A system and method for analyzing the reliability of polishing quality of a microfluidic channel of abrasive particle flow

Abstract

The application discloses a kind of abrasive grain flow microchannel polishing quality reliability analysis systems, including abrasive grain flow polishing system, quality detection system and reliability analysis system, the abrasive grain flow polishing system is used in the polishing processing of microchannel in microfluidic chip;The quality detection system is used to carry out quality detection to the microchannel after processing, and obtains detection value R1;The reliability analysis system carries out reliability analysis to processing quality, and the information basis of analysis is processing parameter P and the microchannel processing quality detection value R1 obtained by detection;The reliability analysis system transmits adjusted processing parameter to the abrasive grain flow polishing system, and is used to control microchannel polishing.The application also discloses a kind of abrasive grain flow microchannel polishing quality reliability analysis method.The application utilizes new learning function, and the correlation coefficient introduced by new learning function, improves the accuracy of model in threshold value vicinity, and can avoid the accuracy in low correlation area, ensures the accuracy of reliability analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to a machining method quality reliability analysis method, and particularly relates to a polishing quality reliability analysis system and method for abrasive grain flow micro-channel. BACKGROUND

[0002] Microfluidic chips made of glass substrates are widely used in biological detection, and micro-channels are the main places for fluid transport and detection analysis of microfluidic chips. The smoothness of the inner wall surface of the micro-channel and the machining quality directly determine the precision and stability of the biological detection of the microfluidic chip. Therefore, various polishing methods can be used to polish the micro-channel, reduce its roughness, and improve its surface smoothness. Furthermore, quality detection is performed after polishing machining of the micro-channel of the microfluidic chip, and it is of great significance to ensure the machining quality reliability.

[0003] For a microfluidic chip, the main micro-channel machining quality includes the micro-channel geometric structure size required for chip detection, such as micro-channel depth, width, channel angle, etc. In addition, it also includes surface size, such as micro-channel surface roughness, surface damage or sub-damage degree, etc. Surface roughness can be detected by a roughness measuring instrument, and surface damage or sub-damage can be detected and analyzed by a white light interferometer, a scanning electron microscope, etc.

[0004] Due to the small size of the micro-channel structure, abrasive grain jet polishing is a new method, but it involves many machining parameters. In actual machining, these machining parameters will cause machining parameter uncertainty due to machine vibration, environmental temperature and humidity changes, mechanical structure wear, etc., thereby causing fluctuations in the machining quality, and even the machining quality cannot meet the use requirements of the microfluidic chip. It is of great practical significance to control the machining quality reliability and ensure that the probability of machining failure is at a low level.

[0005] However, quality detection of each machined microfluidic chip will bring a great workload. In order to avoid a large amount of quality detection work, through the abrasive grain flow micro-channel polishing quality reliability analysis system, the relationship between the machining parameters and the machining quality is obtained through a few machining parameter machining tests, and the polishing quality reliability analysis under machining parameter uncertainty is efficiently performed, so as to predict the machining quality reliability and adjust the machining parameters in time. SUMMARY

[0006] In order to solve the above technical problems in the prior art, the present application provides an abrasive grain flow micro-channel polishing quality reliability analysis system and method, and the specific technical solutions are as follows:

[0007] A reliability analysis system for polishing microchannels using abrasive flow technology includes an abrasive flow polishing system, a quality inspection system, and a reliability analysis system. The abrasive flow polishing system is used for polishing microchannels within a microfluidic chip. The quality inspection system is used to inspect the quality of the processed microchannels and obtain a detection value R1. The reliability analysis system performs reliability analysis on the processing quality, based on the processing parameters P and the obtained microchannel processing quality detection value R1. The reliability analysis system transmits adjusted processing parameters to the abrasive flow polishing system to control the polishing of the microchannels.

