Control logic analysis method and system for intelligent bonding silver wire

Abstract

The invention relates to the technical field of advanced process control, in particular to a control logic analysis method and system for intelligent bonding of a silver wire, and the method comprises the following steps: obtaining real-time force, speed and temperature data in a silver wire bonding process, and obtaining real-time monitoring data; and according to the real-time monitoring data, obtaining purified monitoring data by eliminating abnormal values and noise. According to the method, dynamic data of force, speed and temperature in the silver wire bonding process are collected in real time, abnormal values and noise are eliminated, purer monitoring data are obtained, and the actual production state can be reflected more accurately in the subsequent parameter analysis process; and based on the purified data, the correlation among the bonding force, the speed and the temperature is analyzed, so that a parameter correlation analysis result is generated, and the rationality of control parameter setting is improved.

Description

Technical Field

[0001] The present invention relates to the field of advanced process control technology, and in particular to a control logic analysis method and system for intelligent silver wire bonding. Background Art

[0002] Advanced Process Control (APC) uses complex automated control technologies and algorithms to monitor, analyze, and optimize industrial production processes in real time, ultimately improving production efficiency, stabilizing product quality, reducing energy consumption, and minimizing scrap. The control logic analysis method used for intelligent silver wire bonding optimizes the analysis and control logic for key parameters in the silver wire bonding process.

[0003] However, existing technologies typically rely solely on single-dimensional parameter monitoring and simple empirical data analysis in practical applications, lacking the ability to explore and identify in-depth correlations between multiple parameters. Furthermore, when faced with the complex dynamics of silver wire bonding, control logic often exhibits delays or distortions, making it difficult to maintain stable bonding results. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a control logic analysis method and system for intelligent silver wire bonding.

[0005] In order to achieve the above-mentioned object, the present invention adopts the following technical solution, which is a control logic analysis method for intelligent silver bonding wire, comprising the following steps:

[0006] Acquiring real-time force, speed, and temperature data during the silver wire bonding process to obtain real-time monitoring data; and obtaining purified monitoring data by removing outliers and noise from the real-time monitoring data;

[0007] Determining the correlation between bonding force, speed, and temperature based on the post-cleaning monitoring data to generate a parameter correlation analysis result; adjusting the adaptive neural fuzzy inference system algorithm parameters according to the parameter correlation analysis result to obtain adjusted algorithm parameters;

[0008] Applying the adjusted algorithm parameters to an adaptive neural fuzzy inference system, processing dynamic changes in the silver wire bonding process through the adaptive neural fuzzy inference system, and generating preliminary bonding parameter optimization results; performing an effectiveness evaluation on the preliminary bonding parameter optimization results to detect adaptability and effectiveness in production, and obtaining an optimization effectiveness evaluation result;

[0009] Based on the optimization performance evaluation result, the control logic of the bonding machine is updated to obtain an updated control logic.

[0010] Preferably, the steps of acquiring the real-time monitoring data are:

[0011] Continuously monitor the force, speed, and temperature parameters of the bonding machine, record the changes in force, speed, and temperature parameters in real time, and obtain the original real-time monitoring data set;

[0012] Based on the original real-time monitoring data set, statistical analysis is applied to calculate the average value and standard deviation to obtain real-time monitoring data.

[0013] Preferably, the steps for obtaining the cleaned monitoring data are:

[0014] Based on the real-time monitoring data, the anomaly score of each data point is calculated using the following formula:

[0015]

[0016] Among them, E i is the abnormality score, X i is the value of a single data point, μ is the arithmetic mean of the data set, σ is the standard deviation of the data set, μ med is the median of the data set, and MAD is the median absolute deviation;

[0017] Based on the anomaly scores, data points whose anomaly scores exceed a preset threshold are eliminated, where the preset threshold is set to the 95% quantile of the anomaly score distribution, to obtain purified monitoring data.

[0018] Preferably, the steps for obtaining the parameter correlation analysis result are:

[0019] Based on the post-cleaning monitoring data, the correlation between bonding force, speed and temperature is calculated using the following formula:

[0020]

[0021] Among them, R ij represents the correlation between parameter i and parameter j, F i (t) and F j (t) represents the values ​​of parameter i and parameter j at time t, and Respectively represent the mean of parameter i and parameter j in the time interval [t1, t2], and the integral range t1 to t2 represents the observation time period. and denote the first-order time derivatives of parameters i and j, respectively. To introduce additional direct interaction terms between parameters;

[0022] Based on the correlation, a parameter correlation matrix is ​​constructed, where each element of the correlation matrix represents the correlation degree between two corresponding parameters, and the diagonal elements in the matrix are fixed to 1, thereby generating a parameter correlation analysis result.

[0023] Preferably, the steps for obtaining the adjusted algorithm parameters are:

[0024] Analyzing the parameter correlation analysis results, evaluating the correlation strength between bonding force, speed, and temperature, identifying parameters that need to be optimized, and generating a priority list of key parameters;

[0025] Based on the priority list of key parameters, refining the algorithm parameter adjustment strategy in the adaptive neuro-fuzzy inference system, including adjusting the thresholds of the fuzzy logic rules and the parameters of the membership functions, to obtain a draft adjustment strategy;

[0026] Based on the draft adjustment strategy, algorithm parameters are adjusted, and the impact of the new parameters on performance is tested in a simulation environment to obtain adjusted algorithm parameters.

[0027] Preferably, the steps for obtaining the preliminary bonding parameter optimization results are:

[0028] Loading the adjusted algorithm parameters into the adaptive neural fuzzy inference system, associating the input variables with the membership functions, and initializing the inference engine to generate the loaded adaptive neural fuzzy inference system;

[0029] Based on the loaded adaptive neural fuzzy inference system, the adjustment amount of the bonding parameter is calculated using the following formula:

[0030]

[0031] Among them, D k represents the adjustment of the bonding parameters at time k, A k is the real-time measurement value of the bonding force, B k is the real-time measurement value of bonding speed, C k The impact factor of temperature change;

[0032] Based on the adjustment amount, the control variables are updated in the adaptive neural fuzzy inference system, the membership functions and inference rules are adjusted in real time, the control strategy in the bonding process is iteratively optimized, and the preliminary bonding parameter optimization results are generated.

[0033] Preferably, the steps for obtaining the optimization performance evaluation result are:

[0034] Based on the preliminary bonding parameter optimization results, the adaptability score in the production process is calculated using the following formula:

[0035]

[0036] Among them, P is the adaptability score, RT i is the response time of parameter adjustment in the i-th bonding cycle,

[0037] To prevent division by zero errors, a small constant, T i is the actual cycle time, T ref is the reference cycle time, N is the number of evaluation cycles;

[0038] Based on the adaptability score, the effect of the bonding process is analyzed and the impact of parameter optimization on actual production efficiency is evaluated. If the score is higher than the preset standard, the parameter optimization is considered successful and the optimization efficiency evaluation result is obtained.

