An ESD soft failure prediction method and system based on industrial IoT

Abstract

The present invention proposes an ESD soft failure prediction method and system based on industrial IoT. The ESD soft failure prediction system includes: a data acquisition module for collecting customer return data, repair data, ICT and FCT test data of the product and ESD protection status data monitored by the IoT system; a data storage module for storing the data collected by the data acquisition module; a data preprocessing module for performing data fusion and feature data screening on the data collected by the data acquisition module; a data analysis module for establishing an ESD soft failure prediction model based on the data processed by the data preprocessing module; an application module for analyzing the ICT and FCT test data of the product in real time according to the ESD soft failure prediction model, thereby predicting whether the product has ESD soft failure. The ESD soft failure prediction method and system based on industrial IoT of the present invention are novel in design and highly practical.

Description

Technical Field

[0001] The present invention relates to the field of industrial big data, and in particular to an ESD soft failure prediction method and system based on industrial IoT. Background Art

[0002] During the electronics manufacturing process, ESD failure is a significant type of failure. While ESD damage doesn't necessarily lead to performance or functionality failure, it can affect product reliability, dramatically reducing product lifespan and potentially leading to product rejection. To improve product reliability and reduce rejection rates, it's crucial to identify ESD soft failures before shipment.

[0003] However, the quality testing methods (ICT / FCT) currently in common use can only identify products with functional failures, but cannot screen out products with ESD soft failures. Summary of the Invention

[0004] In response to the above technical problems, the present invention provides an ESD soft failure prediction method and system based on industrial IoT.

[0005] The technical solutions proposed in the present invention are as follows:

[0006] The present invention proposes an ESD soft failure prediction system based on industrial IoT, comprising:

[0007] The data acquisition module is used to collect product return data, repair data, ICT and FCT test data, and ESD protection status data monitored by the IoT system;

[0008] A data storage module, used for storing the data collected by the data collection module;

[0009] The data preprocessing module is used to perform data fusion and feature data screening on the data collected by the data acquisition module;

[0010] A data analysis module is used to establish an ESD soft failure prediction model based on the data processed by the data preprocessing module;

[0011] The application module is used to analyze the product's ICT and FCT test data in real time based on the ESD soft failure prediction model, thereby predicting whether the product has ESD soft failure.

[0012] In the ESD soft failure prediction system of the present invention, the data preprocessing module is used to filter out the list data of target products that were returned due to ESD damage from the customer return data based on the repair data; obtain the ICT and FCT test data and ESD protection status data of the target products based on the list data of the target products, thereby forming an abnormal data set;

[0013] The data preprocessing module is also used to obtain the ICT and FCT test data and ESD protection status data of products in the same batch that have not been rejected by customers based on the production time of the target rejected products, and form a normal data set.

[0014] In the ESD soft failure prediction system of the present invention, data fusion refers to associating the product's customer return data, repair data, ICT and FCT test data, and ESD protection status data according to the product ID;

[0015] The feature data screening process is:

[0016] All ICT and FCT test data in the union of normal and abnormal data sets are normalized. The normalization algorithm is:

[0017] k ij =|(T ij -μ j ) / σ j |;

[0018] Among them, T ij represents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset;

[0019] μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset;

[0020] σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset;

[0021] The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in,

[0022]

[0023] Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =max(k i1 ,k i2 ,…,kim ); with the K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to the K value in the total number of returned products and the proportion of the number of non-returned products corresponding to the K value in the total number of non-returned products as the vertical coordinates, establish a coordinate system and draw corresponding curves, thereby obtaining a curve of the proportion of returned products versus the K value and a curve of the proportion of non-returned products versus the K value; obtain the K value at the position where the distance between the curves of the proportion of returned products versus the K value and the curves of the proportion of non-returned products versus the K value in the vertical coordinate direction is the largest, and record it as k';

[0024] The products whose K value is greater than k' in the abnormal data set are defined as soft failure products; the ICT and FCT test data of soft failure products are extracted to form a soft failure product data set; and B is constructed. n(m+1) ,in,

