Power control method and system for intelligent copper wire bonding based on clustering analysis model

Abstract

The application relates to a power control method and system for intelligent copper wire bonding based on a clustering analysis model, which comprises the following steps: a data acquisition module is used for acquiring multi-dimensional parameters involved in a copper wire bonding process in real time; a data preprocessing module is used for performing normalization, noise reduction and feature extraction processing on the collected data; a clustering analysis module is used for clustering feature data by using a clustering algorithm to generate multiple feature clusters; a power prediction module is used for predicting the required optimal power based on clustering results and historical data by using a deep learning algorithm; and a dynamic power adjustment module is used for predicting and adjusting power output in real time. Through clustering analysis, deep learning and PID control technology means, the power control problem in the copper wire bonding process is effectively solved, the stability and production efficiency of the process are significantly improved, and the application has important application value.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power control and automation, and specifically relates to an intelligent power control method and system based on a clustering analysis model, which is particularly suitable for dynamic adjustment and optimization of precise power output in intelligent equipment, and can be widely applied to the industrial field of copper wire bonding, semiconductor manufacturing, and precise welding requiring high-precision power control. The present application comprehensively uses clustering analysis, big data processing, deep learning, and real-time control algorithms, and provides an efficient and stable solution for power management under complex working conditions. BACKGROUND

[0002] With the wide application of copper wire bonding technology in the fields of semiconductor manufacturing and precise electronic assembly, the demand for power control technology is increasing. However, the existing technology has obvious shortcomings in dynamic adaptability, data mining and utilization, control precision, multi-parameter optimization, and quality evaluation, and it is difficult to meet the requirements of high-precision power control under complex working conditions. Specifically, the following aspects are involved:

[0003] Complexity of copper wire bonding power demand:

[0004] The requirement of copper wire bonding technology for power output is highly dependent on specific working conditions, such as surface roughness, working temperature, and bonding pressure. Fluctuations in these parameters directly affect power demand. However, in traditional processes, fixed power or simple adjustment methods are usually used, lacking dynamic adaptability, which can easily lead to over-power consumption or power deficiency causing bonding failure.

[0005] Lack of working condition classification and power matching:

[0006] Under complex working conditions, different operating conditions require different power outputs. However, traditional power control methods lack intelligent classification of working conditions and can only rely on manual experience or preset rules. As the diversity and randomness of working conditions increase, manual rules are difficult to adapt to dynamic changes, resulting in insufficient precision and adaptability of power regulation.

[0007] Limitations of data processing capabilities:

[0008] A large amount of working condition data generated during the copper wire bonding process, including real-time collected temperature, current, voltage, and pressure information, has not been fully utilized. Existing technology usually relies on fixed parameters and simple real-time feedback for adjustment, ignoring the mining and complex correlation analysis of historical data, and unable to extract deep working condition features from big data.

[0009] Application bottleneck of deep learning algorithms:

[0010] The potential of deep learning in power prediction has gradually emerged, enabling the generation of power prediction models through large amounts of data training. However, the current technology lacks a complete framework that combines deep learning with actual processes, especially in the aspect of coordinated control between power demand prediction and actual power output, and there is no mature solution.

[0011] Dynamic adjustment control precision is insufficient:

[0012] Dynamic adjustment of power output usually relies on simple PID control algorithms, but the error compensation ability of this algorithm for complex conditions is limited. In actual processes, when the working conditions change dramatically, traditional PID controllers are prone to over-regulation or under-regulation, leading to increased deviation between power output and actual demand, further affecting the bonding quality.

[0013] Disconnection between power control and quality evaluation:

[0014] The existing technology usually treats power control and quality evaluation as two independent links, lacking coherence and closed-loop optimization capability. In the copper wire bonding process, the power control effect is not combined with the bonding quality feedback in real time, making it difficult to adaptively adjust the power control strategy, thereby affecting production efficiency and final quality.

