CN115922553A - Method for online monitoring polishing processing state of silicon carbide wafer - Google Patents
Method for online monitoring polishing processing state of silicon carbide wafer Download PDFInfo
- Publication number
- CN115922553A CN115922553A CN202211480337.8A CN202211480337A CN115922553A CN 115922553 A CN115922553 A CN 115922553A CN 202211480337 A CN202211480337 A CN 202211480337A CN 115922553 A CN115922553 A CN 115922553A
- Authority
- CN
- China
- Prior art keywords
- silicon carbide
- carbide wafer
- processing state
- working condition
- polishing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Constituent Portions Of Griding Lathes, Driving, Sensing And Control (AREA)
Abstract
The invention discloses an online monitoring method for the polishing processing state of a silicon carbide wafer, which comprises the following steps: collecting periodic vibration signal data of the complete working condition of the grinding and polishing of the silicon carbide wafer, and constructing a characteristic data set for removing grinding and polishing materials of the silicon carbide wafer; establishing a machining test working condition library, and putting a vibration signal test value into the matched working condition library; carrying out variation modal decomposition on the vibration signal to obtain an IMF (intrinsic mode function) principal component energy spectrum; constructing an actual processing characteristic index; establishing a particle filter state estimation model; and constructing a new characteristic index through the particle points, and iteratively realizing the online monitoring of the polishing processing state of the silicon carbide wafer. According to the invention, through carrying out variational modal decomposition on the original vibration signal, noise can be effectively removed, and a main energy spectrum is obtained, so that the subsequent analysis is more accurate, and the anti-interference capability is enhanced.
Description
Technical Field
The invention belongs to the technical field of processing of high-hardness and brittle material silicon carbide wafers, in particular relates to a processing state monitoring technology, and specifically relates to a particle filter silicon carbide wafer polishing processing state online monitoring method based on vibration signals.
Background
Silicon carbide, as a highly hard and brittle material, is widely used in the field of semiconductor chips, and has increasingly high requirements for the surface engineering processing quality thereof. The most important link in the polishing process of the silicon carbide wafer is a monitoring link of the processing state except a material removal mechanism process, and the most important link restricts the realization of automatic control of the processing process and the like. Currently, off-line testing, in which the quality of a polished workpiece is dependent on downtime, severely results in a great waste of processing time and cost. Therefore, the on-line monitoring of the polishing processing state of the silicon carbide wafer has important engineering value.
At present, two methods, namely mechanism analysis and shallow machine learning analysis, mainly exist in the field of online monitoring. The mechanism analysis has the defects of difficult mathematical physical modeling, low monitoring precision and low operability. The shallow machine learning utilizes a data-driven method, and the learning analysis is carried out on the basis of the acquired physical signals, so that the precision is greatly improved, but the extraction and the processing of the characteristic values are more complicated and difficult, and a more complex monitoring model is difficult to process.
In summary, the monitoring of the polishing processing state of the silicon carbide wafer plays a crucial role in the surface quality, but the current monitoring method has the disadvantages of complex processing process, poor anti-interference capability and low precision, and needs to be solved urgently.
Disclosure of Invention
The invention provides a method for online monitoring the polishing processing state of a silicon carbide wafer based on particle filtering, aiming at overcoming the defects of complex processing process, poor anti-jamming capability and low precision of the existing method for online monitoring the processing state.
The method for monitoring the polishing processing state of the silicon carbide wafer on line comprises the following steps:
a: collecting periodic vibration signal data of the complete working condition of the silicon carbide wafer grinding and polishing, and constructing a silicon carbide wafer grinding and polishing material removal characteristic data set;
b: establishing a machining test working condition library, and putting a vibration signal test value into the matched working condition library;
c: carrying out variation modal decomposition on the vibration signal to obtain an IMF (intrinsic mode function) principal component energy spectrum;
d: constructing an actual processing characteristic index;
e: establishing a particle filter state estimation model;
f: and (4) constructing a new characteristic index through the particle points, and iteratively realizing the online monitoring of the polishing processing state of the silicon carbide wafer.