[0025]

[0026] Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product;

[0027] When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product;

[0028] Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process fp The calculation method is:

[0029]

[0030] Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process;

[0031] i represents the number of monitoring points that alarm simultaneously in the p-th process;

[0032] l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time;

[0033] Determine whether each soft failure product has a process with an alarm index greater than 0. If so, the soft failure product is determined to have ESD soft failure. Delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set. Then combine the processed abnormal data set with the normal data set to obtain a modeling data set. The modeling data set is denoted as C. n(m+p+1) ,

[0034]

[0035] C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model.

[0036] The present invention also proposes an ESD soft failure prediction method based on industrial IoT, comprising the following steps:

[0037] Step S1: Collect product return and repair data, and filter out a list of target returned products due to ESD damage from the return data based on the repair data. Then, obtain the ICT and FCT test data of the target returned products and the ESD protection status data of the IoT monitoring system based on the list data to form an abnormal data set.

[0038] Step S2: Based on the production time of the target product that was rejected by the customer, obtain the ICT and FCT test data and ESD protection status data of the products in the same batch that were not rejected by the customer to form a normal data set;

[0039] A modeling dataset is formed based on the normal and abnormal datasets. Based on the modeling dataset, a prediction model is constructed to predict whether a product has an ESD soft failure based on the product's ICT and FCT test data and ESD protection status data.

[0040] Step S3: Analyze the product's ICT and FCT test data and ESD protection status data in real time based on the ESD soft failure prediction model, so as to predict whether the product has ESD soft failure.

[0041] In the above-mentioned soft failure prediction method of the present invention, step S3 is performed by adopting a machine learning algorithm.

[0042] In the soft failure prediction method of the present invention, step S2 includes the following steps:

[0043] Step 2.1: Normalize all ICT and FCT test data in the union of the normal dataset and the abnormal dataset. The normalization algorithm is:

[0044] k ij=|(T ij -μ j ) / σ j |;

[0045] Among them, T ij represents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset;

[0046] μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset;

[0047] σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset;

[0048] The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in,

[0049]

[0050] Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =max(k i1 ,k i2 ,…,k im ); with the K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to the K value in the total number of returned products and the proportion of the number of non-returned products corresponding to the K value in the total number of non-returned products as the vertical coordinates, establish a coordinate system and draw corresponding curves, thereby obtaining a curve of the proportion of returned products versus the K value and a curve of the proportion of non-returned products versus the K value; obtain the K value at the position where the distance between the curves of the proportion of returned products versus the K value and the curves of the proportion of non-returned products versus the K value in the vertical coordinate direction is the largest, and record it as k';

[0051] Step 2.2: Define the products whose K value is greater than the absolute value of k' in the abnormal data set as soft failure products; extract the ICT and FCT test data of soft failure products to form a soft failure product data set; construct B n(m+1) ,in,

[0052]

[0053] Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product;

[0054] When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product;

[0055] Step 2.3: Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process is fp The calculation method is:

[0056]

[0057] Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process;

[0058] i represents the number of monitoring points that alarm simultaneously in the p-th process;

[0059] l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time;

[0060] Step 2.4: Determine whether each soft failure product has a process with an alarm index greater than 0. If so, determine that the soft failure product has ESD soft failure; delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set; then combine the processed abnormal data set with the normal data set to obtain a modeling data set; the modeling data set is recorded as C n(m+p+1) ,

[0061]

[0062] C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model.

[0063] In the above-mentioned soft failure prediction method of the present invention, the product return data, repair data, ICT and FCT test data and ESD protection status data of the target returned product are derived from the MES system interface and the industrial IoT system.

