Leather shoe sole product qualification rate prediction system based on equipment data

Abstract

The invention relates to the technical field of qualification rate prediction, in particular to a leather shoe sole product qualification rate prediction system based on equipment data, which comprises a data sequence construction module, a period trend analysis module, a production optimization decision module and a parameter adjustment and support module, change parameters of the hardness and the material of the sole are collected, and statistical data calculation is carried out. According to the method, the LSTM network and the Bayesian network are combined to realize dynamic adaptation and real-time prediction of periodic data in the production process, so that the prediction accuracy and the response speed are improved, and tiny changes in the production process are quickly captured and responded, thereby realizing real-time adjustment and optimization of production parameters; the generation of the real-time prediction data provides a transparent window of the real-time state of the production line, so that a decision can be made in advance, the quality problem is avoided, the rejection rate is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of qualified rate prediction, in particular to a leather shoe sole product qualified rate prediction system based on equipment data. Background Art

[0002] The technical field of pass rate prediction involves the use of statistics, machine learning, and data analysis methods to predict the probability of industrial products meeting quality standards during the production process. It usually relies on historical production data, real-time monitoring data, and quality control test results. By estimating the pass rate of future production batches, it helps companies to adjust production parameters in a timely manner, optimize manufacturing processes, reduce defective product rates, and improve production efficiency and product quality.

[0003] Among them, the leather shoe sole product qualification rate prediction system aims to use data collected from leather shoe production equipment, such as temperature, pressure and speed, to predict the production qualification rate of soles by analyzing the data. It can help production managers to understand the status of the production line in a timely manner, predict quality problems, and adjust the production process or parameters accordingly to ensure product quality, mainly to improve the production efficiency of leather shoe soles and reduce the scrap rate, thereby reducing costs.

[0004] Existing technologies suffer from slow response and insufficient adaptability when processing data from the production process. This is especially true when production conditions or raw material quality suddenly change. Models that rely on historical data are unable to adapt to changes immediately, resulting in reduced timeliness and accuracy in predictions. Existing technologies often neglect the processing of immediate feedback from the production process, such as minor equipment failures or environmental changes, and fail to identify and integrate information in a timely manner, leading to quality control failures. For example, if rapid changes in ambient temperature are not reflected quickly, a large number of products will be unqualified, thereby increasing costs and wasting resources. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a leather shoe sole product qualification rate prediction system based on equipment data.

[0006] To achieve the above-mentioned object, the present invention adopts the following technical solution: a leather shoe sole product qualification rate prediction system based on equipment data, the system comprising:

[0007] The data sequence building module collects the variation parameters of sole hardness and material based on the production line equipment data, calculates statistical data, and divides the data set according to production batches to generate equipment sequence data packets;

[0008] The periodic trend analysis module uses the LSTM network to identify the periodic patterns and trends of production line speed and sole thickness based on the equipment sequence data packets, adjusts network parameters and matches data characteristics, and combines with the GRU network to optimize computing efficiency and generate periodic feature recognition records;

[0009] The production optimization decision module uses a Bayesian network to update the probability of production data based on the periodic feature recognition records, reflects the current production status, analyzes the time series deviation in equipment operation, calculates the similarity of equipment behavior, and generates real-time prediction data information;

[0010] The parameter adjustment and support module analyzes the impact of sole hardness and material on production qualification rate based on the real-time prediction data information, adjusts key production parameters, optimizes product qualification rate, and generates qualification rate optimization results.

[0011] The present invention is improved in that the steps of calculating the statistical data are specifically as follows:

[0012] Based on the production line equipment data, collect the sole hardness data and material data of multiple production batches, calculate the average hardness and material value of each batch, and use the formula:

[0013]

[0014] and

[0015]

[0016] Get the average hardness H avg and average material M avg , where n is the number of samples, H i Represents the hardness of the i-th sample, M i Represents the material of the i-th sample;

[0017] Based on the average hardness and average material, the standard deviation of hardness and material is calculated using the formula:

[0018]

[0019] and

[0020]

[0021] Generate hardness standard deviation σ H and material standard deviation σ M , where w i is the weight assigned to each sample, H i Represents the hardness of the i-th sample, M i Represents the material of the i-th sample, H avg is the average hardness, M avg is the average material, n is the number of samples;

[0022] Based on the hardness standard deviation and material standard deviation, the formula:

[0023]

[0024] and

[0025]

[0026] Calculate the coefficient of variation of hardness and material, where CV H is the coefficient of variation of hardness, CV M is the material variation coefficient, H avg is the average hardness, M avg is the average material, σ H is the hardness standard deviation, σ M is the material standard deviation.