[0039] Preferably, the steps for obtaining the updated control logic are:

[0040] Based on the optimization performance evaluation results, the bonding force, bonding speed, temperature fluctuation range and response time are extracted, and the unstable interval in the bonding process is identified. The instantaneous change rate and deviation of each parameter are calculated to obtain a control logic input parameter set;

[0041] According to the control logic input parameter set, the control logic adjustment factor is calculated, and the expression is:

[0042]

[0043] Among them, L is the control logic adjustment factor, J t is the real-time response value of the bonding force, RF t is the target reference value, V t is the bonding speed, V r is the target bonding speed, TMP t is the current temperature, TMP r To set the temperature;

[0044] Based on the control logic adjustment factor, the bonding parameter range is adjusted, and the feedback correction strategy is reset to obtain an updated control logic.

[0045] The present invention provides a control logic analysis system, comprising:

[0046] Data monitoring module, which monitors the force, speed and temperature parameters during the silver wire bonding process in real time, collects real-time monitoring data, and generates real-time monitoring data results;

[0047] The data purification module receives real-time monitoring data results, eliminates outliers and noise in the data, and generates purified monitoring data;

[0048] The correlation analysis module uses the purified monitoring data to analyze the correlation between bonding force, speed and temperature, and generates parameter correlation analysis results;

[0049] The parameter optimization module adjusts the algorithm parameters of the adaptive neural fuzzy inference system based on the parameter correlation analysis results, obtains the adjusted algorithm parameters, uses the adjusted algorithm parameters to process the dynamic changes of the silver wire bonding process, and generates preliminary bonding parameter optimization results;

[0050] The performance evaluation module receives preliminary bonding parameter optimization results, evaluates adaptability and effects in production, generates optimization performance evaluation results, and updates the control logic of the bonding machine based on the optimization performance evaluation results to obtain updated control logic.

[0051] Compared with the prior art, the advantages and positive effects of the present invention are:

[0052] The present invention collects the dynamic data of force, speed and temperature during the silver wire bonding process in real time, eliminates outliers and noise, and obtains purer monitoring data, so that the subsequent parameter analysis process can more accurately reflect the actual production status; and based on the purified data, analyzes the correlation between bonding force, speed and temperature, and then generates parameter correlation analysis results to improve the rationality of control parameter settings. By dynamically adjusting the relevant parameters of the adaptive neural fuzzy inference algorithm, real-time adaptive processing of changes in the bonding process is achieved, avoiding the parameter setting in the bonding process lagging behind the actual working condition changes, so that the control logic can flexibly fit the on-site production needs. In addition, based on the preliminary optimization results, an optimization performance evaluation is performed to further confirm the actual effect of the parameter adjustment, ensure that the parameter optimization truly improves the stability of the silver wire bonding process, and timely update the control logic of the bonding machine accordingly, so that the silver wire bonding control accuracy is improved, and the product consistency and yield are effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] See also Figure 1 The present invention provides a technical solution for a control logic analysis method for intelligent silver bonding wire, comprising the following steps:

[0056] Acquire real-time force, speed, and temperature data during the silver wire bonding process to obtain real-time monitoring data; based on the real-time monitoring data, remove outliers and noise to obtain purified monitoring data;

[0057] Based on the purified monitoring data, the correlation between the bonding force, speed and temperature is determined, and the parameter correlation analysis results are generated; according to the parameter correlation analysis results, the algorithm parameters of the adaptive neural fuzzy inference system are adjusted to obtain the adjusted algorithm parameters;

[0058] The adjusted algorithm parameters are applied to an adaptive neural fuzzy inference system. The adaptive neural fuzzy inference system processes the dynamic changes in the silver wire bonding process and generates preliminary bonding parameter optimization results. The preliminary bonding parameter optimization results are evaluated to test their adaptability and effectiveness in production, and the optimization performance evaluation results are obtained.

[0059] Based on the optimization performance evaluation result, the control logic of the bonding machine is updated to obtain an updated control logic.

[0060] The steps for obtaining real-time monitoring data are as follows:

[0061] Continuously monitor the force, speed, and temperature parameters of the bonding machine, record the changes in force, speed, and temperature parameters in real time, and obtain the original real-time monitoring data set;

[0062] Based on the original real-time monitoring data set, statistical analysis is applied to calculate the mean value and standard deviation to obtain the real-time monitoring data.

[0063] Specifically, based on the uninterrupted collection of force, speed and temperature parameters in the bonding machine, instantaneous readings of force data, speed data and temperature data are first obtained at a frequency of ten times per second, and before collection, the effective range of each parameter is set according to the statistical results of previous production for comparison. For example, the force range is set to 0N to 80N, the speed range is set to 0mm / s to 500mm / s, and the temperature range is set to 20℃ to 120℃. These interval values ​​are determined by checking the arithmetic mean and variance obtained by historical measurement samples and then combining them with actual production experience. It is assumed that the average force in a certain measurement is about 50N and the standard deviation is about 10N. The upper limit of the effective range can be set to the mean value plus 3 times the standard deviation, that is, 80N. If the force data collected subsequently exceeds 80N, it can be regarded as an abnormality. The speed and temperature ranges are also set in a similar way according to their respective mean values ​​and standard deviations. During the on-site collection stage, the data read each time is compared with the set interval, and the data within the range is recorded while the data outside the range is retained so that the abnormal points can be further identified in the subsequent links. The whole process will continue throughout the production cycle. There is no correction or calculation of the original collected values, and finally the original real-time monitoring data set that has not been statistically processed is obtained.

[0064] Based on the original real-time monitoring data set formed above, a set of force, speed and temperature readings collected every second is selected to form a single observation sample. The arithmetic mean and standard deviation are calculated for all observation samples. For example, when the sample size is N, the sum of the force data is divided by N to obtain the average force value. Then pass The standard deviation of the force data is obtained in this way, and the average value and standard deviation of the speed and temperature are calculated in the same way. If any observed value is found to deviate significantly from the estimated range of the corresponding parameter, it will be checked. During the check, it will be compared with the previously determined force range of 0N to 80N, speed of 0mm / s to 500mm / s, and temperature of 20℃ to 120℃ to determine whether the reading is in a reasonable range and whether subsequent processing is required. The upper and lower limits of the interval selected here mainly refer to historical statistical values ​​such as the average force value of 50N and the standard deviation of 10N, the measurement results of the average speed of 250mm / s and the standard deviation of 50mm / s, and the empirical results of the average temperature of 70℃ and the standard deviation of 10℃. Finally, the average value and standard deviation of all observed samples are summarized as real-time monitoring data that can be directly called.

[0065] The steps for obtaining the cleansed monitoring data are as follows:

[0066] Based on real-time monitoring data, the anomaly score of each data point is calculated using the following formula:

[0067]

[0068] Among them, E i is the abnormality score, X i is the value of a single data point, μ is the arithmetic mean of the data set, σ is the standard deviation of the data set, μ med is the median of the data set, and MAD is the median absolute deviation;

[0069] Based on the anomaly score, data points whose anomaly scores exceed the preset threshold are eliminated. The preset threshold is set to the 95% quantile of the anomaly score distribution to obtain the purified monitoring data.

[0070] Specifically, X i Parameter introduction: X i Represents the value of a single data point at a certain moment or within a certain collection period. In actual production monitoring, this value is usually directly read and stored after measurement by a force sensor, speed meter or temperature detector. Several force, speed or temperature readings are accumulated every minute, and these values ​​are arranged in sequence to obtain {X1, X2,…, X N} observation sequence, and then the i-th observation value is recorded as X i For example, if the 256th data recorded in a batch of production is 47.3, then X 256=47.3, in this way a complete input sequence is prepared for the formula.

