A method, system, and apparatus for leak detection in packaged devices based on smart sensors.

Abstract

This invention provides a method, system, and apparatus for leak detection of packaged devices based on smart sensors, belonging to the field of device packaging technology. The method includes: determining the combination type of the packaged sensor to be leak-detected, and obtaining several leak detection processes matching the packaged sensor from a type-packaging process comparison table; extracting historical leak detection information sets matching each leak detection process from a historical database, constructing an effective leak detection function, and selecting the final process; sequentially obtaining a parameter set for leak detection of the packaged sensor according to the final process, and quantifying the parameter set; sequentially filling the quantization results into a parameter analysis table to obtain the degree of influence of each process step on the leak detection result, and obtaining the final leak detection result of the packaged sensor based on all influence degrees and the correlation between adjacent process steps. This effectively determines the leak detection result and ensures the reliability of the smart sensor.

Description

Technical Field

[0001] This invention relates to the field of device packaging technology, and in particular to a method, system and device for leak detection of packaged devices based on smart sensors. Background Technology

[0002] The architecture of the Industrial Internet relies on integrated circuit technology. The accelerated development of the Industrial Internet has driven the development of the integrated circuit industry, particularly in areas such as precision assurance in design and manufacturing, process optimization, and the surge in demand. The Industrial Internet has become a crucial demand side and testing ground for integrated circuits. Simultaneously, integrated circuits are the foundation of the electronic information industry. An integrated circuit is composed of a large number of electronic components and circuits integrated onto a single chip, including microprocessors, memory, and sensors. Therefore, packaging and integrated circuits are inseparable; packaging technology provides protection, heat dissipation, signal connectivity, and other support for integrated circuit applications. Packaging involves encapsulating different types of electronic components or integrated circuit chips within a casing using specific processes and materials to facilitate their use and protection. However, when semiconductors are combined with more advanced electronic devices, the problems associated with basic packaging become increasingly severe. To achieve smarter sensors, fundamental assembly differences between sensor and microelectronic device packaging are being addressed. However, because smart sensors involve different combinations of devices, using a specific device's packaging leak detection method for leak detection would undoubtedly reduce detection efficiency, leading to decreased reliability of the smart sensor.

[0003] Therefore, this invention proposes a leak detection method, system, and device for packaged devices based on intelligent sensors. Summary of the Invention

[0004] This invention provides a leak detection method, system, and device for packaged devices based on smart sensors. It is used to select the final process by matching several leak detection processes based on the sensor type, which effectively improves the leak detection efficiency. Furthermore, the leak detection result is determined by combining quantitative analysis of the parameter set with the influence degree analysis and correlation analysis after the result is filled, which effectively ensures the reliability of the smart sensor.

[0005] This invention provides a leak detection method for packaged devices based on smart sensors, comprising:

[0006] Step 1: Determine the combination type of the device of the leak-tested packaged sensor, and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaged process comparison table;

[0007] Step 2: Extract historical leak detection information sets that match each leak detection process from the historical database, construct an effective leak detection function, and select the final process;

[0008] Step 3: Sequentially obtain the parameter set for leak detection of the leak-packed sensor according to the final process, and quantize the parameter set, wherein the parameter set contains the values ​​of several leak detection parameters under each process step in the final process;

[0009] Step 4: Fill the parameter analysis table with the quantification results to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be leaked based on all the degree of influence and the correlation between adjacent process steps.

[0010] Preferably, an effective leak detection function is constructed, including:

[0011] Based on the operation time of each historical step involved in the same leak detection and combined with the historical leak detection rate, the first loss coefficient of each historical step is determined.

[0012] Based on the first loss coefficient of all historical steps under the same leak detection, the leak detection density function is determined.

[0013] For each leak detection process, each leak detection information in the historical leak detection information set is placed in the order of leak detection to construct a leak detection matrix;

[0014] Construct the first distribution function for each column vector in the leak detection matrix, and obtain the second loss coefficient based on each leak detection parameter;

[0015] Historical curves are constructed based on all second loss coefficients, and parameter alignment is performed with the standard curve of the leak detection process to obtain the adjustment amount for each leak detection parameter.

[0016] Based on the adjustable values ​​of all leak detection parameters involved in the same historical step, the leak detection density function is optimized to obtain the leak detection correlation function of the corresponding leak detection process based on the historical leak detection information set.

[0017] The leak detection correlation function is reset and adjusted to obtain the effective leak detection function.

[0018] Preferably, the leak detection density function is optimized, including:

[0019] The first variable and the leak detection parameters corresponding to each historical step in the leak detection density function are filled into the variable lookup table to obtain several leak detection parameters that match each first variable, and a control group is established.

[0020] Based on the control relationship of each control group, and combined with the adjustment amount of the leak detection parameters involved in the control group, the adjustment value of the variable coefficient for the corresponding first variable in the leak detection density function and the addition of variables are obtained to obtain the adjustment density function;

[0021] Obtain the adjustment density function for each leak detection involved in the corresponding leak detection process, and obtain all value coefficients involved in the same second variable;

[0022] Extract the strong correlation coefficients and weak correlation coefficients from all value coefficients involved in the same second variable, and calculate the absolute value array of the coefficient differences between the strong correlation coefficient, weak correlation coefficient and each value coefficient in turn, to determine whether it is necessary to add auxiliary coefficients for the same second variable.

[0023] If not, then the trend status is determined based on all the value coefficients involved in the same second variable;

[0024] If necessary, coefficient interpolation is performed based on the absolute value array of the coefficient differences to determine the number of additional auxiliary coefficients, and then the trend status of all interpolated coefficients is judged.

[0025] Based on the trend state of each second variable, the leak detection correlation function for the corresponding leak detection process is determined.

[0026] Preferably, quantizing the parameter set includes:

[0027] Obtain the step attributes of each process step in the final process, and construct an attribute vector;

[0028] Obtain a quantization model that matches the attribute vector from the attribute-model database, wherein each layer of the quantization model corresponds to a process step, and each layer of the network contains parameter quantization weights and quantization change information.

[0029] The parameter set is quantized according to the quantization model to obtain the quantization result.

[0030] Preferably, obtaining a quantization model matching the attribute vector from an attribute-model database includes:

[0031] When a first model that is completely identical to the attribute vector exists in the attribute-model database, the first model is regarded as a quantization model;

[0032] When there is no first model in the attribute-model database that is completely consistent with the attribute vector, cluster analysis is performed on all step attributes under the final process to obtain several clusters, and an initial cluster vector is constructed based on the cluster centers of the several clusters.

[0033] The distance between each cluster center and the remaining attributes in the corresponding clustering results is determined sequentially, and the corresponding step attributes are weakened according to the distance.

[0034] ;

[0035] in, This represents the weakening analysis result of the j-th attribute in the corresponding clustering result; n1 represents the total number of attributes involved in the corresponding clustering result; sum1 represents the total number of attributes involved in the final process. This represents the distance between the cluster center of the corresponding clustering result and the j-th remaining attribute; a1 represents the set threshold.

[0036] The attributes with a weakening analysis result of 0 are replaced with 0 and added to the initial cluster vector, and the attributes with a weakening analysis result of non-zero are replaced with their original attributes and added to the initial cluster vector, thus obtaining the current cluster vector;

[0037] The current cluster vector is compared with the attribute-model database. If there is a second model that is completely consistent with the current cluster vector, then the second model is regarded as the quantization model.

[0038] Otherwise, retrieve the third model from the attribute-model database that has the most attributes matching the attribute vector, and simultaneously retrieve the fourth model from the attribute-model database that has the most attributes matching the current cluster vector.

[0039] The parameter quantization weights of the same layer networks in the third and fourth models are obtained respectively, and the required network layers are locked according to max(y1i,y2i), where y1i represents the parameter quantization weight of the i-th layer network in the third model; y2i represents the parameter quantization weight of the i-th layer network in the fourth model.

[0040] All the required network layers are then re-fused to obtain the quantized model.

[0041] Preferably, the degree of impact of each process step on the leak detection results is obtained, including:

[0042] Compare the filling results for each process step in the filling analysis table with the corresponding standard results;

[0043] The degree of impact is determined based on the comparison results.

[0044] Preferably, based on the degree of all influences and the correlation between adjacent process steps, the final leak detection result of the leak-detected packaged sensor is obtained, including:

[0045] Obtain the initial test results of the last process step, and adjust the initial test results based on all the degree of influence and the correlation between adjacent process steps;

[0046] Determine the characteristic parameters of the leak-detected packaged sensor, and sequentially determine the influence coefficient of each characteristic parameter on the last process step;

[0047] The final leak detection result is obtained by adjusting the initial leak detection result based on all the influence coefficients.

[0048] This invention provides a leak detection system for packaged devices based on intelligent sensors, comprising:

[0049] The process matching module is used to determine the combination type of the device of the leak-tested packaged sensor and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaging process comparison table.

[0050] The function building module is used to extract historical leak detection information sets that match each leak detection process from the historical database, build effective leak detection functions, and filter the final process.

[0051] The quantization module is used to sequentially acquire the parameter set for leak detection of the leak-packed sensor according to the final process, and to quantize the parameter set, wherein the parameter set contains the values ​​of several leak detection parameters under each process step.

[0052] The result acquisition module is used to sequentially fill the parameter analysis table with the quantification results to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be detected based on all the degree of influence and the correlation between adjacent process steps.

[0053] This invention provides a leak detection device for packaged devices based on smart sensors, used to perform any of the leak detection methods for packaged devices based on smart sensors described above.

[0054] Compared with the prior art, the beneficial effects of this application are as follows:

[0055] The final process is selected by matching several leak detection processes based on the sensor type, which effectively improves the leak detection efficiency. Furthermore, the leak detection result is determined by combining the quantitative analysis of the parameter set and the influence analysis and correlation of the results after filling in the results, which effectively ensures the reliability of the smart sensor.

[0056] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0057] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a flowchart of a leak detection method for packaged devices based on smart sensors, as described in an embodiment of the present invention.

[0060] Figure 2 This is a structural diagram of a leak detection system for packaged devices based on intelligent sensors, as described in an embodiment of the present invention.

[0061] Figure 3 This is a structural diagram of the packaging device in an embodiment of the present invention. Detailed Implementation

[0062] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0063] This invention provides a leak detection method for packaged devices based on smart sensors, such as... Figure 1 As shown, it includes:

[0064] Step 1: Determine the combination type of the device of the leak-tested packaged sensor, and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaged process comparison table;

[0065] Step 2: Extract historical leak detection information sets that match each leak detection process from the historical database, construct an effective leak detection function, and select the final process;

[0066] Step 3: Sequentially obtain the parameter set for leak detection of the leak-packed sensor according to the final process, and quantize the parameter set, wherein the parameter set contains the values ​​of several leak detection parameters under each process step in the final process;

[0067] Step 4: Fill the parameter analysis table with the quantification results to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be leaked based on all the degree of influence and the correlation between adjacent process steps.

[0068] In this embodiment, the leak-tested packaged sensor refers to a smart sensor that needs to be leak-tested, and the smart sensor contains at least one device. Therefore, the combination type is obtained by obtaining the type of each device in the smart sensor. For example, the leak-tested packaged sensor contains: device 1 and device 2, device 1 is type A and device 2 is type B. In this case, the combination type is A+B.

[0069] In this embodiment, the type-packaging process lookup table contains several leak detection processes under different combination types, and each combination type may have multiple leak detection processes, all of which are pre-stored for easy matching and retrieval. For example, the leak detection process is:

[0070] S1: Place the packaged electronic components into the pretreatment container and pressurize and fill the packaged electronic components with helium;

[0071] S2: After helium filling is completed, the pretreatment tank is subjected to pressure holding treatment;

[0072] S3: Expel the helium gas from the pretreatment tank, and then purge it with dry, hot nitrogen gas;

[0073] S4: Remove the packaged electronic components from the pretreatment tank and place them at the workstation for detailed inspection. The workstation is sealed using fluororubber sealing rings and negative pressure suction.

[0074] The parameters involved in S1 are described as follows: the time for pressurizing and charging helium and the pressure at the moment when pressurizing and charging helium stops;

[0075] The parameters involved in S2 are described as follows: the pressure holding time and the helium leakage during the pressure holding process;

[0076] The parameters involved in S3 are described as follows: helium purging time, time to start purging with dry hot nitrogen, helium concentration before starting purging with dry hot nitrogen, purging speed and duration of purging with dry hot nitrogen;

[0077] The parameters involved in S4 are described as follows: detailed inspection results.

[0078] In this embodiment, the historical database includes leak detection results generated by using different leak detection processes to detect leaks in several smart sensors of the same type, based on different steps. Therefore, the historical leak detection information set can be obtained by matching the leak detection processes. It should be noted that the historical leak detection information set includes leak detection information from several historical leak detections based on the same leak detection process, that is, one leak detection process corresponds to one historical leak detection information set.

[0079] In this embodiment, the leak detection effective function is for the corresponding historical leak detection information set and is constructed based on each process step involved in the corresponding leak detection process as a basic variable.

[0080] In this embodiment, the final process refers to the process obtained by selecting one process from several leak detection processes according to the leak detection effectiveness function.

[0081] In this embodiment, the leak detection parameters refer to the parameters of the leak detection operation involved in the corresponding process step.

[0082] In this embodiment, after the final process is determined, the parameter descriptions involved are obtained by matching the process-parameter database. That is, the database contains different processes and the parameters involved in each step of the process, which are pre-set. Therefore, by capturing the values ​​of the parameter descriptions involved in the leak detection process, a parameter set can be constructed.

[0083] In this embodiment, the quantization result is obtained by processing the parameter set based on the quantization model in order to standardize the values ​​and facilitate subsequent determination of the degree of influence.

[0084] In this embodiment, the degree of influence is obtained by comparing the values ​​involved in the corresponding process steps in the filled parameter analysis table with the standard results. The standard results are pre-set because each smart sensor has its corresponding evaluation standard before leaving the factory, so as to avoid the smart sensor from malfunctioning.

[0085] In this embodiment, the final leak detection result is based on the last process step and is obtained with the help of the previously obtained impact level and correlation, and can be the leak detection rate.

[0086] In this embodiment, the coefficients of each variable in the leak detection effective function are superimposed, and the process corresponding to the function with the largest superposition sum is regarded as the final process.

[0087] In this embodiment, the packaging device for the smart sensor, such as Figure 3 As shown, 1 is a leak detection tank, 2 is a placement platform, 3 is a handle, 4 is a device, and 5 is a material tray.

[0088] The beneficial effects of the above technical solution are: based on the sensor type, several leak detection processes are matched to screen the final process, which effectively improves the leak detection efficiency. Furthermore, by performing quantitative analysis on the parameter set and combining the influence analysis and correlation of the results after filling in the results, the leak detection result is determined, which effectively ensures the reliability of the smart sensor.

[0089] This invention provides a leak detection method for packaged devices based on smart sensors, which constructs an effective leak detection function, including:

[0090] Based on the operation time of each historical step involved in the same leak detection and combined with the historical leak detection rate, the first loss coefficient of each historical step is determined.

[0091] Based on the first loss coefficient of all historical steps under the same leak detection, the leak detection density function is determined.

[0092] For each leak detection process, each leak detection information in the historical leak detection information set is placed in the order of leak detection to construct a leak detection matrix;

[0093] Construct the first distribution function for each column vector in the leak detection matrix, and obtain the second loss coefficient based on each leak detection parameter;

[0094] Historical curves are constructed based on all second loss coefficients, and parameter alignment is performed with the standard curve of the leak detection process to obtain the adjustment amount for each leak detection parameter.

[0095] Based on the adjustable values ​​of all leak detection parameters involved in the same historical step, the leak detection density function is optimized to obtain the leak detection correlation function of the corresponding leak detection process based on the historical leak detection information set.

[0096] The leak detection correlation function is reset and adjusted to obtain the effective leak detection function.

[0097] In this embodiment, the operation time refers to the total time of the corresponding historical steps, and the historical leak detection rate is the leak detection rate under the corresponding historical leak detection.