[0064] The Industrial IoT-based ESD soft failure prediction method and system of the present invention overcomes the inability of ICT / FCT to detect ESD soft failures. During the production process, ESD protection status data collected by the Industrial IoT system can be used to predict whether a product has experienced an ESD soft failure. The Industrial IoT-based ESD soft failure prediction method and system of the present invention are novel in design and highly practical. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 FIG2 shows a functional module block diagram of an ESD soft failure prediction system according to a preferred embodiment of the present invention;

[0066] Figure 2 Shown Figure 1 The flowchart of the ESD soft failure prediction system shown is as follows: DETAILED DESCRIPTION

[0067] The technical problem to be solved by this invention is that the currently commonly used quality testing methods (ICT / FCT) can only identify products with functional failures, but cannot screen out products with ESD soft failures. To address this technical problem, the present invention proposes a method and system for predicting ESD soft failures based on the Industrial Internet of Things (IoT). This system utilizes the Industrial Internet of Things (IIoT), an information management system, and machine learning methods to establish an ESD soft failure prediction model, enabling the identification of ESD soft failures during the production process.

[0068] In order to facilitate those skilled in the art to understand and implement the present invention, the technical objectives, technical solutions and technical effects of the present invention will be further elaborated in detail below with reference to the accompanying drawings and specific embodiments.

[0069] The present invention proposes an ESD soft failure prediction method based on industrial IoT, comprising the following steps:

[0070] Step S1: Collect product return (RMA) data and repair data. Based on the repair data, filter out the list of target returned products due to ESD damage from the return data. Obtain the ICT and FCT test data of the target returned products and the ESD protection status data of the IoT monitoring system based on the list of target returned products to form an abnormal data set.

[0071] In this step, the customer return data of the product includes the product transfer data;

[0072] The product's return RMA and repair data, as well as the ICT and FCT test data and ESD protection status data of the target returned product, are primarily sourced from the MES system interface and the Industrial IoT. Step S2: Based on the production time of the target returned product, obtain the ICT and FCT test data and ESD protection status data of products from the same batch that have not been returned, forming a normal data set.

[0073] A modeling dataset is formed based on the normal and abnormal datasets. Based on the modeling dataset, a prediction model is constructed to predict whether a product has an ESD soft failure based on the product's ICT and FCT test data and ESD protection status data.

[0074] Here, the same batch of products refers to products from the same batch of materials, the same production line, and the same shift. Here, step S2 is performed by adopting a machine learning algorithm.

[0075] Step S2 includes the following steps:

[0076] Step 2.1: Normalize all ICT and FCT test data in the union of the normal dataset and the abnormal dataset. The normalization algorithm is:

[0077] k ij =|(T ij -μ j ) / σ j |;

[0078] Among them, T ij represents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset;

[0079] μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset;

[0080] σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset;

[0081] The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in,

[0082]

[0083] Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =maxk i1 ,k i2 ,…,k im ; With K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to K value in the total number of returned products and the proportion of the number of non-returned products corresponding to K value in the total number of non-returned products as the vertical coordinate, establish a coordinate system and draw corresponding curves, thereby obtaining the proportion of returned products - K value curve and the proportion of non-returned products - K value curve; obtain the K value at the position with the largest distance between the proportion of returned products - K value curve and the proportion of non-returned products - K value curve in the vertical coordinate direction, and record it as k';

[0084] Step 2.2: Define the products whose K value is greater than the absolute value of k' in the abnormal data set as soft failure products; extract the ICT and FCT test data of soft failure products to form a soft failure product data set; construct B n(m+1) ,in,

[0085]

[0086] Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product;

[0087] When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product;

[0088] Step 2.3: Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process is fp The calculation method is:

[0089]

[0090] Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process;

[0091] i represents the number of monitoring points that alarm simultaneously in the p-th process;

[0092] l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time;

[0093] Step 2.4: Determine whether each soft failure product has a process with an alarm index greater than 0. If so, determine that the soft failure product has ESD soft failure; delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set; then combine the processed abnormal data set with the normal data set to obtain a modeling data set; the modeling data set is recorded as C n(m+p+1) ,

[0094]

[0095] C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model.