[0071] μ parameter introduction: μ represents the arithmetic mean of the data set, which is calculated by N The sum is calculated by adding and dividing by the total number of samples N. If the data scale is large, the sum can be calculated in segments at the hardware level first, and then divided by N after accumulation in the storage link. For example, if N = 1200 data are collected, all X i Add to get the cumulative sum Dividing this result by 1200 gives In a specific example, if S = 57600, then

[0072] σ parameter introduction: σ represents the standard deviation of the data set, which is Obtained, used to measure the degree of dispersion of data around the mean value μ, read all observations X1 to X N And subtract μ respectively, square the difference and add it up, finally divide it by N and take the square root to get σ. If N = 1200 in the specific example, the accumulated square difference but

[0073] μ med Parameter introduction: μ med To represent the median of the data set, we need to first calculate {X1,X2,…,X N} Sort in ascending order. If N is an odd number, select the (N+1) / 2th number. If N is an even number, take the average of the N / 2th and (N / 2)+1th values. For example, when N=1200, after sorting, take the average between the 600th and 601th values ​​to get μ med , if the 600th number after sorting is 47.1 and the 601st number is 47.3, then μ med =47.2.

[0074] MAD parameter introduction: MAD represents the median absolute deviation of the data set. It is necessary to convert the original data {X1, X2, ..., X N Each X in i and the median μ med Find the difference and take the absolute value, then median these absolute values ​​again to get the MAD. If we continue with the example of 1200 observations above, the μ obtained after sorting is med =47.2, then calculate all |X i -47.2| forms a new array and sorts the array in ascending order. If the final values ​​corresponding to positions 600 and 601 are 2.3 and 2.5 respectively, then MAD = 2.4.

[0075] Calculation process: for the specific i-th data point X i , first calculate its relationship with μ and μ med The deviations are (X i -μ) and (X i -μ med ), and then use σ and MAD to normalize these two items respectively, and we get and Square them separately, sum them up, and finally take the square root to get E i , where the arithmetic mean μ and standard deviation σ mainly reflect the degree of dispersion of the data to the mean, while the median μ med and MAD highlight the robust evaluation of abnormal distributions.

[0076] The results show that by fully considering the center position of the mean and median from two different perspectives, and combining the two discrete measurement methods of standard deviation and median absolute deviation, a more comprehensive quantitative measurement of the single observation value and the overall distribution position can be formed. i The larger the value, the more significant the difference between the observed value and the overall distribution of the data, and the more significant the difference between the observed value and the overall distribution of the data, the more significant the difference between the observed value and the overall distribution of the data. i Observations above the threshold are removed to obtain a relatively pure monitoring data set.

[0077] Based on the abnormal score sequence calculated previously, a score value list is formed for each data point. During the execution process, the score list is first sorted in ascending order, and the data in it is numbered according to the score size. When all the data are collected, the total amount is counted. In this process, it is necessary to combine all the abnormal scores generated previously to find the quantiles. By traversing the ordered list from the minimum score to the maximum score one by one, the cumulative proportion of the score value in the total number is used as the coordinate to find the score value near the position of about 0.95, and mark it as the elimination threshold. The source of this value refers to the outlier rate analysis process obtained from the statistics of a large number of production batches in the past, that is, the abnormal scores of multiple periods are extracted from the historical monitoring records and their distribution characteristics are calculated respectively, and the lower and upper bounds of the score distribution of all batches are used. The score value at the 95% percentile is used as the basis for comparison by cross-validation, which can usually cover most of the range of the current data in the real production environment. Observations outside this deviation interval rarely appear and are usually accompanied by equipment failure or serious sensor inaccuracy. This percentile can be set as the threshold that needs to be excluded. For example, if the score value corresponding to the 380th score after sorting is 2.6 among 400 scores, 2.6 will be used as the final threshold and recorded. If a score greater than 2.6 is actually monitored, such data will be identified as abnormal, and all data points identified as abnormal will be uniformly saved in a separate sequence. The remaining data points with scores of 2.6 or below will be summarized as valid samples after purification, so that subsequent steps can obtain this part of the more stable data structure, and finally the purified monitoring data will be obtained.

[0078] The steps to obtain the parameter correlation analysis results are:

[0079] Based on the post-cleaning monitoring data, the correlation between bonding force, speed and temperature is calculated using the following formula:

[0080]

[0081] Among them, R ij represents the correlation between parameter i and parameter j, F i (t) and F j (t) represents the values ​​of parameter i and parameter j at time t, and Respectively represent the mean of parameter i and parameter j in the time interval [t1, t2], and the integral range t1 to t2 represents the observation time period. and denote the first-order time derivatives of parameters i and j, respectively. To introduce additional direct interaction terms between parameters;

[0082] Based on the correlation, a parameter correlation matrix is ​​constructed. Each element of the correlation matrix represents the correlation degree between the corresponding two parameters. The diagonal elements in the matrix are fixed to 1 to generate the parameter correlation analysis results.

[0083] Specifically, the benefit of the formula is that by simultaneously considering the centralization term, derivative interaction term, and direct interaction term in the numerator, it can make a multi-level measurement of the correlation between the two parameters from the three aspects of overall trend, dynamic change, and instantaneous coupling, and normalize the two parameters in the denominator in the form of the integral of their respective variances to obtain a comprehensive correlation that has both time series correlation and instantaneous coupling characteristics.

[0084] F i (t) The steps to obtain the parameters are: F i (t) represents the value of the i-th physical quantity at time t. For example, when i=1, it can represent the bonding force. The applied force is detected by installing a sensor on the bonding machine. The corresponding instantaneous readings are collected and recorded once per second. The data at continuous moments can be regarded as a function F1(t). When recording, the physical quantity needs to be matched with the timestamp. The specific value acquisition process is: the k-th collection at time t k When the force sensor reading F1(t k ), if the observation interval is from 0 seconds to 10 seconds, a total of 11 groups of force readings are collected, then {F1(0), F1(1), F1(2), …, F1(10)} is formed. For example, {40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60} can be obtained in actual detection. These data will be brought into the following integration process and combined with other parameters.

[0085] F j (t) The steps to obtain the parameters are: F j (t) represents the difference between F i (t) is the value of another physical quantity corresponding to the bonding force at time t. If i = 1 corresponds to the bonding force, then j = 2 corresponds to the bonding speed. The speed detector collects and records the data at the same time t. For example, the data is also collected in the interval from 0 seconds to 10 seconds.

[0086] {60, 63, 66, 69, 72, 75, 78, 81, 84, 87, 90} as F2(t).

