Dyeing process parameter optimization and recommendation system based on big data analysis

Abstract

This invention discloses a dyeing process parameter optimization and recommendation system based on big data analysis, specifically relating to the field of dyeing process parameter optimization. The system includes a data interface unit, a hybrid modeling unit, a parameter recommendation unit, and a feedback learning unit. The data interface unit collects and processes multi-source heterogeneous data, generating historical data packages and current order data packages. The hybrid modeling unit, based on historical data packages, combines mechanistic constraints and data-driven methods to establish a mapping relationship between process parameters and dyeing results. The parameter recommendation unit, based on current order data, combines a quality prediction model and a multi-objective optimization strategy to generate multiple sets of recommended process parameters. The feedback learning unit analyzes the sources of deviation and updates the model in real time by comparing actual and predicted quality indicators. This system enables intelligent recommendation and dynamic optimization of dyeing process parameters, significantly improving dyeing quality stability, reducing energy consumption and production costs, and possessing good adaptive and continuous optimization capabilities.

Description

Technical Field

[0001] This invention relates to the field of dyeing process parameter optimization technology, and more specifically, to a dyeing process parameter optimization and recommendation system based on big data analysis. Background Technology

[0002] Currently, the setting of dyeing process parameters largely relies on the experience of process engineers or traditional trial-and-error methods, lacking systematicity and scientific rigor. Although some enterprises have introduced automated control systems that can collect some production data, these systems still have some shortcomings in actual use. For example, existing systems are often limited to a single data source or a single type of data, failing to effectively integrate multi-source heterogeneous information such as process parameters, material properties, sensor time-series data, environmental data, and equipment status data. Furthermore, they do not perform process-quality-energy consumption correlation labeling on the collected data, resulting in insufficient basis for parameter optimization. At the same time, traditional parameter recommendation methods are mostly based on static empirical rules or single mechanism models. Mechanism constraints and data-driven approaches are not linked, making it difficult to adapt to dynamic changes in actual production conditions such as changes in material properties and environmental fluctuations, resulting in poor generalization and adaptability.

[0003] On the other hand, existing solutions generally lack effective online learning and feedback mechanisms, or the feedback mechanisms do not clearly define the triggering conditions and update scope, making it impossible to accurately and continuously optimize the prediction model based on actual production results, resulting in model aging and a decline in recommendation effectiveness as production batches increase; in addition, most systems do not clearly define the matching logic between order attributes and optimization modes, making it difficult to achieve the comprehensive goals of energy saving, consumption reduction, and improved production efficiency while ensuring dyeing quality. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a dyeing process parameter optimization and recommendation system based on big data analysis, which solves the problems of data silos, static models, inability to learn, and singular optimization mentioned in the background art through the following solutions.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dyeing process parameter optimization and recommendation system based on big data analysis, comprising a data interface unit for collecting and processing multi-source heterogeneous data, generating historical data packages and current order data packages; the multi-source heterogeneous data includes process parameters, material properties, sensor time-series data, product quality indicators, environmental data, and equipment status data;

[0006] The data interface unit adds process-quality-energy consumption association tags to multi-source data, divides a time window of 1 minute based on the dyeing process stage, and binds the process parameters and sensor timing data within the window with the quality indicators and energy consumption data at the end of the window to form a four-dimensional data block of time-parameter-quality-energy consumption.

[0007] Hybrid modeling unit: used to establish a mapping relationship between process parameters and dyeing results based on historical data packets provided by the data interface unit; the hybrid modeling unit includes:

[0008] The mechanism constraint subunit is used to determine the dynamic feasible region of process parameters based on material properties and dyeing process principles; the data-driven prediction subunit constructs a quality prediction model based on historical data packets under the constraints of the dynamic feasible region.

[0009] The data-driven prediction subunit adopts a long short-term memory network model based on a multi-head attention mechanism. The attention weight module of this model is configured to quantify the impact of different process stages in historical data packets on the final product quality. The specific quantification dimensions include the parameter fluctuation range, duration, and adaptability to material properties of the process stage.

[0010] Parameter recommendation unit: For the current order data package, it generates multiple sets of process parameter recommendation schemes by combining the quality prediction model and multi-objective optimization strategy;

[0011] Feedback Learning Unit: By comparing the actual quality indicators of products produced based on the recommended process parameters with the prediction results of the quality prediction model, the source of deviation is analyzed, and the quality prediction model is updated and optimized online.

