CN-121542897-B - Machine sensor grading method and machine control method based on sensor data

CN121542897BCN 121542897 BCN121542897 BCN 121542897BCN-121542897-B

Abstract

The invention relates to the technical field of production equipment management and control, in particular to a machine sensor grading method and a machine management and control method based on sensor data, wherein the machine sensor grading method comprises the steps of constructing historical data Trace of each sensor in historical FDC data according to the data of the sensor; constructing an abnormality judgment rule of the data Trace, judging whether the real-time data Trace of each sensor is abnormal or not based on the constructed abnormality judgment rule, obtaining a first judgment result, archiving the first judgment result, judging the correlation between the first judgment result of each sensor and corresponding product quality data, and grading each sensor based on the correlation to obtain the corresponding level of each sensor. According to the invention, each sensor is classified according to the sensor original data in the FDC data, key information in the original data is not lost, the classification result is more accurate, and the machine management and control are convenient according to the classification result.

Inventors

WANG SIMIN
SUN PANPAN
Tan Liujuan
XU DONGDONG

Assignees

合肥晶合集成电路股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (9)

1. The machine sensor grading method is characterized by comprising the following steps of: acquiring historical FDC data of a machine; constructing historical data Trace of each sensor in the historical FDC data according to the data of each sensor; constructing an abnormality judgment rule of the data Trace; acquiring real-time acquisition data of each sensor of the machine, and constructing real-time data Trace of each sensor based on the real-time acquisition data; judging whether real-time data Trace of each sensor is abnormal or not based on the constructed abnormality judgment rule, obtaining a first judgment result of each sensor, and archiving; When the first archived judging results reach a preset number, judging the correlation between the first judging results of each sensor and corresponding product quality data, and grading each sensor based on the correlation to obtain the corresponding level of each sensor; the data Trace includes a non-vibration curve, and the abnormality determination rule includes a first abnormality determination rule corresponding to the non-vibration curve, the first abnormality determination rule configured to: And respectively calculating the similarity between the non-vibration curve to be judged and each class in the corresponding data set in the pre-constructed first baseline library, if the similarity between the non-vibration curve to be judged and each class is smaller than the corresponding similarity threshold value, marking that the data of the non-vibration curve to be judged is abnormal, otherwise, marking that the data of the non-vibration curve to be judged is normal.
2. The machine sensor classification method according to claim 1, wherein the first baseline library construction method comprises the steps of: aligning non-vibration curves formed by data acquired by the same sensor at different times in historical data Trace according to the processing steps; Clustering non-vibration curves in historical data Trace in each data set based on a clustering algorithm by taking data of the same Sensor under the same Recipe of the same Tool as one data set, so as to obtain one or more classes in each data set; the similarity threshold is calculated by the following method: and calculating the similarity of the non-vibration curves in each class in each data set of the first baseline library, and taking the average value of the similarity between every two non-vibration curves in each class as the similarity threshold value of the class.
3. The machine sensor grading method according to claim 1, wherein the data Trace includes a vibration curve, the anomaly determination rule includes a second anomaly determination rule corresponding to the vibration curve, the second anomaly determination rule configured to: Calculating the frequency of a vibration curve to be judged, and calculating whether the frequency of the vibration curve is normal relative to the vibration curve in a second baseline library by adopting an abnormality detection algorithm, wherein the second baseline library is formed by the vibration curve in the history data Trace of each sensor; Introducing a sliding time window, calculating local amplitude in each time window of the vibration curve to be judged, and calculating whether the local amplitude in each time window of the vibration curve is normal relative to the vibration curve in the second baseline library by adopting an anomaly detection algorithm; And when the frequency of the vibration curve to be judged and the local amplitude in each time window are normal, marking that the data of the vibration curve to be judged is normal, otherwise, marking that the data of the vibration curve to be judged is abnormal.
4. The method for classifying a machine sensor according to claim 1, wherein in the step of determining a correlation between the first determination result of each sensor and the product quality data, each sensor is classified based on the correlation, and the corresponding level of each sensor is obtained, the determination of the correlation and the classification of the sensor are performed using a random forest model.
5. The method of grading a machine sensor according to claim 1, wherein in the step of grading each sensor based on the correlation, the sensor whose first determination result is strongly correlated with the product quality is classified as a strong risk sensor, the sensor whose first determination result is weakly correlated with the product quality is classified as a weak risk sensor, and the sensor whose first determination result is not correlated with the product quality is classified as a no risk sensor.
6. The machine station control method based on the sensor data is characterized by comprising the following steps of: grading each sensor of the machine by using the machine sensor grading method according to any one of claims 1 to 5 to obtain the corresponding grade of each sensor; acquiring real-time acquisition data of each sensor of the machine, and constructing real-time data Trace of each sensor based on the real-time acquisition data; judging whether real-time data Trace of each sensor is abnormal or not based on the constructed abnormality judgment rule to obtain a second judgment result; and controlling the machine based on the corresponding level of each sensor and the second judging result.
7. The method for machine control based on sensor data as claimed in claim 6, wherein the step of controlling the machine based on the corresponding level of each sensor and the second determination result comprises: When the corresponding level of the sensor is a weak risk sensor and the second judgment result is continuous M times of abnormality, the abnormal control is carried out on the machine, wherein M is larger than N, and both N and M are positive integers; and when the corresponding level of the sensor is a risk-free sensor, periodically checking and evaluating a second judging result of the risk-free sensor, judging whether the working state of the machine is abnormal according to the periodically evaluating result, and implementing corresponding management and control measures.
8. The method for machine control based on sensor data according to claim 7, wherein N has a value of 1.
9. The machine control method based on sensor data as claimed in claim 6, wherein when the second determination result is that the data Trace is abnormal, further performing the following steps: Judging whether preventive maintenance is carried out before occurrence of data Trace abnormality, if so, taking the moment point as a cut-off point, and acquiring FDC data of related sensors before and after the previous multiple preventive maintenance time nodes; Calculating the fluctuation amplitude of FDC data of the relevant sensor before and after each preventive maintenance time node according to the FDC data, and calculating the average fluctuation amplitude of the FDC data of the relevant sensor before and after single preventive maintenance; and adjusting an abnormality judgment rule of the data Trace of the related sensor according to the average fluctuation amplitude.