Further, step B further comprises the steps of:
b1, establishing a working condition library for each step of machining test according to the machining test outline and the operation working condition;
b2, comparing the vibration signal test values in real time, and when the matching changes, putting the current vibration signal into a re-matched working condition library.
Further, step C further comprises the steps of:
denoising by utilizing a wiener filter after C1 variational modal decomposition;
c2, setting a limited bandwidth parameter and initializing the central angular frequency to obtain each estimated central angular frequency;
and C3, obtaining an IMF main component energy spectrum according to different central angular frequencies, and taking the signal as the input of the next step.
Further, the IMF principal component energy spectrum is the respective eigenmode function.
And step D, forming a self-adaptive characteristic index through three steps of reference learning, deviation tolerance calculation and characteristic value construction, and performing normalization processing to obtain an actual characteristic index of the machining state.
Further, the step D of forming the adaptive characteristic index includes:
d1, in a reference learning stage, calculating IMF main component energy spectrum spectrums acquired and processed for the previous n times, and then taking the mean value of the signal spectrums of the n times as a reference value;
d2, in the offset tolerance calculation stage, calculating spectrograms of the signals acquired for 5 times continuously, and analyzing the offset tolerance between the mean frequency spectrum of the signals and the reference frequency spectrum for five times respectively;
and D3, respectively extracting the characteristics of the different working condition libraries through the steps D1 and D2 to obtain characteristic values of the time period, and respectively forming characteristic indexes.
Further, step E comprises the steps of:
e1, analyzing the characteristic indexes of the experimental data in the step D;
e2, selecting a double-exponential model as a degradation model of the model for fitting training, and determining a model parameter initial value of the observation equation according to a data fitting result;
e3, selecting the training data with good fitting effect in the step E2, and substituting the data into the model to initialize parameters.
Further, step F comprises the steps of:
f1, real-time monitoring data is used for updating model parameters and new special diagnosis indexes (machining state indexes) in real time;
and F2, updating and adjusting the distribution of the model parameters.
Further, step F3 is included to perform iterative computation through the new feature index value.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, through carrying out variational modal decomposition on the original vibration signal, noise can be effectively removed, and a main energy spectrum is obtained, so that the subsequent analysis is more accurate, and the anti-interference capability is enhanced.
The invention provides the adaptive characteristic index formed by three steps of reference learning, offset tolerance calculation and characteristic value construction, and adopts particle filtering to iteratively update parameters to realize online monitoring, thereby effectively solving the problem of poor adaptability of the traditional characteristic value and improving the precision.
Drawings
FIG. 1 is a flow chart of the online monitoring method for the polishing processing state of the silicon carbide wafer based on particle filtering according to the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
As shown in FIG. 1, an on-line monitoring method for the polishing process state of a silicon carbide wafer comprises the following steps:
firstly, constructing a characteristic data set for removing a silicon carbide wafer grinding and polishing material, and collecting the periodic vibration original signal data of the complete working condition of the silicon carbide wafer grinding and polishing.
And establishing all working condition libraries according to the operation working conditions according to the processing test outline, putting the vibration signal test value into the working condition library which is successfully matched, comparing in real time, and when the matching changes, putting the current vibration signal into the working condition library which is re-matched, so as to provide the vibration signals of all working conditions for subsequent feature extraction, monitoring and identification.
The method comprises the steps of carrying out variation modal decomposition on a vibration signal, utilizing wiener filtering to remove noise, setting limited bandwidth parameters and central angular frequency, initializing to obtain each estimated central angular frequency, obtaining each intrinsic mode function, namely IMF (intrinsic mode function) principal component energy spectrum according to different central angular frequencies, and taking the signal as the input of the next step.
In the reference learning stage, the IMF main component energy spectrum acquired and processed for the previous n times is calculated, and then the average value of the n times of signal spectra is used as a reference value.
In the offset tolerance calculation stage, spectrograms of 5 continuous acquired signals are calculated, and then offset tolerances between the mean spectrum of the signals and a reference spectrum are analyzed five times respectively.