[0008] The reliability analysis system reads the processing parameters P used in the abrasive polishing system and the microchannel processing quality detection value R1 from the quality inspection system; it uses the processing parameters P and the detection value R1 to construct a processing quality response surface surrogate model A based on the Kriging model, and uses the constructed surrogate model A to calculate the reliability or failure probability of the processing quality through the Monte Carlo simulation test method. The failure probability is as shown in formula (1):

[0009] (1)

[0010] In the formula: The probability of processing failure. The total number of experiments conducted using Monte Carlo methods. for The number of times the processing quality requirements were not met in this test. This indicates that the surface is too rough and the processing quality does not meet the requirements. This indicates that the processing quality meets the requirements. , The required surface roughness threshold;

[0011] The surrogate model A is constructed by updating the active learning function of the Kriging method, as shown in formula (2):

[0012] (2)

[0013] In the formula: To learn the function, The predicted value is based on the constructed proxy model A. The variance of the predicted values, This represents the maximum correlation coefficient, with values ​​between (0,1). The exponential coefficient can be any non-negative rational number. , Parameters used for processing;

[0014] The surrogate model A is constructed by updating the active learning function of the Kriging method, as shown in formula (3):

[0015] (3)

[0016] In the formula: To learn the function, The predicted value is based on the constructed proxy model A. This represents the variance of the predicted values.

[0017] Furthermore, the detection value R1 is obtained by detecting the surface roughness of the microchannel.

[0018] A method for reliability analysis using the aforementioned abrasive flow microchannel polishing quality reliability analysis system includes the following steps:

[0019] Step 1, Initialization: By sampling according to the probability distribution type of the processing parameter P, A population sample S1, , This is a vector representation of the processing parameters;

[0020] Step 2, Initial Training Sample Processing Experiment: In order to establish the initial Kriging surrogate model A, the initial processing experiment needs to... One training sample S2, , The vector representation of the processing parameters for the training sample S2 is used; the microchannel is polished using the processing parameters of the training sample S2, and the surface quality R1 after polishing is measured using a quality inspection system, denoted as... ;

[0021] Step 3, construct or update the Kriging model: construct or update the Kriging model based on the training sample S2 and the detection value R1 to obtain the surrogate model A;

[0022] Step 4, obtain the predicted value using surrogate model A: Substitute the population sample S1 into surrogate model A to obtain the predicted value. Predicted values ,variance and correlation coefficient ;calculate The quantity, to obtain According to formula (1), the predicted failure probability is obtained. ;

[0023] Step 5, add new training points in a semi-parallel manner: simultaneously calculate the two learning functions in formulas (2) and (3) for all population samples S1; make and The processing parameter point that reaches the minimum value is the potential training point;

[0024] Step 6, Stop learning criterion: If the learning function determines the next one or two best points, the corresponding learning function value is compared with the stopping criterion in formula (5);

[0025] (5)

[0026] If formula (5) is satisfied, then learning stops. According to the AK-MCS method theory, the accuracy probability of surrogate model A at this time is... At least greater than ,in The cumulative distribution function is the standard normal distribution.

[0027] Step 7: Update the training sample S2 with the new processing parameters: If the stopping criterion in Step 6 is not met, continue learning and use one or two newly added training points as processing parameters to conduct processing experiments on the abrasive flow polishing system, and use the quality detection system to measure the detection value R1; the learning process returns to Step 3, using the updated parameters... or Use a new training sample S2 and a detection value R1 to update the Kriging surrogate model A; if the stopping criterion in step 6 is met, proceed to step 8;

[0028] Step 8, calculate the coefficient of variation of the failure probability: if the stopping criterion in step 6 is met, then the learning ends; for the Monte Carlo population, Whether the size is suitable remains to be evaluated. The judgment of suitability is as shown in Equation 6:

[0029] (6)

[0030] In the formula, The coefficient of variation is 1. The failure probability is predicted by formula (1). The allowable value for the coefficient of variation is 0.05 or 0.02.

[0031] Step 9, update the total sample S1 in Monte Carlo: if the coefficient of variation If it is too large and does not satisfy formula (6) in step 8, then increase it. Then, generate a new overall sample S1 according to step 1, and then go to step 4 to predict the processing quality of the newly added processing parameters using the surrogate model A, and continue to use the active learning method until the stopping criterion formula (4) is met again; if formula (6) is met, then proceed to step 10.

[0032] Step 10, complete the reliability analysis: if Values ​​below the specified values Then the reliability analysis stops, and the probability of the last failure is estimated. This is the result of the analysis on the reliability of processing quality.

[0033] Furthermore, in step 2, the training sample S2 is sampled from the total sample S1 or sampled according to the probability distribution type of the processing parameter P or randomly sampled.

[0034] Furthermore, the predicted value mentioned in step 4 ,variance All correlation coefficients were calculated using the Kriging method. for Parameter points and training samples S2 The maximum correlation coefficient between points is shown in formula (4):

[0035] (4).

[0036] Beneficial effects

[0037] This invention utilizes a new learning function that not only considers the uncertainty probability (1-95.45%) of potential new training points, but also introduces a correlation coefficient rm(pi) to further consider the degree of correlation between potential new training points and existing training points, thereby enabling the surrogate model of abrasive flow polishing processing quality to have better global fidelity.

[0038] The correlation coefficient introduced by the new learning function causes the two learning functions to obtain different training points during the learning process. By judging whether they are the same point, new training points are added in a semi-parallel manner, which not only improves the accuracy of the model near the threshold, but also avoids the accuracy in the low correlation region, ensuring the accuracy of reliability analysis. In addition, it also reduces the number of iterations required to build a proxy model for abrasive flow polishing processing quality, improving the efficiency of processing quality reliability analysis. Attached Figure Description

[0039] Figure 1 This is a flowchart of the reliability analysis method for abrasive flow microchannel polishing quality according to the present invention. Detailed Implementation

[0040] The invention will now be further described with reference to the accompanying drawings.

[0041] The present invention discloses a reliability analysis system for abrasive flow microchannel polishing quality, comprising an abrasive flow polishing system, a quality inspection system, and a reliability analysis system. The abrasive flow polishing system is used for polishing microchannels within a microfluidic chip. The quality inspection system is used to inspect the quality of the processed microchannels, obtaining a detection value R1, where the parameter being detected is surface roughness. The reliability analysis system performs reliability analysis on the processing quality, based on the processing parameter P and the detected microchannel processing quality value R1. The reliability analysis system can transmit the processing parameters to the abrasive flow polishing system for controlling the microchannel polishing.

[0042] The reliability analysis system reads the processing parameters P used in the abrasive polishing system and the microchannel processing quality detection value R1 from the quality inspection system. Using the processing parameters P and the detection value R1, a processing quality response surface surrogate model A is constructed based on the Kriging model. The reliability or failure probability of the processing quality is then calculated using the Monte Carlo simulation method based on the constructed surrogate model A. The failure probability is shown in Formula 1.

[0043] (1)

[0044] In the formula: The probability of processing failure. The total number of experiments conducted using Monte Carlo methods. for The number of times the processing quality requirements were not met in this test. This indicates that the surface is too rough and the processing quality does not meet the requirements. This indicates that the processing quality meets the requirements. , This is the required surface roughness threshold.

[0045] The surrogate model A is constructed and updated using the active learning function of the Kriging method, as shown in Formula 2:

[0046] (2)

[0047] In the formula: To learn the function, The predicted value is based on the constructed proxy model A. The variance of the predicted values, This represents the maximum correlation coefficient, with values ​​between (0,1). The exponential coefficient can be any non-negative rational number. , These are the parameters used in the processing.

[0048] The surrogate model A is constructed and updated using the active learning function of the Kriging method. The other active learning function is shown in Equation 3:

[0049] (3)

[0050] In the formula: To learn the function, The predicted value is based on the constructed proxy model A. This represents the variance of the predicted values.

[0051] like Figure 1 As shown, the reliability analysis system completes the reliability analysis of processing quality through the following process:

[0052] Step 1, Initialization: By sampling according to the probability distribution type of the processing parameter P, A population sample S1, , This is a vector representation of the processing parameters;

[0053] Step 2, Initial Training Sample Processing Experiment: In order to establish the initial Kriging surrogate model A, the initial processing experiment needs to... One training sample S2, , This is a vector representation of the processing parameters for the training samples; the training sample S2 can be sampled from the overall sample S1, or it can be sampled according to the probability distribution type of the processing parameter P, or randomly sampled; then, the processing parameters of the training sample S2 are used to perform polishing experiments on the microchannel, and the surface quality detection value R1 after polishing is measured using a quality detection system, denoted as... ;

[0054] Step 3, construct or update the Kriging model: construct or update the Kriging model based on the training sample S2 and the detection value R1 to obtain the surrogate model A;

[0055] Step 4, obtain the predicted value using surrogate model A: Substitute the population sample S1 into surrogate model A to obtain the predicted value. Predicted values ,variance and correlation coefficient ;calculate The quantity, to obtain According to Formula 1, the predicted failure probability is obtained. ;

[0056] Among the predicted values ,variance All correlation coefficients were calculated using the Kriging method. for Parameter points and training samples S2 The maximum correlation coefficient between points is shown in Formula 4:

[0057] (4)

[0058] Step 5, add new training points in a semi-parallel manner: simultaneously calculate the two learning functions in Formula 2 and Formula 3 for all population samples S1; making and The processing parameter point that reaches its minimum value is a potential training point; in most cases, Approaching 1, the learning function is primarily based on prediction. and variance Since the minimum point of the two learning functions is usually the same point, only a new point will be added to the current training sample S2 to form a new training sample S2.

[0059] Conversely, it has low correlation with the current training sample S2. The point is based on the minimum point of Formula 2, which is very likely to be different from the minimum point of the U function based on Formula 3. In this case, the potential training points of the two learning functions will be added to the current training sample S2 in parallel to form a new training sample S2.

[0060] Since the number of points added in each learning and training session may be one point or two points, the learning is carried out in a semi-parallel manner.

[0061] Step 6, Stop Learning Criterion: Once the best next one or two points are determined using the learning function, the corresponding learning function value is compared with the stopping criterion in Formula 5;

[0062] (5)

[0063] If Formula 5 is satisfied, then learning stops. According to the AK-MCS methodology, the accuracy probability of surrogate model A at this point is... At least greater than ,in Let be the cumulative distribution function of the standard normal distribution. If we take... ,but They believe that proxy model A can more accurately reflect the actual processing quality.

[0064] Step 7: Update the training sample S2 with the new processing parameters: If the stopping criterion in Step 6 is not met, continue learning and use one or two newly added training points as processing parameters to conduct processing experiments on the abrasive flow polishing system, and use the quality detection system to measure the detection value R1; the learning process returns to Step 3, using the updated parameters... or Use a new training sample S2 and a detection value R1 to update the Kriging surrogate model A; if the stopping criterion in step 6 is met, proceed to step 8;

[0065] Step 8, calculate the coefficient of variation of the failure probability: If the stopping criterion in step 6 is met, then the learning ends and the Kriging surrogate model A is sufficiently accurate; however, for the Monte Carlo population, Whether the size is suitable remains to be evaluated. The judgment of suitability is as shown in Equation 6:

[0066] (6)

[0067] In the formula, The coefficient of variation is 1. The failure probability is predicted using Formula 1. This is the allowable value for the coefficient of variation, typically taken as 0.05 or 0.02; if The scale is too small, and the failure probability is generally considered unreliable; the coefficient of variation is usually lower than [a certain value]. It is acceptable, indicating The size is appropriate, and the failure probability will not be unreliable due to the value being too small;

[0068] Step 9, update the total sample S1 in Monte Carlo: if the coefficient of variation If it's too large and doesn't satisfy formula 6 in step 8, then increase it. Then, generate a new overall sample S1 according to step 1, and then go to step 4 to predict the processing quality of the newly added processing parameters using the surrogate model A, and continue to use the active learning method until the stopping criterion formula 4 is met again; if formula 6 is met, then proceed to step 10.

[0069] Step 10, complete the reliability analysis: if Values ​​below the specified values Then the reliability analysis stops, and the probability of the last failure is estimated. This is the result of the analysis on the reliability of processing quality.

Claims

1. A reliability analysis system for abrasive flow microchannel polishing quality, comprising an abrasive flow polishing system, a quality inspection system, and a reliability analysis system, characterized in that: The abrasive flow polishing system is used for polishing microchannels within a microfluidic chip; the quality inspection system is used to inspect the quality of the processed microchannels and obtain the inspection value R1; the reliability analysis system performs reliability analysis on the processing quality, based on the processing parameter P and the obtained microchannel processing quality inspection value R1; the reliability analysis system transmits the adjusted processing parameters to the abrasive flow polishing system to control the polishing of the microchannels. The reliability analysis system reads the processing parameters P used in the abrasive polishing system and the microchannel processing quality detection value R1 from the quality detection system. Using the processing parameter P and the detection value R1, a processing quality response surface surrogate model A is constructed based on the Kriging model. The reliability or failure probability of the processing quality is calculated using the Monte Carlo simulation test method based on the surrogate model A constructed above. The failure probability is as shown in formula (1): （1） In the formula: The probability of processing failure. The total number of experiments conducted using Monte Carlo methods. for The number of times the processing quality requirements were not met in this test. This indicates that the surface is too rough and the processing quality does not meet the requirements. This indicates that the processing quality meets the requirements. , The required surface roughness threshold; The surrogate model A is constructed by updating the active learning function of the Kriging method, as shown in formula (2): （2） In the formula: To learn the function, The predicted value is based on the constructed proxy model A. The variance of the predicted values, This represents the maximum correlation coefficient, with values ​​between (0,1). The exponential coefficient can be any non-negative rational number. , Parameters used for processing; The surrogate model A is constructed by updating the active learning function of the Kriging method, as shown in formula (3): （3） In the formula: To learn the function, The predicted value is based on the constructed proxy model A. This represents the variance of the predicted values.