[0012] The feedback learning unit includes a deviation source analysis module, used to distinguish deviation types;

[0013] The types of deviations include:

[0014] Random noise: The deviation between the actual quality index and the predicted result is within the preset threshold and there is no continuous fluctuation trend;

[0015] Process execution fluctuations: Deviation values ​​exceed preset thresholds and are directly related to abnormal equipment status data;

[0016] Systemic concept drift: Deviation values ​​exceeding a preset threshold for three or more consecutive batches, and related to batch differences in material properties or sudden changes in environmental data;

[0017] If the deviation type is random noise, it triggers incremental learning of the quality prediction model; if the deviation type is process execution fluctuation, it triggers partial reconstruction of the quality prediction model; if the deviation type is systemic concept drift, it triggers an early warning signal and triggers a full model update after the problem is resolved.

[0018] Preferably, the dyeing process principle in the mechanism constraint subunit includes:

[0019] Based on the fiber type and dye category in the input material properties, the system calls a predefined physicochemical rule library to calculate and output the dynamic feasible domain of the process parameters.

[0020] Preferably, the physicochemical rule base is constructed based on the dye-fiber bonding kinetics theory, the dye liquor diffusion thermodynamic equation, and industry-verified process experience data.

[0021] Preferably, each rule in the physicochemical rule base is encapsulated as an independent calculation unit; the mechanism constraint subunit is configured to dynamically call and combine different rule calculation units according to different material properties through a preset fiber-dye type matching algorithm to generate a customized feasible domain for a specific production order.

[0022] Preferably, the data-driven prediction subunit adopts a long short-term memory network model based on a multi-head attention mechanism. The attention weight module of this model is configured to quantify the degree of influence of different process stages in historical data packets on the final product quality. The specific quantification dimensions include the parameter fluctuation range, duration, and adaptability to material properties of the process stage.

[0023] Preferably, the multi-objective optimization strategy in the parameter recommendation unit includes a quality-first mode, an energy-first mode, an efficiency-first mode, and a comprehensive optimization mode;

[0024] The parameter recommendation unit is configured with order attribute-optimization mode matching logic. By parsing the order attributes in the current order data packet, it automatically selects the optimization mode that matches the order priority. For the selected optimization mode, the weighting coefficient of each optimization objective is calculated using the analytic hierarchy process. Based on the weighting coefficient, multiple objectives are solved in a weighted manner to generate a process parameter recommendation scheme.

[0025] The technical effects and advantages of this invention are as follows:

[0026] 1. This invention achieves high-precision prediction of dyeing quality and accurate definition of the dynamic feasible domain of process parameters by deeply integrating multi-source heterogeneous data such as process parameters, material properties, sensor timing, environment and equipment status, and establishing a hybrid intelligent modeling method that combines mechanism constraints and data-driven approaches. This lays a scientific and reliable foundation for optimization recommendations.

[0027] 2. This invention introduces a multi-objective optimization strategy and an automatic order attribute matching mechanism, and combines the analytic hierarchy process (AHP) to quantify decision weights, thereby achieving flexible and customized process parameter recommendations based on order requirements. This effectively balances multiple production objectives such as energy consumption and efficiency while ensuring core quality.

[0028] 3. By constructing a feedback learning unit with deviation recognition and classification capabilities, and triggering adaptive update strategies such as incremental learning, local reconstruction, or full update, this invention achieves online continuous optimization and self-evolution of the quality prediction model, significantly improving the long-term applicability and stability of the system in dynamic production environments. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0030] Figure 2 This is a schematic diagram of the parameter recommendation unit of the present invention;

[0031] Figure 3 This is a schematic diagram of the feedback learning unit process of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] As attached Figure 1-3 The dyeing process parameter optimization and recommendation system based on big data analysis shown includes:

[0034] Data interface unit: used to collect and process multi-source heterogeneous data, and generate historical data packets and current order data packets; the multi-source heterogeneous data includes process parameters, material properties, sensor time series data, product quality indicators, environmental data and equipment status data.

[0035] It should be noted that the process parameters include dyeing temperature, holding time, pH value of dye liquor, dye concentration, liquor ratio, and stirring rate.

[0036] It should be further noted that the dyeing temperature curve parameter is labeled T, and is acquired using a PT100 platinum resistance temperature sensor with an accuracy of ±0.5℃. The sensor is installed inside the dyeing vat in the chemical circulation pipeline, and the data range is 50-130℃. The specific value can be adjusted according to the type of dye. The holding time parameter is labeled t. ins The PLC control system records the time from when the dye liquor temperature reaches the set value to when cooling begins, with a data range of 10-90 minutes. The pH value of the dye liquor is labeled as pH and is collected using an online pH electrode with a measurement range of 0-14 and an accuracy of ±0.05. The electrode needs to be immersed in the dye liquor in real time, with a data range of 1.0-14.0. The dye concentration parameter is labeled as C. dyeA UV-Vis spectrophotometer with a wavelength range of 400-800 nm was used to perform offline analysis on dye solution samples, with data ranging from 2-60 g / L; the liquor ratio parameter was labeled R. bath The dye liquor volume was calculated using a dyeing vat level sensor with an accuracy of ±1%, combined with fabric weighing results with an accuracy of ±0.1 kg. The data range was 1:5 to 1:20. The stirring rate parameter was labeled as v. stir Read the output frequency of the mixer motor inverter, and calculate the value based on the frequency-speed calibration curve with a calibration error of ±2r / min. The data range is 50-500r / min.