[0098] In this embodiment, the first loss coefficient = Where tz is the operation time of the corresponding historical step; t0 is the setting time of the corresponding historical step; This corresponds to the historical leak detection rate; ts is the sum of the set times for each historical step in the process used for the corresponding leak detection.

[0099] In this embodiment, the leak detection density function = G (the first loss coefficient of each historical step, the coefficient distribution density of all first loss coefficients) is obtained based on the density analysis model. This model is obtained by training the neural network model based on the first loss coefficients of different historical steps and the analysis results of the combination, and thus the leak detection density function can be obtained directly.

[0100] In this embodiment, the leak detection matrix = The self-detection information includes parameter descriptions and values ​​for different steps, and the order of detection is the order of detection sequence.

[0101] In this embodiment, the first distribution function is determined by the positional distribution of all values ​​described by the corresponding parameters in the numerical coordinate system.

[0102] In this embodiment, the second loss coefficient = the first distribution function × the function loss variable, where the function loss variable is obtained by matching from the distribution-variable lookup table, which contains different distribution functions and the loss variables corresponding to those functions, mainly for calculating the second loss coefficient.

[0103] In this embodiment, the horizontal axis of the historical curve represents the corresponding parameter description, and the vertical axis represents the corresponding second loss coefficient.

[0104] In this embodiment, the horizontal axis of the standard curve represents the corresponding parameter description, and the vertical axis represents the corresponding standard loss coefficient. The standard loss coefficients are all preset and are for different parameter descriptions.

[0105] In this embodiment, the amount to be adjusted is: if the second loss coefficient of the corresponding parameter description (i.e., the leak detection parameter) is greater than the standard loss coefficient, then it is calculated according to (second loss coefficient - standard loss coefficient) × the setting amount of the corresponding parameter. The setting amount of the corresponding parameter is obtained by matching from the parameter-quantity lookup table, which includes different parameter descriptions and the matching setting amounts.

[0106] In this embodiment, each historical step includes at least one leak detection parameter, meaning that the parameters involved in each step are pre-defined and known.

[0107] In this embodiment, "reset adjustment" refers to setting the coefficients of the parameter variables involved in the leak detection correlation function to 1 or 0. If the coefficient is less than 0.05, it is set to 0; if the coefficient is greater than 0.95, it is set to 1.

[0108] The beneficial effects of the above technical solution are: the first loss coefficient is initially determined based on the operation time and historical leak detection rate, and then a density function is constructed. Subsequently, the adjustment amount of each parameter is obtained by constructing a matrix and analyzing the second loss coefficient of each parameter, thereby optimizing the function. In addition, the rationality of the function is ensured by combining it with resetting, which provides a basis for the selection of leak detection process.

[0109] This invention provides a leak detection method for packaged devices based on smart sensors, which optimizes the leak detection density function, including:

[0110] The first variable and the leak detection parameters corresponding to each historical step in the leak detection density function are filled into the variable lookup table to obtain several leak detection parameters that match each first variable, and a control group is established.

[0111] Based on the control relationship of each control group, and combined with the adjustment amount of the leak detection parameters involved in the control group, the adjustment value of the variable coefficient for the corresponding first variable in the leak detection density function and the addition of variables are obtained to obtain the adjustment density function;

[0112] Obtain the adjustment density function for each leak detection involved in the corresponding leak detection process, and obtain all value coefficients involved in the same second variable;

[0113] Extract the strong correlation coefficients and weak correlation coefficients from all value coefficients involved in the same second variable, and calculate the absolute value array of the coefficient differences between the strong correlation coefficient, weak correlation coefficient and each value coefficient in turn, to determine whether it is necessary to add auxiliary coefficients for the same second variable.

[0114] If not, then the trend status is determined based on all the value coefficients involved in the same second variable;

[0115] If necessary, coefficient interpolation is performed based on the absolute value array of the coefficient differences to determine the number of additional auxiliary coefficients, and then the trend status of all interpolated coefficients is judged.

[0116] Based on the trend state of each second variable, the leak detection correlation function for the corresponding leak detection process is determined.

[0117] In this embodiment, the leak detection density function is defined as u1x1 + u2x2 + u3x3 + ... + unxn, where x1, x2, x3, ..., xn represent the corresponding first variables, u1, u2, u3, ..., un represent the corresponding value coefficients (related to density; the higher the density, the higher the value coefficient, and the value ranges from 0 to 1), and n represents the number of historical steps. Since each step contains at least one parameter, which is known and predefined, the comparison results between steps and parameters can be obtained directly based on the variable comparison table. Furthermore, the first variable corresponding to each step involves a corresponding number of leak detection parameters. Thus, the control group = {corresponding to the first variable, and several leak detection parameters matching the first variable}.

[0118] In this embodiment, the comparison relationship is the relationship between the variable and the parameter.

[0119] In this embodiment, the variable coefficient adjustment value = the sum of the adjustment amount of the leak detection parameters involved in the first variable × the parameter weights. It should be noted that the weights of all parameters involved in the process are 1, and the parameters and parameter weights involved in each step are pre-planned and can be directly used for calculation.

[0120] In this embodiment, the added variable refers to the new variable added after adjusting the leak detection density function. For example, the adjusted density function is = u1x1 + u2x2 + u3x3 + ... + unxn + un1xn1 + un2xn2, where un1 and un2 are the value coefficients of the new variables xn1 and xn2. The added variable is obtained by inputting the vector composed of the adjusted values ​​of the variable coefficients of all the first variables involved in the leak detection density function into the vector analysis model. The vector analysis model is obtained by training the neural network model based on vectors composed of different values ​​and the added results of the vector. Therefore, the added variable can be obtained directly.

[0121] In this embodiment, each variable in the adjustment density function is collectively referred to as the second variable.

[0122] In this embodiment, all value coefficients of the same second variable are linearly fitted, the values ​​on the fitted line are averaged, and the result is regarded as a strong correlation coefficient. Discrete points that are not on the fitted line are locked, and the straight-line distance between each discrete point and the corresponding fitted line is calculated. The value of the discrete point with the largest straight-line distance is regarded as a weak correlation coefficient.

[0123] In this embodiment, the array of value coefficients = {absolute value of the difference between the coefficients of strong correlation coefficients and corresponding value coefficients, and absolute value of the difference between the coefficients of weak correlation coefficients and corresponding value coefficients};

[0124] If the absolute value of the difference between the weak correlation coefficient and the corresponding value coefficient / the absolute value of the difference between the strong correlation coefficient and the corresponding value coefficient > N1, then it is considered that an auxiliary coefficient needs to be added. Here, N1 is 10, and the auxiliary coefficient = the absolute value of the difference between the weak correlation coefficient and the corresponding value coefficient - the absolute value of the difference between the strong correlation coefficient and the corresponding value coefficient.

[0125] Otherwise, it is considered that no additional auxiliary coefficient is required.

[0126] In this embodiment, curves are plotted for all value coefficients involved in the second variable to analyze the trend of the curves. The trend states include: horizontal trend and fluctuating trend.

[0127] In this embodiment, the result of rounding up (absolute value of the difference between the weak correlation coefficient and the corresponding value coefficient / absolute value of the difference between the strong correlation coefficient and the corresponding value coefficient - N1) / N1 is the number of additions.

[0128] In this embodiment, coefficient interpolation refers to randomly inserting coefficients at different positions among all the value coefficients of the corresponding second variable, and then redrawing the curve to determine the trend state.

[0129] In this embodiment, if the trend is horizontal, the average of all the value coefficients involved is used as the correlation coefficient of the second variable. If the trend is not horizontal, the maximum and minimum values ​​of the value coefficients are removed, the average of the remaining coefficients is calculated, and the variance of the remaining coefficients is subtracted. The result is used as the correlation coefficient of the second variable. At this time, the leak detection correlation function based on the variables and correlation coefficients is: r1x1+r2x2+r3x3+...+rnxn+rn1xn1+rn2xn2, where r1,r2,r3,...,rn,rn1,rn2 are regarded as the correlation coefficients of the corresponding second variables.

[0130] The beneficial effects of the above technical solution are: by establishing a comparison relationship between variables and parameters, the adjustment density function can be easily determined; and by analyzing the value coefficient of each variable, the trend status can be effectively judged, the leak detection correlation function can be obtained, and a foundation can be provided for subsequent screening of the final process.

[0131] This invention provides a leak detection method for packaged devices based on smart sensors, which quantizes the parameter set, including:

[0132] Obtain the step attributes of each process step in the final process, and construct an attribute vector;

[0133] Obtain a quantization model that matches the attribute vector from the attribute-model database, wherein each layer of the quantization model corresponds to a process step, and each layer of the network contains parameter quantization weights and quantization change information.

[0134] The parameter set is quantized according to the quantization model to obtain the quantization result.

[0135] In this embodiment, the process step refers to the execution attribute of the corresponding step, such as helium filling attribute, nitrogen filling attribute, etc., and the attribute vector = {the step attribute of each process step}.

[0136] In this embodiment, the attribute-model database contains vectors with different attribute combinations and quantization models that match the vectors of those combinations. This is to standardize the values ​​of the parameters in the parameter set to facilitate subsequent analysis.

[0137] In this embodiment, the number of network layers in the quantization model is consistent with the number of process steps.

[0138] In this embodiment, the quantization weights and quantization change information of the parameters of each network layer are known, in order to reasonably standardize the values ​​of the corresponding parameters.

[0139] The beneficial effects of the above technical solution are: by constructing attribute vectors, it is easier to match quantification models from the database, realize the quantification of parameter values, and facilitate subsequent analysis.

[0140] This invention provides a leak detection method for packaged devices based on smart sensors, which obtains a quantization model matching the attribute vector from an attribute-model database, including:

[0141] When a first model that is completely identical to the attribute vector exists in the attribute-model database, the first model is regarded as a quantization model;

[0142] When there is no first model in the attribute-model database that is completely consistent with the attribute vector, cluster analysis is performed on all step attributes under the final process to obtain several clusters, and an initial cluster vector is constructed based on the cluster centers of the several clusters.

[0143] The distance between each cluster center and the remaining attributes in the corresponding clustering results is determined sequentially, and the corresponding step attributes are weakened according to the distance.

[0144] ;

[0145] in, This represents the weakening analysis result of the j-th attribute in the corresponding clustering result; n1 represents the total number of attributes involved in the corresponding clustering result; sum1 represents the total number of attributes involved in the final process. This represents the distance between the cluster center of the corresponding clustering result and the j-th remaining attribute; a1 represents the set threshold.

[0146] The attributes with a weakening analysis result of 0 are replaced with 0 and added to the initial cluster vector, and the attributes with a weakening analysis result of non-zero are replaced with their original attributes and added to the initial cluster vector, thus obtaining the current cluster vector;

[0147] The current cluster vector is compared with the attribute-model database. If there is a second model that is completely consistent with the current cluster vector, then the second model is regarded as the quantization model.

[0148] Otherwise, retrieve the third model from the attribute-model database that has the most attributes matching the attribute vector, and simultaneously retrieve the fourth model from the attribute-model database that has the most attributes matching the current cluster vector.

[0149] The parameter quantization weights of the same layer networks in the third and fourth models are obtained respectively, and the required network layers are locked according to max(y1i,y2i), where y1i represents the parameter quantization weight of the i-th layer network in the third model; y2i represents the parameter quantization weight of the i-th layer network in the fourth model.

[0150] All the required network layers are then re-fused to obtain the quantized model.

[0151] In this embodiment, cluster analysis is implemented based on the K-means algorithm, which can then yield clusters.

[0152] In this embodiment, the threshold value is set to 0.25.

[0153] In this embodiment, for example, the initial cluster vector is {cluster 1 cluster 2 cluster 3 blank element 01 blank element 02}. At this time, the current cluster vector obtained according to the weakening analysis result is {cluster 1 cluster 2 cluster 3 0 1}, and the replacement of attributes 0 and 1 is to make it consistent with the number of elements involved in the original attribute vector.

[0154] In this embodiment, the number of attributes refers to the number of attribute elements in the corresponding vector.

[0155] In this embodiment, after obtaining the third and fourth models, and knowing that the quantization weights of each network layer in the model are known, a more reliable network layer can be selected by comparing their magnitudes. The quantization model is obtained by re-fusion of the network layers, that is, by reconfiguring the weights of the network layers according to the current quantization weights of the selected network layers, and the original quantization change information remains unchanged.

[0156] The beneficial effects of the above technical solution are: by constructing attribute vectors to match the quantization model, when there is no perfectly matching model, cluster analysis and substitution of 0 and 1 are used to ensure the rationality of model acquisition and the uniformity of processing each step. Furthermore, by selecting the model with the most matching attribute counts under two vectors and combining the weight comparison to obtain the final network layer, the reliability of the fusion model is ensured, thereby ensuring the accuracy of the quantization parameter values.

[0157] This invention provides a leak detection method for packaged devices based on smart sensors, which obtains the degree of influence of each process step on the leak detection result, including:

[0158] Compare the filling results for each process step in the filling analysis table with the corresponding standard results;

[0159] The degree of impact is determined based on the comparison results.

[0160] In this embodiment, the degree of influence = (1 - sim(filled result, standard result)) / N2, where N2 represents the number of parameters involved in the corresponding process step.

[0161] The beneficial effect of the above technical solution is that the degree of influence can be effectively determined by comparing the filling results with the standard results.

[0162] This invention provides a leak detection method for packaged devices based on smart sensors. Based on the degree of influence of all factors and the correlation between adjacent process steps, the final leak detection result of the packaged sensor to be tested is obtained, including:

[0163] Obtain the initial test results of the last process step, and adjust the initial test results based on all the degree of influence and the correlation between adjacent process steps;

[0164] Determine the characteristic parameters of the leak-detected packaged sensor, and sequentially determine the influence coefficient of each characteristic parameter on the last process step;

[0165] The final leak detection result is obtained by adjusting the initial leak detection result based on all the influence coefficients.

[0166] In this embodiment, the initial detection result refers to the initial leak detection rate of the smart sensor determined based on the last process.

[0167] The result of one adjustment = initial leak detection rate × (1 + ln(1 + sum of the correlation coefficients of each degree of influence and the correlation coefficients of the adjacent process steps involved / number of process steps)), where the correlation coefficients range from 0 to 1. It should be noted that the correlation between adjacent process steps is set in advance. For example, step 1 will affect the result of step 2, but step 2 will not affect the result of step 3. In this case, the correlation coefficient between step 2 and step 3 is 0.

[0168] In this embodiment, the characteristic parameters refer to the sensor's size, sensitivity, integration level, and sensitivity, etc.

[0169] In this embodiment, the influence coefficient of each characteristic parameter on the last process step is obtained from the characteristic-relationship lookup table, and the value range is (0, 0.05).

[0170] Final leak detection result = first adjustment result × (1 + sum of all influence coefficients).

[0171] The beneficial effects of the above technical solution are: the results are adjusted based on the degree of influence and correlation, and then the results are adjusted again based on the influence of the characteristic parameters, so as to ensure the accuracy of the final leak detection results.

[0172] This invention provides a leak detection system for packaged devices based on intelligent sensors, such as... Figure 2 As shown, it includes:

[0173] The process matching module is used to determine the sensing type of the leak-tested packaged sensor and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaging process comparison table.

[0174] The function building module is used to extract historical leak detection information sets that match each leak detection process from the historical database, build effective leak detection functions, and filter the final process.

[0175] The quantization module is used to sequentially acquire the parameter set for leak detection of the leak-packed sensor according to the final process, and to quantize the parameter set, wherein the parameter set contains several leak detection parameters for each process step.

[0176] The result acquisition module is used to sequentially fill the parameter analysis table with the quantification results to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be detected based on all the degree of influence and the correlation between adjacent process steps.

[0177] The beneficial effects of the above technical solution are: based on the sensor type, several leak detection processes are matched to screen the final process, which effectively improves the leak detection efficiency. Furthermore, by performing quantitative analysis on the parameter set and combining the influence analysis and correlation of the results after filling in the results, the leak detection result is determined, which effectively ensures the reliability of the smart sensor.

[0178] This invention provides a leak detection device for packaged devices based on smart sensors, used to perform any of the leak detection methods for packaged devices based on smart sensors described above.

