A product quality risk early warning method and system based on artificial intelligence

Abstract

The application discloses a product quality risk early warning method and system based on artificial intelligence, comprising the following steps: acquiring product chain historical data to construct production stability and environment coefficient, refining historical production data to obtain production quality data, calculating first production deviation and second production deviation, constructing a product quality front-end prediction model, inputting to-be-predicted product chain data into the product quality front-end prediction model to obtain first risk quality and first risk data, constructing a product quality back-end prediction model, inputting to-be-predicted product chain data into the product quality back-end prediction model to obtain second risk quality, correcting the first risk data to obtain a risk coefficient, determining a quality risk result according to the first risk quality, the second risk quality and the risk coefficient, and performing product quality blockchain early warning. The method can not only improve the accuracy of product quality risk early warning, but also has good interpretability and can be directly applied to a product quality risk early warning system.

Description

Technical Field

[0001] The present invention relates to the field of quality early warning, and in particular to an artificial intelligence-based product quality risk early warning method and system. Background Art

[0002] In today's context of rapid development of intelligent manufacturing and industry, product quality control has become a core link for enterprises to enhance their competitiveness and safeguard their market reputation. With the expansion of production scale and the increase in product complexity, the requirements for product quality risk warning are also gradually increasing. Artificial intelligence technology, with its powerful data analysis, pattern recognition and prediction capabilities, provides intelligent solutions for product quality risk warning, which can effectively reduce the incidence of quality accidents, optimize production processes, and promote the transformation of the manufacturing industry towards intelligence and leanness.

[0003] Traditional quality early warning technology has many limitations: on the one hand, it relies on historical data statistics and fixed rule settings, and is difficult to adapt to complex and changeable production environments and dynamic changes in product quality characteristics over time; on the other hand, traditional methods do not have enough depth in data mining, and are unable to fully explore the potential correlation information hidden in massive production data, resulting in a significant reduction in the accuracy and advanceness of early warnings. The present invention designs a product quality risk early warning method and system based on artificial intelligence, by constructing front-end and back-end prediction models for product quality, combining production stability and environmental factors to correct risk data, and introducing blockchain technology to achieve full-chain quality early warning, overcoming the shortcomings of traditional technology, achieving quality monitoring of the entire product life cycle, and improving the accuracy and timeliness of product quality early warnings, providing a more reliable technical guarantee for product quality risk early warnings, and having far-reaching significance for promoting the manufacturing industry to move towards a high-quality development stage and improving the overall quality level and international competitiveness of my country's manufacturing industry. Summary of the Invention

[0004] The purpose of the present invention is to provide a product quality risk early warning method and system based on artificial intelligence.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Acquire historical product chain data, pre-process the historical product chain data, and construct production stability and environmental coefficients based on the historical product chain data; the historical product chain data includes historical production data, historical factory sampling data, historical storage and transportation sales data, and after-sales quality data;

[0008] Refining the historical production data according to the process to obtain production quality data, obtaining a first production deviation and a factory quality deviation based on the production quality data and the historical factory sampling inspection data, and processing the first production deviation to obtain a second production deviation;

[0009] Building a product quality front-end prediction model based on the second production deviation and the factory quality deviation, and inputting the to-be-predicted product chain data into the product quality front-end prediction model to obtain a first risk quality and first risk data;

[0010] Obtain storage and transportation sales deviation characteristics and abnormal operation characteristics based on the historical storage and transportation sales data, build a product quality back-end prediction model in combination with the after-sales quality data, and input the product chain data to be predicted into the product quality back-end prediction model to obtain a second risk quality;

[0011] Correcting the first risk data according to the production stability and the environmental factor to obtain second risk data, determining a risk coefficient according to the second risk data, and determining a quality risk result according to the first risk quality, the second risk quality, and the risk coefficient;

[0012] A product quality blockchain early warning is performed based on the first risk quality, the second risk quality and the quality risk result; the product quality blockchain early warning includes a production chain early warning, a storage, transportation and sales chain early warning and a full chain early warning.

[0013] Furthermore, a method for constructing production stability and environmental coefficients based on the product chain historical data includes:

[0014] Through clustering, historical production data is divided into production equipment status data, production environment data, production process parameters and production material parameters; through clustering, historical storage and transportation sales data is divided into storage and transportation environment data, storage and transportation feature data and abnormal features;

[0015] Divide the production equipment status data into production sections and calculate the production stability of each production section:

[0016] ,

[0017] ,

[0018] ,

[0019] in For production section i production stability, is the failure rate weight, For production section i The failure rate score, is the actual failure rate of the corresponding section, The industry benchmark failure rate, is the inspection weight, For production section i The maintenance frequency score, is the running fatigue weight, For production section i Running fatigue score, The continuous operation time of the equipment in the corresponding section, For the optimal operating time of the equipment, is the current stabilization weight, For production section i The current stability score, For production section i The actual number of maintenance is the critical maintenance frequency, is the fatigue attenuation coefficient, 、 are the variance penalty coefficient and the skewness penalty coefficient, respectively. 、 Production sections i The current variance and power variance, 、 Production sections i The current and power deviations of

[0020] The production stability of the next interval is adjusted according to the production stability of the previous interval, and the average production stability of all intervals is taken as the production stability of the product chain. The expression is:

[0021] ,

[0022] in For production section i Adjusted production stability, For production section i Dependence coefficient with the previous production section;

[0023] Input the production environment data and storage and transportation environment data of the same product chain into the environmental function to obtain the environmental coefficient:

[0024] ,

[0025] ,

[0026] ,

[0027] ,

[0028] ,

[0029] in is the environmental factor, is the production weight, is the production environment coefficient, is the storage and transportation weight, is the storage and transportation environment coefficient, is the mechanical damage weight, is the chemical damage weight, is the vibration attenuation coefficient, is the deviation between the current vibration intensity and the safe vibration intensity, For production environment noise, is the standard production environment noise, Score the static electricity accumulation voltage, For gas j Concentration, including harmful gas concentration and VOCs concentration, is the number of dust particles per unit area, is the electrostatic voltage, 、 is the skewness of common environmental factors, , It is the first common environmental factor set, including light intensity, radiation intensity, temperature, humidity, , It is the first common environmental factor set, including temperature, humidity, ultraviolet intensity, and visible light intensity. is the atmospheric pressure coefficient, Pre is the atmospheric pressure, G is the mechanical shock during storage and transportation, For reference mechanical shock, For the maximum design mechanical shock, is the vibration energy coefficient, is the energy attenuation coefficient, is the smoothing coefficient, is vibration energy, It is the safety value of vibration energy.