2. The abrasive flow microchannel polishing quality reliability analysis system as described in claim 1, characterized in that: The measured value R1 is obtained by detecting the surface roughness of the microchannel.

3. A method for reliability analysis using the abrasive flow microchannel polishing quality reliability analysis system as described in claim 1, comprising the following steps: Step 1, Initialization: By sampling according to the probability distribution type of the processing parameter P, A population sample S1, , This is a vector representation of the processing parameters; Step 2, Initial Training Sample Processing Experiment: In order to establish the initial Kriging surrogate model A, the initial processing experiment needs to... One training sample S2, , The vector representation of the processing parameters for the training sample S2 is used; the microchannel is polished using the processing parameters of the training sample S2, and the surface quality R1 after polishing is measured using a quality inspection system, denoted as... ; Step 3, construct or update the Kriging model: construct or update the Kriging model based on the training sample S2 and the detection value R1 to obtain the surrogate model A; Step 4, obtain the predicted value using surrogate model A: Substitute the population sample S1 into surrogate model A to obtain the predicted value. Predicted values ,variance and correlation coefficient ;calculate The quantity, to obtain According to formula (1), the predicted failure probability is obtained. ; Step 5, add new training points in a semi-parallel manner: calculate the two learning functions in formula (2) and formula (3) simultaneously for all population samples S1; make and The processing parameter point that reaches the minimum value is the potential training point; Step 6, Stop learning criterion: If the learning function determines the next one or two best points, the corresponding learning function value is compared with the stopping criterion in formula (5); （5） If formula (5) is satisfied, then learning stops. According to the AK-MCS method theory, the accuracy probability of surrogate model A at this time is... At least greater than ,in The cumulative distribution function is the standard normal distribution. Step 7: Update the training sample S2 with the new processing parameters: If the stopping criterion in Step 6 is not met, continue learning and use one or two newly added training points as processing parameters to conduct processing experiments on the abrasive flow polishing system, and use the quality detection system to measure the detection value R1; the learning process returns to Step 3, using the updated parameters... or Use a new training sample S2 and a detection value R1 to update the Kriging surrogate model A; if the stopping criterion in step 6 is met, proceed to step 8; Step 8, calculate the coefficient of variation of the failure probability: if the stopping criterion in step 6 is met, then the learning ends; for the Monte Carlo population, Whether the size is suitable remains to be evaluated. The judgment of suitability is as follows (6): （6） In the formula, The coefficient of variation is 1. The failure probability is predicted by formula (1). The allowable value for the coefficient of variation is 0.05 or 0.

02. Step 9, update the total sample S1 in Monte Carlo: if the coefficient of variation If it is too large and does not satisfy formula (6) in step 8, then increase it. Then, generate a new overall sample S1 according to step 1, and then go to step 4 to predict the processing quality of the newly added processing parameters using the surrogate model A, and continue to use the active learning method until the stopping criterion formula (4) is met again; if formula (6) is met, then proceed to step 10. Step 10, complete the reliability analysis: If Values ​​below the specified values Then the reliability analysis stops, and the probability of the last failure is estimated. This is the result of the analysis on the reliability of processing quality.

4. The reliability analysis method as described in claim 3, characterized in that: In step 2, the training sample S2 is sampled from the total sample S1, or sampled according to the probability distribution type of the processing parameter P, or randomly sampled.

5. The reliability analysis method as described in claim 3, characterized in that: The predicted value in step 4 ,variance All correlation coefficients were calculated using the Kriging method. for Parameter points and training samples S2 The maximum correlation coefficient between points is shown in formula (4): （4）。

Images