And respectively extracting the characteristics of different working condition libraries through the two steps to obtain characteristic values of the time period, respectively forming characteristic indexes, and performing normalization processing to obtain the actual characteristic index of the processing state.
And (3) constructing a particle filter state estimation model, selecting a double-exponential model as a degradation model of the particle filter state estimation model for data fitting through characteristic index analysis of experimental data, and substituting training data into the model for parameter initialization.
And updating the model parameters and the new special diagnosis indexes (processing state indexes) in real time according to the real-time monitoring data, adjusting the distribution of the model parameters, and finally performing iterative computation through the new characteristic index values to realize the online monitoring of the polishing processing state of the silicon carbide wafer.
Specific examples are as follows:
firstly, acquiring periodic vibration signal data of a complete working condition of grinding and polishing of a silicon carbide wafer, and constructing a characteristic data set for removing grinding and polishing materials of the silicon carbide wafer;
and establishing all working condition libraries, which are equivalent to n file libraries, according to the operating working conditions, monitoring the operating working conditions during normal operation, comparing the stable processing state with the processing state in the working condition libraries, putting the vibration signal test value into the working condition library which is successfully matched, monitoring and comparing the processing state in real time, and putting the current vibration signal into the working condition library which is matched again after the matching is changed. Therefore, each steady-state working condition has a working condition database, and vibration signals of all working conditions are provided for subsequent feature extraction and life prediction.
Carrying out variation modal decomposition on the vibration signal f to obtain a principal component energy spectrum of an Intrinsic Mode Function (IMF), wherein a multi-component signal f to be decomposed consists of k IMF components u with limited bandwidth k Is formed, and the center frequency of each IMF is corresponding to omega k Intrinsic mode function u k (t) is
u k (t)＝A h (t)cos[Φ h (t)]
Wherein A is k (t) represents u k Instantaneous amplitude of (t), ω k ＝Φ k (t) represents u k (t) instantaneous frequency.
The method comprises the following steps of constructing an actual processing characteristic index: reference learning, offset tolerance calculation and characteristic value construction. In the reference learning stage, the frequency spectrum of the signals acquired n times before the speed reducer is firstly calculated, and then the average value of the frequency spectrums of the signals acquired n times is calculated as a reference value.
In the formula, X i For the spectrum of the signal acquired at the i-th time,
is a reference value.
In the offset tolerance calculation stage, spectrograms of 5 continuous acquired signals are calculated, and then offset tolerances between the 5 signal mean spectrums and the reference spectrums are respectively analyzed. The method comprises the following specific steps
······
T k ＝∑|s i |(i＝1，2，3，...，k)
Wherein S is k Is the shift tolerance value, T, of the k-th variation spectrum in the frequency spectrum k The characteristic value of the k time is formed by superposition and summation of the variation spectrums.
Respectively extracting the characteristics of the working condition library to obtain characteristic values of the time periods to respectively form machining state indexes, accumulating the characteristic values of the time periods and carrying out normalization processing to obtain the actual machining state index T of the tested piece, wherein the normalization maps the data into the range of [ i, j ] (the default is [0,1 ]) by transforming the accumulated characteristic value data
Wherein max is the maximum value in the characteristic values, and min is the minimum value in the characteristic values.
Establishing a particle filter state estimation model, wherein the dynamic change process of a particle filter description system can be represented by a state transfer equation and an observation equation
x k ＝f k (x k-1 ，v k-1 )
z k ＝h k (x k ，n k )
Wherein x is k Is the system state, z k As a system observation value, f k Is a state transfer function, h k As observation function, v k-1 Is process noise, n k To observe the noise.
Fitting the material removal characteristic based on silicon carbide processing by using a bi-exponential model as a degradation model and substituting observation data into the model for parameter initialization, wherein the data fitting method is a least square function method, and the bi-exponential observation equation is
z k ＝ae bt +ce dt
Wherein a and c are model degradation weights, b and d are degradation factors, t is a sampling point, and initial value model parameters of a, b, c and d are determined through data fitting.