[0037] It should be further explained that outliers in process parameters are removed using the 3σ principle, with values ​​exceeding the mean ± 3σ. The results are then combined with production logs to determine whether the outlier is due to equipment failure. Continuous data is converted to the [0,1] interval using Min-Max standardization.

[0038] It should be noted that the material properties include fiber type, dye type, yarn count, fabric density, and pretreatment effect indicators.

[0039] It should be further noted that the fiber type parameter is identified as F. type Wavenumber 4000-400cm -1 The infrared spectrometer was used to detect the fiber composition of the fabric, and the composition was determined by matching a standard spectral library. The data range was represented by codes: 1 = cotton, 2 = polyester, 3 = wool, 4 = viscose; the dye category parameter was identified as D. type Based on the dye instructions provided by the supplier, and verified by high-performance liquid chromatography, the composition was determined. Data ranges are represented by codes: 1 = reactive, 2 = disperse, 3 = acidic; yarn count is identified as N. tex According to GB / T4743-2019 standard, yarn count was measured using a yarn count tester, and the average of three measurements was taken, with a data range of 10-50 tex; the fabric structure density parameter is identified as D. weave Using a fabric density microscope with 20x magnification, five points were counted at different locations on the fabric to determine the number of warp / weft yarns, with a data range of 20-120 yarns / 10cm. The pretreatment effect parameter was labeled as E. wick According to GB / T21655.1-2008 standard, the rise height of water on the fabric within 30 minutes is measured, with a data range of 5-30 cm / 30 minutes.

[0040] It should be further noted that missing material attribute values ​​must be re-collected and cannot be filled in; classification data is stored directly by number and does not require standardization.

[0041] It should be specifically noted that the sensor time-series data includes the time-series data output by the real-time temperature sensor, pH sensor, liquid level sensor and dye concentration sensor, with a collection frequency of 1 minute / time, and the collection duration is synchronized with the dyeing process cycle.

[0042] It should be further noted that the real-time temperature parameter is labeled T. s Temperature data was collected using a PT100 platinum resistance temperature sensor, identical to the dyeing temperature profile, with the acquisition frequency synchronized with the dyeing process cycle. The data range was 50-130℃. Real-time pH values ​​were labeled as pH. s The pH value was collected using an online pH electrode with the same pH value as the dye solution, at a frequency of 1 minute / time, with a data range of 1.0-14.0; the dye tank level parameter was labeled L. tank An ultrasonic level sensor with a measurement range of 0-2m and an accuracy of ±1mm is used to collect data. The sensor is installed on the top of the dyeing vat, and the data range is 0.5-1.8m. The real-time dye concentration parameter is labeled as follows: The data were collected using an online ultraviolet spectral sensor installed in the dye liquor circulation pipeline, with a collection frequency of 1 minute / time and a data range of 2-60 g / L.

[0043] It should be further explained that for sensor time-series data with a missing duration of ≤5 minutes, the average of the five adjacent data points is used to fill the gap; data with a missing duration of more than 5 minutes or outliers need to be marked, and the sensor operation log should be consulted to check whether it is a device failure; the data is stored in time series after standardization.

[0044] It should be specifically noted that the product quality indicators include color fastness, color difference value, fabric hand feel rating, and Lab* color space parameters and K / S dyeing depth values ​​collected periodically by an online spectrometer during the production process.

[0045] It should be further noted that the colorfastness parameter is labeled Q. rub According to GB / T3920-2008 standard, the fabric was tested using a rubbing fastness tester and rated by professionals. The data range was 1-5, with 5 being the best. The color difference value parameter was identified as ΔE, measured using a CIELab* system spectrophotometer with an accuracy of ±0.02ΔE, comparing the fabric with a standard sample. The data range was 0-5.0, with a value not exceeding 1.0 considered acceptable. The fabric hand feel rating parameter was identified as Q. handThe results were blind-tested by three professional evaluators according to the GB / T3819-1997 standard, and the average value was taken. The data range was 1-5, with 5 being the best. The Lab* color space parameters, labeled L*, a*, and b*, were collected using the same spectrophotometer used for color difference measurement. The L* luminance data range was 0-100, the a* red-green data range was -120 to 120, and the b* yellow-blue data range was -120 to 120. The K / S staining depth value parameter, labeled K / S, was calculated using a spectrophotometer according to the GB / T13655-2003 standard. K is the absorption coefficient, and S is the scattering coefficient, with a data range of 0-10. The larger the value, the darker the color.

