CN-121980864-A - Intelligent sensor-based silicon carbide product defect detection method and system

CN121980864ACN 121980864 ACN121980864 ACN 121980864ACN-121980864-A

Abstract

The invention relates to the technical field of silicon carbide defect detection and discloses a silicon carbide product defect detection method and system based on an intelligent sensor, wherein the method comprises the steps of obtaining initial time sequence data; the method comprises the steps of performing spatial feature mapping according to data, extracting a preliminary boundary of a purity nonuniform region, obtaining a purity deviation feature vector of the preliminary boundary, performing multi-time domain evolution analysis based on the vector to obtain an optimized defect transition track sequence, performing time domain differential analysis on the sequence to extract a dynamic evolution path index, judging purity according to the dynamic evolution path index, generating an evolution influence quantization matrix, further fusing associated information to obtain a defect evolution influence factor, and generating a defect detection report based on the factor to obtain a final detection result. The method can realize dynamic evolution tracking of defects under a high-temperature working condition.

Inventors

XU ZHIPENG
LI MINGXIN

Assignees

潍坊新创新材料科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260125

Claims (8)

1. The method for detecting the defects of the silicon carbide product based on the intelligent sensor is characterized by comprising the following steps of: acquiring initial time sequence data; According to the initial time sequence data, performing spatial feature mapping processing to obtain a preliminary boundary of a purity non-uniform region, and obtaining a purity deviation feature vector of the preliminary boundary; performing multi-time domain sequence evolution analysis according to the purity deviation feature vector to obtain an optimized defect transition track sequence; performing time domain differential analysis according to the optimized defect transition track sequence to obtain a dynamic evolution path index; according to the dynamic evolution path index, purity judgment is carried out, and an evolution influence quantization matrix is obtained; Carrying out quantization association fusion according to the evolution influence quantization matrix to obtain a defect evolution influence factor; and generating a defect detection report according to the defect evolution influence factor to obtain a final defect detection report.
2. The intelligent sensor-based silicon carbide product defect detection method according to claim 1, wherein the performing spatial feature mapping processing according to the initial time sequence data to obtain a preliminary boundary of a non-uniform purity region and obtaining a purity deviation feature vector of the preliminary boundary comprises: performing space mapping fusion on the initial time sequence data to form a purity distribution image, and extracting a purity non-uniform region from the purity distribution image to obtain a primary boundary; Performing continuity calculation on the preliminary boundary to obtain a boundary continuity coefficient, and if the boundary continuity coefficient is lower than a preset continuity threshold value, adjusting the preliminary boundary and extracting a purity distribution gradient; Calculating a correlation coefficient between the purity distribution gradient and the initial time sequence data to obtain a correlation coefficient, and calculating a purity deviation amplitude according to the correlation coefficient; and normalizing the purity deviation amplitude to form a purity deviation feature vector.
3. The intelligent sensor-based silicon carbide product defect detection method according to claim 1, wherein the performing multi-time domain sequence evolution analysis according to the purity deviation feature vector to obtain an optimized defect transition track sequence comprises: inputting the purity deviation feature vector into a pre-trained RNN neural network, and outputting an evolution expression sequence; Determining a defect expansion starting point based on the evolution expression sequence, and carrying out space path tracking according to the defect expansion starting point to obtain a defect transition track; and eliminating track discrete points according to the defect transition track to obtain an optimized defect transition track sequence.
4. The intelligent sensor-based silicon carbide product defect detection method according to claim 1, wherein the performing time domain differential analysis according to the optimized defect transition track sequence to obtain a dynamic evolution path index comprises: According to the optimized defect transition track sequence, performing vibration signal difference value calculation to obtain a vibration spectrum change sequence, and extracting a spectrum characteristic sequence of the vibration spectrum change sequence; according to the spectrum characteristic sequence, threshold judgment and cluster extraction are carried out to obtain a crack initiation sequence; And carrying out path fusion calculation according to the crack initiation sequence to obtain a dynamic evolution path index.
5. The intelligent sensor-based silicon carbide product defect detection method according to claim 1, wherein the purity judgment is performed according to the dynamic evolution path index to obtain an evolution influence quantization matrix, comprising: If the dynamic evolution path index is larger than a preset evolution risk threshold, extracting heat treatment time sequence data corresponding to a high-temperature sintering period; According to the heat treatment time sequence data, gradient difference calculation is carried out to obtain a temperature gradient sequence; if a local mutation point higher than a preset temperature difference mutation threshold exists in the temperature gradient sequence, finite element calculation is carried out on the heat treatment time sequence data to obtain a thermal stress concentration subset; and carrying out correlation judgment and matrix generation according to the thermal stress concentration subset to generate an evolution influence quantization matrix.
6. The intelligent sensor-based silicon carbide product defect detection method according to claim 1, wherein the performing quantization association fusion according to the evolution influence quantization matrix to obtain a defect evolution influence factor comprises: extracting evolution characteristics according to the evolution influence quantization matrix to generate a defect early warning input tensor; According to the defect early warning input tensor, judging abnormal characteristics to obtain an early warning signal sequence; according to the early warning signal sequence, carrying out data correlation analysis to generate an optimized dynamic abnormal feature combination; And carrying out feature fusion calculation according to the dynamic abnormal feature combination to obtain a defect evolution influence factor.
7. The intelligent sensor-based silicon carbide product defect detection method of claim 1, wherein the performing defect detection report generation according to the defect evolution influence factor to obtain a final defect detection report comprises: when the defect evolution influence factor is larger than a preset defect threshold value, collecting spectrum signal data; According to the spectrum signal data, information feature vectors are obtained, normalization processing is carried out, and a signal change matrix is obtained; performing significance judgment on the signal change matrix to obtain a significant change feature set; generating a correlation coefficient according to the significant change characteristic set to obtain a quantized correlation coefficient set; and carrying out data mapping and fusion according to the quantized association coefficient set to obtain a final defect detection report.
8. A silicon carbide article defect detection system based on intelligent sensors, comprising: The data acquisition module is used for acquiring initial time sequence data; The feature vector module is used for carrying out space feature mapping processing according to the initial time sequence data to obtain a preliminary boundary of the non-uniform purity region and obtaining a purity deviation feature vector of the preliminary boundary; The track sequence module is used for carrying out multi-time domain sequence evolution analysis according to the purity deviation feature vector to obtain an optimized defect transition track sequence; the evolution path module is used for carrying out time domain differential analysis according to the optimized defect transition track sequence to obtain a dynamic evolution path index; The quantization matrix module is used for judging the purity according to the dynamic evolution path index to obtain an evolution influence quantization matrix; The influence factor module is used for carrying out quantization association fusion according to the evolution influence quantization matrix to obtain a defect evolution influence factor; and the detection report module is used for generating a defect detection report according to the defect evolution influence factor to obtain a final defect detection report.