[0027] The present invention is improved in that the step of acquiring the device sequence data packet is specifically as follows:

[0028] According to the production batch, record the timestamp and corresponding production parameters of each batch, build a data structure, and use the formula:

[0029] S batch =(T,P T ,P P ,H,M)

[0030] Get the batch labeled dataset S batch , where T represents the timestamp, P T and P P Represent temperature and pressure parameters respectively, H and M represent hardness and material data;

[0031] Based on the batch labeled data set, combined with the batch production quality score, the hardness and material data are weighted and the data sequence is packaged, using the formula:

[0032] D final =(T,Q×P T ,Q×P P ,Q×H,Q×M)

[0033] Generate device serial data packet D final , where T represents the timestamp, P T and P P Represent temperature and pressure parameters respectively, H and M represent hardness and material data, and Q represents the weighting coefficient of production quality score on data.

[0034] The present invention is improved in that the steps of identifying the periodic patterns and trends are specifically as follows:

[0035] Based on the equipment sequence data package, the time series data of production line speed and sole thickness are integrated and the formula is used:

[0036] D integrated ={(si ,t i )|s i ∈Speed,t i ∈Thickness,i=1,2,…,n}

[0037] Generate the integrated input dataset D integrated , where s i is the speed of the i-th data point, t i is the thickness of the i-th data point, Speed ​​is the speed data sequence, Thickness is the thickness data sequence, i is the index used to traverse the speed and thickness data points, and n is the total number of data points;

[0038] Based on the integrated input data set, an LSTM network is used to analyze the data and identify periodic patterns and long-term trends in the data using the formula:

[0039] (Patterns, Trends) = LSTM(D integrated ;θ)

[0040] Get periodic patterns Patterns and long-term trends Trends, where θ is the parameter of the LSTM network, D integrated is the integrated input dataset.

[0041] The present invention is improved in that the optimization steps of the computational efficiency are specifically as follows:

[0042] Adjust the weights and biases of the LSTM network and use the gradient descent method to optimize the network's loss function using the formula:

[0043]

[0044] Get the updated parameter set θ new , where θ old is the current set of parameters, including weights and biases, α d is the learning rate, is the gradient of the loss function J with respect to the parameter θ, D integrated is the integrated input dataset;

[0045] Based on the updated parameter set, combined with the GRU network, the processing speed and data processing capacity are optimized, using the formula:

[0046] M integrated =β d LSTM(D integrated ,θ LSTM )+(1-β d )·GRU(D integrated ,θ GRU )

[0047] Get the integrated network output result M integrated , generate periodic feature recognition records, where β d is the mixing parameter, D integrated is the integrated input dataset, θ LSTM and θ GRU are the optimization parameters of LSTM and GRU networks respectively.

[0048] The present invention is improved in that the steps of updating the production data probability are specifically as follows:

[0049] Based on the periodic feature recognition records, the Bayesian network is updated in combination with the current production data to calculate the new posterior probability using the formula:

[0050]

[0051] Get the updated network state, where P(X|D) is the posterior probability, which represents the probability of state X after given data D, P(D|X) is the likelihood under the given state, P(X) is the prior probability, D represents the current data, and X is the state variable in the network;

[0052] Based on the updated network status, new probability information is calculated and the probability information is extracted using the formula:

[0053]

[0054] Get probability information P info , where P(X=x|D) represents the posterior probability of state x given data D, Info(x) is the information function under state x, and X is the state variable in the network.