[0030] Furthermore, the method for obtaining the second production deviation includes:

[0031] The production sections on the product chain are divided into a first pending production section and a second pending production section; the first pending production section is related to product manufacturing, etching, and shaping; the second pending production section is related to product assembly and packaging;

[0032] Fitting the first undetermined production section and the second undetermined production section to obtain a first core production section, a second auxiliary production section, and a third auxiliary production section; the first core production section is an overlapping section of the first undetermined production section and the second undetermined production section; the second auxiliary production section is a non-overlapping section of the first undetermined production section; and the third auxiliary production section is a non-overlapping section of the second undetermined production section;

[0033] The first core production section, the second auxiliary production section, and the third auxiliary production section are transformed according to the production action to obtain the first pending process, the second pending process, and the third pending process. The average process density of each process is calculated to obtain the first process density threshold and the second process density threshold. The first pending process is defined as the first evaluation process. The second pending process is screened according to the first process density threshold to obtain the second evaluation process. The third pending process is screened according to the second process density threshold to obtain the third evaluation process. The average process density is determined by the ratio of the data volume and process length of the corresponding process. The first process density threshold is the average of the first average process density and the second average process density. The second process density threshold is the average of the second average process density and the third average process density.

[0034] The first evaluation process, the second evaluation process, and the third evaluation process are combined into a necessary evaluation process. The production data of the necessary evaluation process is taken as the production quality data. The deviation between the production quality data and the corresponding reference value is calculated to obtain the first production deviation. The deviation between the historical factory sampling data and the corresponding factory standard is calculated to obtain the factory quality deviation.

[0035] The first production deviation is input into the self-attention module to obtain the production deviation weight, and the second production deviation is obtained according to the first production deviation and the corresponding production deviation weight.

[0036] Furthermore, the method for obtaining the first risk quality and the first risk data includes:

[0037] Obtain the quality fluctuation range and quality risk category based on the sampling elimination data, quality elimination category and corresponding factory standards, and determine the corresponding risk probability based on the factory quality deviation, quality fluctuation range and quality risk category;

[0038] The risk category, risk probability, factory quality deviation and second production deviation are combined into a front-end comprehensive set, which is randomly divided into a front-end training set and a front-end test set according to a ratio of 6:4.

[0039] Constructing a product quality front-end prediction model, the product quality front-end prediction model includes an input layer probability prediction layer, a similarity prediction layer, a data prediction layer and an output layer;

[0040] The probability prediction layer adopts gradient boosting tree to learn the non-linear relationship between the risk category, the risk probability and the second production deviation, to predict the first risk quality, and to attach the corresponding production abnormality label; the first risk quality includes the risk category and the risk probability;

[0041] The similarity prediction layer is used to calculate the deviation similarity between the second production deviation and the historical production deviation, to extract the historical factory quality deviation corresponding to the maximum deviation similarity, and to determine the first predicted risk data according to the deviation similarity and the historical factory quality deviation;

[0042] The data prediction layer learns the non-linear relationship between the factory quality deviation and the second production deviation through BP neural network, to predict the second predicted risk data;

[0043] The output layer includes a first full connection layer and a second full connection layer, the first full connection layer is connected with the probability prediction layer to output the first risk quality, and the second full connection layer is connected with the similarity prediction layer and the data prediction layer, to weight and fuse the first predicted risk data and the second predicted risk data and to output the first risk data.

[0044] Further, the method for obtaining the second risk quality comprises:

[0045] The storage and transportation environment data and the storage and transportation characteristic data are compared with the corresponding storage and transportation specifications to obtain the storage and transportation sales deviation characteristics; the storage and transportation characteristic data includes the transportation mode and the storage and transportation sales time;

[0046] The Apriori algorithm is used to mine the association rules of the abnormal characteristics in the historical storage and transportation sales data to obtain the abnormal operation characteristics; the abnormal characteristics include the storage and transportation sales abnormal operation and the transportation accident; the abnormal operation characteristics are the abnormal characteristics related to the after-sales quality data;

[0047] A product quality back-end prediction model based on Bayesian network is constructed, and the specific steps include:

[0048] The network nodes are determined according to the storage and transportation sales deviation characteristics, the abnormal operation characteristics and the after-sales quality data, and the causal paths are determined according to the field experience; the network nodes include the root node, the hidden node and the leaf node; the causal paths include the direct path and the cross path;

[0049] The hybrid conditional probability table is generated by using the expert knowledge base and the historical storage and transportation sales data, the Bayesian network model is generated according to the network nodes, the causal paths and the hybrid conditional probability table, and the Bayesian network model is continuously updated by using the storage and transportation sales deviation characteristics, the abnormal operation characteristics and the after-sales quality data, and the expression is:

[0050] ,

[0051] ,

[0052] in is the conditional probability of incremental update, indicating t +1 time node X At a given parent node Pa The updated conditional probability under the condition for t Current node at the moment X At a given parent node Pa Conditional probability The corresponding sample size, is the number of newly added samples, Based on the new sample The new conditional probability estimate obtained is, is the time decay factor, is the time interval, is the Bayesian shock update probability, The severity of the accident, To control the impact strength, is the new conditional probability after the emergency occurs, is the original conditional probability before the emergency occurs;

[0053] Dynamic pruning is performed based on causal importance, and the optimal sub-network is selected to obtain the product quality back-end prediction model. The dynamic pruning conditions are:

[0054] ,

[0055] in For the edge The pruning score of DC For data completeness, For nodes X and Y Mutual information is the confidence of the model, is the loss function Loss Reaction edge Weight The gradient, is the path activation frequency weight, For the edge The frequency with which the path is activated;

[0056] The product chain data to be predicted is input into the product quality back-end prediction model to obtain a second risk quality, and a corresponding storage, transportation and sales abnormality label is attached; the second risk quality includes risk category and risk probability.

[0057] Furthermore, the method for determining the quality risk result includes:

[0058] The third risk quality is obtained by intersecting the first risk quality and the second risk quality according to the risk category, the third risk quality is updated by weighting the same category risk probability, and the fourth risk quality is obtained by intersecting the first risk quality and the second risk quality with the updated third risk quality;

[0059] The risk coefficient is determined according to the production stability, the environmental coefficient and the first risk data; the risk coefficient is positively correlated with the environmental coefficient and the first risk data, and is negatively correlated with the production stability;

[0060] The fourth risk quality is modified according to the risk coefficient to obtain a quality risk result.

[0061] Further, the method for product quality blockchain early warning comprises:

[0062] The first risk quality is compared with a production risk probability threshold, a risk process is obtained by process tracing for a risk category exceeding the production risk probability threshold, and a production chain early warning is performed on the risk process;

[0063] A second risk quality to be early warned is selected according to a storage and transportation and sales risk probability threshold, a risk root node is determined by root node tracing according to a product quality back-end prediction model, and a storage and transportation and sales chain early warning is performed according to a risk root node category; the root node category comprises a storage and transportation environment category, a storage and transportation mode category and an abnormal feature category;

[0064] A label of the quality risk result is extracted, a production chain early warning is performed according to the production risk probability threshold or a storage and transportation and sales chain early warning is performed according to the storage and transportation and sales risk probability threshold when a certain risk category only contains a production abnormality label or a storage and transportation and sales abnormality label, process tracing and root node tracing are performed when a certain risk category contains both the production abnormality label and the storage and transportation and sales abnormality label, and a whole chain early warning is performed according to a tracing result.

[0065] In a second aspect, a product quality risk early warning system based on artificial intelligence comprises:

[0066] A deviation calculation module is configured to obtain production quality data by refining the production data according to a process, obtain a first production deviation and a factory quality deviation from the production quality data and historical factory shipment inspection data, and process the first production deviation to obtain a second production deviation;

[0067] A production prediction module is configured to construct a product quality front-end prediction model according to the second production deviation and the factory quality deviation, input product chain data to be predicted into the product quality front-end prediction model to obtain a first risk quality and first risk data;

[0068] Storage, transportation and sales forecasting module: used to obtain storage, transportation and sales deviation characteristics and abnormal operation characteristics based on the historical storage, transportation and sales data, build a product quality back-end forecasting model in combination with the after-sales quality data, and input the product chain data to be forecasted into the product quality back-end forecasting model to obtain the second risk quality;

[0069] Correction module: used to construct production stability and environmental coefficient based on product chain historical data, correct the first risk data based on the production stability and environmental coefficient to obtain second risk data, determine the risk coefficient based on the second risk data, and determine the quality risk result based on the first risk quality, the second risk quality, and the risk coefficient;

[0070] Early warning module: used to view, manage and store the first risk quality, the second risk quality and the quality risk results, and perform product quality blockchain early warning based on the first risk quality, the second risk quality and the quality risk results.

[0071] The beneficial effects of the present invention are:

[0072] The present invention is an artificial intelligence-based product quality risk early warning method and system. Compared with the existing technology, the present invention has the following technical effects:

[0073] The present invention can improve the accuracy of product quality risk warning by constructing coefficients, refining data, building models, data correction and blockchain early warning steps, thereby improving the precision of product quality risk warning and optimizing product quality risk warning technology, which can greatly save resources and improve work efficiency. It can realize risk warning of product quality and conduct real-time quality monitoring of the product throughout its life cycle, which helps to improve the accuracy and timeliness of early warning. It has far-reaching significance for promoting enterprises to realize intelligent quality management and efficient production, and for the manufacturing industry to move towards a high-quality development stage. It can adapt to the product quality risk warning needs of different product quality risk systems and different users, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] 图1 This is a flowchart of the steps of an artificial intelligence-based product quality risk early warning method of the present invention. DETAILED DESCRIPTION

[0075] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0076] The present invention provides an artificial intelligence-based product quality risk early warning method and system, comprising the following steps:

[0077] like 图1As shown, in this embodiment, the following steps are included:

[0078] Acquire historical product chain data, pre-process the historical product chain data, and construct production stability and environmental coefficients based on the historical product chain data; the historical product chain data includes historical production data, historical factory sampling data, historical storage and transportation sales data, and after-sales quality data;

[0079] Refining the historical production data according to the process to obtain production quality data, obtaining a first production deviation and a factory quality deviation based on the production quality data and the historical factory sampling inspection data, and processing the first production deviation to obtain a second production deviation;

[0080] Building a product quality front-end prediction model based on the second production deviation and the factory quality deviation, and inputting the to-be-predicted product chain data into the product quality front-end prediction model to obtain a first risk quality and first risk data;

[0081] Obtain storage and transportation sales deviation characteristics and abnormal operation characteristics based on the historical storage and transportation sales data, build a product quality back-end prediction model in combination with the after-sales quality data, and input the product chain data to be predicted into the product quality back-end prediction model to obtain a second risk quality;

[0082] Correcting the first risk data according to the production stability and the environmental factor to obtain second risk data, determining a risk coefficient according to the second risk data, and determining a quality risk result according to the first risk quality, the second risk quality, and the risk coefficient;

[0083] A product quality blockchain early warning is performed based on the first risk quality, the second risk quality and the quality risk result; the product quality blockchain early warning includes a production chain early warning, a storage, transportation and sales chain early warning and a full chain early warning.

[0084] In this embodiment, the method for constructing production stability and environmental coefficient based on the product chain historical data includes:

[0085] Through clustering, historical production data is divided into production equipment status data, production environment data, production process parameters and production material parameters; through clustering, historical storage and transportation sales data is divided into storage and transportation environment data, storage and transportation feature data and abnormal features;

[0086] Divide the production equipment status data into production sections and calculate the production stability of each production section:

[0087] ,

[0088] ,

[0089] ,

[0090] in For production section i production stability, is the failure rate weight, For production section i The failure rate score, is the actual failure rate of the corresponding section, The industry benchmark failure rate, is the inspection weight, For production section i The maintenance frequency score, is the running fatigue weight, For production section i Running fatigue score, The continuous operation time of the equipment in the corresponding section, For the optimal operating time of the equipment, is the current stabilization weight, For production section i The current stability score, For production section i The actual number of maintenance is the critical maintenance frequency, is the fatigue attenuation coefficient, 、 are the variance penalty coefficient and the skewness penalty coefficient, respectively. 、 Production sections i The current variance and power variance, 、 Production sections i The current and power deviations of

[0091] The production stability of the next interval is adjusted according to the production stability of the previous interval, and the average production stability of all intervals is taken as the production stability of the product chain. The expression is:

[0092] ,

[0093] in For production section i Adjusted production stability, For production section i Dependence coefficient with the previous production section;

[0094] Input the production environment data and storage and transportation environment data of the same product chain into the environmental function to obtain the environmental coefficient:

[0095] ,

[0096] ,

[0097] ,

[0098] ,

[0099] ,

[0100] in is the environmental factor, is the production weight, is the production environment coefficient, is the storage and transportation weight, is the storage and transportation environment coefficient, is the mechanical damage weight, is the chemical damage weight, is the vibration attenuation coefficient, is the deviation between the current vibration intensity and the safe vibration intensity, For production environment noise, is the standard production environment noise, Score the static electricity accumulation voltage, For gas j Concentration, including harmful gas concentration and VOCs concentration, is the number of dust particles per unit area, is the electrostatic voltage, 、 is the skewness of common environmental factors, , It is the first common environmental factor set, including light intensity, radiation intensity, temperature, humidity, , It is the first common environmental factor set, including temperature, humidity, ultraviolet intensity, and visible light intensity. is the atmospheric pressure coefficient, Pre is the atmospheric pressure, G is the mechanical shock during storage and transportation, For reference mechanical shock, For the maximum design mechanical shock, is the vibration energy coefficient, is the energy attenuation coefficient, is the smoothing coefficient, is vibration energy, is the vibration energy safety value;

[0101] In the actual assessment, a quality risk warning was conducted on a smartphone motherboard. 1,000 pieces of historical production chain data were collected. The historical production data were classified by clustering. The production equipment status data was divided into 10 production sections according to the production time. The failure rate weight was taken. 0.3, maintenance weight 0.2, running fatigue weight 0.2, current stability weight is 0.3;

[0102] Taking production section 2 as an example, the actual failure rate 0.05, industry benchmark failure rate 0.1, actual number of maintenance 5. Critical maintenance frequency 8. Continuous operation time of the equipment 8h, the optimal operating time of the equipment For 10h, fatigue attenuation coefficient is 0.5, current variance is 0.01, power variance 0.02, current skewness is 0.1, power skewness is 0.2, variance penalty coefficient is 0.5, skewness penalty coefficient is 0.5, and the production stability of production section 2 is calculated is 0.458, calculate the production stability of each production section, according to the production section i Dependency coefficient with the previous production section The production stability of the adjusted interval is calculated, and the production stability of all intervals is calculated to obtain the production stability of the product chain. is 0.76;

[0103] Take production weight 0.6, storage and transportation weight 0.4, mechanical damage weight 0.2, chemical damage weight 0.3, vibration attenuation coefficient is 0.5, energy attenuation coefficient is 0.4, smoothing coefficient The environmental coefficient is 0.1, and the production environment data and storage and transportation environment data are substituted into the calculated environmental coefficient, which is 0.2712.

[0104] In this embodiment, the method for obtaining the second production deviation includes:

[0105] The production sections on the product chain are divided into a first pending production section and a second pending production section; the first pending production section is related to product manufacturing, etching, and shaping; the second pending production section is related to product assembly and packaging;

[0106] Fitting the first undetermined production section and the second undetermined production section to obtain a first core production section, a second auxiliary production section, and a third auxiliary production section; the first core production section is an overlapping section of the first undetermined production section and the second undetermined production section; the second auxiliary production section is a non-overlapping section of the first undetermined production section; and the third auxiliary production section is a non-overlapping section of the second undetermined production section;

[0107] The first core production section, the second auxiliary production section, and the third auxiliary production section are transformed according to the production action to obtain the first pending process, the second pending process, and the third pending process. The average process density of each process is calculated to obtain the first process density threshold and the second process density threshold. The first pending process is defined as the first evaluation process. The second pending process is screened according to the first process density threshold to obtain the second evaluation process. The third pending process is screened according to the second process density threshold to obtain the third evaluation process. The average process density is determined by the ratio of the data volume and process length of the corresponding process. The first process density threshold is the average of the first average process density and the second average process density. The second process density threshold is the average of the second average process density and the third average process density.

[0108] The first evaluation process, the second evaluation process, and the third evaluation process are combined into a necessary evaluation process. The production data of the necessary evaluation process is taken as the production quality data. The deviation between the production quality data and the corresponding reference value is calculated to obtain the first production deviation. The deviation between the historical factory sampling data and the corresponding factory standard is calculated to obtain the factory quality deviation.

[0109] Input the first production deviation into the self-attention module to obtain the production deviation weight, and obtain the second production deviation according to the first production deviation and the corresponding production deviation weight;

[0110] In the actual assessment, production sections 1-4 were divided into the first undetermined production section, including Section 1 - substrate pretreatment before chip soldering, Section 2 - chip mounting, Section 3 - chip reflow, and Section 4 - testing the quality of chip mounting and soldering. Production sections 5-10 were divided into the second undetermined production section, including Section 5 - passive component mounting, Section 6 - passive component reflow, Section 7 - mainboard cleaning, Section 8 - electrical performance testing, Section 9 - installation of protective components, and Section 10 - final appearance inspection and packaging.

[0111] There is overlap between Section 2 - Chip Mounting, Section 3 - Chip Reflow, and Section 6 - Passive Component Reflow. There is also overlap between Section 4 - Chip Mounting and Soldering Quality Inspection and Section 8 - Electrical Performance Testing. Regional fitting was performed to obtain the first core production section (including Sections 2, 3, 4, 6, and 8), the second auxiliary production section (including Section 1), and the third auxiliary production section (including Sections 5, 7, 9, and 10).

[0112] Each of the initial 10 production sections includes multiple processes. Taking section 2 of the first core production section (including processes such as substrate positioning, chip positioning, and chip placement) as an example, the chip positioning process data volume is 100 and the process length is 10. The chip positioning process density is calculated to be 10. The density of all processes in all sections of the first core production section is calculated to obtain a first average process density of 12.2. Similarly, the density of all processes in all sections of the second auxiliary production section is calculated to obtain a second average process density of 9.4. The density of all processes in all sections of the third auxiliary production section is calculated to obtain a third average process density of 7.6. The first process density threshold is 10.8 and the second process density threshold is 8.5. The processes in the second pending process with a density greater than the first process density threshold are screened to obtain the second evaluation process. The processes in the third pending process with a density greater than the second process density threshold are screened to obtain the third evaluation process to form the necessary evaluation processes.

[0113] The deviation between the production quality data and the corresponding reference value in the evaluation process is calculated as the first production deviation. The first production deviation is input into the self-attention module to obtain the production deviation weight. The first production deviation is multiplied by the corresponding production deviation weight to obtain the second production deviation.

[0114] In this embodiment, the method for obtaining the first risk quality and the first risk data includes:

[0115] Obtain the quality fluctuation range and quality risk category based on the sampling elimination data, quality elimination category and corresponding factory standards, and determine the corresponding risk probability based on the factory quality deviation, quality fluctuation range and quality risk category;

[0116] The risk category, risk probability, factory quality deviation and second production deviation are combined into a front-end comprehensive set, which is randomly divided into a front-end training set and a front-end test set according to a ratio of 6:4.

[0117] Constructing a product quality front-end prediction model, the product quality front-end prediction model includes an input layer probability prediction layer, a similarity prediction layer, a data prediction layer and an output layer;

[0118] The probability prediction layer uses a gradient boosting tree to learn the nonlinear relationship between risk category, risk probability, and second production deviation, predicts the first risk quality, and attaches a corresponding production anomaly label; the first risk quality includes the risk category and risk probability;

[0119] The similarity prediction layer is used to calculate the deviation similarity between the second production deviation and the historical production deviation, extract the historical factory quality deviation corresponding to the historical production deviation with the largest deviation similarity, and determine the first predicted risk data based on the deviation similarity and the historical factory quality deviation;

[0120] The data prediction layer learns the nonlinear relationship between the factory quality deviation and the second production deviation through the BP neural network to predict the second prediction risk data;

[0121] The output layer includes a first fully connected layer and a second fully connected layer, the first fully connected layer is connected to the probability prediction layer to output the first risk quality, the second fully connected layer is connected to the similarity prediction layer and the data prediction layer, and the first predicted risk data and the second predicted risk data are weightedly fused to output the first risk data;

[0122] In the actual evaluation, 100 random inspection data points included 10 eliminated data points, including 4 solder joint defects, 3 chip pin short circuits, and 3 electrical performance parameter substandards. The corresponding factory standard comparison determined the quality fluctuation range to be: ±10% of the standard value for solder joint strength, ±5% of the standard value for chip pin resistance, and ±8% of the standard value for electrical performance parameters. The corresponding quality risk categories and probabilities were 0.6 for solder defects, 0.3 for short circuit failures, and 0.1 for performance issues. These formed the front-end comprehensive data set for training and testing the front-end prediction model for product quality.

[0123] Input the product chain data to be predicted into the product quality front-end prediction model:

[0124] The probability prediction layer obtains the first risk quality: welding defect (short circuit risk) 0.5, placement deviation 0.4;

[0125] The similarity prediction layer calculates the maximum deviation similarity between the second production deviation of the product chain data to be predicted and the historical production deviation, which is 0.8. The corresponding historical factory quality deviations are extracted: solder point equipment temperature deviation 2.2°C, solder point pressure deviation -0.1MPa, and mounting position deviation 0.06mm. The historical factory quality deviations are calculated. (Maximum deviation similarity -1) The first risk data is obtained: solder point equipment temperature deviation 1.8°C, solder point pressure deviation -0.082MPa, and placement position deviation 0.049mm;

[0126] The data prediction layer obtains the second risk data: solder point equipment temperature deviation 2°C, solder point pressure deviation -0.15MPa, and placement position deviation 0.06mm;

[0127] The first predicted risk data and the second predicted risk data are weighted and fused to output first risk data: a welding spot device temperature deviation of 1.9℃, a welding spot pressure deviation of-0.116MPa, and a mounting position deviation of 0.0545mm.

[0128] In the embodiment, the method for obtaining the second risk quality comprises:

[0129] The storage and transportation environment data and the storage and transportation characteristic data are compared with the corresponding storage and transportation specifications to obtain storage and transportation sales deviation characteristics; the storage and transportation characteristic data comprises a transportation mode and a storage and transportation sales time;

[0130] An Apriori algorithm is used to perform association rule mining on abnormal characteristics in historical storage and transportation sales data to obtain abnormal operation characteristics; the abnormal characteristics comprise storage and transportation sales abnormal operations and transportation accidents; the abnormal operation characteristics are abnormal characteristics related to after-sales quality data;

[0131] A product quality back-end prediction model based on a Bayesian network is constructed, and the specific steps comprise:

[0132] Network nodes are determined according to the storage and transportation sales deviation characteristics, the abnormal operation characteristics and the after-sales quality data, and a causal path is determined according to domain experience; the network nodes comprise root nodes, hidden nodes and leaf nodes; the causal path comprises a direct path and a cross path;

[0133] A hybrid conditional probability table is generated using an expert knowledge base and historical storage and transportation sales data, a Bayesian network model is generated according to the network nodes, the causal path and the hybrid conditional probability table, the Bayesian network model is continuously updated using the storage and transportation sales deviation characteristics, the abnormal operation characteristics and the after-sales quality data, and the expression is:

[0134] ,

[0135] ,

[0136] wherein is an incrementally updated conditional probability, represents t a node X at a time point Pa updated under the condition that parent nodes are given, t is a current node X at a time point Pa under the condition that parent nodes are given, is a corresponding sample quantity, is a newly added sample quantity, is a new conditional probability estimate value obtained based on the newly added sample is the time decay factor, is the time interval, is the Bayesian shock update probability, The severity of the accident, To control the impact strength, is the new conditional probability after the emergency occurs, is the original conditional probability before the emergency occurs;

[0137] Dynamic pruning is performed based on causal importance, and the optimal sub-network is selected to obtain the product quality back-end prediction model. The dynamic pruning conditions are:

[0138] ,

[0139] in For the edge The pruning score of DC For data completeness, For nodes X and Y Mutual information is the confidence of the model, is the loss function Loss Reaction edge Weight The gradient, is the path activation frequency weight, For the edge The frequency with which the path is activated;

[0140] Inputting the product chain data to be predicted into the product quality backend prediction model to obtain a second risk quality, and attaching a corresponding storage, transportation and sales abnormality label; the second risk quality includes a risk category and a risk probability;