[0015] Insufficient comprehensive optimization of multi-dimensional parameters:

[0016] The copper wire bonding process involves complex relationships between multiple parameters (such as temperature, pressure, roughness) and power. However, the existing technology lacks a comprehensive optimization mechanism for the correlation between multi-dimensional parameters, usually using single variable adjustment, resulting in limited optimization of the overall performance of the system.

[0017] Single power calculation:

[0018] Traditional power calculation is mostly based on empirical formulas, usually only considering the linear combination of a small number of parameters, ignoring nonlinear factors and complex correlations. In actual working conditions, due to the existence of various nonlinear factors, it is difficult to accurately describe the power demand using only traditional formulas, resulting in inaccurate adjustment results. SUMMARY

[0019] The present application provides an intelligent power control method and system for bonding copper wire based on a clustering analysis model. Through the combination of data acquisition module, data preprocessing module, clustering analysis module, power prediction module, dynamic power adjustment module, power calculation module, quality evaluation and optimization module, and by combining clustering analysis, big data processing, deep learning algorithm and PID control technology, the problems of low power control precision, poor dynamic adaptability and unstable power regulation in existing copper wire bonding processes are solved, thereby improving bonding quality, reducing energy consumption and improving production efficiency. The following is a detailed introduction to each module:

[0020] 1. Data acquisition module:

[0021] Function: The data acquisition module is responsible for real-time acquisition of key parameters affecting power demand during copper wire bonding, including temperature, pressure, and surface roughness.

[0022] Implementation:

[0023] By installing embedded sensors on the bonding equipment, real-time monitoring of various data during the process is achieved, and the data is transmitted to the central control unit through wired (RS485) or wireless (Wi-Fi) methods. To ensure high reliability and real-time performance of the data, the sensor acquisition cycle is in milliseconds, ensuring that changes in critical working conditions can be accurately captured.

[0024] Steps:

[0025] After system startup, initialize each sensor to ensure calibration.

[0026] Temperature acquisition: The sensor detects the ambient temperature in real time.

[0027] Pressure acquisition: The pressure sensor is installed on the mechanical arm of the bonding machine, and the applied pressure is recorded in real time, combined with synchronous processing of the mechanical arm movement.

[0028] Roughness acquisition: The laser displacement sensor scans the copper wire surface to obtain three-dimensional profile data and calculate the roughness value.

[0029] Store the collected multi-dimensional data in the local cache and upload it to the central processing unit.

[0030] 2. Data preprocessing module

[0031] Function: The data preprocessing module is responsible for denoising, normalization, and feature extraction preprocessing of the original data transmitted by the data acquisition module, ensuring the accuracy and consistency of the data.

[0032] Implementation:

[0033] Noise filtering: Remove high-frequency noise in the data through a low-pass filter (the cutoff frequency of the filter is set to 100 Hz to ensure that the noise is removed and the useful signal is preserved).

[0034] Data normalization: Normalize various dimensions of data to the same range to avoid the influence of different dimensions. The normalization formula is:

[0035] ;

[0036] Where: is the original data, and Maximum and minimum values of data, respectively;

[0037] Feature extraction: Extract the most relevant features from multiple sensor data through Principal Component Analysis (PCA) dimensionality reduction technique;

[0038] The calculation formula of PCA dimensionality reduction:

[0039] ;

[0040] Where:

[0041] is the feature vector matrix, is the feature data before dimensionality reduction, is the feature data after dimensionality reduction;

[0042] Steps:

[0043] Read real-time data from the data acquisition module;

[0044] Noise filtering: Remove high-frequency noise in the data through a low-pass filter (the cutoff frequency of the filter is set to 100 Hz to ensure that the noise is removed and the useful signal is retained);

[0045] Data normalization: By unifying various dimensions of data to the same range, the influence of different dimensions is avoided;

[0046] Feature extraction: Extract the most relevant features from multiple sensor data through Principal Component Analysis (PCA) dimensionality reduction technique.

[0047] 3. Clustering analysis module

[0048] Function: Group data through clustering algorithm, identify different working condition modes, and help subsequent power prediction and adjustment;

[0049] Implementation:

[0050] K-means clustering algorithm: Use K-means algorithm for clustering analysis of data, form multiple categories according to different feature combinations;

[0051] Objective of minimizing error within K-means clustering cluster:

[0052] ;

[0053] Where:

[0054] is the number of clusters,

[0055] is the cluster ,

[0056] clusters cluster centers,

[0057] data points;

[0058] Steps:

[0059] Receive pre-processed data from the data pre-processing module;

[0060] Set the initial value of K (e.g., K=3), and randomly select initial cluster centers using the K-means algorithm;

[0061] Calculate the distance of data points to each cluster center, and assign data points to the nearest cluster;

[0062] Update the cluster center position until the cluster center no longer changes;

[0063] Output the final clustering result as the input for subsequent power prediction;

[0064] Example: The system performs clustering analysis on copper wire surface roughness, applied pressure, and temperature data, resulting in three clusters corresponding to normal working conditions, overheating working conditions, and low power working conditions.

[0065] 4. Power prediction module

[0066] Function: Using the clustering analysis results, through deep learning algorithms and minimizing loss functions, optimize the prediction accuracy of the model to achieve accurate power prediction and optimal power demand under different working conditions;

[0067] Implementation:

[0068] Deep neural network (DNN) model: input is the feature data obtained from clustering analysis, output is the predicted power value;

[0069] Training method: use the minimizing loss function to optimize the prediction accuracy of the model to achieve accurate power demand prediction;

[0070] Minimizing loss function:

[0071] ;

[0072] Where:

[0073] predicted power,

[0074] actual power,

[0075] number of data points;

[0076] Example: For a specific copper wire surface roughness, applied pressure and temperature conditions, the model predicts the required power as 8W, which will be used for subsequent power adjustments.

[0077] 5. Dynamic power adjustment module

[0078] Function: Adjust the working power of the device in real time according to the prediction value provided by the power prediction module, to ensure that the optimal power requirement is maintained during the copper wire bonding process;

[0079] Implementation:

[0080] PID controller: dynamically adjust control parameters according to power error, optimize power output;

[0081] Steps:

[0082] Receive the output of the power prediction module to determine the target power value;

[0083] Adjust using the PID controller, calculate the new power setting through the controller, and adjust the device power output in real time;

[0084] Continuously monitor the actual power and adjust the control parameters according to the feedback information;

[0085] Example: The power predicted by the power prediction module is 8W, while the current power is 7.5W, the PID controller will calculate the power difference and adjust the device power to the target value.

[0086] 6. Power calculation module

[0087] Function: Based on real-time temperature, pressure, surface roughness data, generate an empirical model of power demand, determine the empirical coefficient in the empirical model through experiments, map the physical quantities under different working conditions with the actual power, as the basic input of the dynamic power adjustment module;

[0088] Implementation:

[0089] Real-time measurement of temperature, pressure, surface roughness data through sensors;

[0090] Determine the empirical coefficient through the empirical model, map the physical quantities under different working conditions with the actual power, as the basic input of the dynamic power adjustment module, the power calculation formula:

[0091] ;

[0092] Where:

[0093] and is the empirical coefficient,

[0094] is the pressure,

[0095] Surface roughness,

[0096] Ambient temperature;

[0097] Step:

[0098] Real-time acquisition of temperature, pressure, and surface roughness data from sensors;

[0099] Actual power calculation according to power calculation formula;

[0100] Pass the calculation result to the dynamic power adjustment module for power output adjustment;

[0101] 7. Quality evaluation and optimization module

[0102] Function: According to multiple quality evaluation indexes (such as bonding strength, thermal stability, surface smoothness), calculate the quality score by weighted average method, and adjust the system parameters according to the evaluation results, further optimize the clustering analysis and power prediction model, so as to improve the stability and production efficiency of the system;

[0103] Implementation:

[0104] Monitor bonding quality parameters (bonding strength, thermal stability, surface smoothness);

[0105] Calculate the quality score by weighted average method;

[0106] According to the evaluation results, feedback and optimize the system parameters, adjust the clustering analysis and power prediction model, to continuously improve the quality of the final product and the stability of the process;

[0107] Example: Through quality detection, it is found that the strength of a batch of copper wire bonding is low, the system will optimize the power setting according to the historical data to ensure the quality of the next bonding;

[0108] The quality score formula is as follows:

[0109] ;

[0110] Among them:

[0111] Bonding strength,

[0112] Thermal stability,

[0113] Surface smoothness,

[0114] , , The weight coefficient is denoted as w.

[0115] The power control method for intelligent copper wire bonding based on a clustering analysis model includes the following steps:

[0116] Step one, data collection: real-time collection of working condition data related to the copper wire bonding process through a data collection module, including temperature, pressure, surface roughness environmental factors, acquisition of these data through sensors and monitoring equipment, and transmission to a data preprocessing module;

[0117] Step two, data preprocessing: noise removal and normalization processing of the original data collected from the data collection module, conversion of feature data of different dimensions to a uniform scale through a standardization method to ensure that the clustering analysis and power prediction module can correctly process the data;

[0118] Step three, clustering analysis: using the K-means clustering algorithm to analyze the data, and dividing the collected working condition data into several clusters according to similarity, each cluster representing a specific working state or working condition mode, through this way, the clustering analysis module provides classification features for subsequent power prediction and dynamic adjustment, and provides input based on actual working conditions for power prediction;

[0119] Step four, power prediction: based on the working condition cluster features generated by the clustering analysis module, using a deep neural network for power prediction, the prediction model trained according to historical data is inputted into the clustering analysis results, and the power demand in the bonding process is predicted through a regression algorithm, the power prediction result provides a reference for the system to adjust the actual power;

[0120] Step five, dynamic power adjustment: adjusting the actual power through a dynamic power adjustment module, which uses a PID control algorithm to calculate the error between the target power and the actual power in real time, and adjusts the power output to ensure that the actual power is close to the predicted power, avoiding bonding quality problems caused by power fluctuations;

[0121] Step six, power calculation: based on environmental parameters, pressure, temperature and roughness factors, calculate the reference power value, this module uses empirical formula combined with process data in actual production to determine the required power value under different working conditions, and provides input for the dynamic power adjustment module;

[0122] Step seven, quality evaluation and optimization: comprehensive evaluation of copper wire bonding quality according to multiple quality evaluation indexes (such as bonding strength, thermal stability, surface smoothness), calculation of quality score through weighted average method, and adjustment of system parameters according to evaluation results to further optimize clustering analysis and power prediction model, thereby improving the stability and production efficiency of the system.

[0123] Advantages

[0124] Compared with the prior art, the present application provides a power control method and system for intelligent copper wire bonding based on a clustering analysis model, which has the following beneficial effects:

[0125] Improving the power accuracy and stability of the copper wire bonding process:

[0126] The present application realizes accurate adjustment of power control by integrating data acquisition, preprocessing, clustering analysis and deep learning technology. By real-time monitoring of environmental parameters (such as temperature, pressure and surface roughness) during the copper wire bonding process, and predicting the required power value according to historical data and real-time calculation results, the system can dynamically adjust the power output to ensure the stability of the power during each bonding process, which effectively avoids bonding defects caused by power fluctuations, and improves the stability and consistency of copper wire bonding.

[0127] Enhancing the adaptability and intelligence of the system:

[0128] The system uses clustering analysis algorithm to divide the working condition data into multiple characteristic clusters. Based on different working conditions, the system can adaptively predict and adjust the power. The clustering analysis module can identify different production conditions and predict the power through a deep learning model to ensure that the system can accurately calculate the required power value and dynamically adjust it under each specific working condition, thereby improving the intelligence level and flexibility of the system.

[0129] Improving production efficiency and resource utilization:

[0130] The system avoids excessive or insufficient power output through accurate power prediction and dynamic adjustment, reducing energy waste. By optimizing the power control of copper wire bonding, the system improves production efficiency and resource utilization, reduces energy consumption and production cost.

[0131] Ensuring the repeatability and stability of the process:

[0132] The system not only can adjust the power output in real time, but also can continuously optimize the power control strategy according to the working condition changes and quality feedback during each production process. Through learning from historical data, the system can repeatedly produce consistent power control schemes under the same or similar working conditions, ensuring the repeatability and stability of the copper wire bonding process, and providing reliable guarantee for large-scale production.

[0133] Simplifying the power control in the production process:

[0134] By integrating various complex algorithms and technologies (such as data preprocessing, clustering analysis, deep learning, PID control) into a unified system, the present application simplifies the power control process in the production process, and the operator only needs to master the power demand in the copper wire bonding process through the real-time feedback and adjustment suggestions provided by the system, reducing the complexity and error probability of human operation, improving the convenience and safety of operation;

[0135] Provide comprehensive fault warning and optimization feedback:

[0136] The system can timely find the situation that the process parameters deviate from the predetermined range by comparing the key parameters in the copper wire bonding process with the historical data, and give corresponding adjustment suggestions. Through the continuous feedback and optimization of the quality evaluation module, the system can warn potential faults or quality problems in the production process, and timely adjust and correct, so as to reduce the defect rate in production and improve the qualified rate of products. BRIEF DESCRIPTION OF DRAWINGS

[0137] Figure 1 is a data acquisition module flow chart;

[0138] Figure 2 is a data preprocessing module flow chart;

[0139] Figure 3 is a clustering analysis module flow chart;

[0140] Figure 4 is a power prediction module flow chart;

[0141] Figure 5 is a dynamic power adjustment module flow chart;

[0142] Figure 6 is a power calculation module flow chart;

[0143] Figure 7 is a quality evaluation and optimization module flow chart. DETAILED DESCRIPTION

[0144] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application:

[0145] Data acquisition:

[0146] By installing embedded sensors on the bonding equipment, real-time monitoring of various data in the process is realized, and the data is transmitted to the central control unit through wired (RS485) or wireless (Wi-Fi) mode. In order to ensure the high reliability and real-time performance of the data, the sensor acquisition cycle is millisecond level, which ensures that the changes of key working conditions can be accurately captured;

[0147] Step:

[0148] After the system is started, initialize each sensor to ensure calibration;

[0149] Temperature collection: the sensor detects the ambient temperature in real time;

[0150] Pressure collection: the pressure sensor is installed on the mechanical arm of the bonder, and the applied pressure is recorded in real time, combined with the synchronous processing of the mechanical arm movement;

[0151] Roughness collection: the laser displacement sensor scans the surface of the copper wire to obtain three-dimensional profile data and calculate the roughness value;

[0152] Store the collected multi-dimensional data in the local cache and upload it to the central processing unit;

[0153] Data preprocessing:

[0154] Remove high-frequency noise in the data through a low-pass filter (the cutoff frequency of the filter is set to 100 Hz to ensure that the noise is removed and the useful signal is retained), and unify the data of various dimensions to the same range to avoid the influence of different dimensions. Use the normalization formula:

[0155] ;

[0156] Where: is the original data, and are the maximum and minimum values of the data, respectively, and the most relevant features are extracted from multiple sensor data through principal component analysis (PCA) dimensionality reduction technique;

[0157] The calculation formula of PCA dimensionality reduction is:

[0158] ;

[0159] Where:

[0160] is the feature vector matrix, is the feature data before dimensionality reduction, is the feature data after dimensionality reduction;

[0161] Steps:

[0162] Read real-time data from the data collection module;

[0163] Noise filtering: remove high-frequency noise in the data through a low-pass filter (the cutoff frequency of the filter is set to 100 Hz to ensure that the noise is removed and the useful signal is retained);

[0164] Data normalization: unify the data of various dimensions to the same range to avoid the influence of different dimensions;

[0165] Feature extraction: Extract the most relevant features from multiple sensor data through principal component analysis (PCA) dimensionality reduction technique;

[0166] Clustering analysis:

[0167] Use K-means algorithm to cluster the data, form multiple categories according to different feature combinations, and use them as input features for subsequent power prediction;

[0168] K-means clustering cluster error minimization objective:

[0169] ;

[0170] Where:

[0171] is the number of clusters, is the cluster , is the cluster center, is the data point;

[0172] Steps:

[0173] Receive processed data from the data preprocessing module;

[0174] Initial setting of K value (such as K=3), random selection of initial cluster center using K-means algorithm;

[0175] Calculate the distance of data points to each cluster center, and assign the data points to the nearest cluster;

[0176] Update the cluster center position until the cluster center no longer changes;

[0177] Output the final clustering result as the input of subsequent power prediction;

[0178] Power prediction:

[0179] Using the clustering analysis results, through deep learning algorithm and minimizing loss function, optimize the prediction accuracy of the model to achieve accurate power prediction, and minimize the loss function:

[0180] ;

[0181] Where: is the predicted power, is the actual power, is the number of data points.

[0182] Dynamic power adjustment:

[0183] According to the predicted value provided by the power prediction module, the working power of the device is adjusted in real time according to PID control to ensure that the optimal power requirement is maintained during the copper wire bonding process;

[0184] Steps:

[0185] Receive the output of the power prediction module to determine the target power value;

[0186] Adjust using the PID controller, calculate the new power setting through the controller, and adjust the device power output in real time;

[0187] Continuously monitor the actual power and adjust the control parameters according to the feedback information;

[0188] Power calculation:

[0189] Based on real-time temperature, pressure, and surface roughness data, an empirical model of power demand is generated. The empirical model determines the coefficients through experiments, maps the physical quantities under different working conditions with the actual power, and serves as the basic input of the dynamic power adjustment module;

[0190] Power calculation formula:

[0191] ;

[0192] Where: and are empirical coefficients, is the pressure, is the surface roughness, is the ambient temperature;

[0193] Steps:

[0194] Real-time acquisition of temperature, pressure, and surface roughness data from sensors;

[0195] Calculate the actual power according to the power calculation formula;

[0196] Pass the calculation result to the dynamic power adjustment module for power output adjustment.

[0197] Quality evaluation and optimization module

[0198] According to multiple quality evaluation indicators (such as bonding strength, thermal stability, and surface smoothness), calculate the quality score by weighted average method, and adjust the system parameters according to the evaluation results to further optimize the clustering analysis and power prediction model, thereby improving the stability and production efficiency of the system;

[0199] Steps:

[0200] Monitor the bonding quality parameters (bonding strength, thermal stability, and surface smoothness);

[0201] Weighted average method to calculate the quality score;

[0202] The quality score formula is as follows:

[0203] ;

[0204] Wherein: is the bonding strength, is the thermal stability, is the surface smoothness, , , is the weight coefficient;

[0205] According to the evaluation results, feedback optimization system parameters, adjust the clustering analysis and power prediction model, to continuously improve the final product quality and process stability;

[0206] The power control method of intelligent copper wire bonding based on clustering analysis model includes the following steps:

[0207] Step one, data acquisition: real-time collection of working condition data related to the copper wire bonding process through the data acquisition module, including temperature, pressure, surface roughness environmental factors; Through sensors and monitoring equipment to obtain these data, and transmit to the data preprocessing module;

[0208] Step two, data preprocessing: noise removal and normalization processing of the original data collected from the data acquisition module; Through the standardization method, the characteristic data of different dimensions are converted to a unified scale to ensure that the clustering analysis and power prediction module can correctly process the data;

[0209] Step three, clustering analysis: use K-means clustering algorithm to analyze the data, and divide the collected working condition data into several clusters according to the similarity; Each cluster represents a specific working state or working condition mode; In this way, the clustering analysis module provides classification features for subsequent power prediction and dynamic adjustment, and provides input based on actual working conditions for power prediction;

[0210] Step four, power prediction: based on the working condition cluster characteristics generated by the clustering analysis module, use deep neural network for power prediction; The prediction model trained by the module according to historical data, the input is the clustering analysis result, and the power demand in the bonding process is predicted through the regression algorithm; The power prediction result provides a reference for the system to adjust the actual power;

[0211] Step five, dynamic power adjustment: adjust the actual power through the dynamic power adjustment module; This module uses PID control algorithm to calculate the error between target power and actual power in real time, and adjusts the power output to ensure that the actual power is close to the predicted power, avoiding bonding quality problems caused by power fluctuations;

[0212] Step six, power calculation: based on environmental parameters, pressure, temperature and roughness factors, calculate the reference power value; this module uses empirical formula combined with actual production process data to determine the required power value under different working conditions and provides input for the dynamic power adjustment module;

[0213] Step seven, quality evaluation and optimization: according to multiple quality evaluation indexes (such as bonding strength, thermal stability, surface smoothness), comprehensively evaluate the copper wire bonding quality; calculate the quality score by weighted average method, and adjust the system parameters according to the evaluation results, further optimize the clustering analysis and power prediction model, so as to improve the stability and production efficiency of the system.

Claims

1. A power control system for intelligent copper wire bonding based on clustering analysis model, characterized in that, Comprise: Data acquisition module for real-time acquisition of multi-dimensional parameters involved in copper wire bonding process; Data preprocessing module for normalizing, denoising and feature extraction processing of the collected data; Cluster analysis module for clustering the feature data by clustering algorithm to generate multiple feature clusters; Power prediction module based on clustering results and historical data, using deep learning algorithm to predict the required optimal power; Dynamic power adjustment module for real-time power prediction and power output adjustment; Power calculation module for collecting data and calculating actual power value according to empirical formula; The power calculation module specifically includes that the power calculation module calculates the actual power value according to the historical data and the empirical formula, and the module generates an empirical model of power demand through the combined analysis of temperature, pressure and surface roughness three working condition parameters, and the coefficients in the empirical model are determined by experiment, which maps the physical quantities under different working conditions and the actual power as the basic input of the dynamic power adjustment module; The empirical formula for calculating the actual power is: P r = a · F b · R s + β · T e ; where: a and β are empirical coefficients, F b is the pressure, R s is the surface roughness, T e is the ambient temperature; Quality evaluation and optimization module for comprehensive evaluation of bonding quality and optimization of system parameters according to the evaluation results.

2. The system of claim 1, wherein, The data acquisition module specifically includes that the data acquisition module is used to acquire the environmental and process parameters related to the copper wire bonding process in real time, including temperature, pressure and surface roughness, wherein: Temperature: Real-time temperature T of the working environment is monitored using a temperature sensor e The temperature will affect the thermal expansion and electrical conductivity of the copper wire, thereby affecting the power requirement; Pressure: The pressure sensor measures the pressure F applied during the bonding process b A pressure that is too high can cause damage to the copper wire, and a pressure that is too low can result in a weak bond; Surface roughness: Surface roughness R of the bonded copper wire was obtained using a laser displacement sensor s The surface roughness affects the electrical contact effect, thereby affecting the power consumption.

3. The system of claim 1, wherein, The data preprocessing module specifically includes that the data preprocessing module cleans, normalizes and noise filters the original data obtained from the acquisition module to remove irrelevant information and extract valuable features, wherein: Normalization: Since different sensors output data with different dimensions, normalization is part of the standardization process to ensure that each data feature is within the same range, thereby avoiding the influence of a certain feature on the result; Noise filtering: remove high-frequency noise in the data through a low-pass filter, the source of noise is sensor failure, environmental interference or unstable work of electronic equipment, the cutoff frequency of the filter is set to 100Hz to ensure that the noise is removed and the useful signal is retained; Feature extraction: extract the most relevant features from multiple sensor data through principal component analysis (PCA) dimension reduction technology, PCA projects the original data to a new orthogonal space, retains most of the data variance and reduces computational complexity; The calculation formula of PCA dimension reduction is: X' = W·X; Wherein: W is the eigenvector matrix, X is the feature data before dimension reduction, X ′ is the feature data after dimension reduction.

4. The system of claim 1, wherein, The cluster analysis module specifically includes that the feature data obtained from the data acquisition module and the data preprocessing module are classified by using unsupervised learning method, and the data are divided into several clusters by clustering algorithm, so as to identify different working condition modes, this module uses K-means clustering algorithm, groups according to the similarity between each feature data point, each cluster represents a specific working condition mode, which will be used as input features in subsequent power prediction, the minimum error minimization target of K-means clustering cluster is: where: k is the number of clusters, C i is the cluster i, μ i is the cluster center of cluster i, x is the data point.

5. The system of claim 1, wherein, The power prediction module specifically includes that the power prediction module performs power prediction through a deep learning algorithm based on the cluster characteristics provided by the cluster analysis module, the module adopts a deep neural network model, a power prediction model trained according to historical data, the input is the output characteristics of the cluster analysis module, and the output is a power prediction value; the prediction accuracy of the model is optimized by minimizing the loss function, so as to realize accurate power demand prediction; The minimum loss function is: where: P w,i is the predicted power, P t,i is the actual power, n is the number of data points.

6. The system of claim 1, wherein, The dynamic power adjustment module specifically includes that the dynamic power adjustment module dynamically adjusts the actual power output through a PID control algorithm based on the output of the power prediction module, ensures the accurate control and stability of the power in the copper wire bonding process, the module calculates the error between the target power and the actual power, and adjusts the power output through a proportional, integral and differential feedback regulator, so that the actual power and the predicted power tend to be consistent.

7. The system of claim 1, wherein, The quality evaluation and optimization module specifically includes that the quality evaluation and optimization module comprehensively evaluates the copper wire bonding quality based on multiple quality evaluation indexes, evaluates the bonding quality in combination with process parameters and operating conditions, the module calculates the quality score by using a weighted average method, and feeds back the system parameters according to the evaluation results to adjust the cluster analysis and power prediction model, so as to continuously improve the final product quality and process stability; The quality score formula is as follows: Q s = ω1·Q m + ω2·Q t + ω3·Q r wherein: Q m is the bonding strength, Q t is the thermal stability, Q r is the surface smoothness, ω1, ω2, ω3 are weight coefficients.

8. A power control method for intelligent copper wire bonding based on a clustering analysis model, characterized in that, The method comprises the following steps: Step one, data acquisition: real-time collection of working condition data related to the copper wire bonding process through a data acquisition module, including three environmental factors of temperature, pressure and surface roughness, acquisition of these data through sensors and monitoring equipment, and transmission to a data preprocessing module; Step two, data preprocessing: noise removal and normalization processing of the original data collected from the data acquisition module, conversion of feature data of different dimensions into a unified scale through a standardization method, so as to ensure that the cluster analysis and power prediction module can correctly process data; Step three, cluster analysis: analysis of data by using a K-means clustering algorithm, and division of the collected working condition data into several clusters according to similarity, each cluster representing a specific working state or working condition mode, through this way, the cluster analysis module provides classification characteristics for subsequent power prediction and dynamic adjustment, and provides actual working condition-based input for power prediction; Step four, power prediction: power prediction using a deep neural network based on the working condition cluster characteristics generated by the cluster analysis module, the module uses a prediction model trained according to historical data, the input is the cluster analysis result, and the power demand in the bonding process is predicted through a regression algorithm, the power prediction result provides a reference for the system to adjust the actual power; Step five, dynamic power adjustment: adjustment of the actual power through a dynamic power adjustment module, the module uses a PID control algorithm, calculates the error between the target power and the actual power in real time, and adjusts the power output, so as to ensure that the actual power is close to the predicted power, and avoid bonding quality problems caused by power fluctuation; Step six, power calculation: based on environmental parameters, pressure, temperature and roughness factors, the actual power value is calculated by the power calculation module, which uses empirical formula combined with actual production process data to determine the required power value under different working conditions and provides input for the dynamic power adjustment module; Step seven, quality evaluation and optimization: according to multiple quality evaluation indexes, the copper wire bonding quality is comprehensively evaluated, the quality score is calculated by weighted average method, and the system parameters are adjusted according to the evaluation results to further optimize the clustering analysis and power prediction model, thereby improving the stability and production efficiency of the system.

Images