[0096] Step S3: Analyze the product's ICT and FCT test data and ESD protection status data in real time based on the ESD soft failure prediction model, so as to predict whether the product has ESD soft failure.

[0097] like Figure 1-Figure 2 As shown, Figure 1 FIG2 shows a functional module block diagram of an ESD soft failure prediction system according to a preferred embodiment of the present invention; Figure 2 Shown Figure 1 The flowchart of the ESD soft failure prediction system shown is as follows:

[0098] The present invention also proposes an ESD soft failure prediction system based on industrial IoT, comprising:

[0099] The data acquisition module 100 is used to collect product return data, repair data, ICT and FCT test data, and ESD protection status data monitored by the IoT system;

[0100] Here, the product return data includes product pass data; ICT and FCT test data can include ESD protection status data and ICT / FCT quality data;

[0101] The data storage module 200 is used to store the data collected by the data collection module 100;

[0102] Preferably, the data storage module 200 is used to uniformly store the data collected by the data collection module 100 according to product numbers.

[0103] The data preprocessing module 300 is used to perform data fusion and feature data screening on the data collected by the data collection module 100 in order to improve the efficiency and quality of data mining;

[0104] The data preprocessing module 300 is configured to filter out a list of target products that were rejected due to ESD damage from the customer rejection data based on the repair data; obtain the ICT and FCT test data and ESD protection status data of the target products based on the list of target products, thereby forming an abnormal data set;

[0105] The data preprocessing module 300 is further configured to obtain the ICT and FCT test data and ESD protection status data of products of the same batch that have not been rejected by the customer based on the production time of the target rejected product, and form a normal data set.

[0106] Here, products from the same batch refer to products from the same batch of materials, the same production line, and the same shift.

[0107] Here, data fusion refers to associating the product's ID with the product's customer return data, repair data, and ICT and FCT test data; feature data screening refers to screening through the product's ICT and FCT test data.

[0108] After data fusion, the data pre-processing module 300 can obtain the ESD protection status of each site the product passes through during the production process and flag information indicating whether it is an ESD soft failure.

[0109] The data analysis module 400 is used to establish an ESD soft failure prediction model based on the data processed by the data preprocessing module 300;

[0110] Here, the data analysis module 400 works by adopting a machine learning algorithm to establish a classification prediction model for the preprocessed data set, and uses ICT / FCT as an intermediate bridge to screen the data to improve the accuracy of the model.

[0111] The process of feature data screening and ESD soft failure prediction model establishment is as follows:

[0112] All ICT and FCT test data in the union of normal and abnormal data sets are normalized. The normalization algorithm is:

[0113] k ij =|(T ij -μ j ) / σ j |;

[0114] Among them, T ijrepresents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset;

[0115] μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset;

[0116] σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset;

[0117] The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in,

[0118]

[0119] Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =maxk i1 ,k i2 ,…,k im ; With K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to K value in the total number of returned products and the proportion of the number of non-returned products corresponding to K value in the total number of non-returned products as the vertical coordinate, establish a coordinate system and draw corresponding curves, thereby obtaining the proportion of returned products - K value curve and the proportion of non-returned products - K value curve; obtain the K value at the position with the largest distance between the proportion of returned products - K value curve and the proportion of non-returned products - K value curve in the vertical coordinate direction, and record it as k';

[0120] The products whose K value is greater than k' in the abnormal data set are defined as soft failure products; the ICT and FCT test data of soft failure products are extracted to form a soft failure product data set; and B is constructed. n(m+1) ,in,

[0121]

[0122] Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product;

[0123] When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product;

[0124] Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process fp The calculation method is:

[0125]

[0126] Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process;

[0127] i represents the number of monitoring points that alarm simultaneously in the p-th process;

[0128] l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time;

[0129] Determine whether each soft failure product has a process with an alarm index greater than 0. If so, the soft failure product is determined to have ESD soft failure. Delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set. Then combine the processed abnormal data set with the normal data set to obtain a modeling data set. The modeling data set is denoted as C. n(m+p+1) ,

[0130]

[0131] C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model.