And updating the model parameters and the new special diagnosis indexes (processing state indexes) in real time according to the real-time monitoring data, adjusting the distribution of the model parameters, and finally performing iterative computation through the new characteristic index values to realize the online monitoring of the polishing processing state of the silicon carbide wafer.
According to the invention, through carrying out variational modal decomposition on the original vibration signal, noise can be effectively removed, and a main energy spectrum is obtained, so that the subsequent analysis is more accurate, and the anti-interference capability is enhanced.
The invention provides the adaptive characteristic index formed by three steps of reference learning, offset tolerance calculation and characteristic value construction, and adopts particle filtering to iteratively update parameters to realize online monitoring, thereby effectively solving the problem of poor adaptability of the traditional characteristic value and improving the precision.
Claims (9)
1. An on-line monitoring method for the polishing processing state of a silicon carbide wafer is characterized by comprising the following steps:
a: collecting periodic vibration signal data of the complete working condition of the grinding and polishing of the silicon carbide wafer, and constructing a characteristic data set for removing grinding and polishing materials of the silicon carbide wafer;
b: establishing a machining test working condition library, and putting a vibration signal test value into the matched working condition library;
c: carrying out variation modal decomposition on the vibration signal to obtain an IMF (inertial measurement framework) main component energy spectrum;
d: constructing an actual processing characteristic index;
e: establishing a particle filter state estimation model;
f: and constructing a new characteristic index through the particle points, and iteratively realizing the online monitoring of the polishing processing state of the silicon carbide wafer.
2. The on-line monitoring method for the polishing processing state of the silicon carbide wafer according to claim 1, characterized in that the step B further comprises the steps of:
b1, establishing a working condition library for each step of machining test according to the machining test outline and the operation working condition;
b2, comparing the vibration signal test values in real time, and when the matching changes, putting the current vibration signal into a re-matched working condition library.
3. The on-line monitoring method for the polishing processing state of the silicon carbide wafer according to claim 1, characterized in that the step C further comprises the steps of:
denoising by utilizing a wiener filter after C1 variational modal decomposition;
c2, setting a limited bandwidth parameter and initializing the central angular frequency to obtain each estimated central angular frequency;
and C3, obtaining an IMF main component energy spectrum according to different central angular frequencies, and taking the signal as the input of the next step.
4. A method for on-line monitoring the polishing process state of a silicon carbide wafer according to claim 1 or 3, characterized in that the IMF principal component energy spectrum is each eigenmode function.
5. The on-line monitoring method for the polishing processing state of the silicon carbide wafer according to claim 1, wherein the step D is to form an adaptive characteristic index through three steps of reference learning, offset tolerance calculation and characteristic value construction, and to perform normalization processing to obtain an actual characteristic index of the processing state.
6. The on-line monitoring method for the polishing processing state of the silicon carbide wafer according to claim 5, wherein the forming step of the adaptive characteristic index in the step D is:
d1, in a reference learning stage, calculating IMF main component energy spectrum spectrums acquired and processed for the previous n times, and then taking the mean value of the signal spectrums of the n times as a reference value;
d2, in the offset tolerance calculation stage, calculating spectrograms of the signals acquired for 5 times continuously, and analyzing the offset tolerance between the mean frequency spectrum of the signals and the reference frequency spectrum for five times respectively;
and D3, respectively extracting the characteristics of the different working condition libraries through the steps D1 and D2 to obtain characteristic values of the time period, and respectively forming characteristic indexes.
7. The on-line monitoring method for the polishing processing state of the silicon carbide wafer according to claim 1, characterized in that the step E comprises the steps of:
e1, analyzing the characteristic indexes of the experimental data in the step D;
e2, selecting a double-exponential model as a degradation model of the model for fitting training, and determining a model parameter initial value of the observation equation according to a data fitting result;
e3, selecting the training data which are well fitted in the step E2, and substituting the data into the model to carry out parameter initialization.