Description

Machine sensor grading method and machine control method based on sensor data Technical Field The invention relates to the technical field of production equipment management and control, in particular to a machine sensor grading method and a machine management and control method based on sensor data. Background The machine sensor data is used as direct data of the monitoring machine, the current running state (namely the working state) of the machine can be reflected in real time, and the quality of products can be directly affected when the state of the machine is abnormal, so that the machine sensor data is very important to analysis and excavation. With the development of big data technology and advanced intelligent technology, trillion levels of sensor data storage and analysis are made possible. Therefore, each semiconductor factory is not limited to analysis of the product yield, but hopes to reach the purpose of early warning through researching the mass data of the bottom machine. However, the data volume and data dimension of the machine sensor are huge, and the management by people is time and labor consuming, namely, as the complexity of the semiconductor production process is increased year by year and the data which can be collected by the sensor is more and more huge, engineering personnel are required to spend a large amount of manual analysis work on the work of anomaly monitoring and anomaly removal, and due to huge analysis dimension (which can be more than hundreds of thousands of parameters) and parameter characteristic difference, the comprehensive and effective management and control of FDC data cannot be achieved manually, and the abnormal state of the machine cannot be found in time often. In addition, the current engineer establishes an FDC statistical model and sets corresponding control rules by selecting part of key parameters through process experience. On the one hand, part of parameters cannot fully monitor the state of the machine, and part of abnormal states may not be detected due to neglecting other parameters. On the other hand, when partial information is lost by adopting FDC statistics/characteristic data, the rule is not properly controlled due to insufficient artificial experience, and false alarm is frequently caused. Therefore, it is necessary to provide a machine management and control method capable of comprehensively monitoring and accurately early warning. Disclosure of Invention The invention aims to provide a machine sensor grading method and a machine control method based on sensor data, wherein the machine control effect can be greatly improved by accurately grading the sensor and taking the abnormal detection result of the sensor and the grading result thereof as detection basis. In order to solve the technical problems, the technical scheme adopted by the invention is that the machine sensor grading method comprises the following steps: acquiring historical FDC data of a machine; constructing historical data Trace of each sensor in the historical FDC data according to the data of each sensor; constructing an abnormality judgment rule of the data Trace; acquiring real-time acquisition data of each sensor of the machine, and constructing real-time data Trace of each sensor based on the real-time acquisition data; judging whether real-time data Trace of each sensor is abnormal or not based on the constructed abnormality judgment rule, obtaining a first judgment result of each sensor, and archiving; And when the first determination results of the archiving reach the preset quantity, judging the correlation between the first determination results of the sensors and the corresponding product quality data, and grading the sensors based on the correlation to obtain the corresponding grade of each sensor. Further, the data Trace includes a non-vibration curve, and the abnormality determination rule includes a first abnormality determination rule corresponding to the non-vibration curve, the first abnormality determination rule being configured to: And respectively calculating the similarity between the non-vibration curve to be judged and each class in the corresponding data set in the pre-constructed first baseline library, if the similarity between the non-vibration curve to be judged and each class is smaller than the corresponding similarity threshold value, marking that the data of the non-vibration curve to be judged is abnormal, otherwise, marking that the data of the non-vibration curve to be judged is normal. Further, the construction method of the first baseline library comprises the following steps: aligning non-vibration curves formed by data acquired by the same sensor at different times in historical data Trace according to the processing steps; Clustering non-vibration curves in historical data Trace in each data set based on a clustering algorithm by taking data of the same Sensor under the same Recipe of the same Tool as one data set, so as to obta