A project follow-up method for a digital audit supervision system

Abstract

The present invention relates to a project follow-up method of a digital audit supervision system, comprising the following steps: obtaining the shredding process parameters and tobacco quality parameters on a tobacco production line, constructing a response surface model and establishing a nonlinear mapping relationship; optimizing and solving the best combination of shredding process parameters to obtain target shredding process parameters for achieving target tobacco quality; sending the optimized target shredding process parameters to the shredding process of the tobacco production line; monitoring tobacco quality parameters during the shredding process; performing big data analysis on tobacco quality parameters, shredding process parameters and product characteristic data during the shredding process; obtaining target process parameter combinations under different product characteristic targets, establishing a knowledge base and case base for shredding process parameter optimization; and incorporating the shredding process parameter optimization process into a digital audit supervision system for management. The present invention realizes intelligent optimization and automatic regulation of shredding process parameters, and improves the stability and consistency of tobacco quality.

Description

Technical Field

[0001] The present invention relates to the technical field of tobacco auditing, and in particular to a project follow-up method of a digital audit supervision system. Background Art

[0002] In the exploration of digital transformation of the audit department, auditing and supervision of tobacco cutting process is an important link. Through the project follow-up of the digital audit supervision system, real-time monitoring and analysis of tobacco cutting process parameter optimization can be achieved, but there are many technical contradictions and difficulties in the optimization analysis of tobacco cutting process parameters. First, there is a complex coupling relationship between parameters such as cutting tool angle, rotation speed and feed speed. Adjusting one of the parameters may affect other parameters, thereby affecting the quality of tobacco; secondly, different tobacco raw materials show different physical properties during the cutting process, such as moisture content, fiber length and hardness, which will affect the parameter setting during the cutting process; in addition, tobacco filling value and combustion performance are key indicators for evaluating tobacco quality, but there may be contradictions between these two indicators. For example, increasing the tobacco filling value may reduce its combustion performance. At the same time, different tobacco products have different requirements for tobacco quality, such as cigarettes, cigars and pipe tobacco, and different cutting process parameters need to be set for different products; finally, in actual production, factors such as the wear of tobacco cutting equipment and changes in environmental temperature and humidity will also affect the stability of the cutting process parameters, which require real-time monitoring and adjustment.

[0003] In summary, the optimization of tobacco cutting process parameters faces multiple technical challenges. It is necessary to comprehensively consider factors such as raw material characteristics, product requirements, and equipment conditions, and balance the relationship between various parameters to achieve stable control of tobacco quality. Through the digital audit supervision system, the real-time collection, analysis, and optimization of tobacco cutting process parameters can be realized, providing data support for production management decisions; at the same time, the system can also monitor and warn key indicators such as equipment operation status, raw material quality, and product quality, and promptly discover and solve problems in the production process to ensure the stability and consistency of tobacco quality; in addition, the digital audit supervision system can also use big data analysis and machine learning algorithms to explore potential laws and trends in production data, providing new ideas and directions for process optimization and product innovation. Summary of the invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a project follow-up method of a digital audit supervision system. The project follow-up method of a digital audit supervision system realizes the intelligent optimization and automatic regulation of shredded tobacco process parameters, improves the stability and consistency of tobacco quality, and provides effective technical support for tobacco product production.

[0005] A project follow-up method of a digital audit supervision system according to the present invention comprises the following steps:

[0006] S1. Obtain the shredded tobacco production line's shredded tobacco process parameters and shredded tobacco quality parameters, construct a response surface model, and establish a nonlinear mapping relationship;

[0007] S2. According to the constructed response surface model and the established nonlinear mapping relationship, the best combination of shredded tobacco process parameters is optimized and solved to obtain the target shredded tobacco process parameters for achieving the target shredded tobacco quality;

[0008] S3, sending the optimized target shredding process parameters to the shredding process of the tobacco production line, collecting quality parameters in real time and performing process monitoring, and triggering re-optimization when quality parameters deviate;

[0009] S4. Monitor the quality parameters of tobacco shreds during the shredding process, and control the physical and chemical properties of tobacco shreds whose quality parameters fluctuate irregularly or are uncontrollable;

[0010] S5. Obtain product characteristic data, conduct big data analysis on tobacco quality parameters, cutting process parameters and product characteristic data during the cutting process, and continuously revise and improve the response surface model based on the intrinsic relationship between the parameters;

[0011] S6. Obtain target process parameter combinations under different product characteristic targets, establish a knowledge base and case library for optimizing slitting process parameters, and form a standardized slitting process formula;

[0012] S7. Incorporate the process of optimizing the wire cutting process parameters into the digital audit supervision system for management, track the progress of the optimization project in real time, conduct audit supervision on key nodes and milestones, and form an audit report on the optimization results.

[0013] Preferably, the step S1 specifically includes:

[0014] The shredding process parameters include the shredding cutter angle, the shredding cutter rotation speed and the shredded tobacco feeding speed;

[0015] The tobacco quality parameters include tobacco filling value and combustion performance;

[0016] Collecting real-time data of the cutter angle, the cutter rotation speed and the tobacco feed speed, measuring the tobacco filling value and the combustion performance index, and normalizing the collected raw data;

[0017] The moving average filtering method was used to remove the noise in the normalized data, and the box plot method was used to identify outliers;

[0018] A response surface model is constructed according to the filtered data, and a nonlinear mapping relationship between the shredded tobacco process parameters and the shredded tobacco quality parameters is established through a support vector machine algorithm.

[0019] Preferably, the step S2 specifically includes:

[0020] The shredding process parameters include the shredding cutter angle, the shredding cutter rotation speed and the shredded tobacco feeding speed;

[0021] The tobacco quality parameters include tobacco filling value and combustion performance;

[0022] According to the constructed response surface model and the established nonlinear mapping relationship, the genetic algorithm intelligent optimization algorithm is adopted, and the optimization includes the following steps:

[0023] The preset chromosome encoding method encodes the three process parameters of the shredder angle, the shredder rotation speed and the shredded tobacco feeding speed into a binary string, and sets the length of the binary string according to the parameter accuracy requirements;

[0024] Obtaining a set binary string, setting a fitness function of a genetic algorithm, and converting the tobacco filling value and the combustion performance into dimensionless indicators using a normalization method;

[0025] After the fitness function is set, the roulette wheel selection operation is used to select excellent individuals, and the crossover probability of the roulette wheel selection operation is set to a preset value, and the single-point crossover method is used for gene recombination;

[0026] According to the selected excellent individuals, the genetic algorithm is iteratively run, and the population size of the genetic algorithm is set to a preset value, and the maximum number of iterations is set to a preset value;

[0027] When the genetic algorithm reaches the convergence condition or the maximum number of iterations, the cutting tool angle, cutting tool rotation speed and tobacco feed speed corresponding to the optimal individual are decoded and obtained as the target cutting process parameters for the target tobacco quality.

[0028] Preferably, the step S2 further includes:

[0029] Encoding the shredding process parameters as parameter combinations into individuals, and randomly generating an initial population;

[0030] Calculate the fitness of each individual based on the response surface model, select excellent individuals and generate new offspring individuals through crossover mutation;

[0031] Iterate multiple evolutionary operations to optimize the parameter combination, and finally select the individual with the highest fitness, and use its corresponding parameter combination as the best combination of optimal process parameters.

[0032] Preferably, the step S3 specifically includes:

[0033] Obtaining the optimized target shredding process parameters, and sending the target shredding process parameters to the programmable logic controller of the tobacco production line via industrial Ethernet;

[0034] Controlling the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco according to the target cutting process parameters, and monitoring the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco in real time;

[0035] If there is a deviation between the monitored actual parameters and the target wire cutting process parameters, a PID control algorithm is used to dynamically adjust the input signal of the actuator;

[0036] Use online detection equipment to continuously collect tobacco filling value and combustion performance, and use Kalman filter algorithm to obtain the best estimated quality parameters;

[0037] The exponentially weighted moving average method is used to calculate the trend of the optimal estimated quality parameters, and the control chart method is combined for anomaly detection;

[0038] If the tobacco filling value or combustion performance of three consecutive samples exceeds the target value by ±5%, the shredding process parameter re-optimization program will be triggered and the optimal parameter combination will be recalculated.