Description

Intelligent sensor-based silicon carbide product defect detection method and system Technical Field The invention relates to the technical field of silicon carbide defect detection, in particular to a silicon carbide product defect detection method and system based on an intelligent sensor. Background The silicon carbide product is widely applied in the fields of aerospace, semiconductor manufacturing, chemical reactor and other high-requirement fields due to the excellent high-temperature wear resistance, corrosion resistance and thermal stability. The materials are used for constructing key structural parts operating in high-temperature environment, and have extremely high processing and quality control requirements, and particularly, the quality stability in a sintering stage has a decisive influence on the service life of finished products. The existing method generally adopts final state detection means, such as spectrum analysis, surface image recognition and the like, to carry out defect classification and purity evaluation on the sintered silicon carbide product. Although the method can be used for identifying surface defects and purity anomalies, the method is limited by fixed detection positions and missing time sequence information, and the dynamic change process of defects in the whole sintering process is difficult to reflect. Especially under high temperature, vacuum and airtight sintering environment, the traditional mounting type sensor cannot work stably for a long time, so that defect evolution behavior of products in the sintering process is difficult to capture and quantify effectively due to uneven temperature distribution, inconsistent material shrinkage or structural perturbation, a defect expansion mechanism and a risk level lack of real-time judgment basis, and the whole detection chain presents problems of response lag and invisible evolution. In conclusion, the prior art has the problem that the dynamic evolution of defects is difficult to monitor under the high-temperature working condition. Disclosure of Invention The invention provides a method and a system for detecting defects of a silicon carbide product based on an intelligent sensor, which are used for realizing dynamic evolution tracking of defects under a high-temperature working condition. In order to solve the technical problems, the present invention provides a method for detecting defects of a silicon carbide product based on an intelligent sensor, comprising: acquiring initial time sequence data; According to the initial time sequence data, performing spatial feature mapping processing to obtain a preliminary boundary of a purity non-uniform region, and obtaining a purity deviation feature vector of the preliminary boundary; performing multi-time domain sequence evolution analysis according to the purity deviation feature vector to obtain an optimized defect transition track sequence; performing time domain differential analysis according to the optimized defect transition track sequence to obtain a dynamic evolution path index; according to the dynamic evolution path index, purity judgment is carried out, and an evolution influence quantization matrix is obtained; Carrying out quantization association fusion according to the evolution influence quantization matrix to obtain a defect evolution influence factor; and generating a defect detection report according to the defect evolution influence factor to obtain a final defect detection report. Preferably, the performing spatial feature mapping according to the initial time sequence data to obtain a preliminary boundary of the non-uniform purity region, and obtaining a purity deviation feature vector of the preliminary boundary, includes: performing space mapping fusion on the initial time sequence data to form a purity distribution image, and extracting a purity non-uniform region from the purity distribution image to obtain a primary boundary; Performing continuity calculation on the preliminary boundary to obtain a boundary continuity coefficient, and if the boundary continuity coefficient is lower than a preset continuity threshold value, adjusting the preliminary boundary and extracting a purity distribution gradient; Calculating a correlation coefficient between the purity distribution gradient and the initial time sequence data to obtain a correlation coefficient, and calculating a purity deviation amplitude according to the correlation coefficient; and normalizing the purity deviation amplitude to form a purity deviation feature vector. Preferably, the performing multi-time domain sequence evolution analysis according to the purity deviation feature vector to obtain an optimized defect transition track sequence includes: inputting the purity deviation feature vector into a pre-trained RNN neural network, and outputting an evolution expression sequence; Determining a defect expansion starting point based on the evolution expression sequence, and carrying out s