[0055] The present invention is improved in that the step of acquiring the real-time prediction data information is specifically as follows:

[0056] Based on the current production status, the time series data of the production equipment is collected and analyzed, and the AR model is used to identify deviations in the data using the formula:

[0057]

[0058] Get the time series deviation S deviation , where T data (t) is the data point at time t, a t is the coefficient at time t, ∈ is the model error term, t is the time index, and n is the total number of data points;

[0059] Based on the time series deviation, the behavioral similarity between devices is calculated using the cosine similarity formula:

[0060]

[0061] Get behavior similarity A similarity , where S deviation is the time series deviation, T reference is the time series reference data under normal operating conditions, i is the index used to traverse the time series data points, and n is the total number of data points;

[0062] Based on the behavioral similarity, linear regression analysis was performed using the formula:

[0063] P forecast =b0+b1·A similarity

[0064] Get real-time prediction data information P forecast , where b0 and b1 are the parameters of the regression model, A similarity is the behavioral similarity.

[0065] The present invention is improved in that the steps for obtaining the pass rate optimization result are specifically as follows:

[0066] Based on the real-time prediction data information, the sole hardness and material data are collected to analyze the impact on the product qualification rate. The linear regression analysis is performed using the formula:

[0067] RZ=β0+β1H+β2M

[0068] Get the predicted pass rate RZ, where β0 is the intercept, β1 and β2 are the influence coefficients of hardness and material on the pass rate, H is the sole hardness data, and M is the material data;

[0069] Based on the predicted pass rate, adjust key production parameters and use the gradient ascent method to capture the optimal parameter configuration and optimize the pass rate using the formula:

[0070]

[0071] Get the new parameter setting P new , where P current is the current production parameter, γ is the learning rate, is the pass rate gradient; based on the new parameter settings, re-evaluate the pass rate and verify the adjustment effect using the formula:

[0072] R optimized =f(P new )

[0073] Get the optimized qualified rate R optimized , where f is the function of evaluating the pass rate, P new are the optimized production parameters.

[0074] Compared with the prior art, the advantages and positive effects of the present invention are:

[0075] In the present invention, by combining the LSTM network and the Bayesian network, dynamic adaptation and real-time prediction of periodic data in the production process are achieved, which improves the accuracy and response speed of the prediction, so that small changes in the production process can be quickly captured and responded to, thereby realizing the instant adjustment and optimization of production parameters. The generation of real-time prediction data provides a transparent window into the immediate status of the production line, so that decisions can be made in advance and quality problems can be avoided, which not only reduces the scrap rate but also improves production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a system flow chart of the present invention;

[0077] Figure 2 This is a flow chart for calculating statistical data in the present invention;

[0078] Figure 3 This is a flow chart for obtaining device sequence data packets in the present invention;

[0079] Figure 4 A flow chart for identifying periodic patterns and trends in the present invention;

[0080] Figure 5 This is a flowchart for optimizing computational efficiency in the present invention;

[0081] Figure 6 A flowchart of the probability update of production data in the present invention;

[0082] Figure 7 This is a flowchart for obtaining real-time prediction data information in the present invention;

[0083] Figure 8 This is a flow chart for obtaining the qualified rate optimization results in the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0085] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0086] Example

[0087] See also Figure 1 The present invention provides a technical solution: a leather shoe sole product qualification rate prediction system based on equipment data includes:

[0088] The data sequence building module collects the variation parameters of sole hardness and material based on the production line equipment data, calculates statistical data, and divides the data set according to production batches to generate equipment sequence data packets;

[0089] The cycle trend analysis module uses an LSTM network based on device sequence data packets to identify periodic patterns and trends in production line speed and sole thickness. It adjusts network parameters to match data characteristics and combines it with a GRU network to optimize computational efficiency and generate cycle feature recognition records.

[0090] The production optimization decision module uses a Bayesian network to update production data probabilistically based on periodic feature recognition records, reflecting the current production status, analyzing time series deviations in equipment operations, calculating the similarity of equipment behavior, and generating real-time prediction data information.

[0091] The parameter adjustment and support module analyzes the impact of sole hardness and material on production qualification rate based on real-time prediction data information, adjusts key production parameters, optimizes product qualification rate, and generates qualification rate optimization results.