[0141] In the actual evaluation, a Bayesian network is constructed, wherein the root node includes temperature and humidity deviation, transportation mode code, transportation time deviation, number of loading and unloading collisions, stacking violation index and transportation accident level; the implicit nodes include environmental stress accumulation factor and mechanical damage coupling coefficient; the environmental stress accumulation factor and the mechanical damage coupling coefficient are constructed by storage and transportation environment data and transportation accidents respectively; the leaf nodes include after-sales quality data, appearance defects, functional failure and life attenuation; the direct paths include "temperature and humidity deviation → environmental stress accumulation factor → material aging → life attenuation", "loading and unloading collision → mechanical damage coefficient → structural deformation → appearance defect loading and unloading collision → mechanical damage coefficient → structural deformation → appearance defect"; the cross paths include "transportation time × accident level → fatigue damage → functional failure";

[0142] Take the time decay factor 0.1, control the impact strength 2, path activation frequency weight 0.2, with the node X At a given parent node Pa Conditional probability Update, for example, the number of samples =100, the number of new samples =20, the new conditional probability estimate 0.8, time interval is 1, the updated conditional probability is 0.72, with the edge For example, data completeness DC is 0.6, mutual information The confidence level of the model is 0.9. is 0.95, gradient is 0.2, the frequency of path activation is 1.5, calculate the pruning score The edge is retained if it is 0.525 (greater than the pruning score threshold of 0.5). , update the product quality back-end prediction model, input the product chain data to be predicted into the product quality back-end prediction model to obtain the second risk quality: circuit erosion damage 0.5, collision loosening (short circuit risk) 0.44, and performance degradation 0.1.

[0143] In this embodiment, the method for determining the quality risk result includes:

[0144] According to the risk category, the first risk quality and the second risk quality are intersected to obtain a third risk quality, the third risk quality is updated by weighted calculation of the risk probabilities of the same category, and the fourth risk quality is obtained by taking the union of the non-intersecting parts of the first risk quality and the second risk quality and the updated third risk quality;

[0145] determining a risk coefficient based on production stability, an environmental coefficient, and the first risk data; wherein the risk coefficient is positively correlated with the environmental coefficient and the first risk data, and negatively correlated with production stability;

[0146] Correct the corresponding risk probability of the fourth risk quality according to the risk coefficient to obtain the quality risk result;

[0147] In the actual evaluation, the third risk quality is obtained by taking the intersection of the first risk quality and the second risk quality, and the short circuit risk is (0.44+0.5) / 2=0.47, and the fourth risk quality is obtained: mounting deviation 0.4, circuit erosion damage 0.5, performance degradation 0.1, and short circuit risk 0.47;

[0148] The first risk data (solder point equipment temperature deviation 1.9°C, solder point pressure deviation -0.116MPa, placement position deviation 0.0545mm) is adjusted based on the sum of the production stability of 0.76 and the environmental factor of 0.2712, which is 1.0312. The second risk data (solder point equipment temperature deviation 1.96°C, solder point pressure deviation -0.12MPa, placement position deviation 0.056mm) is obtained. The risk coefficient is determined based on the second risk data and the corresponding floating ranges of 1.5°C, 0.075MPa, and 0.05mm. ,Adjusting the fourth risk quality, we get the quality risk results: mounting deviation 0.4632, circuit erosion damage 0.579, performance degradation 0.1158, and short circuit risk 0.5443.

[0149] In this embodiment, the method for performing product quality blockchain early warning includes:

[0150] Compare the first risk quality with the production risk probability threshold, trace the process to obtain the risk process for the risk category that exceeds the production risk probability threshold, and issue a production chain early warning for the risk process;

[0151] The second risk quality to be warned is selected based on the storage, transportation and sales risk probability threshold, and the root node is traced back to determine the risk root node based on the product quality backend prediction model. Storage, transportation and sales chain warnings are issued based on the risk root node category; the root node categories include storage and transportation environment category, storage and transportation method category, and abnormal feature category.

[0152] Extract labels from quality risk results. When a risk category only contains production anomaly labels or storage, transportation, and sales anomaly labels, issue a production chain warning based on the production risk probability threshold, or issue a storage, transportation, and sales chain warning based on the storage, transportation, and sales risk probability threshold. When a risk category contains both production anomaly labels and storage, transportation, and sales anomaly labels, perform process tracing and root node tracing, and issue a full-chain warning based on the tracing results.

[0153] In the actual assessment, the chip welding process is given a production chain early warning based on the first risk quality: welding defect (short circuit risk) 0.5, placement deviation 0.4 (the risk probability threshold is both 0.5);

[0154] Based on the second risk quality: circuit erosion damage 0.5, collision loosening (short circuit risk) 0.44, and performance degradation 0.1 (risk probability thresholds are 0.5, 0.5, and 0.3, respectively), we identify the risk root nodes related to temperature and humidity, and issue early warnings for storage, transportation, and sales chains related to storage and transportation environment processes.

[0155] Extract labels for quality risk results: mounting deviation 0.4632 (production abnormality label), circuit erosion damage 0.579 (storage, transportation and sales abnormality label), performance degradation 0.1158 (storage, transportation and sales abnormality label), short circuit risk 0.5443 (production abnormality label, storage, transportation and sales abnormality label). Process traceability is carried out, and production chain warnings are issued for chip mounting and positioning processes. Storage and transportation chain warnings are issued for storage and transportation environment processes. Full-chain warnings are issued for processes corresponding to reflow soldering, transportation methods, and abnormal operations.

[0156] Second, an AI-based product quality risk early warning system includes:

[0157] Deviation calculation module: used to extract the production data according to the process to obtain production quality data, obtain a first production deviation and a factory quality deviation based on the production quality data and historical factory sampling inspection data, and process the first production deviation to obtain a second production deviation;

[0158] Production prediction module: used to build a product quality front-end prediction model based on the second production deviation and the factory quality deviation, and input the product chain data to be predicted into the product quality front-end prediction model to obtain the first risk quality and first risk data;

[0159] Storage, transportation and sales forecasting module: used to obtain storage, transportation and sales deviation characteristics and abnormal operation characteristics based on the historical storage, transportation and sales data, build a product quality back-end forecasting model in combination with the after-sales quality data, and input the product chain data to be forecasted into the product quality back-end forecasting model to obtain the second risk quality;

[0160] Correction module: used to construct production stability and environmental coefficient based on product chain historical data, correct the first risk data based on the production stability and environmental coefficient to obtain second risk data, determine the risk coefficient based on the second risk data, and determine the quality risk result based on the first risk quality, the second risk quality, and the risk coefficient;

[0161] Early warning module: used to view, manage and store the first risk quality, the second risk quality and the quality risk results, and perform product quality blockchain early warning based on the first risk quality, the second risk quality and the quality risk results.

