each type of sensor monitoring signal, the characteristic frequency bands corresponding to each type of sensor monitoring signal in each process are set. By using the frequency domain characteristics of the standard sensor monitoring signal sequence of each process in the characteristic frequency band, candidate frequency domain abnormal processes are determined. Based on the candidate time-domain anomalous processes and the candidate frequency-domain anomalous processes, the processes that are simultaneously determined to be anomalous in both the time and frequency domains are score-enhanced. The comprehensive anomalous score of each process is calculated. The source process of the anomalous process is determined based on the comprehensive anomalous score. The anomalous propagation path is deduced along the processing sequence based on the directed causal transit matrix. The source process information, the anomalous propagation path, and the anomalous trigger time are output, and an anomalous source tracing report is generated.

2. The method for tracing antenna manufacturing anomalies based on multi-source data analysis according to claim 1, characterized in that: The standard sensor monitoring signal sequences obtained for each process include: The process set consists of all the processes in the entire antenna manufacturing process, where each element in the process set represents a process in the entire antenna manufacturing process. The sensor monitoring signals collected from each process are processed for time synchronization so that the sampled values ​​of all sensor monitoring signals have the same sampling start time and sampling frequency. Record the set of discrete sampling times after time synchronization, where each element in the set of discrete sampling times represents a discrete sampling time. The vibration signal, current signal, force signal and acoustic emission signal of the same process at the same discrete sampling time are combined into the sensing and monitoring signal vector of the process at the discrete sampling time. The sensor monitoring signal vector is filtered and denoised to remove environmental noise and interference components from various types of sensor monitoring signals. The amplitude of the filtered sensor monitoring signal vector is then normalized to obtain a standard sensor monitoring signal vector. The standard sensor monitoring signal sequence for each process is formed by the standard sensor monitoring signal vectors at each discrete sampling time in chronological order of sampling time.

3. The antenna fabrication anomaly tracing method based on multi-source data analysis according to claim 1, characterized in that: The abnormal triggering time of the recording candidate time-domain abnormal process includes: Calculate the comprehensive signal anomaly of each process at all discrete sampling times based on the standard sensor monitoring signal sequence of each process. For any two processes, cross-correlation is performed based on the vector inner product between the standard sensor monitoring signal vectors of the two processes at each discrete sampling time to obtain the cross-correlation function between the two processes. The peak value of the cross-correlation function is extracted as the temporal correlation strength value between the two processes, and the number of hysteresis sampling points corresponding to the peak value is recorded. Construct a directed causal transfer matrix between processes based on the processing sequence of the processes in the entire antenna manufacturing process and the time-series correlation strength value; Based on the comprehensive signal anomaly degree of each process and the directed causal transmission matrix, calculate the anomaly contribution degree of each process after deducting the anomaly contribution transmitted from the preceding process to the process. The mean and standard deviation of the abnormal contribution of each process are calculated, and the deviation multiple of the abnormal contribution of each process is calculated based on the mean and standard deviation. Processes whose deviation multiple of abnormal contribution exceeds the preset abnormal contribution deviation threshold are marked as candidate abnormal sources in the time domain. For a process marked as a candidate time-domain anomaly source, the anomaly propagation time is calculated using the number of lag sampling points and the sampling frequency based on the signal timing relationship between the process and the adjacent preceding process. The anomaly trigger time of each candidate time-domain anomaly source process is then deduced and recorded based on the anomaly propagation time.