[0179] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A leak detection method for packaged devices based on smart sensors, characterized in that, include: Step 1: Determine the combination type of the device of the leak-tested packaged sensor, and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaged process comparison table; Step 2: Extract historical leak detection information sets that match each leak detection process from the historical database, construct an effective leak detection function, and select the final process; Step 3: Sequentially obtain the parameter set for leak detection of the leak-packed sensor according to the final process, and quantize the parameter set, wherein the parameter set contains the values ​​of several leak detection parameters under each process step in the final process; Step 4: Fill the parameter analysis table with the quantification results in sequence to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be leaked based on all the degree of influence and the correlation between adjacent process steps. The quantization of the parameter set includes: Obtain the step attributes of each process step in the final process, and construct an attribute vector; Obtain a quantization model that matches the attribute vector from the attribute-model database, wherein each layer of the quantization model corresponds to a process step, and each layer of the network contains parameter quantization weights and quantization change information. The parameter set is quantized according to the quantization model to obtain the quantization result. The process of obtaining a quantization model matching the attribute vector from the attribute-model database includes: When a first model that is completely identical to the attribute vector exists in the attribute-model database, the first model is regarded as a quantization model; When there is no first model in the attribute-model database that is completely consistent with the attribute vector, cluster analysis is performed on all step attributes under the final process to obtain several clusters, and an initial cluster vector is constructed based on the cluster centers of the several clusters. The distances between each cluster center and the remaining attributes in the corresponding clustering results are determined sequentially, and the corresponding step attributes are weakened according to the distances. ; in, This represents the weakening analysis result of the j-th attribute in the corresponding clustering result; n1 represents the total number of attributes involved in the corresponding clustering result; sum1 represents the total number of attributes involved in the final process. This represents the distance between the cluster center of the corresponding clustering result and the j-th remaining attribute; a1 represents the set threshold. The attributes with a weakening analysis result of 0 are replaced with 0 and added to the initial cluster vector, and the attributes with a weakening analysis result of non-zero are replaced with their original attributes and added to the initial cluster vector, thus obtaining the current cluster vector; The current cluster vector is compared with the attribute-model database. If there is a second model that is completely consistent with the current cluster vector, then the second model is regarded as the quantization model. Otherwise, retrieve the third model from the attribute-model database that has the most attributes matching the attribute vector, and simultaneously retrieve the fourth model from the attribute-model database that has the most attributes matching the current cluster vector. The parameter quantization weights of the same layer networks in the third and fourth models are obtained respectively, and the required network layers are locked according to max(y1i,y2i), where y1i represents the parameter quantization weight of the i-th layer network in the third model; y2i represents the parameter quantization weight of the i-th layer network in the fourth model. All required network layers are re-fused to obtain a quantized model; The construction of an effective leak detection function includes: Based on the operation time of each historical step involved in the same leak detection and combined with the historical leak detection rate, the first loss coefficient of each historical step is determined. Based on the first loss coefficient of all historical steps under the same leak detection, the leak detection density function is determined. For each leak detection process, each leak detection information in the historical leak detection information set is placed in the order of leak detection to construct a leak detection matrix; Construct the first distribution function for each column vector in the leak detection matrix, and obtain the second loss coefficient based on each leak detection parameter; Historical curves are constructed based on all second loss coefficients, and parameter alignment is performed with the standard curve of the leak detection process to obtain the adjustment amount for each leak detection parameter. Based on the adjustable values ​​of all leak detection parameters involved in the same historical step, the leak detection density function is optimized to obtain the leak detection correlation function of the corresponding leak detection process based on the historical leak detection information set. The leak detection correlation function is reset and adjusted to obtain the effective leak detection function; The optimization of the leak detection density function includes: The first variable and the leak detection parameters corresponding to each historical step in the leak detection density function are filled into the variable lookup table to obtain several leak detection parameters that match each first variable, and a control group is established. Based on the control relationship of each control group, and combined with the adjustment amount of the leak detection parameters involved in the control group, the adjustment value of the variable coefficient for the corresponding first variable in the leak detection density function and the addition of variables are obtained to obtain the adjustment density function; Obtain the adjustment density function for each leak detection involved in the corresponding leak detection process, and obtain all value coefficients involved in the same second variable; Extract the strong correlation coefficients and weak correlation coefficients from all value coefficients involved in the same second variable, and calculate the absolute value array of the coefficient differences between the strong correlation coefficient, weak correlation coefficient and each value coefficient in turn, to determine whether it is necessary to add auxiliary coefficients for the same second variable. If not, then the trend status is determined based on all the value coefficients involved in the same second variable; If necessary, coefficient interpolation is performed based on the absolute value array of the coefficient differences to determine the number of additional auxiliary coefficients, and then the trend status of all interpolated coefficients is judged. Based on the trend state of each second variable, the leak detection correlation function for the corresponding leak detection process is determined.

2. The leak detection method for packaged devices based on smart sensors according to claim 1, characterized in that, To determine the impact of each process step on the leak detection results, including: Compare the filling results for each process step in the filling analysis table with the corresponding standard results; The degree of impact is determined based on the comparison results.

3. The leak detection method for packaged devices based on smart sensors according to claim 2, characterized in that, Based on the degree of influence and the correlation between adjacent process steps, the final leak detection result of the leak-detected packaged sensor is obtained, including: Obtain the initial test results of the last process step, and adjust the initial test results based on all the degree of influence and the correlation between adjacent process steps; Determine the characteristic parameters of the leak-detected packaged sensor, and sequentially determine the influence coefficient of each characteristic parameter on the last process step; The final leak detection result is obtained by adjusting the initial leak detection result based on all the influence coefficients.