[0162] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A product quality risk early warning method based on artificial intelligence, characterized in that: The following steps are involved: S1. Acquire historical product chain data, pre-process the historical product chain data, and construct production stability and environmental coefficients based on the historical product chain data; the historical product chain data includes historical production data, historical factory inspection data, historical storage, transportation and sales data, and after-sales quality data; S2. Refining the historical production data according to the process to obtain production quality data, obtaining a first production deviation and a factory quality deviation based on the production quality data and the historical factory sampling inspection data, and processing the first production deviation to obtain a second production deviation; S3. Build a product quality front-end prediction model based on the second production deviation and the factory quality deviation, and input the to-be-predicted product chain data into the product quality front-end prediction model to obtain a first risk quality and first risk data; S4. Determine storage and transportation sales deviation characteristics and abnormal operation characteristics based on the historical storage and transportation sales data, build a product quality backend prediction model based on the after-sales quality data, and input the product chain data to be predicted into the product quality backend prediction model to obtain a second risk quality; S5. Correct the first risk data according to the production stability and the environmental factor to obtain second risk data, determine a risk coefficient according to the second risk data, and determine a quality risk result according to the first risk quality, the second risk quality, and the risk coefficient; S6. Perform a product quality blockchain early warning based on the first risk quality, the second risk quality, and the quality risk result; The product quality blockchain early warning includes production chain early warning, storage and transportation sales chain early warning and full chain early warning; The method for constructing production stability and environmental coefficients based on the product chain historical data includes: Through clustering, historical production data is divided into production equipment status data, production environment data, production process parameters and production material parameters; through clustering, historical storage and transportation sales data is divided into storage and transportation environment data, storage and transportation feature data and abnormal features; Divide the production equipment status data into production sections and calculate the production stability of each production section: ， ， ， in For production section i production stability, is the failure rate weight, For production section i The failure rate score, is the actual failure rate of the corresponding section, The industry benchmark failure rate, is the inspection weight, For production section i The maintenance frequency score, is the running fatigue weight, For production section i Running fatigue score, The continuous operation time of the equipment in the corresponding section, For the optimal operating time of the equipment, is the current stabilization weight, For production section i The current stability score, For production section i The actual number of maintenance is the critical maintenance frequency, is the fatigue attenuation coefficient, 、 are the variance penalty coefficient and the skewness penalty coefficient, respectively. 、 Production sections i The current variance and power variance, 、 Production sections i The current and power deviations of The production stability of the next interval is adjusted according to the production stability of the previous interval, and the average production stability of all intervals is taken as the production stability of the product chain. The expression is: ， in For production section i Adjusted production stability, For production section i Dependence coefficient with the previous production section; Input the production environment data and storage and transportation environment data of the same product chain into the environmental function to obtain the environmental coefficient: ， ， ， ， ， in is the environmental factor, is the production weight, is the production environment coefficient, is the storage and transportation weight, is the storage and transportation environment coefficient, is the mechanical damage weight, is the chemical damage weight, is the vibration attenuation coefficient, is the deviation between the current vibration intensity and the safe vibration intensity, For production environment noise, is the standard production environment noise, Score the static electricity accumulation voltage, For gas j Concentration, including harmful gas concentration and VOCs concentration, is the number of dust particles per unit area, is the electrostatic voltage, 、 is the skewness of common environmental factors, , It is the first common environmental factor set, including light intensity, radiation intensity, temperature, humidity, , It is the first common environmental factor set, including temperature, humidity, ultraviolet intensity, and visible light intensity. is the atmospheric pressure coefficient, Pre is the atmospheric pressure, G is the mechanical shock during storage and transportation, For reference mechanical shock, For the maximum design mechanical shock, is the vibration energy coefficient, is the energy attenuation coefficient, is the smoothing coefficient, is vibration energy, It is the safety value of vibration energy.

2. The product quality risk early warning method based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the second production deviation includes: The production sections on the product chain are divided into a first pending production section and a second pending production section; the first pending production section is related to product manufacturing, etching, and shaping; the second pending production section is related to product assembly and packaging; Fitting the first undetermined production section and the second undetermined production section to obtain a first core production section, a second auxiliary production section, and a third auxiliary production section; the first core production section is an overlapping section of the first undetermined production section and the second undetermined production section; the second auxiliary production section is a non-overlapping section of the first undetermined production section; and the third auxiliary production section is a non-overlapping section of the second undetermined production section; The first core production section, the second auxiliary production section, and the third auxiliary production section are transformed according to the production action to obtain the first pending process, the second pending process, and the third pending process. The average process density of each process is calculated to obtain the first process density threshold and the second process density threshold. The first pending process is defined as the first evaluation process. The second pending process is screened according to the first process density threshold to obtain the second evaluation process. The third pending process is screened according to the second process density threshold to obtain the third evaluation process. The average process density is determined by the ratio of the data volume and process length of the corresponding process. The first process density threshold is the average of the first average process density and the second average process density. The second process density threshold is the average of the second average process density and the third average process density. The first evaluation process, the second evaluation process, and the third evaluation process are combined into a necessary evaluation process. The production data of the necessary evaluation process is taken as the production quality data. The deviation between the production quality data and the corresponding reference value is calculated to obtain the first production deviation. The deviation between the historical factory sampling data and the corresponding factory standard is calculated to obtain the factory quality deviation. The first production deviation is input into the self-attention module to obtain the production deviation weight, and the second production deviation is obtained according to the first production deviation and the corresponding production deviation weight.

3. The product quality risk early warning method based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the first risk quality and the first risk data includes: Obtain the quality fluctuation range and quality risk category based on the sampling elimination data, quality elimination category and corresponding factory standards, and determine the corresponding risk probability based on the factory quality deviation, quality fluctuation range and quality risk category; The risk category, risk probability, factory quality deviation and second production deviation are combined into a front-end comprehensive set, which is randomly divided into a front-end training set and a front-end test set according to a ratio of 6:

4. Constructing a product quality front-end prediction model, the product quality front-end prediction model includes an input layer probability prediction layer, a similarity prediction layer, a data prediction layer and an output layer; The probability prediction layer uses a gradient boosting tree to learn the nonlinear relationship between risk category, risk probability, and second production deviation, predicts the first risk quality, and attaches a corresponding production anomaly label; the first risk quality includes the risk category and risk probability; The similarity prediction layer is used to calculate the deviation similarity between the second production deviation and the historical production deviation, extract the historical factory quality deviation corresponding to the historical production deviation with the largest deviation similarity, and determine the first predicted risk data based on the deviation similarity and the historical factory quality deviation; The data prediction layer learns the nonlinear relationship between the factory quality deviation and the second production deviation through the BP neural network to predict the second prediction risk data; The output layer includes a first fully connected layer and a second fully connected layer. The first fully connected layer is connected to the probability prediction layer to output the first risk quality. The second fully connected layer is connected to the similarity prediction layer and the data prediction layer to weightedly fuse the first predicted risk data and the second predicted risk data and output the first risk data.

4. The product quality risk early warning method based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the second risk quality comprises: Comparing the storage and transportation environment data and storage and transportation characteristic data with the corresponding storage and transportation specifications to obtain storage and transportation sales deviation characteristics; the storage and transportation characteristic data includes transportation mode and storage and transportation sales time; The Apriori algorithm is used to perform association rule mining on abnormal features in historical storage, transportation and sales data to obtain abnormal operation features; the abnormal features include abnormal storage, transportation and sales operations and transportation accidents; the abnormal operation features are abnormal features related to after-sales quality data; Build a product quality backend prediction model based on Bayesian network. The specific steps include: Determine network nodes based on storage, transportation, and sales deviation characteristics, abnormal operation characteristics, and after-sales quality data, and determine causal paths based on domain experience; the network nodes include root nodes, implicit nodes, and leaf nodes; the causal paths include direct paths and cross paths; The expert knowledge base and historical storage, transportation and sales data are used to generate a mixed conditional probability table. A Bayesian network model is generated based on network nodes, causal paths and the mixed conditional probability table. The Bayesian network model is continuously updated using storage, transportation and sales deviation characteristics, abnormal operation characteristics and after-sales quality data. The expression is: ， ， in is the conditional probability of incremental update, indicating t +1 time node X At a given parent node Pa The updated conditional probability under the condition for t Current node at the moment X At a given parent node Pa Conditional probability The corresponding sample size, is the number of newly added samples, Based on the new sample The new conditional probability estimate obtained is, is the time decay factor, is the time interval, is the Bayesian shock update probability, The severity of the accident, To control the impact strength, is the new conditional probability after the emergency occurs, is the original conditional probability before the emergency occurs; Dynamic pruning is performed based on causal importance, and the optimal sub-network is selected to obtain the product quality back-end prediction model. The dynamic pruning conditions are: ， in For the edge The pruning score of DC For data completeness, For nodes X and Y Mutual information is the confidence of the model, is the loss function Loss Reaction edge Weight The gradient, is the path activation frequency weight, For the edge The frequency with which the path is activated; The product chain data to be predicted is input into the product quality back-end prediction model to obtain a second risk quality, and a corresponding storage, transportation and sales abnormality label is attached; the second risk quality includes risk category and risk probability.

5. The product quality risk early warning method based on artificial intelligence according to claim 1, characterized in that: The method for determining quality risk results comprises: According to the risk category, the first risk quality and the second risk quality are intersected to obtain a third risk quality, the third risk quality is updated by weighted calculation of the risk probabilities of the same category, and the fourth risk quality is obtained by taking the union of the non-intersecting parts of the first risk quality and the second risk quality and the updated third risk quality; determining a risk coefficient based on production stability, an environmental coefficient, and the first risk data; wherein the risk coefficient is positively correlated with the environmental coefficient and the first risk data, and negatively correlated with production stability; The quality risk result is obtained by correcting the risk probability of the corresponding category in the fourth risk quality according to the risk coefficient.

6. The product quality risk early warning method based on artificial intelligence according to claim 1, characterized in that: The method for performing product quality blockchain early warning includes: Compare the first risk quality with the production risk probability threshold, trace the process to obtain the risk process for the risk category that exceeds the production risk probability threshold, and issue a production chain early warning for the risk process; The second risk quality to be warned is selected based on the storage, transportation and sales risk probability threshold, and the root node is traced back to determine the risk root node based on the product quality backend prediction model. Storage, transportation and sales chain warnings are issued based on the risk root node category; the root node categories include storage and transportation environment category, storage and transportation method category, and abnormal feature category. Extract labels of quality risk results. When a risk category only contains production abnormality labels or storage, transportation and sales abnormality labels, issue a production chain warning based on the production risk probability threshold or a storage, transportation and sales chain warning based on the storage, transportation and sales risk probability threshold. When a risk category contains both production abnormality labels and storage, transportation and sales abnormality labels, perform process traceability and root node traceability, and issue a full-chain warning based on the traceability results.

7. A product quality risk early warning system based on artificial intelligence, used to execute the method according to any one of claims 1 to 6, characterized in that: include: Deviation calculation module: used to extract the production data according to the process to obtain production quality data, obtain a first production deviation and a factory quality deviation based on the production quality data and historical factory sampling inspection data, and process the first production deviation to obtain a second production deviation; Production prediction module: used to build a product quality front-end prediction model based on the second production deviation and the factory quality deviation, and input the product chain data to be predicted into the product quality front-end prediction model to obtain the first risk quality and first risk data; Storage, transportation and sales forecasting module: used to obtain storage, transportation and sales deviation characteristics and abnormal operation characteristics based on the historical storage, transportation and sales data, build a product quality back-end forecasting model in combination with the after-sales quality data, and input the product chain data to be forecasted into the product quality back-end forecasting model to obtain the second risk quality; Correction module: used to construct production stability and environmental coefficient based on product chain historical data, correct the first risk data based on the production stability and environmental coefficient to obtain second risk data, determine the risk coefficient based on the second risk data, and determine the quality risk result based on the first risk quality, the second risk quality, and the risk coefficient; Early warning module: used to view, manage and store the first risk quality, the second risk quality and the quality risk results, and perform product quality blockchain early warning based on the first risk quality, the second risk quality and the quality risk results.

Images