[0087] The steps to obtain the parameters are: It represents the average value of parameter i in the time interval [t1, t2], which is calculated as follows: When i = 1 represents the bond force, you can perform numerical integration over the 0 to 10 second interval or use the trapezoidal method to approximate the integral value for discrete points and then divide by 10 to get the average force. For example, if but

[0088] The steps to obtain the parameters are: It represents the average value of parameter j in the time interval [t1, t2]. The acquisition steps are the same as Consistent, perform numerical integration or equivalent summation on the measured data such as speed or temperature and then divide by (t2-t1). If, in the example of force and speed, the speed data F2(t) has been recorded and summed from 0 to 10 seconds by the discrete method and multiplied by the step size of 1 second between adjacent sampling points, if the result is but

[0089] dF i (t) / dt and dF j The steps to obtain the (t) / dt parameter are: i (t) / dt and dF j (t) / dt represents the first-order time derivative of parameters i and j at time t, and the derivative value at the current moment is estimated between adjacent sampling points by discrete difference method, which can be written as When k traverses from 0 to 9, a series of derivative values ​​can be obtained. For example, in the case of force data {40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60}, the time step is 1 second, and the derivative can be approximated as {2, 2, 2, 2, 2, 2, 2, 2, 2}. Similarly, the derivative of velocity {60, 63, 66, 69, 72, 75, 78, 81, 84, 87, 90} can also be obtained in the same way as {3, 3, 3, 3, 3, 3, 3, 3}.

[0090] The steps for obtaining the t1 and t2 parameters are as follows: t1 and t2 define the start and end time of the observation. For example, in a certain production link, 0 seconds is the initial collection point and 10 seconds is the end collection point, so t1 = 0 and t2 = 10.

[0091] Calculation process:

[0092] Take the force parameter F1(t) = 40 + 2t and the speed parameter F2(t) = 60 + 3t as an example, let the time interval [t1, t2] be [0, 10],

[0093]

[0094]

[0095] Molecular part:

[0096]

[0097] The sum of the numerators = 500 + 60 + 38000 = 38560;

[0098] Denominator:

[0099]

[0100] in:

[0101]

[0102] Therefore

[0103] and then

[0104] The results show that the two parameters have a high correlation in the defined force and velocity example. By integrating the centralization offset, derivative interaction and direct product terms in the numerator and normalizing the variance integral in the denominator, a correlation value greater than 1 is finally obtained. In this step, when R ij The larger the value, the more significant the coupling between the two physical quantities. With the help of this calculation, the correlation between force, speed and temperature can be determined, and a reference can be provided for subsequent parameter optimization or control strategy.

[0105] Based on the various R ij During the execution process, the correlation between force, speed and temperature needs to be counted in a table, so as to calculate the R values ​​corresponding to the three groups of parameters: force and speed, force and temperature, and speed and temperature. 12 、R 13 and R 23 The force is obtained according to the aforementioned collection method within the period of 0 seconds to 10 seconds, and an observation sequence of the interval is formed. The speed and temperature are also collected and sorted according to the same time node. Then, for each pair of parameters, the corresponding correlation is calculated using the aforementioned integral operation. If the statistical value is large, it means that the relationship between force and speed or other parameters is closer. If the statistical value is small, it means that the correlation is weak. During the recording process, the threshold reference range formed in the previous article is compared, for example, 0 to 5 represents low correlation, 5 to 10 represents moderate correlation, and 10 and above represent a relatively higher correlation level. This range is an empirical threshold formed after referring to multiple production operation data and making classified statistics on the correlation between force, speed and temperature of each batch. For example, when the R after the force and speed operation is calculated, 12 The calculated value of 12.04 falls within the range of 10 and above. Therefore, when establishing the correlation matrix, the rows and columns of the matrix represent different parameters, the diagonal positions are marked as 1, and the corresponding R is filled in the remaining positions. ijnumerical values, so that each matrix element can intuitively represent the correlation degree of the corresponding two parameters in the time interval, and finally generate a complete parameter correlation analysis result.

[0106] The steps to obtain the adjusted algorithm parameters are:

[0107] Analyze the parameter correlation analysis results, evaluate the correlation strength between bonding force, speed and temperature, identify parameters that need to be optimized, and generate a priority list of key parameters;

[0108] Based on the priority list of key parameters, the algorithm parameter adjustment strategy in the adaptive neuro-fuzzy inference system is refined, including adjusting the thresholds of fuzzy logic rules and the parameters of membership functions, and a draft adjustment strategy is obtained;

[0109] Based on the draft adjustment strategy, the algorithm parameters are adjusted, the impact of the new parameters on the performance is tested in a simulation environment, and the adjusted algorithm parameters are obtained.

[0110] Specifically, according to the parameter correlation analysis results obtained previously and the corresponding values ​​between the three groups of parameters of force, speed and temperature, first combine these correlation values ​​to compare with the segmentation standards established in the internal historical records, for example, a correlation greater than 10 is considered a strong correlation, 5 to 10 is considered a moderate correlation, and less than 5 is considered a relatively low correlation. This segmentation standard is formed and summarized through the accumulation of observations from multiple production experiments in the early stage. Then, the correlation between force and speed, force and temperature, and speed and temperature are analyzed item by item. During the analysis, the correlation values ​​of each group of parameters are first summarized to determine the level of interaction between the three, and the groups with higher values ​​are marked as the parts that need the most attention, and then refer to the force data generated previously. , speed, and temperature sampling records, search for the measurement conditions of each parameter in different production stages, compare the search results with the degree of correlation, if the fluctuation synchronization of force and speed is greater within the same time period, it means that the correlation between the two is more prominent, on the contrary, if the value is smaller, it will be ranked at the back first, and then based on these comparison results, summarize the parameter groups that need to be optimized, and then select the single variables with more significant influence in each group of parameters and rank them in a higher priority position, for example, when the correlation between force and speed is the largest, force and speed are listed at the front of the list at the same time, and if the temperature has a lower correlation with the other two, it is ranked at the back. After completing the grouping and sorting, a priority list of key parameters is generated.

[0111] Based on the previously obtained priority list of key parameters, the bond force and speed, which are the highest priority in the list, are first examined. During execution, these two parameters are selected as the key adjustments for the adaptive neuro-fuzzy inference system. Specifically, the preconditions and subsequent inference actions of the current fuzzy logic rules are listed one by one. Then, specific numerical ranges for the membership functions are set. For example, the force range is divided into five segments within the range of 0N to 80N, and the speed range is divided into five segments from 0mm / s to 500mm / s. The size and number of these intervals are determined based on statistical data collected from previous sampling in the same production environment. If the boundaries of the membership functions need to be adjusted, the position of the connection points can be fine-tuned within the range, and the fuzzy logic rules can be configured to trigger offset revisions. The offset revision value can be assigned different degrees of revision based on the correlation degree obtained previously. If the force exceeds 70N or the speed exceeds 450mm / s after revision, further restrictions are required based on the production safety warning range. The peak area of ​​the membership function is set according to actual needs. Once all adjustments are summarized in text and numerical form, a draft adjustment strategy for the key parameters is obtained.