4. A leak detection system for packaged devices based on intelligent sensors, characterized in that, include: The process matching module is used to determine the combination type of the device of the leak-tested packaged sensor and obtain several leak detection processes that match the leak-tested packaged sensor from the type-packaging process comparison table. The function building module is used to extract historical leak detection information sets that match each leak detection process from the historical database, build effective leak detection functions, and filter the final process. The quantization module is used to sequentially acquire the parameter set for leak detection of the leak-packed sensor according to the final process, and to quantize the parameter set, wherein the parameter set contains the values ​​of several leak detection parameters under each process step. The result acquisition module is used to fill the parameter analysis table with the quantification results in sequence to obtain the degree of influence of each process step on the leak detection result, and obtain the final leak detection result of the packaged sensor to be leaked based on all the degree of influence and the correlation between adjacent process steps. The quantization of the parameter set includes: Obtain the step attributes of each process step in the final process, and construct an attribute vector; Obtain a quantization model that matches the attribute vector from the attribute-model database, wherein each layer of the quantization model corresponds to a process step, and each layer of the network contains parameter quantization weights and quantization change information. The parameter set is quantized according to the quantization model to obtain the quantization result; The process of obtaining a quantization model matching the attribute vector from the attribute-model database includes: When a first model that is completely identical to the attribute vector exists in the attribute-model database, the first model is regarded as a quantization model; When there is no first model in the attribute-model database that is completely consistent with the attribute vector, cluster analysis is performed on all step attributes under the final process to obtain several clusters, and an initial cluster vector is constructed based on the cluster centers of the several clusters. The distance between each cluster center and the remaining attributes in the corresponding clustering results is determined sequentially, and the corresponding step attributes are weakened according to the distance. ; in, This represents the weakening analysis result of the j-th attribute in the corresponding clustering result; n1 represents the total number of attributes involved in the corresponding clustering result; sum1 represents the total number of attributes involved in the final process. This represents the distance between the cluster center of the corresponding clustering result and the j-th remaining attribute; a1 represents the set threshold. The attributes with a weakening analysis result of 0 are replaced with 0 and added to the initial cluster vector, and the attributes with a weakening analysis result of non-zero are replaced with their original attributes and added to the initial cluster vector, thus obtaining the current cluster vector; The current cluster vector is compared with the attribute-model database. If there is a second model that is completely consistent with the current cluster vector, then the second model is regarded as the quantization model. Otherwise, retrieve the third model from the attribute-model database that has the most attributes matching the attribute vector, and simultaneously retrieve the fourth model from the attribute-model database that has the most attributes matching the current cluster vector. The parameter quantization weights of the same layer networks in the third and fourth models are obtained respectively, and the required network layers are locked according to max(y1i,y2i), where y1i represents the parameter quantization weight of the i-th layer network in the third model; y2i represents the parameter quantization weight of the i-th layer network in the fourth model. All required network layers are re-fused to obtain a quantized model; The construction of an effective leak detection function includes: Based on the operation time of each historical step involved in the same leak detection and combined with the historical leak detection rate, the first loss coefficient of each historical step is determined. Based on the first loss coefficient of all historical steps under the same leak detection, the leak detection density function is determined. For each leak detection process, each leak detection information in the historical leak detection information set is placed in the order of leak detection to construct a leak detection matrix; Construct the first distribution function for each column vector in the leak detection matrix, and obtain the second loss coefficient based on each leak detection parameter; Historical curves are constructed based on all second loss coefficients, and parameter alignment is performed with the standard curve of the leak detection process to obtain the adjustment amount for each leak detection parameter. Based on the adjustable values ​​of all leak detection parameters involved in the same historical step, the leak detection density function is optimized to obtain the leak detection correlation function of the corresponding leak detection process based on the historical leak detection information set. The leak detection correlation function is reset and adjusted to obtain the effective leak detection function; The optimization of the leak detection density function includes: The first variable and the leak detection parameters corresponding to each historical step in the leak detection density function are filled into the variable lookup table to obtain several leak detection parameters that match each first variable, and a control group is established. Based on the control relationship of each control group, and combined with the adjustment amount of the leak detection parameters involved in the control group, the adjustment value of the variable coefficient for the corresponding first variable in the leak detection density function and the addition of variables are obtained to obtain the adjustment density function; Obtain the adjustment density function for each leak detection involved in the corresponding leak detection process, and obtain all value coefficients involved in the same second variable; Extract the strong correlation coefficients and weak correlation coefficients from all value coefficients involved in the same second variable, and calculate the absolute value array of the coefficient differences between the strong correlation coefficient, weak correlation coefficient and each value coefficient in turn, to determine whether it is necessary to add auxiliary coefficients for the same second variable. If not, then the trend status is determined based on all the value coefficients involved in the same second variable; If necessary, coefficient interpolation is performed based on the absolute value array of the coefficient differences to determine the number of additional auxiliary coefficients, and then the trend status of all interpolated coefficients is judged. Based on the trend state of each second variable, the leak detection correlation function for the corresponding leak detection process is determined.

5. A leak detection device for packaged devices based on intelligent sensors, comprising: A computer-readable storage medium, characterized in that it is used to perform the leak detection method for a packaged device based on a smart sensor as described in any one of claims 1-3.

Images