[0132] Product return and repair data, as well as ICT and FCT test data for target returned products, are primarily sourced from MES system interfaces and the Industrial IoT. This ICT and FCT test data primarily includes the ICT and FCT test data for ESD-sensitive components in the target returned products before shipment.

[0133] The application module 500 is used to analyze the ICT and FCT test data of the product in real time according to the ESD soft failure prediction model, so as to predict whether the product has ESD soft failure.

[0134] The Industrial IoT-based ESD soft failure prediction method and system of the present invention overcomes the inability of ICT / FCT to detect ESD soft failures. During the production process, ESD protection status data collected by the Industrial IoT system can be used to predict whether a product has experienced an ESD soft failure. The Industrial IoT-based ESD soft failure prediction method and system of the present invention are novel in design and highly practical.

[0135] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. An ESD soft failure prediction system based on industrial IoT, characterized by: include: A data acquisition module (100) is used to collect product return data, repair data, ICT and FCT test data, and ESD protection status data monitored by the IoT system; A data storage module (200) for storing data collected by the data collection module (100); A data preprocessing module (300) is used to perform data fusion and feature data screening on the data collected by the data collection module (100); A data analysis module (400) is used to establish an ESD soft failure prediction model based on the data processed by the data preprocessing module (300); An application module (500) is used to analyze the ICT and FCT test data of a product in real time according to an ESD soft failure prediction model, thereby predicting whether the product has an ESD soft failure; A data pre-processing module (300) is used to filter out list data of target products rejected by customers due to ESD damage from customer rejection data based on the repair data; Obtain the ICT and FCT test data and ESD protection status data of the target rejected products based on the list data of the target rejected products to form an abnormal data set; The data preprocessing module (300) is further used to obtain ICT and FCT test data and ESD protection status data of products in the same batch that have not been rejected by customers based on the production time of the target product rejected by customers, and form a normal data set; Data fusion refers to associating the product's return data, repair data, ICT and FCT test data, and ESD protection status data based on the product ID; The process of feature data screening and ESD soft failure prediction model establishment is as follows: All ICT and FCT test data in the union of normal and abnormal data sets are normalized. The normalization algorithm is: k ij =|(T ij -m j ) / s j |; Among them, T ij represents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset; μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset; σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset; The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in, Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =max(k i1 ,k i2 ,…,k im ); with the K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to the K value in the total number of returned products and the proportion of the number of non-returned products corresponding to the K value in the total number of non-returned products as the vertical coordinates, establish a coordinate system and draw corresponding curves, thereby obtaining a curve of the proportion of returned products versus the K value and a curve of the proportion of non-returned products versus the K value; obtain the K value at the position where the distance between the curves of the proportion of returned products versus the K value and the curves of the proportion of non-returned products versus the K value in the vertical coordinate direction is the largest, and record it as k'; The products whose K value is greater than the absolute value of k' in the abnormal data set are defined as soft failure products; the ICT and FCT test data of soft failure products are extracted to form a soft failure product data set; and B is constructed. n(m+1) ,in, Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product; When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product; Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process fp The calculation method is: Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process; i represents the number of monitoring points that alarm simultaneously in the p-th process; l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time; Determine whether each soft failure product has a process with an alarm index greater than 0. If so, the soft failure product is determined to have ESD soft failure. Delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set. Then combine the processed abnormal data set with the normal data set to obtain a modeling data set. The modeling data set is denoted as C. n(m+p+1) , C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model.