[0112] Based on the draft adjustment strategy obtained above, the algorithm parameters of the corresponding adaptive neuro-fuzzy inference system are updated during execution according to the force and velocity priority order. First, the newly set thresholds and the values ​​of each membership function segment are loaded into the internal program. Then, a set of input data similar to real production conditions is established in the simulation environment. For example, different force and velocity combinations are entered in multiple batches. For each batch, several benchmark values ​​are selected within the force range of 0N to 80N, and corresponding benchmark values ​​are selected within the speed range of 0mm / s to 500mm / s. These two are paired to generate multiple sets of test data. Each set of data is dynamically inferred using the existing system logic. The resulting output values ​​are observed and the stability and amplitude changes of these output values ​​at each stage are recorded. If the force and velocity coordination in some batches is found to exceed the preset range, the corresponding offset correction amount is gradually increased or decreased in the simulation environment. After several rounds of iteration and recording, the algorithm parameters that can maintain numerical balance under different pairing scenarios are obtained. Finally, the numerical outputs of all test batches are summarized and compared to obtain the adjusted algorithm parameters.

[0113] The steps to obtain the preliminary bonding parameter optimization results are:

[0114] Loading the adjusted algorithm parameters into the adaptive neuro-fuzzy inference system, associating the input variables with the membership functions, and initializing the inference engine to generate the loaded adaptive neuro-fuzzy inference system;

[0115] Based on the loaded adaptive neuro-fuzzy inference system, the adjustment amount of the bonding parameters is calculated using the following formula:

[0116]

[0117] Among them, D k represents the adjustment of the bonding parameters at time k, A k is the real-time measurement value of the bonding force, B k is the real-time measurement value of bonding speed, C k The impact factor of temperature change;

[0118] Based on the adjustment amount, the control variables are updated in the adaptive neuro-fuzzy inference system, the membership functions and inference rules are adjusted in real time, the control strategy in the bonding process is iteratively optimized, and the preliminary bonding parameter optimization results are generated.

[0119] Specifically, load the adjusted algorithm parameters into the adaptive neural fuzzy inference system. First, specify the corresponding name and initial value range for each parameter in the system. Disassemble the content of the previously obtained algorithm parameter file and extract the quantitative indicators related to force, speed and temperature. Read them one by one in the order listed in the design document and place them in the internal variable buffer. In this process, it is also necessary to compare the force range. Compare each record with the pre-set 0N to 80N range to confirm that all its records are within the detectable range. The speed is also compared with the 0mm / s to 500mm / s range in the same way, and the temperature is compared with the 20℃ to 120℃ range. If an abnormal value exceeds the range, it is necessary to call the abnormal rejection strategy obtained in the previous step, mark the abnormal data and skip the subsequent steps. When all parameters are read, the corresponding input variable list is established in the internal structure, and these lists are compared with the previously configured Good membership functions are matched accordingly. For example, force may be divided into several semantic ranges, representing lower segments, middle segments and higher segments respectively. The specific numerical boundaries of these segments are determined based on the statistical analysis of historical production data. Several segments are also set in the division of speed and temperature. Then the initialization process of the inference engine is started. During the initialization process, each membership function is checked in turn to see if it is in a usable state. If the definition interval of a membership function is missing or its parameters do not meet the basic conditions, it needs to be corrected to a range that is consistent with the current production requirements through calibration logic. The calibration logic includes repeated detection of the left and right endpoints of each membership function and determination of whether there is a numerical conflict. Finally, an overall check is performed on all loading and matching results. Once the check is correct, the loaded force, speed, temperature and its membership function are registered and enter the operational state, thereby generating a loaded adaptive neuro-fuzzy inference system within the system.

[0120] The benefit of the formula lies in that it compares the difference between force and speed through the numerator and introduces the temperature factor to correct the denominator. At the same time, a cube root fraction containing the product and temperature is added to the second half. This can take into account the balance between force and speed in absolute and relative values, and incorporate the influencing factor corresponding to temperature into the cube root operation, so that the adjustment amount calculated in real time can better reflect the joint effect of multiple parameters.

[0121] A k The steps to obtain the parameters are: A k It represents the bonding force recorded at time k. Its value is usually in the actual range of 0N to 80N. For example, when k=5, the data obtained by reading the force sensor is 65.4N. After verifying that 65.4N is in the range of 0N to 80N, A5=65.4 is set. In this way, the bonding force value at the 5th second is obtained. k All are obtained in this form.

[0122] B k The steps to obtain the parameters are: B k Indicates the bonding speed at time k. The speed calculation process can be based on the acquisition of the ratio of motion displacement to time or obtained through the pulse counting method of a digital encoder. For example, in a production cycle, the speed value detected at the 5th second is 256.7 mm / s, which is within the set range, then B5 = 256.7.

[0123] C k The steps to obtain the parameters are: C k Indicates the impact factor brought by temperature change. First, the temperature sensor obtains the temperature T of the environment or contact surface at time k. k The temperature range is generally set at 20℃ to 120℃. After acquisition, it is converted into C through a quantitative expression. k , which can be written as C k =α×(T k -T min ), where T min is the lowest temperature reference, α is the conversion coefficient set to map the temperature difference to the same order of magnitude as force and speed. α needs to be adjusted based on previous measurement data to match the actual production situation. For example, if α = 0.2, T min =20℃, and the current detection temperature T k =60℃, then C k =0.2×(60-20)=8.

[0124] Calculation process:

[0125] Taking the example of time k=5, the force sensor reading A5=65.4, the speed monitoring value B5=256.7, and the temperature influence factor C5=8, substitute it into the formula

[0126] First calculate the numerator |65.4-256.7|=191.3;

[0127] Then calculate the denominator Then |65.4+256.7|=322.1, so So the first part of the denominator

[0128] Divide the numerator and denominator to get the first term:

[0129] Then calculate the second term: |65.4×256.7|=16780.18, add 8 to get 16788.18, and then take Therefore

[0130] Finally, D5≈67.57+0.0389=67.6089≈67.61; the result shows that at the time of 5 seconds, due to the large difference between force and speed and the moderate level of temperature influence factor, the obtained D5 value is relatively high. In the subsequent steps, instructions can be issued to the control system to further adjust the membership function and inference rules based on D5. If this value continues to increase, it means that the gap between force and speed continues to widen or the temperature factor increases. It is necessary to reduce the excessive fluctuation of force or speed from the source. If it decreases, it means that the parameter change has stabilized. When D is calculated at every moment in the whole production process, k When the system is running, it can continuously make dynamic corrections based on these values ​​to achieve the effect of real-time updating of control variables.

[0131] Based on the adjustment amount obtained previously, the control variables are updated in the adaptive neuro-fuzzy inference system and the membership function and inference rules are continuously revised. During the execution process, the force, speed, and temperature influencing factors at each acquisition moment are first compared. For example, the force value is between 0N and 80N, and the speed is between 0mm / s and 500mm / s. These numerical ranges are obtained from the previous statistical analysis and verified many times. When the real-time reading of force or speed is close to the preset upper limit, the system determines its offset amplitude based on the adjustment amount. If it exceeds the empirical threshold set near 70N or 450mm / s, the revision of the membership function is increased in the current iteration round. For example, the force range originally divided into the "medium and high" segment is fine-tuned to the high segment or the speed is subdivided, so that the inference process is more sensitive to force or speed changes at subsequent moments. Similarly, if the temperature When the change in the impact factor is greater than the reference value of 5, the inference rule linked to the temperature is contracted or expanded. This contraction or expansion refers to directly changing the upper and lower boundaries of the rule triggering conditions, and dynamically writing them into the control variables to coordinate with other parameters. After one round of iteration is completed, the system will record the force, speed, and temperature at the current moment, and then conduct a comprehensive comparison based on the previously calculated adjustment amount to determine whether the revised membership function and rule threshold at the current moment are reasonable. If the force and speed changes in some intervals are still within an acceptable range but the temperature change has exceeded the originally set interval, a second or more rounds of iterations are required until all conditions are finally met and the intermediate adjustment of this stage is completed. In this way, the bonding process at subsequent moments can be executed through the updated control variables, and finally a preliminary bonding parameter optimization result is generated.