4. The antenna fabrication anomaly tracing method based on multi-source data analysis according to claim 1, characterized in that: The process of determining candidate frequency domain anomalies includes: Based on the processing type of each step in the antenna manufacturing process and the physical characteristics of each type of sensing and monitoring signal, corresponding characteristic frequency band ranges are set for each type of standard sensing and monitoring signal in each step. The discrete Fourier transform is performed on the standard sensing and monitoring signal sequence of each process to obtain the frequency domain standard sensing and monitoring signal vector of each process in the frequency domain, and the frequency domain standard sensing and monitoring signal vector is expressed in amplitude-phase form; Extract the amplitude spectrum of each type of standard sensor monitoring signal in the characteristic frequency band of each process, and calculate the frequency domain energy concentration of each type of standard sensor monitoring signal in the characteristic frequency band based on the amplitude spectrum; Calculate the total energy of each type of standard sensor monitoring signal in the full frequency band for each process, and calculate the energy ratio of each type of standard sensor monitoring signal in the characteristic frequency band based on the frequency domain energy concentration and the total energy. Based on the processing type of each process, set the expected energy ratio range corresponding to each type of standard sensor monitoring signal, compare the energy ratio of each type of standard sensor monitoring signal with the corresponding expected energy ratio range, and calculate the frequency domain anomaly index of each type of standard sensor monitoring signal in the characteristic frequency band. The frequency domain anomaly indices of the four types of standard sensor monitoring signals in each process are weighted and fused to obtain the comprehensive frequency domain anomaly index of each process; according to the preset frequency domain anomaly index threshold, the process whose comprehensive frequency domain anomaly index exceeds the frequency domain anomaly index threshold is marked as a frequency domain candidate anomaly source.

5. The antenna fabrication anomaly tracing method based on multi-source data analysis according to claim 1, characterized in that: The generated anomaly tracing report includes: The processes that will be marked as time-domain candidate anomalies will form a time-domain candidate anomaly source set, and the processes that will be marked as frequency-domain candidate anomalies will form a frequency-domain candidate anomaly source set. If both the time-domain candidate anomaly source set and the frequency-domain candidate anomaly source set are empty, it is determined that there is no anomaly in the current processing and the result of no processing anomaly is output. For each process, calculate the dual-evidence collaborative anomaly score for the process, enhance the score of processes that appear simultaneously in the time domain candidate anomaly source set and the frequency domain candidate anomaly source set, and take the process with the highest dual-evidence collaborative anomaly score as the anomaly source process. According to the directed causal transmission matrix, starting from the abnormal source process, each subsequent process is traversed along the processing sequence. For each subsequent process of the abnormal source process, if the abnormal transmission coefficient of the abnormal source process to the subsequent process is greater than the preset propagation judgment threshold, the subsequent process is marked as an abnormally affected process and included in the abnormal propagation path. Output the process number of the process originating the anomaly and the corresponding anomaly trigger time, and output the process information affected by the anomaly in the anomaly propagation path, generating an anomaly tracing report that includes the anomaly originating process information, the anomaly propagation path and the anomaly trigger time.

6. An antenna manufacturing anomaly tracing system based on multi-source data analysis, used to execute the antenna manufacturing anomaly tracing method based on multi-source data analysis as described in any one of claims 1-5, characterized in that: The system includes: a signal acquisition and preprocessing module, a time-domain feature extraction and time-domain candidate determination module, a frequency-domain feature extraction and frequency-domain candidate determination module, and a time-frequency fusion and source tracing output module; The signal acquisition and preprocessing module is used to acquire and preprocess the sensor monitoring signals of multiple processes in the entire antenna manufacturing process, and to obtain the standard sensor monitoring signal sequence of each process through time synchronization acquisition and preprocessing. The time-domain feature extraction and time-domain candidate determination module is used to calculate time-domain features and determine time-domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process, and at the same time record the anomaly trigger time of the time-domain candidate anomaly source; The frequency domain feature extraction and frequency domain candidate determination module is used to calculate frequency domain features and determine frequency domain candidate anomaly sources based on the standard sensor monitoring signal sequence of each process. The time-frequency fusion and source tracing output module is used to calculate the dual-evidence collaborative anomaly score based on the time-domain candidate anomaly source and the frequency-domain candidate anomaly source, determine the anomaly source process, deduce the anomaly propagation path, and generate an anomaly source tracing report.