2. An ESD soft failure prediction method based on industrial IoT, characterized in that: The following steps are involved: Step S1: Collect product return and repair data, and filter out a list of target returned products due to ESD damage from the return data based on the repair data. Then, obtain the ICT and FCT test data of the target returned products and the ESD protection status data of the IoT monitoring system based on the list data to form an abnormal data set. Step S2: Based on the production time of the target product that was rejected by the customer, obtain the ICT and FCT test data and ESD protection status data of the products in the same batch that were not rejected by the customer to form a normal data set; Composing a modeling dataset based on normal and abnormal datasets; Based on the modeling data set, a prediction model is constructed to predict whether a product has ESD soft failure based on the product's ICT and FCT test data and ESD protection status data; Step S3: Analyze the product's ICT and FCT test data and ESD protection status data in real time based on the ESD soft failure prediction model to predict whether the product has ESD soft failure. Step S2 includes the following steps: Step 2.1: Normalize all ICT and FCT test data in the union of the normal dataset and the abnormal dataset. The normalization algorithm is: k ij =|(T ij -m j ) / s j |; Among them, T ij represents the jth ICT and FCT test data of the i-th product in the union of the normal dataset and the abnormal dataset; μ j represents the mean of the j-th ICT and FCT test data of all products in the normal dataset; σ j represents the standard deviation of the j-th ICT and FCT test data of all products in the normal dataset; The total number of products in the union of the normal data set and the abnormal data set is recorded as n, and the total number of ICT and FCT test data items of the products is recorded as m. The 1st to mth ICT and FCT test data after normalization of the first product, the 1st to mth ICT and FCT test data after normalization of the second product, ..., the 1st to mth ICT and FCT test data after normalization of the nth product are used to construct A. nm ;in, Calculate the K values ​​of all products in the union of the normal data set and the abnormal data set respectively; record the K value of the i-th product in the union of the normal data set and the abnormal data set as K i , where K i =max(k i1 ,k i2 ,…,k im ); with the K value as the horizontal coordinate, and the proportion of the number of returned products corresponding to the K value in the total number of returned products and the proportion of the number of non-returned products corresponding to the K value in the total number of non-returned products as the vertical coordinates, establish a coordinate system and draw corresponding curves, thereby obtaining a curve of the proportion of returned products versus the K value and a curve of the proportion of non-returned products versus the K value; obtain the K value at the position where the distance between the curves of the proportion of returned products versus the K value and the curves of the proportion of non-returned products versus the K value in the vertical coordinate direction is the largest, and record it as k'; Step 2.2: Define the products whose K value is greater than the absolute value of k' in the abnormal data set as soft failure products; extract the ICT and FCT test data of soft failure products to form a soft failure product data set; construct B n(m+1) ,in, Here, y1 represents the value of whether the first product in the union of the normal data set and the abnormal data set is a soft failure product; y2 represents the value of whether the second product in the union of the normal data set and the abnormal data set is a soft failure product; ...; y n A value indicating whether the nth product in the union of the normal data set and the abnormal data set is a soft failure product; When y1, y2, ..., y n When any one of them is 0, it means that the corresponding product is not a soft failure product; when y1, y2, ..., y n When any one of them is 1, it means that the corresponding product is a soft failure product; Step 2.3: Calculate the alarm index of each product passing through each process; the alarm index r of the fth product passing through the pth process is fp The calculation method is: Where j represents the maximum number of monitoring points that alarm simultaneously on the p-th process; i represents the number of monitoring points that alarm simultaneously in the p-th process; l i Indicates the duration of the alarm when i monitoring points on the p-th process give an alarm at the same time; Step 2.4: Determine whether each soft failure product has a process with an alarm index greater than 0. If so, determine that the soft failure product has ESD soft failure; delete the data of the soft failure products without ESD soft failure from the abnormal data set to obtain a processed abnormal data set; then combine the processed abnormal data set with the normal data set to obtain a modeling data set; the modeling data set is recorded as C n(m+p+1) , C n(m+p+1) The first m+p columns are used as independent variables, and the m+p+1th column is used as the dependent variable. The machine learning classification algorithm is used for analysis to construct an ESD soft failure prediction model. The product return data, repair data, ICT and FCT test data, and ESD protection status data of the target returned products are derived from the MES system interface and the industrial IoT system.

3. The soft failure prediction method according to claim 2, characterized in that: Step S3 is performed by employing a machine learning algorithm.

Images