[0092] The equipment sequence data package includes the measurement time point, parameter measurement value, and batch number. The periodic feature recognition record includes periodic pattern data, trend change index, and efficiency optimization record. The real-time prediction data information includes status update frequency, deviation responsiveness, and similarity comparison results. The pass rate optimization results include parameter adjustment effects, production quality feedback results, and pass rate statistics.

[0093] See also Figure 2 , the calculation steps of statistical data are as follows:

[0094] Based on the production line equipment data, collect the sole hardness data and material data of multiple production batches, calculate the average hardness and material value of each batch, and use the formula:

[0095]

[0096] and

[0097]

[0098] Get the average hardness H avg and average material M avg , where n is the number of samples, H i Represents the hardness of the i-th sample, M i Represents the material of the i-th sample;

[0099] Based on the average hardness and average material, calculate the standard deviation of hardness and material using the formula:

[0100]

[0101] and

[0102]

[0103] Generate hardness standard deviation σ H and material standard deviation σ M , where w i is the weight assigned to each sample, H i Represents the hardness of the i-th sample, M i Represents the material of the i-th sample, H avg is the average hardness, M avg is the average material, n is the number of samples;

[0104] Based on the hardness standard deviation and material standard deviation, the formula:

[0105]

[0106] and

[0107]

[0108] Calculate the coefficient of variation of hardness and material, where CV H is the coefficient of variation of hardness, CV M is the material variation coefficient, H avg is the average hardness, M avg is the average material, σ H is the hardness standard deviation, σ M is the material standard deviation.

[0109] Assume that there is a production batch containing 5 samples with hardness values ​​of:

[0110] H=[80,82,78,81,79]

[0111] The material values ​​are:

[0112] M=[30,32,31,29,30]

[0113] Calculate the average hardness H avg :

[0114]

[0115] Calculate the average material M avg :

[0116]

[0117] The average hardness is 80, indicating that the hardness of this batch of samples is relatively uniform; the average material is 30.4, providing the center position of the material data.

[0118] w i Assume it is 1, use the above average value and the same hardness and material data.

[0119] Calculate the standard deviation σ of hardness H :

[0120]

[0121] Calculate the standard deviation σ of the material M :

[0122]

[0123] Standard deviation σ H =1.581 and σ M =1.118 indicates that the data distribution of hardness and material fluctuates.

[0124] Using the standard deviation and mean above:

[0125] Calculate the coefficient of variation CV of hardness H :

[0126]

[0127] Calculate the coefficient of variation CV of the material M :

[0128]

[0129] Coefficient of variation (CV) H =1.98% and CV M =3.68% shows the relative stability and volatility of hardness and material.

[0130] See also Figure 3, the specific steps for obtaining device serial data packets are:

[0131] According to the production batch, record the timestamp and corresponding production parameters of each batch, build the data structure, and use the formula:

[0132] S batch =(T,P T ,P P ,H,M)

[0133] Get the batch labeled dataset S batch , where T represents the timestamp, P T and P P Represent temperature and pressure parameters respectively, H and M represent hardness and material data;

[0134] Based on the batch-labeled dataset, combined with the batch production quality score, the hardness and material data are weighted and packaged into a data sequence, using the formula:

[0135] D final =(T,Q×P T ,Q×P P ,Q×H,Q×M)

[0136] Generate device serial data packet D final , where T represents the timestamp, P T and P P Represent temperature and pressure parameters respectively, H and M represent hardness and material data, and Q represents the weighting coefficient of production quality score on data, reflecting the quality impact of production parameters and material data.

[0137] Assume that a production batch is produced at time T = "2024-07-108:00:00";

[0138] P T =350℃;

[0139] P P =150Pa;

[0140] H = 75;

[0141] M = 50;

[0142] Once the data is collected and structured according to the formula, the following data structure will be formed:

[0143] S batch ={("2024-07-10 8:00:00", 350, 150, 75, 50)}

[0144] The data structure provides the basis for the next step of weighted processing and data analysis.