[0039] Preferably, the step S4 specifically includes:

[0040] The physicochemical properties of the cut tobacco include nicotine content and reducing sugar content;

[0041] A statistical process control chart is used to monitor the changing trend of tobacco quality parameters, and the statistical process control chart applies the eight abnormality criteria of the sigma control chart;

[0042] Determine whether the difference determination conditions are met according to the eight difference determination criteria;

[0043] If the abnormality judgment condition is not met, it is determined that the tobacco quality parameters do not have irregular fluctuations or are uncontrollable, and the subsequent processing is continued;

[0044] If the abnormality judgment conditions are met, it is determined that the quality parameters of the tobacco are irregularly fluctuating or uncontrollable, triggering the detection and analysis procedure for the physical and chemical properties of the tobacco, and the detection and analysis results are fed back to the upstream leaf blending process for raw material optimization.

[0045] Preferably, the step S5 specifically includes:

[0046] The product characteristic data include the taste, aroma and smoke index of the cigarette;

[0047] Obtain taste data, aroma data and smoke index of cigarette products, and obtain cutting process parameters and tobacco quality parameters from the production line;

[0048] Performing outlier detection on the taste data, the aroma data, the smoke index, the shredded tobacco process parameters, and the shredded tobacco quality parameters, using a box plot method to identify outliers, and performing smoothing on data that exceeds 1.5 times the interquartile range;

[0049] According to the smoothed data, principal component analysis is used to reduce the dimension and extract the main characteristic variables;

[0050] By using the main characteristic variables and the Apriori algorithm to perform association rule analysis, the association rules between the shredded tobacco process parameters, the shredded tobacco quality parameters and the product characteristics are mined, and the rule of "if the shredded tobacco angle is within the specified range and the filling value is within the specified range, the taste score is above the specified value" is obtained;

[0051] The adaptive response surface methodology was used to modify the model, new sampling points were generated through Box-Behnken experimental design, and the weighted least squares method was used to fit the new response surface equation.

[0052] Preferably, the step S6 specifically includes:

[0053] Obtain historical production data and optimization results, and use FCM algorithm for fuzzy clustering;

[0054] Outlier detection was performed on the data based on the fuzzy clustering results, outliers were identified by the box plot method, and the median was used to replace outliers when the data exceeded 1.5 times the interquartile range;

[0055] Establishing a rule-based expert knowledge base, wherein the expert knowledge base includes rules in the form of if-then;

[0056] A case library is constructed using a case-based reasoning algorithm, wherein the case library includes product characteristic targets, process parameter combinations, and actual production results, and the similarity between cases is calculated using cosine similarity;

[0057] Receive input of target product characteristics, perform rule reasoning based on the expert knowledge base, retrieve similar cases from the case library, and derive target process parameters by combining a fuzzy comprehensive evaluation method.

[0058] Preferably, the step S7 specifically includes:

[0059] Receiving initialization request information carrying a wire cutting process parameter optimization project identifier, setting a baseline progress and key performance indicators according to the initialization request information, and the digital management platform sending a project initialization completion signal according to the baseline progress and the key performance indicators;

[0060] Obtaining process parameters, quality indicators and optimization result data sent by the real-time data acquisition system, monitoring key nodes and milestones in real time, and determining whether the process parameters, quality indicators and optimization result data meet preset time and quality thresholds;

[0061] If the process parameters, the quality indicators and the optimization result data do not meet the preset time and quality thresholds, an audit task is automatically initiated through the workflow engine;

[0062] Receive the audit report generation instruction sent by the text analysis module, and automatically generate a phased audit report and a final optimization result report according to the audit report generation instruction.

[0063] The project follow-up method of a digital audit supervision system described in the present invention has the advantages of:

[0064] The project follow-up method of a digital audit supervision system of the present invention can quickly discover abnormalities in the production process by real-time collection and analysis of tobacco quality parameters, and timely adjust the cutting process parameters to ensure the stability and consistency of tobacco quality. The optimized cutting process parameter combination can maximize the production efficiency of tobacco while ensuring the improvement of product quality; it can automatically construct a response surface model, establish a nonlinear mapping relationship, and continuously optimize and improve the model through big data analysis, thereby improving the intelligence level of the production process; it can monitor tobacco quality parameters in real time, and can timely trigger re-optimization when quality parameters deviate, effectively avoiding the occurrence of quality problems; the control of the physical and chemical properties of tobacco further enhances the stability of the production process. Qualitative and controllable; through big data analysis, the system can continuously correct and improve the response surface model and accumulate knowledge on the optimization of shredding process parameters; the knowledge base and case library for the optimization of shredding process parameters established provide standardized shredding process formulas for future production, improving production efficiency and consistency of product quality; the process of optimizing shredding process parameters is included in the digital audit supervision system for management, which can track the progress of the optimization project in real time and ensure that the project is carried out as planned; audit supervision of key nodes and milestones can timely discover and solve problems in the project execution process, improve management efficiency, and the audit report formed provides decision-making support for management, which helps to continuously optimize the production process and improve product quality. A project follow-up method of a digital audit supervision system realizes the intelligent optimization and automatic regulation of shredding process parameters, improves the stability and consistency of tobacco quality, and provides effective technical support for tobacco product production. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of a project follow-up method of a digital audit supervision system described in the present invention. DETAILED DESCRIPTION

[0066] like Figure 1 As shown, a project follow-up method of a digital audit supervision system according to the present invention comprises the following steps:

[0067] S1. Obtain the shredded tobacco production line's shredded tobacco process parameters and shredded tobacco quality parameters, construct a response surface model, and establish a nonlinear mapping relationship;

[0068] S2. According to the constructed response surface model and the established nonlinear mapping relationship, the best combination of shredded tobacco process parameters is optimized and solved to obtain the target shredded tobacco process parameters for achieving the target shredded tobacco quality;

[0069] S3, sending the optimized target shredding process parameters to the shredding process of the tobacco production line, collecting quality parameters in real time and performing process monitoring, and triggering re-optimization when quality parameters deviate;

[0070] S4. Monitor the quality parameters of tobacco shreds during the shredding process, and control the physical and chemical properties of tobacco shreds whose quality parameters fluctuate irregularly or are uncontrollable;

[0071] S5. Obtain product characteristic data, conduct big data analysis on tobacco quality parameters, cutting process parameters and product characteristic data during the cutting process, and continuously revise and improve the response surface model based on the intrinsic relationship between the parameters;

[0072] S6. Obtain target process parameter combinations under different product characteristic targets, establish a knowledge base and case library for optimizing slitting process parameters, and form a standardized slitting process formula;

[0073] S7. Incorporate the process of optimizing the wire cutting process parameters into the digital audit supervision system for management, track the progress of the optimization project in real time, conduct audit supervision on key nodes and milestones, and form an audit report on the optimization results.

[0074] Furthermore, in this embodiment, step S1 specifically includes:

[0075] The cutting process parameters include the cutting tool angle, cutting tool speed and tobacco feed speed;

[0076] Tobacco quality parameters include tobacco filling value and combustion performance;

[0077] Collect real-time data of the cutter angle, cutter speed and tobacco feed speed, measure the tobacco filling value and combustion performance indicators, and normalize the collected raw data;

[0078] The moving average filtering method was used to remove the noise in the normalized data, and the box plot method was used to identify outliers;

[0079] A response surface model was constructed based on the filtered data, and a nonlinear mapping relationship between the cutting process parameters and the tobacco quality parameters was established using the support vector machine algorithm.

[0080] Furthermore, the central composite design method was used to determine the experimental plan, and a quadratic polynomial function was used to fit the response surface;

[0081] The significance and goodness of fit of the response surface model were evaluated by variance analysis to determine the value range and combination scheme of each process parameter;

[0082] A multivariate linear regression model was established, and significant variables were screened by stepwise regression method to obtain the relationship between cutting process parameters and tobacco quality parameters.

[0083] Using the grid search method, the process parameter combination traversed in the predefined parameter space is selected as the optimal process parameter if the difference between the predicted value and the target value is within the preset range;

[0084] According to the optimal process parameters, the angle of the cutting tool, the speed of the cutting tool and the feeding speed of the tobacco are adjusted to achieve closed-loop control of the cutting process;

[0085] Specifically, a high-precision sensor is used to collect real-time data on the angle of the cutting tool, the rotation speed of the cutting tool, and the feed speed of the tobacco, and a special detection device is used to measure the filling value of the tobacco and the index of the combustion performance, and the collected process parameters and quality parameters are stored in the database as a modeling sample;

[0086] The collected raw data were normalized to eliminate the dimension effect, the moving average filtering method was used to remove the noise in the data, and the box-and-whisker plot method was used to identify outliers;

[0087] The response surface model was constructed based on the processed sample data, the central composite design method was used to determine the experimental plan, and the response surface was fitted using a quadratic polynomial function;

[0088] The significance and goodness of fit of the model were evaluated by variance analysis, the value range and combination scheme of each process parameter were determined, and the polynomial function relationship between process parameters and quality indicators was fitted using the least squares method;

[0089] The relationship between the cutting process parameters and tobacco quality parameters was established using a multiple linear regression model, and significant variables were screened using the stepwise regression method to improve the model's explanatory power and prediction accuracy.

[0090] Based on the established multiple linear regression model, the current shredded tobacco process parameters are input to predict the corresponding tobacco filling value and combustion performance;

[0091] A grid search method is used to traverse possible combinations of process parameters in a predefined parameter space. For each set of parameters, the quality index is predicted using the established regression model, and the parameter combination that makes the predicted value closest to the target value is selected as the optimal process parameter.

[0092] According to the optimized process parameters, the angle of the cutting tool, the speed of the cutting tool and the feeding speed of the tobacco are adjusted to achieve closed-loop control of the cutting process;

[0093] Here is an example:

[0094] A high-precision laser displacement sensor is used to measure the angle of the shredder with an accuracy of 0.01 degrees, and a photoelectric encoder is used to monitor the tool speed with a resolution of 1rpm. The tobacco feed speed is measured by a weight sensor with an accuracy of 0.1kg / h. The special testing equipment uses a pneumatic method to measure the tobacco filling value with an accuracy of 0.1cm³ / g. The combustion performance is measured by an automatic ignition device with an accuracy of 0.1s. The acquisition frequency is set to 10Hz, and the data is collected continuously for 8 hours to obtain 28,800 sets of raw data.

[0095] The original data was normalized to min-max, the value range was adjusted to the interval [0,1], and a 5-point moving average filter was used to remove high-frequency noise. The smoothness of the filtered data was improved by 30%;

[0096] Outliers were identified by the box plot method, and data points exceeding 1.5 times the interquartile range were marked as abnormal, and a total of 87 abnormal points were eliminated;

[0097] The central composite design method was used to generate 25 experimental schemes, including 8 vertices, 6 center points, and 11 axis points;

[0098] Using a quadratic polynomial function Fitting response surface, where y is the quality index, b0 is the intercept (constant term) of the model, and b i is the first-order coefficient, x i and x j For different process parameters, b ij is the second-order interaction coefficient, indicating that the process parameter x i and x j The interaction effect between ii is a quadratic coefficient, representing the process parameter x i The squared contribution of the parameter itself, i.e., the nonlinear effect of the parameter on the response variable;

[0099] The coefficient of determination R² calculated by ANOVA was 0.92, indicating that the model had good goodness of fit;

[0100] The stepwise regression method was used to screen variables, the F statistic threshold for entering the model was set to 4.0, and three insignificant variables were eliminated. Finally, a multivariate linear regression model containing seven independent variables was obtained. The adjusted R² value of the model was 0.89, and the root mean square error of the prediction was 0.05.

[0101] A grid search method was used to generate 1000 sets of process parameter combinations in the predefined parameter space: wire cutting angle [15°, 25°], tool speed [1000rpm, 2000rpm], feed rate [500kg / h, 1000kg / h];

[0102] For each set of parameters, the regression model was used to predict the filling value and combustion performance, and the parameter combination that minimized the Euclidean distance between the predicted value and the target value (filling value 3.2 cm³ / g, combustion time 25 s) was selected as the optimal process parameter;

[0103] Finally, the optimized process parameters were obtained: cutting angle 20.3°, tool speed 1750rpm, and feed speed 720kg / h;

[0104] The optimized parameters are sent to the PLC control system, the servo motor is used to adjust the angle of the wire cutting tool, the frequency converter is used to adjust the tool speed, and the hopper feed valve is used to adjust the feed speed, thus realizing closed-loop control of the wire cutting process.

[0105] Furthermore, in this embodiment, step S2 specifically includes:

[0106] The cutting process parameters include the cutting tool angle, cutting tool speed and tobacco feed speed;

[0107] Tobacco quality parameters include tobacco filling value and combustion performance;

[0108] According to the constructed response surface model and the established nonlinear mapping relationship, the genetic algorithm intelligent optimization algorithm is adopted, and the optimization includes the following steps:

[0109] The preset chromosome encoding method encodes the three process parameters of the cutting tool angle, cutting tool speed and tobacco feed speed into a binary string, and sets the length of the binary string according to the parameter accuracy requirements;

[0110] Obtaining a set binary string, setting the fitness function of the genetic algorithm, and converting the tobacco filling value and the combustion performance into dimensionless indicators by using a normalization method;

[0111] After the fitness function is set, the roulette wheel selection operation is used to select excellent individuals, and the crossover probability of the roulette wheel selection operation is set to a preset value, and the single-point crossover method is used for gene recombination;

[0112] According to the selected excellent individuals, the genetic algorithm is iteratively run, and the population size of the genetic algorithm is set to a preset value, and the maximum number of iterations is set to a preset value;

[0113] When the genetic algorithm reaches the convergence condition or the maximum number of iterations, the cutting tool angle, cutting tool rotation speed and tobacco feed speed corresponding to the optimal individual are decoded and obtained, and used as the target cutting process parameters for the target tobacco quality;

[0114] Specifically, according to the constructed response surface model and the established nonlinear mapping relationship, the chromosome encoding method of the genetic algorithm is set, and the three process parameters of the cutting tool angle, the cutting tool speed and the tobacco feed speed are encoded into binary strings. The chromosome length is set according to the parameter accuracy requirements, each parameter occupies 32 bits, and the total length is 96 bits;

[0115] According to the actual value range of process parameters, the initial population is generated using uniformly distributed random numbers, each parameter is discretized, and the continuous value is mapped to the discrete space to ensure that the generated initial solution meets the actual process constraints;

[0116] The fitness function of the genetic algorithm is set, and the tobacco filling value and combustion performance are converted into dimensionless indicators by using the normalization method;

[0117] Use fuzzy comprehensive evaluation method to determine weight coefficients, combine expert opinions and historical data to improve the rationality of weight setting;

[0118] The filling value and combustion performance corresponding to the current chromosome are calculated through the response surface model, and the fitness value is obtained by comparing with the preset target value;

[0119] Roulette wheel selection was used to select excellent individuals, the crossover probability was set to 0.8, single-point crossover was used for gene recombination, the crossover point was generated by random numbers, the mutation probability was set to 0.01, gene mutation was achieved by bit inversion, the mutation position was determined by uniform random selection, and a new chromosome combination was generated;

[0120] After each iteration, the response surface model is used to calculate the filling value and combustion performance corresponding to the current optimal individual, and then compared with the historical optimal solution and updated;

[0121] Set the convergence judgment condition and terminate the iteration early when the change of the optimal solution for 30 consecutive generations is less than 0.0001;

[0122] The genetic algorithm was run iteratively, with the population size set to 100 and the maximum number of iterations set to 1000;

[0123] When the convergence condition or the maximum number of iterations is reached, the iteration is stopped, and the optimal individual corresponding to the shredder angle, shredder rotation speed and tobacco feed speed are decoded and obtained as the optimal process parameter combination;

[0124] The final optimization results were verified using the response surface model, and the tobacco filling value and combustion performance under the optimized parameters were calculated and compared with the target values ​​to evaluate the optimization effect.

[0125] Here is an example:

[0126] In the implementation of the genetic algorithm, the three process parameters of the cutter angle, cutter speed and tobacco feed speed are first encoded into a 96-bit binary string, where the angle range is 10°-30°, the speed range is 1000-2000rpm, and the feed speed range is 500-1000kg / h. Each parameter uses 32-bit encoding, and the accuracy reaches 0.01°, 0.1rpm and 0.1kg / h respectively;

[0127] The initial population is created with 100 individuals through a uniformly distributed random number generator, and each individual represents a set of possible process parameter combinations;

[0128] The fitness function was normalized, the filling value target was set to 3.2 cm³ / g, the combustion performance target was set to 25 s, and the weight coefficients were determined to be 0.6 and 0.4 respectively through the fuzzy comprehensive evaluation method;

[0129] In the roulette wheel selection operation, the probability of each individual being selected is proportional to its fitness value. The single-point crossover operation determines the crossover position by generating a random integer between 1 and 95, and swaps all genes after that position. The bit reversal mutation generates a random number between 0 and 1. If it is less than 0.01, it mutates. The mutation position is also determined by the random number;

[0130] After each iteration, the fitness value of the best individual in the population is calculated, compared with the historical best value and updated. When the change of the best solution for 30 consecutive generations is less than 0.0001 or reaches 1000 generations, the iteration is terminated;

[0131] The final optimization results were verified by the response surface model, and the filling value and combustion performance under the optimized parameters were calculated and compared with the target values ​​of 3.2 cm³ / g and 25 s. If the deviation was within the range of ±5%, the optimization was considered successful.

[0132] Furthermore, in this embodiment, step S2 also includes:

[0133] The shredding process parameters are encoded as parameter combinations into individuals, and the initial population is randomly generated;

[0134] Calculate the fitness of each individual based on the response surface model, select excellent individuals and generate new offspring individuals through crossover mutation;

[0135] Iterate multiple evolution operations to optimize the parameter combination, and finally select the individual with the highest fitness, and use its corresponding parameter combination as the best combination of optimal process parameters;

[0136] According to the three parameters of the cutting tool angle, the cutting tool rotation speed and the tobacco feed speed, a sixty-bit binary string is generated by using a binary coding method, wherein the cutting tool angle, the cutting tool rotation speed and the tobacco feed speed occupy twenty bits respectively;

[0137] An initial population is created by using a uniformly distributed random number generator, the initial population includes one hundred individuals, each of which represents a set of process parameter combinations;

[0138] A response surface model is obtained, and a fitness value of each individual is calculated according to the response surface model, wherein the fitness value is obtained by weighted summation of a tobacco filling value target and a combustion performance target after normalization;

[0139] The roulette wheel selection method is used to select excellent individuals. If excellent individuals are selected, gene recombination is carried out. Gene recombination is achieved by single-point crossover, and the crossover point is determined by a random number.

[0140] For the initial population, the evolution operation is iteratively run, where the population size is kept at 100. If the change of the optimal solution for 30 consecutive generations is less than the preset threshold, the iterative operation is terminated;

[0141] Obtain the individual with the highest fitness from the population, and decode the individual with the highest fitness to obtain the optimal process parameter combination;

[0142] Specifically, the three parameters of the cutter angle, cutter speed and tobacco feed speed are encoded into a 60-bit binary string using a binary encoding method, wherein the cutter angle, cutter speed and tobacco feed speed occupy 20 bits respectively. 100 individuals are created as the initial population through a uniformly distributed random number generator, and each individual represents a set of possible process parameter combinations.

[0143] The fitness value of each individual was calculated according to the response surface model, the filling value target was set to 3.2 cm³ / g, the combustion performance target was set to 25 s, the two indicators were converted into dimensionless values ​​by normalization processing method, and the comprehensive fitness was obtained by weighted summation. The weight coefficient was determined by hierarchical analysis method;

[0144] Set a reasonable parameter range. When the response surface model predicts a value outside the range, set the fitness to the minimum value. Sort the individuals according to the calculated fitness values ​​and select the top 50% of the individuals to enter the next generation.

[0145] Roulette wheel selection method was used to select excellent individuals, the crossover probability was set to 0.8, single-point crossover was used for gene recombination, the crossover point was determined by random numbers, the mutation probability was set to 0.01, gene mutation was achieved by bit inversion, and the mutation position was determined by uniform random selection method;

[0146] For invalid solutions generated by crossover and mutation, a repair strategy is adopted to replace values ​​beyond the parameter range with boundary values;

[0147] The elite individuals in the parent generation and the newly generated offspring individuals are combined to form a new generation population. The evolution operation is iterated, the population size is kept at 100, the maximum number of iterations is set to 1000, and the iteration is terminated early when the change in the optimal solution for 30 consecutive generations is less than 0.0001. The cutter angle, cutter speed and tobacco feed speed corresponding to the individual with the highest fitness are decoded and used as the optimal process parameter combination;

[0148] Use the response surface model to calculate the filling value and combustion performance under the optimal parameter combination, compare with the actual production results, and evaluate the optimization effect;

[0149] Here is an example:

[0150] In the implementation process, the wire cutting tool angle (range 10°-30°), rotation speed (range 1000-2000rpm) and feed speed (range 500-1000kg / h) are first encoded into a 60-bit binary string, with each parameter occupying 20 bits, and the accuracy reaches 0.00002°, 0.001rpm and 0.0005kg / h respectively;

[0151] A linear congruential generator was used to create 100 random individuals as the initial population, and the response surface model used a quadratic polynomial function. Fitting, where y is the quality index, b0 is the intercept (constant term) of the model, and b i is the first-order coefficient, x i and x j For different process parameters, b ij is the second-order interaction coefficient, indicating that the process parameter x i and x j The interaction effect between ii is a quadratic coefficient, representing the process parameter x i The squared contribution of the parameter itself, i.e., the nonlinear effect of the parameter on the response variable;

[0152] The weights of filling value and combustion performance were determined to be 0.6 and 0.4 respectively through the hierarchical analysis method. The effective range of tobacco filling value was set to 2.8-3.6 cm³ / g, and the effective range of combustion performance was set to 20-30s. When the range was exceeded, the adaptability was set to 0.

[0153] Individuals are arranged in descending order of fitness, and the first 50 enter the next generation directly. In the roulette wheel selection, the probability of each individual being selected is proportional to its fitness value;

[0154] The single-point crossover operation determines the crossover position by generating a random integer between 1 and 59. The bit reversal mutation generates a random number between 0 and 1, and mutates if it is less than 0.01. For invalid solutions, such as angles exceeding 30°, they are set to 30°;

[0155] The new generation population consists of 50 elite individuals and 50 offspring individuals. During the iteration process, the optimal solution of each generation is recorded, and the iteration stops when the change is less than 0.0001 for 30 consecutive generations or reaches 1000 generations;

[0156] Finally, the optimal parameter combination was obtained, such as a shredding angle of 20.5°, a rotation speed of 1750rpm, and a feed rate of 650kg / h. This combination was input into the response surface model, and the theoretical filling value and combustion performance were calculated and compared with the actual production data. If the error was within ±5%, the optimization was considered successful.

[0157] Furthermore, in this embodiment, step S3 specifically includes:

[0158] Obtaining the optimized target shredding process parameters, and sending the target shredding process parameters to the programmable logic controller of the tobacco production line via industrial Ethernet;

[0159] Controlling the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco according to the target cutting process parameters, and monitoring the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco in real time;

[0160] If there is a deviation between the actual parameters monitored and the target wire cutting process parameters, the PID control algorithm is used to dynamically adjust the input signal of the actuator;

[0161] Use online detection equipment to continuously collect tobacco filling value and combustion performance, and use Kalman filter algorithm to obtain the best estimated quality parameters;

[0162] The exponentially weighted moving average method is used to calculate the trend of the optimal estimated quality parameters, and the control chart method is combined for anomaly detection;

[0163] If the filling value or combustion performance of the tobacco slices sampled three times continuously exceeds the target value by ±5%, the shredding process parameter re-optimization program is triggered and the optimal parameter combination is recalculated; the optimal parameter combination is recalculated by calling the genetic algorithm module;

[0164] Specifically, the optimized target shredding process parameters are sent to the programmable logic controller of the tobacco production line through industrial Ethernet, and the parameters are transmitted in 24-bit floating point format to ensure that the accuracy loss when converted to analog signals is less than 0.01%. The three parameters of shredding tool angle, shredding tool speed and tobacco feed speed are converted into corresponding analog signals respectively, and the actuator is driven to make automatic adjustments.

[0165] High-precision sensors are used to monitor the angle of the cutting tool, the speed of the cutting tool and the feeding speed of the tobacco in real time. The PID control algorithm is used to dynamically adjust the input signal of the actuator according to the deviation between the actual parameters and the target parameters to achieve precise adjustment of the parameters. The control accuracy reaches ±0.1°, ±1rpm and ±0.1kg / h respectively, and the actual parameter values ​​are transmitted back to the central control unit.

[0166] The process parameters and quality parameters are uniformly stored in the time series database, and the moving average filtering method is used to remove high-frequency noise;

[0167] Use online detection equipment to continuously collect tobacco filling value and combustion performance, with the sampling frequency set to once per minute. Use Kalman filter algorithm to combine historical data and current measurement values ​​to obtain the best estimated quality parameters, store them in the real-time database, and display them in real time and record trends on the human-machine interface;

[0168] Use the exponentially weighted moving average method to calculate the trend of quality parameters and combine it with the control chart method for anomaly detection;

[0169] According to the preset tolerance range of quality parameters, when the filling value or combustion performance of three consecutive samples exceeds the target value by ±5%, the shredding parameter re-optimization program is automatically triggered, and the genetic algorithm module is called to recalculate the optimal parameter combination;

[0170] The genetic algorithm module sets the population size to 100, the number of iterations to 500, the crossover probability to 0.8, the mutation probability to 0.01, and uses roulette wheel selection and single-point crossover to send the new parameters to the production line for execution;

[0171] Here is an example:

[0172] The optimized wire cutting process parameters are transmitted through the PROFINET interface of the Siemens S7-1500 series PLC, using the IEEE754 standard single-precision floating point format to ensure accuracy;

[0173] The angle parameter of the wire cutting tool (range 10°-30°) is converted into a 4-20mA analog signal to drive the servo motor, the speed parameter (range 1000-2000rpm) is converted into a 0-10V signal to control the frequency converter, and the feed speed parameter (range 500-1000kg / h) is converted into a PWM signal to adjust the opening of the feed valve; a photoelectric encoder is used to measure the actual angle with an accuracy of 0.01°; a Hall sensor detects the speed with a resolution of 1rpm; a mass flow meter monitors the feed speed with an accuracy of 0.1kg / h; the PID controller parameters are set to Kp=0.5, Ki=0.2, Kd=0.1, and the sampling period is 100ms to achieve stable parameter control;

[0174] Process parameters and quality data are stored in the InfluxDB time series database at a frequency of 1 Hz, and a 5-point moving average filter is used. The smoothness of the filtered data is increased by 30%;

[0175] The online detection equipment uses the pneumatic method to measure the filling value with an accuracy of 0.1cm³ / g; the automatic ignition device determines the combustion performance with an accuracy of 0.1s;

[0176] In the Kalman filter algorithm, the process noise covariance Q=0.001, the measurement noise covariance R=0.1, and the initial estimation error covariance P=1;

[0177] The EWMA was calculated using a smoothing factor of 0.2 and combined with the Shewhart control chart (3σ limits) for anomaly detection;

[0178] When reoptimization is triggered, the genetic algorithm uses 60-bit binary encoding, a 10% elite retention rate in roulette wheel selection, and single-point crossover positions are randomly generated.

[0179] Furthermore, in this embodiment, step S4 specifically includes:

[0180] The physical and chemical properties of cut tobacco include nicotine content and reducing sugar content;

[0181] Statistical process control charts are used to monitor the changing trends of tobacco quality parameters. Statistical process control charts apply the eight exception criteria of sigma control charts.

[0182] Determine whether the judgment conditions are met according to the eight judgment criteria;

[0183] If the abnormality judgment condition is not met, it is determined that the tobacco quality parameters do not have irregular fluctuations or are uncontrollable, and the subsequent processing is continued;

[0184] If the abnormality judgment conditions are met, it is determined that the tobacco quality parameters have irregular fluctuations or are uncontrollable, triggering the detection and analysis program for the physical and chemical properties of the tobacco, and the detection and analysis results are fed back to the upstream blending process for raw material optimization;

[0185] The detection and analysis procedure includes the following steps:

[0186] Collect tobacco samples from the production line, and the tobacco samples are collected at fixed time intervals by an automatic sampling system;

[0187] Acquiring spectral data of the tobacco sample, the spectral data is obtained by quickly scanning the tobacco sample with an online near-infrared spectrometer;

[0188] Establish a quantitative analysis model between spectral data and content, and the quantitative analysis model is established using partial least squares regression method;

[0189] Calculate the specific content value of the tobacco sample, which is obtained by calculating the quantitative analysis model;

[0190] The physical and chemical properties data of the tobacco samples were compared with the historical normal batch data, and the comparison was carried out using the principal component analysis method;

[0191] Calculate the Mahalanobis distance between the tobacco sample and the historical normal batch. If the Mahalanobis distance exceeds the preset threshold, it is determined that the tobacco quality is abnormal due to fluctuations in the properties of the raw materials. According to the results of the principal component analysis, the category of the tobacco quality abnormality is determined, and the test and analysis results are fed back to the upstream blending process for raw material optimization. The raw material optimization is to adjust the raw material ratio according to the category of the tobacco quality abnormality;

[0192] If the Mahalanobis distance does not exceed the preset threshold, the data is collected again for the detection and analysis procedure;

[0193] Specifically, the statistical process control chart is used to monitor the changing trend of tobacco quality parameters, and the eight abnormality criteria of the sigma control chart are applied, including 7 consecutive points showing an upward or downward trend, any point exceeding the 3σ control limit, and at least one of two consecutive points exceeding the 2σ limit. When any abnormality condition is met, the detection and analysis procedure of the physical and chemical properties of tobacco is triggered;

[0194] Use an automatic sampling system to collect tobacco samples from the production line at regular time intervals to ensure that the samples are representative;

[0195] The tobacco samples were quickly scanned by an online near-infrared spectrometer to obtain the spectral data of nicotine content and reducing sugar content;

[0196] Regularly calibrate the near-infrared spectrometer using standard samples of known content, establish a calibration curve, use partial least squares regression to establish a quantitative analysis model between spectral data and content, calculate the specific content value, and store the test results in the batch management database;

[0197] Use principal component analysis to compare the physical and chemical properties of the current batch with the historical normal batch data, determine the number of principal components through eigenvalue decomposition, and calculate sample scores and variable loadings;

[0198] Calculate the Mahalanobis distance. If the distance value exceeds the preset threshold, it is determined that the fluctuation of raw material properties is the cause of the abnormal quality of tobacco. According to the results of principal component analysis, the abnormality is divided into categories such as raw material component deviation and processing parameter abnormality.

[0199] If it is determined that the quality is abnormal due to fluctuations in the properties of the raw materials, the test results and abnormality judgment information will be fed back to the recipe optimization module of the upstream blending process through the manufacturing execution system, and the automatic recipe optimization program based on the feedback data will be started. The raw material ratio will be adjusted according to the abnormality category, such as increasing or decreasing the proportion of tobacco leaves of a specific variety. The optimized recipe will be trial-produced in small batches, and the physical and chemical properties will be tested again to verify the optimization effect.

[0200] If the Mahalanobis distance does not exceed the preset threshold, the data is collected again for the detection and analysis procedure;

[0201] Here is an example:

[0202] Minitab software was used to implement statistical process control, with the fill value target set at 3.2 cm³ / g, the upper control limit at 3.5 cm³ / g, and the lower control limit at 2.9 cm³ / g. When seven consecutive points showed a monotonic trend or any point exceeded the ±3σ range, anomaly detection was automatically triggered;

[0203] The automatic sampling system collects 100 g of tobacco samples from the production line every 15 minutes. The samples are scanned using a Bruker MPA near-infrared spectrometer with a wavelength range of 4000-12500 cm -1 , resolution 4cm -1 , scanning times 32 times; calibrate with 5 standard samples of known content every 8 hours, establish a quadratic polynomial calibration curve, and the R² value must be greater than 0.99;

[0204] The quantitative model was established using the PLS regression algorithm in OPUS software, 50 samples were selected for modeling, and the cross-validation RMSECV was less than 0.1%. The test results were stored in the Oracle database in real time;

[0205] Use MATLAB software to perform principal component analysis, and determine the number of principal components when the cumulative contribution rate is greater than 85%, usually 3-5;

[0206] Calculate the T² statistic and SPE statistic, and determine it as abnormal when both exceed the 99% confidence interval;

[0207] Determine the abnormal type based on the load matrix. For example, a large load value of the first principal component indicates an abnormal total amount of the component, and a large load value of the second principal component indicates an abnormal proportion of the component.

[0208] Transmit the abnormal information to the leaf distribution system controlled by Siemens S7-1500PLC through OPCUA protocol;

[0209] The optimization algorithm uses a genetic algorithm with a population size of 50, 100 iterations, a crossover probability of 0.8, a mutation probability of 0.01, and the optimization goal is to make the predicted physical and chemical indicators closest to the target value.

[0210] Furthermore, in this embodiment, step S5 specifically includes:

[0211] Product characteristic data include cigarette taste, aroma and smoke index;

[0212] Obtain the taste data, aroma data and smoke index of cigarette products, and obtain the shredded process parameters and tobacco quality parameters from the production line; the taste data and aroma data are collected by electronic tongue equipment and electronic nose equipment; the smoke index is measured by gas chromatography-mass spectrometry;

[0213] Outlier detection was performed on taste data, aroma data, smoke and smog index, shredded tobacco process parameters, and tobacco quality parameters. Outliers were identified using the boxplot method, and data exceeding 1.5 times the interquartile range were smoothed.

[0214] According to the smoothed data, principal component analysis is used to reduce the dimension and extract the main characteristic variables;

[0215] By using the main characteristic variables and the Apriori algorithm to perform association rule analysis, we can mine the association rules between the cutting process parameters, tobacco quality parameters and product characteristics, and obtain the rule that "if the cutting angle is within the specified range and the filling value is within the specified range, the taste score is above the specified value";

[0216] The adaptive response surface methodology was used to modify the model, new sampling points were generated through Box-Behnken experimental design, and the weighted least squares method was used to fit the new response surface equation.

[0217] If the prediction error is less than the preset threshold, the adaptive response surface method correction model is applied to the actual production process control;

[0218] Specifically, the taste and aroma data of cigarette products are collected through electronic tongue and electronic nose devices, and the smoke index is measured by gas chromatography-mass spectrometry. At the same time, the shredding process parameters and tobacco quality parameters are collected from the production line in real time, and all data are uniformly stored in a distributed database;

[0219] Perform outlier detection on the stored data, use the box plot method to identify outliers, and smooth the data that exceeds 1.5 times the interquartile range;

[0220] The principal component analysis method is used to reduce the dimension of the preprocessed multidimensional data, extract the main characteristic variables, and reduce the complexity of the data;

[0221] The key parameters were screened by two-way ANOVA, the F value and p value were calculated, and the significance level was selected as 0.05 to determine the parameters that significantly affected the product characteristics;

[0222] The Apriori algorithm is used to analyze association rules, and the support and confidence are dynamically adjusted according to the distribution of historical data. The initial minimum support is set to 0.1 and the minimum confidence is set to 0.8;

[0223] Mining the association rules between shredded tobacco process parameters, tobacco quality parameters and product characteristics, we obtained a rule such as "If the shredded tobacco angle is between 20° and 22° and the filling value is between 3.1 and 3.3 cm³ / g, the taste score is above 8.5";

[0224] The 10-fold cross-validation method was used to evaluate the generalization ability of the rules, and the rules with an accuracy rate exceeding 85% were retained;

[0225] Based on the verified association rules, the adaptive response surface method was used to modify and improve the original model, new sampling points were generated through Box-Behnken experimental design, and the weighted least squares method was used to fit the new response surface equation.

[0226] Set the maximum number of iterations to 50, and define the convergence criterion as prediction error less than 1%. After each iteration, dynamically adjust the sampling point distribution according to the prediction error;

[0227] Continuously iterate and optimize the model parameters until the convergence criterion or the maximum number of iterations is reached, and the prediction accuracy of the optimized model is verified using new production batch data. If the prediction error is less than 3%, the model is applied to the actual production process control;

[0228] Here is an example:

[0229] Taste and aroma data were collected using AlphaMOS's ASTREEII electronic tongue and HERACLESNEO electronic nose, with samples taken every hour and each measurement lasting 5 minutes. Smoke and smoke indicators were measured using an Agilent 7890B gas chromatograph and 5977A mass spectrometer, with samples collected every 2 hours.

[0230] Production line parameters are collected in real time by Siemens S7-1500 PLC with a sampling frequency of 1 Hz. All data are stored in the Apache Hadoop distributed file system and Spark SQL is used for data preprocessing.

[0231] Outlier detection was performed using the box plot method, with the interquartile range multiple set to 1.5, and the 5-point moving average method was used to smooth out the data points that were out of range;

[0232] Principal component analysis is implemented using the sklearn library, with the cumulative contribution rate threshold set to 85%, and usually 3-5 principal components are retained;

[0233] Two-way ANOVA was performed using SAS software, with a significance level of α = 0.05 and an F critical value of 3.84. The Apriori algorithm was implemented using the arules package of the R language, with an initial minimum support of 0.1, a minimum confidence of 0.8, and dynamic adjustment every 100 data points with an adjustment step of 0.01;

[0234] The 10-fold cross validation was implemented using Python's sklearn library, and the random seed was set to 42 to ensure repeatable results;

[0235] The adaptive response surface method uses the ResponseSurface toolbox of MATLAB. The Box-Behnken design selects three factors and five levels, generating a total of 15 sampling points. The weight in the weighted least squares method is inversely proportional to the prediction error.

[0236] During the iteration process, for example, if the prediction error changes less than 0.1% for 5 consecutive times, it is considered to have converged;

[0237] The final model was verified on new batch data with a prediction error of 2.7%, which met the application requirements. It was then deployed to the production line PLC for real-time parameter optimization control.

[0238] Furthermore, in this embodiment, step S6 specifically includes:

[0239] Obtain historical production data and optimization results, and use the FCM algorithm for fuzzy clustering; the cluster number of the FCM algorithm is set to 5, and the fuzzy index is set to 2;

[0240] Outlier detection was performed on the data based on the fuzzy clustering results, outliers were identified by the box plot method, and the median was used to replace outliers when the data exceeded 1.5 times the interquartile range;

[0241] Establish a rule-based expert knowledge base, which includes if-then rules; if the product aroma index requirement is high, the shredding angle should be within the range of 18°-20°;

[0242] A case-based reasoning algorithm is used to build a case library, which includes product characteristic targets, process parameter combinations, and actual production results. The similarity between cases is calculated using cosine similarity.

[0243] Receive the input of target product characteristics, perform rule reasoning based on the expert knowledge base, retrieve similar cases from the case library, and combine the fuzzy comprehensive evaluation method to obtain the target process parameters; the fuzzy comprehensive evaluation method uses the weighted average method for comprehensive evaluation;

[0244] Specifically, based on historical production data and optimization results, the mapping relationship between product characteristics and wire cutting process parameters was established, and the FCM (Fuzzy C-Means) algorithm was used for fuzzy clustering, with the number of clusters set to 5 and the fuzzy index set to 2;

[0245] The data were tested for outliers, the box plot method was used to identify outliers, and the data exceeding 1.5 times the interquartile range were replaced by the median;

[0246] The weighted average of the process parameters of each group is taken as the standard process parameter combination for this type of product, and the weight is determined based on the distance from the sample to the cluster center;

[0247] Build a rule-based expert knowledge base to convert the experience and knowledge of process experts into if-then rules, such as "If the product aroma index requirement is high, the shredding angle should be within the range of 18°-20°";

[0248] The Delphi method was used to collect expert opinions, the confidence of the rules was determined through three rounds of questionnaire iterations, the importance weights of the rules were calculated using the analytic hierarchy process, a judgment matrix was constructed, and consistency tests were performed;

[0249] Use case-based reasoning algorithms to build a case library, taking each successful production batch as a case, including product characteristic targets, process parameter combinations, and actual production results;

[0250] Cases are represented in the form of attribute-value pairs. The similarity between cases is calculated by cosine similarity. A kd-tree index structure is established to improve retrieval efficiency. The update frequency is set to be monthly. The screening criteria for new cases are that the product quality meets the standards and the process parameters are highly stable.

[0251] When producing new products, the target product characteristics are input, and the optimal process parameter combination is obtained through knowledge base rule reasoning and case base similar case retrieval combined with fuzzy comprehensive evaluation method;

[0252] The evaluation factors include parameter feasibility, expected product quality and production efficiency. A trapezoidal membership function is constructed and the weighted average method is used for comprehensive evaluation.

[0253] Convert the optimized parameter combination into a standard process recipe, including specific equipment adjustment instructions and operating steps;

[0254] This combination is sent to the production line as the initial value for execution, and the online optimization program is started for fine-tuning;

[0255] Here is an example:

[0256] Use Python's scikit-fuzzy library to implement the FCM algorithm, set the number of clusters to 5, the fuzzy index to 2, the maximum number of iterations to 100, and the termination condition to be the objective function change less than 0.001;

[0257] Outlier detection uses the boxplot function of the scipy library, replacing data outside 1.5 times the interquartile range with the median;

[0258] The standard process parameters of each cluster are weighted averaged by the average function of the numpy library, with the weight being 1 / (the distance from the sample to the cluster center + 0.1);

[0259] The Delphi questionnaire was distributed through the SurveyMonkey platform, with a total of 20 experts participating. After three rounds of iterations, the convergence criterion was 75% expert consensus;

[0260] The analytic hierarchy process was implemented using the ahp package of the R language, and a 5 × 5 judgment matrix was constructed. The consistency ratio CR was considered to have passed the test when it was less than 0.1;

[0261] The case library is stored in MongoDB, and each case contains 15 attribute-value pairs;

[0262] The cosine similarity calculation uses the cosine_similarity function of the sklearn library, and the kd tree index is implemented by scipy.spatial.KDTree, setting the leaf node size to 10;

[0263] The case update program runs automatically on the last working day of each month. The screening criteria are that the product quality inspection pass rate is greater than 98% and the process parameter fluctuation range is less than ±2%;

[0264] The fuzzy comprehensive evaluation uses five evaluation factors, the membership function uses the trapezoid function, and the weight vector is obtained by the hierarchical analysis method;

[0265] The final standard process recipe includes 10 key parameter setting values ​​and 5 core operation steps, which are sent to the Siemens S7-1500 PLC for execution via the OPC UA protocol.

[0266] Furthermore, in this embodiment, step S7 specifically includes:

[0267] Receive the initialization request information carrying the slitting process parameter optimization project identifier, set the baseline progress and key performance indicators according to the initialization request information, and the digital management platform sends a project initialization completion signal according to the baseline progress and key performance indicators; the initialization request information is sent by the digital management platform when it is connected to the network;

[0268] Obtain the process parameters, quality indicators and optimization result data sent by the real-time data acquisition system, conduct real-time monitoring of key nodes and milestones, and determine whether the process parameters, quality indicators and optimization result data meet the preset time and quality thresholds; the process parameters, quality indicators and optimization result data are collected by the real-time data acquisition system from the production line, laboratory and quality inspection department;

[0269] If the process parameters, quality indicators and optimization result data do not meet the preset time and quality thresholds, the workflow engine will automatically initiate an audit task; the audit task will be assigned to the relevant responsible persons by the workflow engine;

[0270] Receive the audit report generation instruction sent by the text analysis module, and automatically generate the interim audit report and the final optimization result report according to the audit report generation instruction; the audit report generation instruction is generated by the text analysis module according to the keywords extracted by the TF-IDF algorithm and the summary generated by the TextRank algorithm; the interim audit report and the final optimization result report contain keywords, summaries and numerical indicators;

[0271] Specifically, a digital management platform for the wire cutting process parameter optimization project was built to initialize the project, set the baseline progress and key performance indicators (KPIs), divide the optimization process into four stages: data collection, model building, parameter optimization, and result verification. Key nodes and milestones were set for each stage, and the project progress was displayed through a Gantt chart. Progress information was automatically updated using Microsoft Project.

[0272] A real-time data collection system is used to obtain process parameters, quality indicators, and optimization result data from production lines, laboratories, and quality inspection departments. Data is cleaned and converted through Talend Open Studio. Cleaning rules include removing missing values, detecting outliers, and unifying units. The processed data is stored in the InfluxDB time series database.

[0273] Set up automated audit rules to monitor key nodes and milestones in real time, and set specific time and quality thresholds. For example, if the progress delay exceeds 5% or the quality indicator deviation exceeds 3%, a yellow warning will be triggered, and if it exceeds 10%, a red warning will be triggered;

[0274] If anomalies or delays are detected, an audit task is automatically initiated through the Camunda workflow engine and assigned to the relevant responsible person for processing;

[0275] Establish an audit result feedback mechanism, including problem tracking, improvement measures formulation and implementation effect evaluation, and use Jira software to record and track the progress of problem solving;

[0276] Use the NLTK library to perform text analysis, use the TF-IDF algorithm to extract keywords, and the TextRank algorithm to generate summaries. Combined with numerical indicators, the interim audit report and the final optimization results report are automatically generated.

[0277] The audit report structure includes an executive summary, key findings, risk assessment, improvement recommendations, and follow-up action plan;

[0278] Publish and archive through a digital platform that uses role-based access control (RBAC) for user rights management and AES-256 encryption algorithm to protect sensitive data;

[0279] Here is an example:

[0280] The wire cutting process parameter optimization project was managed through Microsoft Project, with a total duration of 12 weeks, and the four phases were 2 weeks, 3 weeks, 5 weeks, and 2 weeks respectively;

[0281] Key performance indicators include shred uniformity (target value ≥ 95%), filling value (target value 3.2 ± 0.1 cm³ / g) and production capacity (target value ≥ 500 kg / h);

[0282] Talend Open Studio sets data cleaning rules: fields with missing value rates > 5% are deleted, outliers are defined as exceeding 3 times the standard deviation, and units are unified to the international system of units;

[0283] InfluxDB uses shard storage, one shard per hour, with a retention period of 3 months. In the audit rules, a 5% delay in progress triggers an email notification, and a 10% delay triggers an SMS alert;

[0284] Camunda workflow is set to automatically escalate after 48 hours. Issues in Jira are divided into four states: "pending", "in progress", "resolved" and "closed";

[0285] NLTK uses the punkt tokenizer, the TF-IDF algorithm to extract the top-10 keywords, and the TextRank algorithm to generate a summary of about 100 words;

[0286] The audit report adopts a five-part structure, with an executive summary of no more than 500 words, 3-5 key findings, a 5×5 matrix for risk assessment, no fewer than 3 suggestions for improvement, and a follow-up action plan with responsible persons and deadlines;

[0287] RBAC sets 4 levels of permissions: read-only, edit, approve, and manage. AES-256 encryption uses hardware acceleration, and the key is updated every quarter.

[0288] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the device or element referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention.

[0289] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A project follow-up method for a digital audit supervision system, characterized in that: The following steps are involved: S1. Obtain the shredded tobacco production line's shredded tobacco process parameters and shredded tobacco quality parameters, construct a response surface model, and establish a nonlinear mapping relationship; S2. According to the constructed response surface model and the established nonlinear mapping relationship, the best combination of shredded tobacco process parameters is optimized and solved to obtain the target shredded tobacco process parameters for achieving the target shredded tobacco quality; S3, sending the optimized target shredding process parameters to the shredding process of the tobacco production line, collecting quality parameters in real time and performing process monitoring, and triggering re-optimization when quality parameters deviate; S4. Monitor the quality parameters of tobacco shreds during the shredding process, and control the physical and chemical properties of tobacco shreds whose quality parameters fluctuate irregularly or are uncontrollable; S5. Obtain product characteristic data, conduct big data analysis on tobacco quality parameters, cutting process parameters and product characteristic data during the cutting process, and continuously revise and improve the response surface model based on the intrinsic relationship between the parameters; S6. Obtain target process parameter combinations under different product characteristic targets, establish a knowledge base and case library for optimizing slitting process parameters, and form a standardized slitting process formula; S7. Incorporate the process of optimizing the cutting process parameters into the digital audit supervision system for management, track the progress of the optimization project in real time, conduct audit supervision on key nodes and milestones, and form an audit report on the optimization results; The step S1 specifically includes: The shredded tobacco process parameters include the shredded tobacco angle, shredded tobacco rotation speed and shredded tobacco feeding speed; The shredded tobacco quality parameters include shredded tobacco filling value and combustion performance; Collecting real-time data of the cutter angle, the cutter rotation speed and the tobacco feed speed, measuring the tobacco filling value and the combustion performance index, and normalizing the collected raw data; The moving average filtering method was used to remove the noise in the normalized data, and the box plot method was used to identify outliers; A response surface model is constructed according to the filtered data, and a nonlinear mapping relationship between the shredded tobacco process parameters and the shredded tobacco quality parameters is established through a support vector machine algorithm.

2. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized by: The step S2 specifically includes: The shredded tobacco process parameters include the shredded tobacco angle, shredded tobacco rotation speed and shredded tobacco feeding speed; The shredded tobacco quality parameters include shredded tobacco filling value and combustion performance; According to the constructed response surface model and the established nonlinear mapping relationship, the genetic algorithm intelligent optimization algorithm is adopted. The optimization includes the following steps: The preset chromosome encoding method encodes the three process parameters of the shredder angle, the shredder rotation speed and the shredded tobacco feeding speed into a binary string, and sets the length of the binary string according to the parameter accuracy requirements; Obtaining a set binary string, setting a fitness function of a genetic algorithm, and converting the tobacco filling value and the combustion performance into dimensionless indicators using a normalization method; After the fitness function is set, the roulette wheel selection operation is used to select excellent individuals, and the crossover probability of the roulette wheel selection operation is set to a preset value, and the single-point crossover method is used for gene recombination; According to the selected excellent individuals, the genetic algorithm is iteratively run, and the population size of the genetic algorithm is set to a preset value, and the maximum number of iterations is set to a preset value; When the genetic algorithm reaches the convergence condition or the maximum number of iterations, the cutting tool angle, cutting tool rotation speed and tobacco feed speed corresponding to the optimal individual are decoded and obtained as the target cutting process parameters for the target tobacco quality.

3. According to the project follow-up method of the digital audit supervision system of claim 2, it is characterized in that: The step S2 further comprises: Encoding the shredding process parameters as parameter combinations into individuals, and randomly generating an initial population; Calculate the fitness of each individual based on the response surface model, select excellent individuals and generate new offspring individuals through crossover mutation; Iterate multiple evolutionary operations to optimize the parameter combination, and finally select the individual with the highest fitness, and use its corresponding parameter combination as the best combination of optimal process parameters.

4. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized in that: The step S3 specifically includes: Obtaining the optimized target shredding process parameters, and sending the target shredding process parameters to the programmable logic controller of the tobacco production line via industrial Ethernet; Controlling the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco according to the target cutting process parameters, and monitoring the angle of the cutting tool, the rotation speed of the cutting tool and the feeding speed of the tobacco in real time; If there is a deviation between the monitored actual parameters and the target wire cutting process parameters, a PID control algorithm is used to dynamically adjust the input signal of the actuator; Use online detection equipment to continuously collect tobacco filling value and combustion performance, and use Kalman filter algorithm to obtain the best estimated quality parameters; The exponentially weighted moving average method is used to calculate the trend of the optimal estimated quality parameters, and the control chart method is combined for anomaly detection; If the tobacco filling value or combustion performance of three consecutive samples exceeds the target value by ±5%, the shredding process parameter re-optimization program will be triggered and the optimal parameter combination will be recalculated.

5. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized in that: The step S4 specifically includes: The physicochemical properties of the cut tobacco include nicotine content and reducing sugar content; A statistical process control chart is used to monitor the changing trend of tobacco quality parameters, and the statistical process control chart applies the eight abnormality criteria of the sigma control chart; Determine whether the difference determination conditions are met according to the eight difference determination criteria; If the abnormality judgment condition is not met, it is determined that the tobacco quality parameters do not have irregular fluctuations or are uncontrollable, and the subsequent processing is continued; If the abnormality judgment conditions are met, it is determined that the quality parameters of the tobacco are irregularly fluctuating or uncontrollable, triggering the detection and analysis procedure for the physical and chemical properties of the tobacco, and the detection and analysis results are fed back to the upstream leaf blending process for raw material optimization.

6. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized in that: The step S5 specifically includes: The product characteristic data include the taste, aroma and smoke index of the cigarette; Obtain taste data, aroma data and smoke index of cigarette products, and obtain cutting process parameters and tobacco quality parameters from the production line; Performing outlier detection on the taste data, the aroma data, the smoke index, the shredded tobacco process parameters, and the shredded tobacco quality parameters, using a box plot method to identify outliers, and performing smoothing on data that exceeds 1.5 times the interquartile range; According to the smoothed data, principal component analysis is used to reduce the dimension and extract the main characteristic variables; By using the main characteristic variables and the Apriori algorithm to perform association rule analysis, the association rules between the shredded tobacco process parameters, the shredded tobacco quality parameters and the product characteristics are mined, and the rule of "if the shredded tobacco angle is within the specified range and the filling value is within the specified range, the taste score is above the specified value" is obtained; The adaptive response surface methodology was used to modify the model, new sampling points were generated through Box-Behnken experimental design, and the weighted least squares method was used to fit the new response surface equation.

7. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized in that: The step S6 specifically includes: Obtain historical production data and optimization results, and use FCM algorithm for fuzzy clustering; According to the fuzzy clustering results, the data were tested for outliers, outliers were identified by the box plot method, and the median was used to replace outliers when the data exceeded 1.5 times the interquartile range; Establishing a rule-based expert knowledge base, wherein the expert knowledge base includes rules in the form of if-then; A case library is constructed using a case-based reasoning algorithm, wherein the case library includes product characteristic targets, process parameter combinations, and actual production results, and the similarity between cases is calculated using cosine similarity; Receive input of target product characteristics, perform rule reasoning based on the expert knowledge base, retrieve similar cases from the case library, and derive target process parameters by combining a fuzzy comprehensive evaluation method.

8. According to the project follow-up method of the digital audit supervision system of claim 1, it is characterized by: The step S7 specifically includes: Receiving initialization request information carrying a wire cutting process parameter optimization project identifier, setting a baseline progress and key performance indicators according to the initialization request information, and the digital management platform sending a project initialization completion signal according to the baseline progress and the key performance indicators; Obtaining process parameters, quality indicators and optimization result data sent by the real-time data acquisition system, monitoring key nodes and milestones in real time, and determining whether the process parameters, quality indicators and optimization result data meet preset time and quality thresholds; If the process parameters, the quality indicators and the optimization result data do not meet the preset time and quality thresholds, an audit task is automatically initiated through the workflow engine; Receive the audit report generation instruction sent by the text analysis module, and automatically generate a phased audit report and a final optimization result report according to the audit report generation instruction.

Images