[0132] The steps to obtain the optimization performance evaluation results are as follows:

[0133] Based on the preliminary bonding parameter optimization results, the adaptability score in the production process is calculated using the following formula:

[0134]

[0135] Among them, P is the adaptability score, RT i is the response time of parameter adjustment in the i-th bonding cycle,

[0136] To prevent division by zero errors, a small constant, T i is the actual cycle time, T ref is the reference cycle time, N is the number of evaluation cycles;

[0137] Based on the adaptability score, the effect of the bonding process is analyzed and the impact of parameter optimization on actual production efficiency is evaluated. If the score is higher than the preset standard, the parameter optimization is considered successful and the optimization efficiency evaluation result is obtained.

[0138] Specifically, the formula is beneficial in that it comprehensively considers the coupling relationship between response time and cycle time, quantifies the response time of each bonding cycle and the production cycle ratio, and superimposes 1 / RT in the molecule. i +∈with(T i / T ref ) 2 The product of makes both shorter response time and reasonable cycle bring higher scores, while the denominator N plays an averaging role, which can make a unified evaluation of the entire production batch.

[0139] RT i The steps to obtain the parameters are: RT i It represents the response time required for the parameters to be adjusted during the i-th bonding cycle. This is usually quantified by the built-in timing function during the execution of the device. The specific method is to put a start mark before each round of bonding. When the system finds that the parameters such as force or speed exceed the specified range, it starts to record the response timing. When the relevant parameters are adjusted and restored to the acceptable range through the fuzzy inference rules mentioned above, the timing ends and RT is obtained. i If a measurement is taken for 6.2 seconds, then RT i =6.2.

[0140] The steps for obtaining the ∈ parameter are as follows: ∈ is a small constant to prevent division by zero errors. For example, if ∈=0.001 and it is found to be sufficient to cope with the shortest response time of 0.002, this value can be retained to ensure the safety of the operation.

[0141] T i The steps to obtain the parameters are: T i represents the actual total time of the i-th bonding cycle. The equipment starts timing when each bonding round starts and stops timing when the round ends. If the entire process of matching the bonding force and speed and forming the solder joint takes 4.6 seconds in a certain measurement, then T i =4.6.

[0142] T ref The steps to obtain the parameters are: T ref Indicates the reference cycle time, which can generally be obtained from historical average time or industry standards. For example, after testing multiple batches of similar production equipment, it is found that the average time to complete a bond under most normal working conditions is about 4.5 seconds. In this case, T ref =4.5.

[0143] The steps for obtaining the N parameter are as follows: N is the number of evaluation cycles, which is used to indicate the total number of cycles that need to be included in the calculation. In a certain production link or a specific time window, the number of bonding cycles that have occurred is confirmed by counting. For example, if 10 cycles are confirmed after 10 consecutive processings, then N = 10.

[0144] Calculation process:

[0145] For example, in a certain production, N=5 bonding cycles are completed, and the response time series {RT1, RT2

[0146] ,RT3,RT4,RT5}={6.2,4.7,5.0,5.3,4.9}, take∈=0.001, and record the total time of the cycle {T1,T2

[0147] ,T3,T4,T5}={4.6,4.9,4.5,5.2,4.8}, and set the reference period T ref =4.5;

[0148] Molecular part:

[0149] First calculate separately (4.6 / 4.5) 2 ≈1.0237,(4.9 / 4.5) 2 ≈1.1852,(4.5 / 4.5) 2 =1,(5.2 / 4.5) 2 ≈1.3378,(4.8 / 4.5) 2 ≈1.1378;

[0150] Then calculate the denominator (RT1+0.001)=6.201, etc., and take the reciprocal to get: {1 / 6.201, 1 / 4.701, 1 / 5.001, 1 / 5.301, 1 / 4.901}≈{0.1613, 0.2127, 0.19996, 0.1885, 0.2040};

[0151] Multiply the above correspondences and add them together:

[0152] 0.1613×1.0237+0.2127×1.1852+0.19996×1+0.1885×1.3378+0.2040×1.1378≈0.1652+0.2524+0.19996+0.2522+0.2320

[0153] =1.10176;

[0154] The denominator is N=5;

[0155] therefore The results show that in these five cycles, after comprehensively measuring the response time and actual cycle time combined with the reference cycle of 4.5 seconds, the system obtained an adaptability score of approximately 0.220352. When this value is higher than the pre-set score threshold, such as 0.2, it indicates that the adjustment response within these cycles is relatively timely and the cycle time is close to the reference standard. On the contrary, if it is lower than 0.2, it means that the response in some cycles is too slow or the cycle is too long. Subsequently, the control strategy can be adjusted or other parameters can be optimized based on this value.

[0156] Based on the adaptability score obtained above, during the execution process, the timing data of each bonding cycle is first listed together with the score value obtained above, for example, the RT of each time is recorded. i and T i If the score exceeds the pre-set standard of 0.2, this part of the record is marked as a qualified cycle and prioritized in the summary. If the score is found to be below 0.2, it is marked as a cycle that needs to be reviewed again, and the system adjustments mentioned above are used to backtrack the force, speed, and temperature factors. Specifically, after each cycle, the data exceeding or falling below the threshold are placed in two lists. The force and speed distributions during the same period are compared to see if there are frequent jumps. At the same time, the temperature is checked for sudden increases. If the inspection finds a single force jump exceeding 20N or a sudden increase in speed exceeding 100mm / s, it can be determined that there is still room for optimization in this cycle. Then, this information is used in the system control interface to pre-select the parameters for the next production link. The priority list of parameters to be adjusted is handed over to the operator for final review. This allows the reasons for the current poor evaluation to be identified one by one, and the parameters of the qualified cycle can be compared with the reference values ​​to see if the bonding cycle can be further reduced. After the data of all cycles are verified, the marked lists are summarized to form the final optimization performance evaluation results.

[0157] The steps to obtain the updated control logic are:

[0158] Based on the optimization performance evaluation results, the bonding force, bonding speed, temperature fluctuation range and response time are extracted, and the unstable intervals in the bonding process are identified. The instantaneous change rate and deviation of each parameter are calculated to obtain the control logic input parameter set.

[0159] According to the control logic input parameter set, the control logic adjustment factor is calculated, and the expression is:

[0160]

[0161] Among them, L is the control logic adjustment factor, J t is the real-time response value of the bonding force, RF t is the target reference value, V t is the bonding speed, V ris the target bonding speed, TMP t is the current temperature, TMP r To set the temperature;

[0162] Based on the control logic adjustment factor, the bonding parameter range is adjusted, and the feedback correction strategy is reset to obtain an updated control logic.

[0163] Specifically, based on the optimization performance evaluation results obtained above, three key information, namely force, speed and temperature change range, are selected and fully summarized in combination with the corresponding response time. During the implementation process, the force collection record is first read, and each force record is compared with the effective range of 0N to 80N and those out of range are marked for subsequent inspection. Then the speed collection record is read and compared with the interval of 0mm / s to 500mm / s. Then the temperature data is read and compared with the interval of 20℃ to 120℃. For periods with more frequent temperature fluctuations, they are marked as higher fluctuation segments. Then, these types of data are retrieved to see if there are large deviations in the same time period. If the force deviation exceeds 10N or the instantaneous speed change rate is greater than 60mm / s / s in the same time period, and the temperature fluctuation exceeds the set empirical threshold of 3℃, then the time period is marked as an unstable interval. The unstable interval often corresponds to the entire production cycle. Sudden interference or equipment performance fluctuations require additional analysis of their instantaneous rate of change and deviation. To obtain the instantaneous rate of change, the ratio of the difference between the parameters at adjacent sampling moments to the time interval is calculated point by point. For example, the instantaneous rate of change of force can be obtained by dividing the difference in the force values ​​at adjacent moments by the sampling time. If this value is higher than the standard set at around 15N / s, it is considered that the force changes drastically. The instantaneous rate of change of speed is calculated by calculating the increase or decrease in speed per unit time. If it is higher than 80mm / s / s, it means that the acceleration is too fast during this period. Similarly, if the temperature increases or decreases by more than 2℃ / s within the temperature comparison range, it is considered a serious fluctuation. Finally, these instantaneous rate of change and deviation are summarized as the control logic input parameter set, and the average deviation of each unstable interval is recorded in a table in digital form. After sorting all intervals in sequence, the force, speed, and temperature fluctuation ranges and corresponding response times can be integrated into a complete input parameter set.

[0164] The benefit of this formula lies in that it comprehensively considers the maximum value between the integral error and dynamic rate of change of force and velocity relative to the target reference, and adds correction terms for temperature and its rate of change. This allows the control logic to account for the instantaneous impact of temperature while addressing force and velocity deviations. Through this multi-dimensional coupling design, the final adjustment range can be determined based on the degree of deviation and change rate of different physical quantities, preventing excessive fluctuations in any single item.

[0165] J t The steps to obtain the parameters are: tThe real-time response value of the bonding force is obtained by installing a force sensor on the bonding machine and recording the instantaneous data of the force at a fixed sampling interval. If the sampling is performed with a step length of 1 second from 0 to 10 seconds, {J0, J1, …, J 10}.

[0166] RF t The steps to obtain the parameters are: RF t It represents the target reference value, which is generally given by the process planning and is set by comprehensively evaluating the average force demand and product quality requirements of typical production batches. The reference value will change or remain constant over time, depending on the process requirements at different stages. If the process goal is to make the force rise steadily in the initial stage, then RF t It can be composed of a numerical sequence, for example, starting from 40N and increasing at 2N per second, and then leveling off at 60N after 10 seconds. These interval information are loaded into the software as a time-force comparison, and when reading, the RF value can be found in each second. t What is it? For example, if the reference force should be 52N at the 6th second, then RF6=52.

[0167] V t The steps to obtain the parameters are: V t Indicates the real-time acquisition value of the bonding speed, ranging from 0 mm / s to 500 mm / s.

[0168] V r The steps to obtain the parameters are: V r is the target bonding speed, same as the RF mentioned above t Similarly, this value is given by the process standard or the historical optimal solution. Assuming that the first 5 seconds require a medium speed and the next 5 seconds need to accelerate to a higher speed, then at the t second, {V r}sequence to know the target speed at that time. For example, the previous production data shows that the optimal speed should be maintained at around 300mm / s within a certain period of time. Then {V r} will be segmented in this area. When formally executed, the actual speed V t The difference from 300 mm / s will be reflected in the formula for operations such as integration and square root, and will affect the subsequent maximum derivative comparison. If the speed deviation is large, it will cause the final L to increase, thereby prompting the system to make corresponding revisions.

[0169] DJ t / dt,dR t / dt,dV t / dt,dV r The steps to obtain the / dt parameters are as follows: These derivatives represent the first-order rate of change of the bond force, target reference value, speed, and target speed in the time dimension. The calculation is usually done by differential method, with For example, if the force difference between the 4th second and the 5th second is 2N and the time difference is 1 second, then dJ t / dt≈2N / s, if a target speed sequence increases from 300mm / s to 305mm / s within 1 second, then dV r / dt=5mm / s 2 , complete collection can obtain multiple time derivative values ​​of these four quantities. In the formula, we need to find their absolute values ​​and then take the maximum.

[0170] TMP t and TMP r The steps to obtain the parameters are: TMP t is the current temperature, TMP r To set the temperature, usually in the range of 20℃ to 120℃, if a specific temperature range is required to achieve reliable bonding during the production period, the TMP will be set before the equipment is started. r Load the system, and the hardware temperature measuring element records TMP every second t , and check its fluctuation amount, then |TMP t -TMP r |Incorporate into the formula and also obtain dTMP t / dt to add 1+|dTMP to the denominator t / dt|, if the temperature at a certain moment is 65°C and the set temperature is 70°C, then |65-70|=5.

[0171] Calculation process:

[0172] For example, in a production stage, within the time interval [0,5] seconds, data is collected once per second, and the force and target force are recorded as J t and RF t , the speed and target speed are recorded as V t and V r , the temperature is the same as the set temperature TMP t With TMP r , first calculate:

[0173]

[0174] And take the square root of the sum, assuming the sum = 50, then

[0175] Then calculate the maximum value of the derivative, for example |dJ t / dt|max=5N / s,|dV t / dt|max=30mm / s^{2},|dR t / dt|max=2N / s,|dV r / dt|max=10mm / s^{2}, then max(…)=30;

[0176] Therefore, the first part of the numerator is ≈ 7.071, and the denominator is (30+1)=31;

[0177] Then the temperature difference: |TMP t -TMP r |, if the current temperature difference = 5°C, and |dTMP t / dt|=2℃ / s, then the denominator=(1+2)=3, this term=5 / 3≈1.667;

[0178] therefore The results show that within this 5-second time domain, the cumulative deviation values ​​of force and velocity are squared and summed with the maximum value of the derivative to form a denominator correction. Adding the contribution of temperature difference, we finally get an L of about 1.895. If this value exceeds the internal experience setting of 2.0, the bonding range can be redivided or the temperature adjustment space can be expanded when it is included in the upper-level process. If it is lower than a certain benchmark, it means that the fluctuations of various system parameters are small.