8. The on-line monitoring method for the polishing processing state of the silicon carbide wafer as set forth in claim 7, wherein the step F comprises the steps of:
f1, real-time monitoring data updates model parameters and new special diagnosis indexes (processing state indexes) in real time;
and F2, updating and adjusting the distribution of the model parameters.
9. The method according to claim 8, further comprising the step of performing iterative calculation by using the new characteristic index value in step F3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211480337.8A CN115922553A (en) | 2022-11-24 | 2022-11-24 | Method for online monitoring polishing processing state of silicon carbide wafer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211480337.8A CN115922553A (en) | 2022-11-24 | 2022-11-24 | Method for online monitoring polishing processing state of silicon carbide wafer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115922553A true CN115922553A (en) | 2023-04-07 |
Family
ID=86655108
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211480337.8A Pending CN115922553A (en) | 2022-11-24 | 2022-11-24 | Method for online monitoring polishing processing state of silicon carbide wafer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115922553A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116842415A (en) * | 2023-09-01 | 2023-10-03 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
| CN117549205A (en) * | 2024-01-11 | 2024-02-13 | 东晶电子金华有限公司 | Quartz wafer polishing method |
-
2022
- 2022-11-24 CN CN202211480337.8A patent/CN115922553A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116842415A (en) * | 2023-09-01 | 2023-10-03 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
| CN116842415B (en) * | 2023-09-01 | 2023-12-26 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
| CN117549205A (en) * | 2024-01-11 | 2024-02-13 | 东晶电子金华有限公司 | Quartz wafer polishing method |
| CN117549205B (en) * | 2024-01-11 | 2024-04-02 | 东晶电子金华有限公司 | Quartz wafer polishing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115922553A (en) | Method for online monitoring polishing processing state of silicon carbide wafer | |
| CN110928237B (en) | Vibration signal-based numerical control machining center flutter online identification method | |
| CN113927371A (en) | Cutter wear prediction method based on multi-sensor feature fusion | |
| CN114619292B (en) | Milling cutter wear monitoring method based on fusion of wavelet denoising and attention mechanism with GRU network | |
| CN111504645B (en) | Rolling bearing fault diagnosis method based on frequency domain multipoint kurtosis | |
| CN111660141A (en) | Milling cutter wear state identification method based on spindle driving current and irrelevant to working conditions | |
| CN110682159A (en) | Cutter wear state identification method and device | |
| CN111832624A (en) | Tool remaining life prediction method based on anti-migration learning | |
| CN106970593A (en) | It is a kind of that the method that processing flutter suppresses online is realized by speed of mainshaft intelligent control | |
| CN112207631B (en) | Method for generating tool detection model, method, system, device and medium for detecting tool detection model | |
| Yang et al. | Application of bispectrum diagonal slice feature analysis in tool wear states monitoring | |
| CN116861313B (en) | Kalman filtering working condition identification method and system based on vibration energy trend | |
| CN114492527A (en) | Fuzzy neural network and principal component analysis based surface roughness online prediction method | |
| CN115409067A (en) | Method for predicting residual life of numerical control machine tool assembly | |
| CN115741235A (en) | Wear prediction and health management method based on five-axis machining center cutter | |
| CN114001654B (en) | Workpiece end face pose evaluation method | |
| CN111611943A (en) | Gear classification method based on sound analysis | |
| Li et al. | An improved multi-ridge extraction method based on differential synchro-squeezing wavelet transform | |
| CN114193229B (en) | Variable working condition cutter wear prediction method based on causal inference | |
| CN112903296B (en) | Rolling bearing fault detection method and system | |
| CN111160464B (en) | Industrial high-order dynamic process soft measurement method based on multi-hidden-layer weighted dynamic model | |
| CN106881630B (en) | High-speed milling flutter online recognition method based on adaptive-filtering Yu AR models | |
| CN115771061B (en) | Milling cutter abrasion monitoring method based on frequency attention mechanism | |
| CN117078118B (en) | Intelligent detection system for quality of workpiece produced by numerical control machine tool | |
| CN111459140B (en) | Fermentation process fault monitoring method based on HHT-DCNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |