Electronic component quality prediction method based on big data analysis

Abstract

The embodiment of the invention provides an electronic component quality prediction method based on big data analysis. The electronic component quality prediction method comprises the steps that historical quality data and detection records of all batches of electronic components in a supply chain are collected and normalized; constructing a quality feature vector based on the normalized data, and performing clustering analysis to identify a potential risk batch; dynamically adjusting warehousing sampling inspection strategy parameters according to the risk level in the clustering analysis result; and applying the adjusted sampling inspection strategy to the warehousing detection process of the current batch to reduce the quality abnormity missing detection risk. Through the scheme of the embodiment of the invention, the problem of how to regulate and control the warehousing sampling inspection strategy according to the batch difference data in the supply chain so as to solve the problem of rising of the quality abnormality missing inspection risk can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electronic information engineering, and particularly relates to an electronic component quality prediction method based on big data analysis. BACKGROUND

[0002] The electronic component quality prediction method based on big data analysis is to collect and analyze a large amount of use data, detection records, supplier information, etc. of electronic components, and use data mining and machine learning techniques to predict the quality of components, so as to identify potential quality problems in advance and improve the reliability and stability of products. One key problem of the method is that in the supply chain, different batches of components may have performance or quality differences, and how to dynamically regulate the warehouse sampling inspection strategy according to these batch difference data to reduce the risk of missing quality abnormalities due to insufficient sampling is an important challenge currently faced. SUMMARY

[0003] Therefore, the electronic component quality prediction method based on big data analysis is provided, which at least partially solves the problems in the prior art.

[0004] The electronic component quality prediction method based on big data analysis comprises:

[0005] collecting and normalizing historical quality data and detection records of each batch of electronic components in the supply chain;

[0006] constructing a quality feature vector based on the normalized data and performing cluster analysis to identify potential risk batches;

[0007] dynamically adjusting warehouse sampling inspection strategy parameters according to the risk level in the cluster analysis result;

[0008] applying the adjusted sampling inspection strategy to the current batch of warehouse detection processes to reduce the risk of missing quality abnormalities.

[0009] According to one embodiment, the collecting and normalizing historical quality data and detection records of each batch of electronic components in the supply chain further comprises:

[0010] obtaining detection indexes of each batch of electronic components, such as failure times (Fi), qualified rates (Qi), and transportation times (Ti);

[0011] standardizing the detection indexes of all batches so that the value of each index falls within the interval [0, 1];

[0012] calculating the normalized weight coefficient W of each batch based on the following formula: W = (Fi + a*Qi) / β, wherein a is a correction coefficient and β is a total reference parameter;

[0013] The normalized data is constructed in a matrix form for subsequent processing.

[0014] According to one embodiment, the constructing a quality feature vector based on the normalized data and performing clustering analysis further comprises:

[0015] Extracting the data feature of each batch to form a vector V = [W, Ri, Di, Fi], wherein Ri is a supply channel stability coefficient, Di is a transportation environment difference value;

[0016] Calculating the similarity between vectors by distance algorithm, using Euclidean distance D = √∑(Vi Vj) 2 ;

[0017] Setting a dynamic threshold Th = μ + λσ, wherein μ is the average value, σ is the standard deviation, and λ is the adjustment coefficient;

[0018] Vectors with a similarity greater than the threshold are divided into the same cluster to identify potential risk batches.

[0019] According to one embodiment, the dynamically adjusting the warehouse entry inspection strategy parameters according to the risk level in the clustering analysis result further comprises:

[0020] Assigning a risk level Rk ∈ {1, 2, 3} based on the clustering result, wherein 1 is low risk and 3 is high risk;

[0021] Defining an inspection density function S = k * (Rk^a), wherein k is a basic parameter and a is a power adjustment factor;

[0022] Introducing a historical false detection rate δ for compensation calculation, S = S + γ * δ, and γ is a compensation coefficient;

[0023] Applying the adjusted inspection proportion to the detection task of the current batch.

[0024] According to one embodiment, the applying the adjusted inspection strategy to the warehouse entry detection process of the current batch further comprises:

[0025] Establishing a real-time detection feedback mechanism to obtain inspection result feedback information;

[0026] Dynamically updating the risk level Rk = Rk × (1 / ρ * E) according to the current batch detection data, wherein E is the abnormal detection proportion and ρ is the attenuation coefficient;

[0027] If Rk ≤ 1, the normal detection process is restored; otherwise, high-frequency inspection is continued;

[0028] The updated strategy is fed back to the previous step to realize adaptive closed-loop optimization.

[0029] According to one embodiment, the constructing quality feature vector and performing cluster analysis further comprises:

[0030] Combining external data sources to increase dimensions, such as supplier credit rating (Cr), environmental temperature and humidity fluctuation index (Ew);

[0031] Adding new features to the original vector V to form an extended vector V = [V, Cr, Ew];

[0032] Performing clustering by improved Kmeans algorithm, and the initial center point is obtained by calculating the historical average;

[0033] Calculating the standard deviation σ c of each cluster, and setting the judgment standard: if σ c > σ avg, the cluster needs to be paid special attention.

[0034] According to one embodiment, the dynamically adjusting the warehouse inspection strategy parameters further comprises:

[0035] Formulating a hierarchical sampling scheme according to the risk level, and setting the sampling frequency f = (1 + η * Rk) p, η and p are parameter coefficients;

[0036] Introducing the concept of time window in the high-frequency inspection process, if no quality problem is found within T consecutive time, the degraded detection mode is triggered;

[0037] If there are abnormal records during the period, the degraded period is extended and manual verification is started;

[0038] Integrating the above rules into the decision tree model to realize intelligent judgment.

[0039] According to one embodiment, the constructing quality feature vector according to the normalized data further comprises:

[0040] Extracting time-varying features, and calculating the stability coefficient Stab = (ΔFi / Fi_avg) of each batch, wherein ΔFi is the failure change amount of adjacent batches, and Fi_avg is the average failure number;

[0041] Adding a nonlinear transformation factor φ to reflect the non-stable trend of batch quality;

[0042] Constructing a multi-dimensional tensor representation method: T = [V, Stab, φ];

[0043] Using a deep learning model to train it, and improving the feature expression ability.

[0044] According to one embodiment, the dynamically adjusting the warehouse inspection strategy parameters further comprises:

[0045] Constructing a cost-benefit model, and setting the cost C = (N * P * K), wherein N is the sample quantity, P is the unit detection cost, and K is the frequency coefficient;

[0046] The optimization goal is to minimize the probability of missed detection Pm, with the constraint condition C≤B, where B is the upper limit of the budget;

[0047] The optimal K value is solved by the Lagrange multiplier method;

[0048] The sampling strategy parameters are adjusted according to the solution result, and the constraint condition library is updated.

[0049] According to an embodiment, the application of the adjusted sampling strategy to the current batch of warehouse-in detection process further comprises:

[0050] Fuzzy decision logic is introduced when performing sampling, and priority P=1exp(μ×Rk) is defined, where μ is a sensitivity coefficient;

[0051] The sampling item distribution is determined in combination with the sample characteristics, and a probability matrix M=[M1, M2,..., Mn] is set, where Mj represents the probability of the jth detection item being selected;

[0052] The actual result is recorded after each sampling, and the weight parameters are continuously optimized using reinforcement learning;

[0053] The updated strategy is synchronously updated to all related system modules to ensure consistency.

[0054] The electronic component quality prediction method based on big data analysis provided by the embodiments of the present disclosure comprises: collecting and normalizing the historical quality data and detection records of each batch of electronic components in the supply chain; constructing a quality feature vector based on the normalized data and performing clustering analysis to identify potential risk batches; dynamically adjusting the warehouse-in sampling strategy parameters according to the risk level in the clustering analysis result; applying the adjusted sampling strategy to the current batch of warehouse-in detection process to reduce the risk of missed detection of quality abnormalities. Through the scheme of the embodiments of the present disclosure, the problem of how to adjust the warehouse-in sampling strategy according to the batch difference data in the supply chain to solve the problem of rising risk of missed detection of quality abnormalities can be solved. BRIEF DESCRIPTION OF DRAWINGS

[0055] In the drawings, like reference numerals designate like parts throughout the several views, and similar components or elements in the figures are designated with like reference numerals. These drawings are not necessarily to scale. It should be understood that these drawings are merely schematic and that actual implementation can differ from that shown.

[0056] Figure 1 is a flowchart of an electronic component quality prediction method based on big data analysis;

[0057] Figure 2is a further flow chart of collecting and normalizing historical quality data and detection records of batches of electronic components in a supply chain;

[0058] Figure 3 is a further flow chart of constructing quality feature vectors based on the normalized data and performing clustering analysis;

[0059] Figure 4 is a further flow chart of dynamically adjusting parameters of an incoming inspection sampling strategy according to risk levels in the clustering analysis results;

[0060] Figure 5 is a further flow chart of applying the adjusted sampling strategy to an incoming inspection process of a current batch;

[0061] Figure 6 is a further flow chart of constructing quality feature vectors and performing clustering analysis;

[0062] Figure 7 is a further flow chart of dynamically adjusting parameters of an incoming inspection sampling strategy;

[0063] Figure 8 is a further flow chart of constructing quality feature vectors based on the normalized data;

[0064] Figure 9 is a further flow chart of dynamically adjusting parameters of an incoming inspection sampling strategy;

[0065] Figure 10 is a further flow chart of applying the adjusted sampling strategy to an incoming inspection process of a current batch. DETAILED DESCRIPTION

[0066] In order to make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the embodiments of the present disclosure are further described below in conjunction with the accompanying drawings, and the schematic embodiments of the present disclosure and the descriptions thereof are only used to explain the present disclosure, but not to limit the present disclosure.

[0067] Next, referring to the accompanying drawings, the electronic component quality prediction method based on big data analysis of the present application specifically includes collecting and normalizing the historical quality data and detection records of each batch of electronic components in the supply chain. In this step, first, historical data needs to be collected from multiple sources, including quality reports submitted by suppliers, data from detection laboratories, pre-warehouse and post-warehouse detection results, etc., to ensure that all relevant parameters are covered. Then, clean up these data, handle missing values, outliers and inconsistent formats, etc., to make the data uniform and standardized. At the same time, normalize the indicators from different sources so that the numerical values are distributed on a unified scale, avoiding the influence of dimension difference on subsequent analysis effect. For example, in a large electronic manufacturing enterprise, the procurement department integrates the quality records of electronic components from multiple suppliers in different countries and regions into a structured database, and uses algorithms to normalize them to facilitate the next step of analysis.

[0068] Next, construct a quality feature vector based on the normalized data and perform clustering analysis to identify potential risk batches. In this step, select several feature indicators related to product quality, such as material purity, electrical performance parameters, appearance defect rate, environmental test results, etc., to construct a quality feature vector. Then use clustering algorithms such as K-means or hierarchical clustering method to classify the feature vectors, find batches with similar characteristics and judge whether they belong to the abnormal group. This process can reveal the differences between different batches, thereby identifying those batch groups that have appeared quality abnormalities in historical data. For example, in one embodiment, by analyzing the detection records of 30,000 batches of electronic capacitor products in the past five years, it is found that some batches have obvious deviation in temperature resistance index, and after clustering analysis they are marked as high-risk batches.

[0069] Then, dynamically adjust the warehouse sampling inspection strategy parameters according to the risk level in the clustering analysis results. The core of this link is to develop different sampling schemes according to the risk level obtained by clustering analysis. For example, for batches with high risk, increase the sampling quantity and increase the types of detection items; while for batches with low risk, simplify the process and improve efficiency. In order to realize dynamic regulation, the system may integrate an early warning module, when it detects that the risk score of a batch exceeds the threshold, the system will automatically recommend or enforce stricter sampling standards. For example, after a home appliance manufacturer receives the supplier information of a batch of imported chips, it uses the existing model to determine that the batch may have hidden defects, and performs 100% full inspection when it is warehoused, and adds a special test item to improve accuracy.

[0070] Finally, the adjusted sampling strategy is applied to the current batch of incoming inspection process to reduce the risk of missing quality abnormalities. This step requires the system to call the pre-established analysis model upon receiving new batch data, automatically generate corresponding sampling parameters based on real-time risk level calculation, including sample size, test item settings, and testing methods, and input these parameters into the automated testing equipment or manual inspection process. In addition, feedback and optimization should be performed on the execution results to gradually improve the prediction accuracy and adaptability of the system. For example, in a specific case, after a communication equipment manufacturer introduced the system, when a new batch of sensor modules was received, the system automatically increased the original 2% sampling rate to 5%, and by increasing the test items of key parameters, it successfully intercepted 2 potential quality problems, effectively improving the overall production pass rate.

[0071] Next, the collection and normalization of historical quality data and test records of each batch of electronic components in the supply chain are described further. First, the test indicators of each batch of electronic components are obtained, such as failure times (Fi), pass rate (Qi), and transportation time (Ti). These indicators are used to evaluate the overall quality of different batches of products and the impact of logistics links. For example, in the electronic manufacturing industry, a batch of products may have a higher failure rate, but at the same time, the pass rate is also higher, so it needs to be quantified to support subsequent analysis.

[0072] Secondly, the test indicators of all batches are standardized to make the value of each indicator fall within the interval [0, 1]. This process can eliminate dimensional differences and ensure that different indicators can be effectively compared and calculated. Specifically, the minimum-maximum normalization method can be used to map the numerical value to the interval, ensuring that the data is on a unified scale.

[0073] Next, the normalized weight coefficient W of each batch is calculated based on the following formula: W = (Fi + a * Qi) / β, where Fi represents the failure times of a batch, generally a non-negative integer; Qi is the pass rate of the batch, ranging from 0 to 1; a is a correction coefficient used to adjust the weight of the pass rate in the overall score, typically taking a value between 0.5 and 2; β is the total reference parameter, which is the average or weighted mean of multiple reference batches, usually selecting a suitable positive number as the denominator to control the value of W not to exceed a reasonable range. The significance of the formula is to consider both unqualified cases and pass rates to form a more comprehensive quality evaluation system to more accurately reflect the true quality state of the batch.

[0074] Finally, the normalized data is structured into a matrix form for subsequent processing. This structure facilitates input into models such as classification algorithms, regression analysis, etc. In one embodiment, a batch of capacitors is measured to have Fi = 3, Qi = 0.9, a = 1, b = 2, then its weight coefficient W = (3 + 1 * 0.9) / 2 = 1.95, indicating that the batch quality is weak and needs to be focused on.

[0075] Next, the construction of quality feature vectors based on normalized data and further clustering analysis of the present application are described. First, the data features of each batch are extracted to form a vector V = [W, Ri, Di, Fi], where W represents the welding quality index, Ri represents the supply channel stability coefficient, Di represents the transportation environment difference value, and Fi represents the functional detection result. This vector integrates multi-dimensional quality information into a unified mathematical representation, facilitating subsequent calculations. For example, in the production process of a batch of electronic components, W can take values from 0 to 100, Ri from 0.8 to 1.0, Di from 0 to 50, and Fi from 0.9 to 1.0. The selection of these parameters is based on actual statistical analysis of data distribution.

[0076] Then, the similarity between vectors is calculated by distance algorithm, using Euclidean distance D = √Σ(Vi-Vj) 2 This formula measures the difference between quality feature vectors of different batches, and the smaller the value, the more similar the batches. Vi and Vj are elements of any vector, and the closer the values, the higher the similarity. This formula can quickly evaluate the relative position of data points in a big data environment, and is suitable for most continuous data comparison scenarios. For example, when the difference between the parameters in two vectors is small, the value of D will be significantly lower than other combinations.

[0077] Next, a dynamic threshold Th = m + l * s is set, where m is the average distance of all vectors, s is the standard deviation, and l is the adjustment coefficient. m and s are used to reflect the overall distribution trend and dispersion, respectively, and their values are obtained based on historical data statistics. l is usually between 1 and 2, and can be manually adjusted to balance the false positive rate and detection efficiency. This threshold is dynamically adjusted according to data fluctuations, effectively addressing differences in data characteristics between different batches.

[0078] Finally, vectors with a similarity greater than the threshold are divided into the same cluster to identify potential risk batches. If the distance between a batch and other batches is less than or equal to Th, it may contain the same quality problems or risks and needs to be focused on. For example, in a certain instance, if the Wi, Ri, Di, etc. parameters of a batch are highly similar to known risk batches, it will be classified as a suspected high-risk batch.

[0079] Next, the application of dynamically adjusting the warehouse entry inspection strategy parameters according to the risk level in the clustering analysis result is further described. Based on the clustering result, the risk level Rk∈{1, 2, 3} is assigned, where 1 is low risk and 3 is high risk. This step will identify different quality trends of component batches through data clustering and assign them corresponding levels to represent the likelihood of potential quality problems. For example, in one embodiment, the clustering result shows that the electrical characteristics of some batches deviate less from the standard value, and are assigned Rk=1, while others perform poorly in thermal stability and are classified as Rk=3.

[0080] The sampling density function S=k*(Rk^a) is defined, where k is the basic parameter, usually taking a value between 0.1 and 1, and a is the power adjustment factor, usually taking a value between 0.5 and 2, and the optimal value needs to be determined according to the actual historical data. This formula is used to quantify the detection demand degree under different risk levels, i.e. high risk level will lead to higher sampling rate. For example, if Rk=3 and a=1, then S=k*3, which is significantly higher than the value in the low risk case, thereby strengthening the quality control of it.

[0081] The historical false detection rate δ is introduced for compensation calculation, S=S+γ*δ, where γ is the compensation coefficient, usually set between 0.01 and 0.1. This step reduces false positives caused by data fluctuations by adding historical error rates to improve detection accuracy. For example, a batch has a high false detection rate in the past, so the sampling rate is higher through compensation to improve the overall pass rate.

[0082] The adjusted sampling rate is applied to the detection task of the current batch. This step integrates all the previous calculations to generate a specific sampling plan suitable for the current batch. For example, for a high-risk batch, if the adjusted S value is 0.8, 80% of all components need to be sampled for detailed detection, thereby effectively controlling the quality risk.

[0083] Next, the application of the adjusted sampling strategy to the current batch of warehouse entry detection process is further described. A real-time detection feedback mechanism is established to obtain sampling result feedback information. This step aims to continuously collect data generated during the sampling process through the system to ensure that the quality status of the current batch can be understood in a timely manner. For example, when electronic components are entered into the warehouse, data is collected through sensors and detection equipment and transmitted to the analysis system for processing.

[0084] According to the current batch detection data, the risk level Rk is dynamically updated as Rk = Rk x (1 + p x E), where E is the abnormal detection ratio and p is the decay coefficient. This formula represents adjusting the original risk level based on the proportion of abnormal detection. E usually ranges from 0 to 1, measuring the proportion of abnormalities found in the current batch. p is a negative or zero coefficient that controls the magnitude of the risk level change, usually set between -0.1 and 0. When E increases, Rk will decrease accordingly, reflecting the effectiveness of the current batch quality control.

[0085] If Rk ≤ 1, return to the normal detection process; otherwise, continue to maintain high-frequency sampling. This step is based on the risk level to determine whether to maintain high-intensity sampling or return to normal. For example, if the abnormality ratio is low in a batch of sampling, causing Rk to decrease to below 1, the system will automatically reduce the sampling frequency to save resources and improve efficiency.

[0086] The updated strategy is fed back to the previous steps to achieve adaptive closed-loop optimization. This process ensures that the information obtained after each detection is absorbed by the system and improves subsequent operations. Specifically, the detection data is used to optimize the data collection and risk prediction model in the early stage, improving the intelligence and response speed of the overall quality detection system.

[0087] Next, the construction of the quality feature vector and further clustering analysis of the present invention are described. First, according to claim 5, external data sources are combined to increase dimensions, such as supplier credit rating (Cr) and environmental temperature and humidity fluctuation index (Ew). This step improves the comprehensiveness and predictive ability of the feature vector by introducing additional information. Cr usually ranges from 0 to 100, with higher values indicating stronger supplier reliability; Ew reflects the possible impact of the environment on components, which can range from 0 to 10. For example, in one embodiment, if a supplier has a Cr of 85 and the average temperature and humidity fluctuation index of the products it supplies is 2.5, these parameters will be included in the subsequent analysis.

[0088] Next, the new features are added to the original vector V to form an extended vector V = [V, Cr, Ew]. This step expands the feature space, making subsequent analysis more in-depth. For example, the original feature vector V may include temperature, voltage, and current, and after adding Cr and Ew, it can more comprehensively reflect the actual use environment and supply chain of the product.

[0089] Then, clustering is performed through an improved Kmeans algorithm, and the initial center point is obtained by calculating the historical mean. This improved method improves the stability of the algorithm and avoids inconsistency caused by random initialization. The historical mean is usually based on data from the past few months, ensuring that the initial position is representative. For example, in actual operation, the mean value from the feature vectors of past batches is extracted as the initial center.

[0090] The standard deviation σ c of each cluster is calculated, and a judgment criterion is set: if σ c > σ avg, the cluster needs to be paid special attention. σ avg is the average of the standard deviations of all clusters, which measures the overall dispersion. When the standard deviation of a cluster is significantly greater than the average, it means that the members within the cluster have large differences, and further inspection and analysis of quality problems are needed. For example, in a certain analysis, the σ c of a certain cluster is 3.2, and the σ avg is 1.8, so this cluster is considered as the focus of analysis.

[0091] Next, the dynamic adjustment of the warehouse entry sampling strategy parameters of the present application is further described. A hierarchical sampling scheme is developed according to the risk level, with a sampling frequency f = (1 + η * R k) p, where η and p are parameter coefficients. Among them, R k represents the risk level of a certain type of electronic components, and the higher the value represents the greater the possibility of quality problems, usually ranging from 0 to 1; η is an adjustment coefficient, used to control the influence of risk on the sampling frequency, generally set to a value between 0.5 and 2; p is the exponential coefficient, which determines the influence of risk level change on the sampling frequency, usually set to a value between 0.5 and 1. When R k takes a larger value, the result of the formula will increase significantly, thereby increasing the sampling frequency and achieving more stringent quality control.

[0092] The concept of time window is introduced in the process of high-frequency sampling. If no quality problems are found within T consecutive time, the degraded detection mode is triggered. Among them, T is a pre-set time threshold, representing the length of time without abnormal conditions, which can be set according to specific industry standards and historical data. For example, in the electronic component industry, T can be set to 7 days or 14 days. If no problems occur within the specified time, it means that the quality of the current batch is relatively stable, and the sampling intensity can be appropriately reduced to reduce unnecessary resource waste.

[0093] If there is an abnormal record during the period, the degraded period is extended and manual review is started. Abnormal records may include the discovery of non-conforming products or quality fluctuation phenomena. In this case, not only does the sampling interval need to be extended, but manual review also needs to be introduced to ensure that quality problems are not missed. For example, if the resistance deviation of a certain type of resistor exceeds the standard during sampling, the detection interval is automatically extended and a special person is arranged for review to prevent the problem from expanding.

[0094] The above rules are incorporated into the decision tree model to achieve intelligent judgment. By training data to build a decision tree, factors such as risk level, time window, and abnormal records are used as judgment nodes, and the system can autonomously choose whether to enter the degraded detection or maintain the original state according to the actual situation. For example, in a certain embodiment, an electronic component manufacturer uses a decision tree model to dynamically adjust the sampling method based on supplier historical records, product types, and past quality inspection results, resulting in a 30% improvement in quality inspection efficiency and a 15% reduction in defect rate.

[0095] Next, the application according to the data after normalization to build quality feature vector is further described. Time-varying features are extracted, and the stability coefficient Stab = (ΔFi / Fi_avg) of each batch is calculated, where ΔFi is the failure change between adjacent batches, indicating the difference in the number of failures between adjacent batches, usually ranging from 0 to 1, and if ΔFi tends to 0, it means high stability; Fi_avg is the average number of failures, which is used to quantify the average level of the batch to ensure that the data of different batches are compared on the same scale. Stability coefficient helps to identify the trend of fluctuation between batches, so as to judge whether the quality is stable. For example, in a certain electronic component production line, if the first batch of failures is 10 and the second batch is 15, then ΔFi is 5, Fi_avg is 12.5, and Stab is 0.4, indicating low stability. A nonlinear transformation factor φ is added, which is used to describe the nonlinear change of quality caused by environmental or process disturbance, and its value range can be set to 0 to 1, and by adjusting the nonlinear strength, the model can better capture complex patterns. For example, when a batch is in a certain temperature and humidity environment, φ can be set to 0.6 to reflect the unstable trend. A multi-dimensional tensor representation method is constructed: T = [V, Stab, φ], where V represents the normalized original quality variable, Stab is the stability coefficient calculated above, and φ is the nonlinear factor, which forms a multi-dimensional data structure by combination. Use deep learning model to train it, improve feature expression ability, deep neural network can automatically extract high-level features, enhance prediction accuracy. For example, use convolutional neural network to analyze time series features and identify potential quality risk patterns to improve electronic component life prediction accuracy.

[0096] Next, the application of dynamically adjusting the warehouse entry sampling strategy parameters is further described. First, a cost-benefit model is constructed. This model is used to evaluate the impact of different sampling strategies on overall cost and detection effect, where the cost C is equal to the number of samples N multiplied by the unit detection cost P and multiplied by the frequency coefficient K. N represents the number of electronic components in each batch for detection, P is the average cost required for each electronic component detection, and K represents the number of times or frequency of each sampling. This formula is designed to consider various cost factors in the detection process. For example, in one embodiment, when the supplier provides 1000 electronic component batches, the single detection cost is 5 yuan, and K is set to 2, then the total cost C is 1000 × 5 × 2 = 10000 yuan.

[0097] Secondly, the optimization goal is to minimize the probability of missed detection Pm. This goal represents the desire to reduce the probability of quality issues occurring due to missed detection as much as possible. In practice, this can be achieved by increasing the frequency of detection or the sample size, but is limited by the budget. For example, in a specific scenario, if the rate of missed detection is higher than a set threshold, it may be necessary to increase the frequency of detection to reduce the risk of quality incidents. The constraint is C≤B, where B is the maximum cost that the enterprise can accept, to ensure that the entire quality inspection process does not exceed the financial plan. For example, if the enterprise budget limit B is set to 12000 yuan, then the cost must be guaranteed not to exceed this value when designing the detection scheme.

[0098] Subsequently, the optimal K value is solved by the Lagrange multiplier method. This is the most commonly used method in mathematics, which can be used to find the extreme value of a function under certain conditions. In this scenario, the objective function is Pm, which is constrained by C≤B, so the Lagrange multiplier is used to incorporate the constraint into the objective function, and the optimal K value is obtained. For example, by calculation, it is determined that when K is set to 1.8, the probability of missed detection reaches the lowest point while satisfying the budget condition. This allows the enterprise to control expenses while achieving the best detection efficiency.

[0099] Finally, the sampling strategy parameters are adjusted according to the results of the solution, and the constraint condition library is updated. This step ensures that the model can continuously optimize and adapt to changes in the environment, such as fluctuations in detection costs, changes in the supply chain, or the introduction of new products, etc. For example, when the single detection cost of a new type of component rises, the value of K can be recalculated and adjusted, and this information can be recorded for future reference and updating of decision-making basis.

[0100] Next, the application of the adjusted sampling strategy to the current batch of incoming inspection process is described. Fuzzy decision logic is introduced when performing sampling, and the priority P = 1exp(μ×Rk) is defined, where μ is the sensitivity coefficient and Rk represents the quality risk value of the kth sample. The value of μ in the formula is usually between 0 and 5, used to adjust the sensitivity to the risk value, and the optimal value depends on the specific business needs and data distribution, determined by experimental adjustment. The meaning of the formula is that as the risk value Rk increases, the priority P becomes higher, so that high-risk samples are more likely to be detected first. For example, in the incoming inspection of a batch of capacitor components, if a batch has a high frequency of failure records, the system will assign higher priority to individual samples in that batch.

[0101] The sampling item distribution is determined in combination with the sample characteristics, and a probability matrix M = [M1, M2,..., Mn] is set, where Mj represents the probability of the jth detection item being selected. The values of this matrix are usually between 0 and 1, and the sum of all elements is 1. The value of Mj reflects the importance of different detection items to quality control, and is set according to the sample's material, purpose, etc. For example, for high-precision sensors, the probability of being selected for pressure testing should be higher than that for appearance testing.

[0102] The actual results are recorded after each inspection, and the weight parameters are continuously optimized using reinforcement learning. By comparing the predictions with the actual results, the model can adjust the weights of each detection item so that future inspections are closer to the actual quality situation. For example, if it is found that a certain detection item has limited ability to identify unqualified products during a period of operation, the system will reduce the probability value of this item to improve efficiency.

[0103] The updated strategy is synchronized to all relevant system modules to ensure consistency. This process ensures that the data flow and judgment criteria of the entire detection system remain uniform, avoiding misjudgment or repeated checks due to module differences. For example, when the algorithm adjusts the probability matrix of a certain detection item, all modules responsible for data collection, analysis, and reporting will synchronize the corresponding configuration parameters, forming an efficient closed-loop management.

[0104] The electronic component quality prediction method based on big data analysis of the present application includes: first, collecting and normalizing the historical quality data and detection records of each batch of electronic components in the supply chain to eliminate the differences and redundancies between different sources of data, ensuring the accuracy and consistency of subsequent analysis results; then, based on the processed data, a quality feature vector is constructed, and clustering analysis technology is used to identify batches with similar characteristics, further identifying risk batches that may have potential quality problems; on this basis, the system evaluates the risk level based on the results of clustering analysis and dynamically adjusts the parameters of the warehouse inspection strategy, such as sampling ratio, detection index weight, or detection frequency, so that more stringent detection measures are taken for high-risk batches, while for low-risk batches, the detection process is optimized to reduce unnecessary resource waste; finally, the optimized inspection strategy is applied to the current batch of warehouse detection processes, and the model is continuously corrected and optimized through a real-time feedback mechanism, thereby effectively improving the detection ability of quality abnormalities and significantly reducing the risk of missed detection of quality abnormalities. This method realizes dynamic regulation and control of the warehouse inspection strategy through deep mining and intelligent analysis of big data, improves detection efficiency and guarantees product quality, solving the problem of missed detection caused by ignoring batch differences in traditional static inspection strategies.

[0105] In the method, program, system, device, etc. of the embodiments of the present application, they can be executed or implemented in a single or multiple networked computers, or can be practiced in a distributed computing environment. In the embodiments of the present specification, in these distributed computing environments, tasks can be performed by remote processing devices connected through a communication network.

[0106] Those skilled in the art should understand that the embodiments of the present specification can be provided as a method, a system or a computer program product. Therefore, those skilled in the art can conceive that the implementation of the functional modules / units or controllers and the related method steps illustrated by the above-mentioned embodiments can be realized by software, hardware and a combination of software and hardware.

[0107] Unless explicitly stated, the actions or steps of the methods, programs recited in the embodiments of the present application do not have to be performed in a specific order and still achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0108] In this document, multiple embodiments of the present application are described, but for the sake of brevity, the description of each embodiment is not exhaustive and identical or similar features or parts between different embodiments can be omitted. In this document, "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means applicable to at least one embodiment or example according to the present application, but not all embodiments. The above terms do not necessarily mean referring to the same embodiment or example. Those skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of different embodiments or examples without contradiction.

[0109] The exemplary systems and methods of the present application have been specifically illustrated and described herein through the use of the above-mentioned embodiments, which are merely examples of the best modes of implementing the systems and methods. Those skilled in the art can understand that various changes can be made to the embodiments of the systems and methods described herein without departing from the spirit and scope of the present application defined in the appended claims.

Claims

1. A method for predicting the quality of electronic components based on big data analysis, characterized in that, include: Historical quality data and testing records of each batch of electronic components in the supply chain are collected and normalized. Based on the normalized data, a quality feature vector is constructed and cluster analysis is performed to identify potentially risky batches. The parameters of the warehousing sampling strategy are dynamically adjusted based on the risk level in the cluster analysis results. The adjusted sampling strategy will be applied to the current batch's incoming inspection process to reduce the risk of missing quality anomalies.

2. The method for predicting the quality of electronic components based on big data analysis according to claim 1, characterized in that, The collection and normalization of historical quality data and testing records of each batch of electronic components in the supply chain further includes: Obtain the testing indicators for each batch of electronic components, such as failure count (Fi), pass rate (Qi), and transportation time (Ti); All batches of test indicators were standardized so that the value of each indicator fell within the range of [0,1]. The normalized weighting coefficient W for each batch is calculated based on the following formula: W = (Fi + α * Qi) / β, where α is the correction coefficient and β is the total baseline parameter. The normalized data is constructed into a matrix for subsequent processing.

3. The method for predicting the quality of electronic components based on big data analysis according to claim 2, characterized in that, The construction of quality feature vectors based on the normalized data and the subsequent cluster analysis further include: Extract the data features of each batch to form a vector V = [W, Ri, Di, Fi], where Ri is the supply channel stability coefficient and Di is the transportation environment difference value; The similarity between vectors is calculated using a distance algorithm, employing Euclidean distance D = √Σ(Vi Vj). 2 ; Set the dynamic threshold Th = μ + λσ, where μ is the mean, σ is the standard deviation, and λ is the adjustment coefficient; Vectors with similarity greater than a threshold are grouped into the same cluster to identify potentially risky batches.

4. The method for predicting the quality of electronic components based on big data analysis according to claim 3, characterized in that, The step of dynamically adjusting the parameters of the warehousing sampling strategy based on the risk level in the cluster analysis results further includes: Risk levels Rk∈{1,2,3} are assigned based on clustering results, where 1 represents low risk and 3 represents high risk; Define the sampling density function S = k*(Rk^a), where k is the basic parameter and a is the power adjustment factor; The historical false detection rate δ is introduced for compensation calculation, S=S+γ*δ, where γ is the compensation coefficient; The adjusted sampling ratio will be applied to the testing tasks for the current batch.

5. The method for predicting the quality of electronic components based on big data analysis according to claim 4, characterized in that, The application of the adjusted sampling strategy to the current batch of inbound inspection process further includes: Establish a real-time detection and feedback mechanism to obtain feedback information on sampling results; The risk level Rk = Rk × (1ρ * E) is dynamically updated based on the current batch of test data, where E is the abnormal detection rate and ρ is the attenuation coefficient. If Rk≤1, then resume the regular testing process; otherwise, continue with high-frequency sampling. The updated strategy is fed back to the preceding steps to achieve adaptive closed-loop optimization.

6. The method for predicting the quality of electronic components based on big data analysis according to claim 5, characterized in that, The construction of the quality feature vector and the cluster analysis further include: Incorporate external data sources to add dimensions, such as supplier credit rating (Cr) and environmental temperature and humidity fluctuation index (Ew); The new features are added to the original vector V to form the extended vector V = [V,Cr,Ew]; Clustering is performed using an improved K-means algorithm, with the initial centroids calculated from the historical mean. Calculate the standard deviation σ_c for each cluster and set the judgment criterion: if σ_c > σ_avg, then the cluster needs to be monitored separately.

7. The method for predicting the quality of electronic components based on big data analysis according to claim 6, characterized in that, The dynamically adjusted inbound sampling inspection strategy parameters further include: A graded sampling plan is formulated based on the risk level, and the sampling frequency is set as f = (1 + η * Rk)^p, where η and p are parameter coefficients. In the high-frequency sampling process, the concept of a time window is introduced. If no quality problem is found within a continuous period of T, a downgraded detection mode is triggered. If any abnormal records are found during the period, the downgrade period will be extended and a manual review will be initiated. The above rules are incorporated into the decision tree model to achieve intelligent judgment.

8. The method for predicting the quality of electronic components based on big data analysis according to claim 7, characterized in that, The step of constructing a quality feature vector based on the normalized data further includes: Extract time-varying features and calculate the stability coefficient Stab = (ΔFi / Fi_avg) for each batch, where ΔFi is the change in failure between adjacent batches and Fi_avg is the average number of failures. A nonlinear transformation factor φ is added to reflect the unstable trend of batch quality; Construct a multidimensional tensor representation: T = [V, Stab, φ]; Use a deep learning model to train it and improve its feature representation ability.

9. The method for predicting the quality of electronic components based on big data analysis according to claim 8, characterized in that, The dynamically adjusted inbound sampling inspection strategy parameters further include: Construct a cost-benefit model, with cost C = (N × P × K), where N is the number of samples, P is the unit testing cost, and K is the frequency coefficient; The optimization objective is to minimize the false negative probability Pm, with the constraint C≤B, where B is the upper limit of the budget. The optimal value of K is determined using the Lagrange multiplier method. Adjust the sampling strategy parameters based on the solution results and update the constraint library.

10. The method for predicting the quality of electronic components based on big data analysis according to claim 9, characterized in that, The application of the adjusted sampling strategy to the current batch of inbound inspection process further includes: When performing random sampling, fuzzy decision logic is introduced, and the priority is defined as P = 1exp(μ×Rk), where μ is the sensitivity coefficient. The distribution of sampling items is determined by combining the characteristics of the samples, and a probability matrix M = [M1, M2, ..., Mn] is set, where Mj represents the probability that the j-th detection item is selected; The actual results are recorded after each sampling, and the weight parameters are continuously optimized using reinforcement learning. The update strategy will be synchronized to all relevant system modules to ensure consistency.

Images