[0145] Continuing with the above hypothetical data, assume that the quality score of this batch is Q = 1.2;

[0146] Apply the formula to weight each parameter:

[0147] Q×P T =1.2×350=420

[0148] Q×P P =1.2×150=180

[0149] Q×H=1.2×75=90

[0150] Q×M=1.2×50=60

[0151] The data structure is:

[0152] D final ={("2024-07-10 8:00:00", 420, 180, 90, 60)}

[0153] The data package now contains adjusted temperature, pressure, hardness and material data to more accurately reflect the quality of the product.

[0154] See also Figure 4 , the steps for identifying periodic patterns and trends are as follows:

[0155] Based on the equipment sequence data package, the time series data of production line speed and sole thickness are integrated and the formula is used:

[0156] D integrated ={(s i ,t i )|s i ∈Speed,t i ∈Thickness,i=1,2,…,n}

[0157] Generate the integrated input dataset D integrated , where s i is the speed of the i-th data point, t i is the thickness of the i-th data point, Speed ​​is the speed data sequence, Thickness is the thickness data sequence, i is the index used to traverse the speed and thickness data points, and n is the total number of data points;

[0158] Based on the integrated input data set, the LSTM network is used to analyze the data and identify periodic patterns and long-term trends in the data using the formula:

[0159] (Patterns, Trends) = LSTM(D integrated ;θ)

[0160] Get periodic patterns Patterns and long-term trends Trends, where θ is the parameter of the LSTM network, D integrated is the integrated input dataset.

[0161] Assume the following velocity and thickness data:

[0162] Speed=[20,22,21,23]m / s

[0163] Thickness=[5,5.5,5,5.5]mm

[0164] The integrated dataset D integrated Will contain:

[0165] D integrated ={(20,5),(22,5.5),(21,5),(23,5.5)}

[0166] Dataset D integrated It will be used as input to the deep learning model.

[0167] Assumption D integrated Given a dataset and with θ optimized after training, the LSTM model identifies the following patterns and trends:

[0168] Patterns = ["Rise Period", "Fall Period"]

[0169] Trends = ["Overall Upward Trend"]

[0170] This indicates that the LSTM model has identified periodic rising and falling patterns from the input data and observed an overall upward trend.

[0171] See also Figure 5 , the optimization steps for computational efficiency are as follows:

[0172] Adjust the weights and biases of the LSTM network and use the gradient descent method to optimize the network's loss function using the formula:

[0173]

[0174] Get the updated parameter set θ new , where θ old is the current set of parameters, including weights and biases, α d is the learning rate, is the gradient of the loss function J with respect to the parameter θ, D integrated is the integrated input dataset;

[0175] Based on the updated parameter set, combined with the GRU network, the processing speed and data processing capacity are optimized, using the formula:

[0176] M integrated =β d LSTM(D integrated ,θ LSTM )+(1-β d )·GRU(D integrated ,θ GRU )

[0177] Get the integrated network output result M integrated , generate periodic feature recognition records, where β d is the mixing parameter, D integrated is the integrated input dataset, θ LSTM and θ GRU are the optimization parameters of LSTM and GRU networks respectively.

[0178] Assumptions:

[0179] θ old =[0.5,0.5];

[0180] α d =0.1;

[0181]

[0182] Calculation process:

[0183] θ new =[0.5,0.5]-0.1×[0.2,0.2]=[0.5-0.02,0.5-0.02]=[0.48,0.48]

[0184] Updated parameter θ new It is [0.48, 0.48], which shows that each parameter is slightly reduced, reducing the value of the loss function and improving the accuracy of the model prediction.

[0185] Assumptions:

[0186] β d =0.6;

[0187] LSTM output is 0.8;

[0188] The GRU output is 0.5.

[0189] Calculation process:

[0190] M integrated =0.6·0.8+(1-0.6)·0.5=0.48+0.2=0.68

[0191] M integrated The value is 0.68, which is a weighted average of the outputs of the LSTM and GRU models. It optimizes the bias of a single model and improves the stability and accuracy of the overall prediction.

[0192] See also Figure 6 ,The specific steps for updating the production data probability are:

[0193] Based on the periodic feature recognition records, the Bayesian network is updated in combination with the current production data to calculate the new posterior probability using the formula:

[0194]

[0195] Get the updated network state, where P(X|D) is the posterior probability, which represents the probability of state X after given data D, indicating how the likelihood of the state changes after observing new data, P(D|X) is the likelihood under the given state, P(X) is the prior probability, D represents the current data, and X is the state variable in the network;

[0196] Based on the updated network status, new probability information is calculated and the probability information is extracted using the formula:

[0197]

[0198] Get probability information P info , where P(X=x|D) represents the posterior probability of state x given data D, which is used to evaluate the likelihood of each state after specific data observations. Info(x) is the information function under state x, and X is the state variable in the network.

[0199] Assume there are two states X = {x1, x2}:

[0200] Prior probability:

[0201] P(x1)=0.6, P(x2)=0.4

[0202] Likelihood:

[0203] P(D|x1)=0.5,P(D|x2)=0.2

[0204] Calculate the posterior probability:

[0205] Normalization constant:

[0206]

[0207] Posterior probability:

[0208]

[0209] It means that after observing data D, the probabilities of states x1 and x2 are updated to 0.789 and 0.211 respectively.

[0210] Assumptions:

[0211] Information function: Info(x1)=10Info(x2)=5.

[0212] Calculate probability information:

[0213] P info =0.789×10+0.211×5=7.89+1.055=8.945

[0214] Value P info =8.945 represents the overall information volume of the system under the current data observation, reflecting the expected performance of the current production status.

[0215] See also Figure 7 ,The specific steps for obtaining real-time prediction data information are as follows:

[0216] Based on the current production status, we collect and analyze the time series data of production equipment and use the AR model to identify deviations in the data using the formula:

[0217]

[0218] Get the time series deviation S deviation , where T data (t) is the data point at time t, a t is the coefficient at time t, ∈ is the model error term, t is the time index, and n is the total number of data points;

[0219] Based on the time series deviation, the behavioral similarity between devices is calculated using the cosine similarity formula:

[0220]

[0221] Get behavior similarity A similarity , where S deviation is the time series deviation, T reference is the time series reference data under normal operating conditions, i is the index used to traverse the time series data points, and n is the total number of data points;

[0222] Based on behavioral similarity, linear regression analysis is used, using the formula:

[0223] P forecast =b0+b1·A similarity

[0224] Get real-time prediction data information P forecast, where b0 and b1 are the parameters of the regression model, A similarity is the behavioral similarity.

[0225] Assume there are data at three time points, a1 = 0.5, a2 = 0.3, a3 = 0.2;

[0226] The data points are:

[0227] T data (1) = 100, T data (2) = 150, T data (3) = 120

[0228] Error∈=10.

[0229] Calculation process:

[0230] S deviation =(0.5×100)+(0.3×150)+(0.2×120)+10=50+45+24+10=129

[0231] It means that based on the given data and coefficients, the calculated deviation is 129, which reflects the degree of deviation from the model expectation.

[0232] Assumptions:

[0233] S deviation is [129, 140, 150];

[0234] T reference is [130, 135, 145].

[0235] Calculation process:

[0236] molecular:

[0237] 129×130+140×135+150×145=16770+18900+21750=57420

[0238] Denominator:

[0239]

[0240] Final calculation:

[0241]

[0242] It means that, based on the given data and calculations, the current behavior of the device is very similar to the normal reference model, with a similarity close to 1.

[0243] Assume b0 = 10 and b1 = 50, similarity A similarity =1.00:

[0244] P forecast =10+50×1.00=60

[0245] This means that the prediction result is 60, which indicates that the equipment is operating well or the production efficiency meets expectations based on the model and the given similarity.

[0246] See also Figure 8 , the specific steps for obtaining the pass rate optimization results are:

[0247] Based on real-time prediction data information, collect sole hardness and material data, analyze the impact on product qualification rate, use linear regression analysis, and adopt the formula:

[0248] RZ=β0+β1H+β2M

[0249] The predicted pass rate RZ is obtained, where β0 is the intercept, reflecting the basic pass rate, β1 and β2 are the influence coefficients of hardness and material on the pass rate, H is the sole hardness data, and M is the material data;

[0250] Based on the predicted pass rate, adjust key production parameters and use the gradient ascent method to capture the optimal parameter configuration and optimize the pass rate using the formula:

[0251]

[0252] Get the new parameter setting P new , where P current is the current production parameter, γ is the learning rate, is the pass rate gradient; based on the new parameter settings, re-evaluate the pass rate and verify the adjustment effect using the formula:

[0253] R optimized =f(P new )

[0254] Get the optimized qualified rate R optimized , where f is the function of evaluating the pass rate, P new are the optimized production parameters.