7. The antenna fabrication anomaly tracing system based on multi-source data analysis according to claim 6, characterized in that: The signal acquisition and preprocessing module includes: The process set definition unit is used to define the process set, perform time synchronization processing on the sensing and monitoring signals of each process, and define the set of discrete sampling times; The sensing and monitoring signal vector combination unit is used to combine vibration signals, current signals, force signals and acoustic emission signals of the same process at the same time into a sensing and monitoring signal vector; The filtering, denoising, and normalization processing unit is used to filter, denoise, and normalize the amplitude of the sensing and monitoring signal vectors to obtain the standard sensing and monitoring signal sequences for each process.

8. The antenna fabrication anomaly tracing system based on multi-source data analysis according to claim 6, characterized in that: The temporal feature extraction and temporal candidate determination module includes: The integrated signal anomaly calculation unit is used to calculate the integrated signal anomaly of each process based on the standard sensor monitoring signal sequence of each process. The time-series correlation strength calculation unit is used to perform cross-correlation calculations on the standard sensor monitoring signal sequences of any two processes to obtain the time-series correlation strength value and the number of hysteresis sampling points corresponding to the peak value. A directed causal transitivity matrix construction unit is used to construct a directed causal transitivity matrix based on the processing order and temporal correlation strength value; The anomaly contribution calculation unit is used to calculate the anomaly contribution of each process after deducting the effects of the preceding transmission based on the comprehensive signal anomaly degree and the directed causal transfer matrix. The time-domain candidate anomaly source marking unit is used to mark time-domain candidate anomaly sources according to the deviation multiple of the anomaly contribution of each process; The abnormal triggering time recording unit is used to infer and record the abnormal triggering time of the time-domain candidate abnormal source process based on the number of lag sampling points and the sampling frequency.

9. The antenna fabrication anomaly tracing system based on multi-source data analysis according to claim 6, characterized in that: The frequency domain feature extraction and frequency domain candidate determination module includes: The characteristic frequency band setting unit is used to set the characteristic frequency band range corresponding to each type of standard sensor monitoring signal according to the processing type of each process and the physical characteristics of each type of sensor monitoring signal; The Discrete Fourier Transform (DFT) unit is used to perform DFT on the standard sensing and monitoring signal sequences of each process to obtain the frequency domain standard sensing and monitoring signal vector. The frequency domain energy concentration calculation unit is used to extract the amplitude spectrum of various types of standard sensor monitoring signals in the characteristic frequency band and calculate the frequency domain energy concentration. The energy percentage calculation unit is used to calculate the total energy of various types of standard sensor monitoring signals in the entire frequency band and the energy percentage in the characteristic frequency band. The frequency domain anomaly index calculation unit is used to calculate the frequency domain anomaly index of various types of standard sensor monitoring signals based on the expected energy ratio range and the actual energy ratio of each process. The comprehensive frequency domain anomaly index calculation unit is used to perform weighted fusion of the frequency domain anomaly indices of the four types of standard sensor monitoring signals in each process to obtain the comprehensive frequency domain anomaly index. The frequency domain candidate anomaly source marking unit is used to mark processes whose comprehensive frequency domain anomaly index exceeds the threshold as frequency domain candidate anomaly sources according to a preset threshold.

10. The antenna fabrication anomaly tracing system based on multi-source data analysis according to claim 6, characterized in that: The time-frequency fusion and source tracing output module includes: The candidate anomaly source set definition unit is used to define the time-domain candidate anomaly source set and the frequency-domain candidate anomaly source set; The no-anomaly determination unit is used to determine whether there are candidate anomaly sources. If all are empty, the no-anomaly result is output. The dual-evidence collaborative anomaly scoring calculation unit is used to calculate the dual-evidence collaborative anomaly score for all processes, enhance the score of processes that appear in both candidate sets, and take the process with the highest score as the source of the anomaly. The anomaly propagation path deduction unit is used to deduce the anomaly propagation path step by step from the anomaly source process along the processing sequence based on the directed causal transmission matrix. The source tracing result output unit is used to output information on the source process of the anomaly, the propagation path of the anomaly, and the trigger time of the anomaly, and generate an anomaly source tracing report.

Images