[0179] Based on the control logic adjustment factor obtained above, the value of the factor is first compared with the force and speed range measured previously to confirm whether the current force range of 0N to 80N still has enough margin, and whether the speed range of 0mm / s to 500mm / s meets the production requirements. If it is found that the force is often close to 70N in some unstable ranges and accompanied by obvious speed fluctuations, the force and speed usage range will be appropriately narrowed or widened according to the adjustment factor obtained above. For example, if the adjustment factor exceeds 2.5 and the force has exceeded 60N for many times in multiple records, the maximum force range can be limited to 75N before the start of a new round of production tasks to avoid excessive force increase. If the speed also deviates, the force and speed range will be appropriately narrowed or widened according to the adjustment factor obtained above. If the cumulative difference exceeds 400mm / s for dozens of times, the speed limit is set accordingly, and the temperature control range is recalibrated from certain ranges of 20℃ to 120℃, and the feedback correction strategy for these ranges is updated in the system. For example, an additional reference value is set for the temperature feedback to prevent the temperature from rising too quickly. If it is detected that the temperature fluctuation has decreased during this process, the original range can be maintained unchanged. After all ranges are updated, the revision is recorded to form an updated control logic, which is used as the basis for executing the bonding action in the next round of production. In this way, the previous control logic and the super-threshold changes detected this time are combined, and finally an updated logic that can adapt to the current production conditions is obtained.

[0180] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A control logic analysis method for intelligent silver bonding wire, characterized in that: The following steps are involved: Acquiring real-time force, speed, and temperature data during the silver wire bonding process to obtain real-time monitoring data; and obtaining purified monitoring data by removing outliers and noise from the real-time monitoring data; determining, based on the post-cleaning monitoring data, the correlation between bonding force, speed, and temperature, and generating a parameter correlation analysis result; Adjusting the algorithm parameters of the adaptive neural fuzzy inference system according to the parameter correlation analysis results to obtain adjusted algorithm parameters; Applying the adjusted algorithm parameters to an adaptive neural fuzzy inference system, processing dynamic changes in the silver wire bonding process through the adaptive neural fuzzy inference system, and generating preliminary bonding parameter optimization results; performing an effectiveness evaluation on the preliminary bonding parameter optimization results to detect adaptability and effectiveness in production, and obtaining an optimization effectiveness evaluation result; Based on the optimization performance evaluation result, the control logic of the bonding machine is updated to obtain an updated control logic.

2. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the real-time monitoring data are as follows: Continuously monitor the force, speed, and temperature parameters of the bonding machine, record the changes in force, speed, and temperature parameters in real time, and obtain the original real-time monitoring data set; Based on the original real-time monitoring data set, statistical analysis is applied to calculate the average value and standard deviation to obtain real-time monitoring data.

3. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the cleaned monitoring data are as follows: Based on the real-time monitoring data, the anomaly score of each data point is calculated using the following formula: Among them, E i is the abnormality score, X i is the value of a single data point, μ is the arithmetic mean of the data set, σ is the standard deviation of the data set, μ med is the median of the data set, and MAD is the median absolute deviation; Based on the anomaly scores, data points whose anomaly scores exceed a preset threshold are eliminated, where the preset threshold is set to the 95% quantile of the anomaly score distribution, to obtain purified monitoring data.

4. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the parameter correlation analysis results are: Based on the post-cleaning monitoring data, the correlation between bonding force, speed and temperature is calculated using the following formula: Among them, R ij represents the correlation between parameter i and parameter j, F i (t) and F j (t) represents the values ​​of parameter i and parameter j at time t, and Respectively represent the mean of parameter i and parameter j in the time interval [t1, t2], and the integral range t1 to t2 represents the observation time period. and denote the first-order time derivatives of parameters i and j, respectively. To introduce additional direct interaction terms between parameters; Based on the correlation, a parameter correlation matrix is ​​constructed, where each element of the correlation matrix represents the correlation degree between two corresponding parameters, and the diagonal elements in the matrix are fixed to 1, thereby generating a parameter correlation analysis result.

5. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the adjusted algorithm parameters are as follows: Analyzing the parameter correlation analysis results, evaluating the correlation strength between bonding force, speed, and temperature, identifying parameters that need to be optimized, and generating a priority list of key parameters; Based on the priority list of key parameters, refining the algorithm parameter adjustment strategy in the adaptive neuro-fuzzy inference system, including adjusting the thresholds of the fuzzy logic rules and the parameters of the membership functions, to obtain a draft adjustment strategy; Based on the draft adjustment strategy, algorithm parameters are adjusted, and the impact of the new parameters on performance is tested in a simulation environment to obtain adjusted algorithm parameters.

6. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the preliminary bonding parameter optimization results are: Loading the adjusted algorithm parameters into the adaptive neural fuzzy inference system, associating the input variables with the membership functions, and initializing the inference engine to generate the loaded adaptive neural fuzzy inference system; Based on the loaded adaptive neural fuzzy inference system, the adjustment amount of the bonding parameter is calculated using the following formula: Among them, D k represents the adjustment of the bonding parameters at time k, A k is the real-time measurement value of the bonding force, B k is the real-time measurement value of bonding speed, C k The impact factor of temperature change; Based on the adjustment amount, the control variables are updated in the adaptive neural fuzzy inference system, the membership functions and inference rules are adjusted in real time, the control strategy in the bonding process is iteratively optimized, and the preliminary bonding parameter optimization results are generated.

7. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the optimization performance evaluation results are as follows: Based on the preliminary bonding parameter optimization results, the adaptability score in the production process is calculated using the following formula: Among them, P is the adaptability score, RT i is the response time of parameter adjustment in the i-th bonding cycle, and is a small constant to prevent division by zero errors, T i is the actual cycle time, T ref is the reference cycle time, N is the number of evaluation cycles; Based on the adaptability score, the effect of the bonding process is analyzed and the impact of parameter optimization on actual production efficiency is evaluated. If the score is higher than the preset standard, the parameter optimization is considered successful and the optimization efficiency evaluation result is obtained.

8. The control logic analysis method for intelligent silver bonding wire according to claim 1, characterized in that: The steps for obtaining the updated control logic are: Based on the optimization performance evaluation results, the bonding force, bonding speed, temperature fluctuation range and response time are extracted, and the unstable interval in the bonding process is identified. The instantaneous change rate and deviation of each parameter are calculated to obtain a control logic input parameter set; According to the control logic input parameter set, the control logic adjustment factor is calculated, and the expression is: Among them, L is the control logic adjustment factor, J t is the real-time response value of the bonding force, RF t is the target reference value, V t is the bonding speed, V r is the target bonding speed, TMP t is the current temperature, TMP r To set the temperature; Based on the control logic adjustment factor, the bonding parameter range is adjusted, and the feedback correction strategy is reset to obtain an updated control logic.

9. The control logic analysis system for the control logic analysis method for intelligent silver bonding wire according to any one of claims 1 to 8, characterized in that: include: Data monitoring module, which monitors the force, speed and temperature parameters during the silver wire bonding process in real time, collects real-time monitoring data, and generates real-time monitoring data results; The data purification module receives real-time monitoring data results, eliminates outliers and noise in the data, and generates purified monitoring data; The correlation analysis module uses the purified monitoring data to analyze the correlation between bonding force, speed and temperature, and generates parameter correlation analysis results; The parameter optimization module adjusts the algorithm parameters of the adaptive neural fuzzy inference system based on the parameter correlation analysis results, obtains the adjusted algorithm parameters, uses the adjusted algorithm parameters to process the dynamic changes of the silver wire bonding process, and generates preliminary bonding parameter optimization results; The performance evaluation module receives preliminary bonding parameter optimization results, evaluates adaptability and effects in production, generates optimization performance evaluation results, and updates the control logic of the bonding machine based on the optimization performance evaluation results to obtain updated control logic.

Images