[0255] Assume that the following sample data is obtained:

[0256] H = 70;

[0257] M = 30;

[0258] β0=50%;

[0259] β1=0.1%;

[0260] β2=0.2%.

[0261] Calculation process:

[0262] RZ=50+0.1×70+0.2×30=50+7+6=63%

[0263] It means that under the given hardness and material, the predicted pass rate is 63%.

[0264] Assume the following data:

[0265] γ is 0.05;

[0266] 5%;

[0267] P current is 100.

[0268] Calculation process:

[0269] P new =100+0.05×5=100+0.25=100.25

[0270] It means that based on the current parameters, after optimization and adjustment, the new parameter is set to 100.25, which is expected to improve the pass rate.

[0271] Assume that through model evaluation, when the parameter is set to P new =100.25, the pass rate increased to 65%;

[0272] Assuming that the function f is linear, it is directly related to P new proportional;

[0273] R optimized =65%

[0274] It shows that the prediction qualification rate after parameter optimization is 65%, indicating the improvement of sole quality after adjusting the parameters.

[0275] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A leather shoe sole product qualification rate prediction system based on equipment data, characterized in that: The system comprises: The data sequence building module collects the change parameters of the sole hardness and material based on the production line equipment data, calculates statistical data, and divides the data set according to the production batch to generate the equipment sequence data packet; The periodic trend analysis module uses the LSTM network to identify the periodic patterns and trends of the production line speed and the sole thickness based on the equipment sequence data packets, adjusts the network parameters and matches the data characteristics, combines the GRU network, optimizes the calculation efficiency, and generates periodic feature recognition records; The production optimization decision module uses a Bayesian network to update the probability of production data based on the periodic feature recognition records, reflects the current production status, analyzes the time series deviation in equipment operation, calculates the similarity of equipment behavior, and generates real-time prediction data information; The parameter adjustment and support module analyzes the influence of the hardness and material of the sole on the production qualification rate based on the real-time prediction data information, adjusts key production parameters, optimizes the product qualification rate, and generates the qualification rate optimization result.

2. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The calculation steps of the statistical data are specifically as follows: Based on the production line equipment data, collect the sole hardness data and material data of multiple production batches, calculate the hardness and material average of each batch, and use the formula: and Get the average hardness H avg and average material M avg , where n is the number of samples, H i represents the hardness of the i-th sample, M i Represents the material of the i-th sample; Based on the average hardness and average material, the standard deviation of hardness and material is calculated using the formula: and Generate hardness standard deviation σ H and material standard deviation σ M , where w i is the weight given to each sample, H i represents the hardness of the i-th sample, M i represents the material of the i-th sample, H avg is the average hardness, M avg is the average material, n is the number of samples; Based on the hardness standard deviation and material standard deviation, the formula: and Calculate the coefficient of variation of hardness and material, where CV H is the coefficient of variation of hardness, CV M is the material variation coefficient, H avg is the average hardness, M avg is the average material, σ H is the hardness standard deviation, σ M is the material standard deviation.

3. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The steps for obtaining the device sequence data packet are specifically as follows: According to the production batch, record the timestamp and corresponding production parameters of each batch, build a data structure, and use the formula: S batch =(T,P T ,P P ,H,M) Get the batch labeled dataset S batch , where T represents the timestamp, P T and P P Respectively represent temperature and pressure parameters, H and M represent hardness and material data; Based on the batch labeled data set, combined with the batch production quality score, the hardness and material data are weighted and the data sequence is packaged, using the formula: D final =(T,Q×P T ,Q×P P ,Q×H,Q×M) Generate device serial data packet D final , where T represents the timestamp and P T and P P Represent temperature and pressure parameters respectively, H and M represent hardness and material data, and Q represents the weighting coefficient of production quality score on data.

4. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The steps for identifying the periodic patterns and trends are as follows: Based on the equipment sequence data package, the time series data of production line speed and sole thickness are integrated, and the formula is used: D integrated ={(s i ,t i )|s i ∈Speed,t i ∈Thickness,i=1,2,…,n} Generate the integrated input dataset D integrated , where s i is the velocity of the ith data point, t i is the thickness of the i-th data point, Speed ​​is the speed data sequence, Thickness is the thickness data sequence, i is the index used to traverse the speed and thickness data points, and n is the total number of data points; Based on the integrated input data set, the LSTM network is used to analyze the data to identify periodic patterns and long-term trends of the data using the formula: (Patterns,Trends)=LSTM(D integrated ;θ) Get the periodic patterns Patterns and long-term trends Trends, where θ is the parameter of the LSTM network, D integrated is the integrated input dataset.

5. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The optimization steps of the computational efficiency are specifically as follows: The weights and biases of the LSTM network are adjusted, and the loss function of the network is optimized using the gradient descent method, using the formula: Get the updated parameter set θ new , where θ old is the current set of parameters, including weights and biases, α d is the learning rate, is the gradient of the loss function J with respect to the parameter θ, D integrated is the integrated input dataset; Based on the updated parameter set, combined with the GRU network, the processing speed and data processing capacity are optimized, using the formula: M integrated =b d ·LSTM(D integrated ,i LSTM )+(1-β d )·GRU(D integrated ,i GRU ) Get the integrated network output result M integrated , generate periodic feature recognition records, where β d is the mixing parameter, D integrated is the integrated input dataset, θ LSTM and θ GRU are the optimization parameters of LSTM and GRU networks respectively.

6. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The steps of updating the production data probability are specifically as follows: Based on the periodic feature recognition records, the Bayesian network is updated in combination with the current production data to calculate the new posterior probability using the formula: Get the updated network state, where P(X|D) is the posterior probability, which represents the probability of state X after given data D, P(D|X) is the likelihood under a given state, P(X) is the prior probability, D represents the current data, and X is the state variable in the network; Based on the updated network status, new probability information is calculated and the probability information is extracted using the formula: Get probability information P info , where P(X=x|D) represents the posterior probability of state x given data D, Info(x is the information function under state x, and X is the state variable in the network.

7. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The steps for obtaining the real-time prediction data information are specifically as follows: Based on the current production status, the time series data of the production equipment is collected and analyzed, and the AR model is used to identify deviations in the data, using the formula: Get the time series deviation S deviation , where T data (t) is the data point at time t, a t is the coefficient at time t, ∈ is the model error term, t is the time index, and n is the total number of data points; Based on the time series deviation, the behavior similarity between devices is calculated using the cosine similarity formula: Get the behavior similarity A similarity , where S deviation is the time series deviation, T reference is the time series reference data under normal operating conditions, i is the index used to traverse the time series data points, and n is the total number of data points; Based on the behavioral similarity, linear regression analysis was used, using the formula: P forecast =b0+b1·A similarity Get real-time prediction data information P forecast , where b0 and b1 are the parameters of the regression model, A similarity is the behavioral similarity.

8. The leather shoe sole product qualification rate prediction system based on equipment data according to claim 1 is characterized in that: The steps for obtaining the pass rate optimization result are specifically as follows: Based on the real-time prediction data information, the sole hardness and material data are collected to analyze the impact on the product qualification rate. The linear regression analysis is used and the formula is adopted: RZ=β0+β1H+β2M The predicted qualified rate RZ is obtained, where β0 is the intercept, β1 and β2 are the influence coefficients of hardness and material on the qualified rate, H is the sole hardness data, and M is the material data; Based on the predicted pass rate, adjust the key production parameters, use the gradient ascent method, capture the optimal parameter configuration, optimize the pass rate, and use the formula: Get the new parameter setting P new , where P current is the current production parameter, γ is the learning rate, is the pass rate gradient; Based on the new parameter settings, the pass rate was re-evaluated to verify the adjustment effect, using the formula: R optimized =f(P new ) Get the optimized pass rate R optimized , where f is the function for evaluating the pass rate, P new are the optimized production parameters.

Images