[0046] It should be further explained that the color fastness is rated from 1 to 5, with 5 being the best; the color difference is rated from 0 to 5.0, with a value not exceeding 1.0 being acceptable; the hand feel rating is rated from 1 to 5, with 5 being the best; missing values ​​need to be retested; outliers should be judged whether to retain them based on the test records; and numerical indicators need to be standardized.

[0047] It should be specifically noted that the environmental data includes the hardness, pH value, conductivity, and temperature and humidity of the production water.

[0048] It should be further noted that the water hardness parameter is labeled as H. water According to GB / T6909-2018 standard, EDTA titration was used to test inlet water samples, with data ranging from 0-500 mg / L, calculated as CaCO3; the pH value parameter of the water was labeled as pH. water A laboratory pH meter with an accuracy of ±0.01 was used to test the inlet water sample, with data ranging from 6.0 to 8.5; the water conductivity parameter was denoted as σ. water A conductivity meter with a measurement range of 0-1000 μS / cm and an accuracy of ±1% was used to test the inlet water sample, with data ranging from 50-1000 μS / cm. The ambient temperature and humidity parameters were labeled as T. enu RH enu Temperature and humidity data are collected using sensors installed in the workshop away from heat sources. The temperature accuracy is ±0.5℃, and the data range is 15-35℃. The humidity accuracy is ±3%RH, and the data range is 30-80%RH.

[0049] It should be further explained that the environmental data processing method is continuous data standardization; outliers are investigated in conjunction with water quality reports and environmental monitoring records, and are removed after confirmation.

[0050] It should be noted that the equipment status data includes the operating frequency of the circulating pump, the valve opening degree, the actual power of the heating element, and the working pressure of the nozzle.

[0051] It should be further noted that the operating frequency parameter of the circulating pump is identified as f. pumpThe output frequency of the circulating pump inverter can be directly read, with a data range of 20-50Hz; the valve opening parameter is marked as O. value The output signal of the solenoid valve controller is read and converted according to the signal-opening calibration curve. The data range is 0-100%, where 0 is fully closed and 100 is fully open. The actual power parameter of the heating element is marked as P. heat A power analyzer with an accuracy of ±0.5% was used to measure the active power of the heating element power supply circuit, with a data range of 0-120kW; the nozzle working pressure parameter is labeled as P. nozzle The pressure is collected using a pressure sensor with a measurement range of 0-1 MPa and an accuracy of ±0.01 MPa. The sensor is installed at the nozzle inlet, and the data range is 0.1-0.5 MPa.

[0052] It should be further explained that outliers in equipment status data are determined in conjunction with equipment operation logs, and outliers caused by faults are removed; after data standardization, they are associated with the corresponding production batches for storage.

[0053] It should be further explained that historical data is collected from production data within the past 12 months, and stored in categories according to order number and production batch. Each batch of data must include the complete dyeing process cycle, from the fabric entering the dyeing vat to the fabric leaving the vat, and all indicators. For current order data, after receiving a new order, material attributes are collected first, and then environmental data and initial equipment status data are collected before production. During production, sensor timing data and process parameters are collected every minute. After production is completed, product quality indicators are collected.

[0054] It should be specifically noted that the data interface unit adds process-quality-energy consumption association tags to the multi-source data, divides the dyeing process into a 1-minute time window, and binds the process parameters and sensor timing data within the window with the quality indicators and energy consumption data at the end of the window to form a four-dimensional data block of time-parameter-quality-energy consumption.

[0055] Hybrid modeling unit: used to establish a mapping relationship between process parameters and dyeing results based on historical data packets provided by the data interface unit; the hybrid modeling unit includes:

[0056] The mechanism constraint subunit is used to determine the dynamic feasible region of process parameters based on material properties and dyeing process principles; the data-driven prediction subunit constructs a quality prediction model based on historical data packets under the constraints of the dynamic feasible region.

[0057] Parameter recommendation unit: For the current order data package, it generates multiple sets of process parameter recommendation schemes by combining the quality prediction model and multi-objective optimization strategy.

[0058] It should be specifically noted that the dyeing process principle in the mechanism constraint subunit includes: based on the fiber type and dye category in the input material properties, calling a predefined physicochemical rule library, calculating and outputting the dynamic feasible domain of the process parameters.

[0059] It should be further explained that the physicochemical rule base is built based on the dye-fiber bonding kinetics theory, the dye liquor diffusion thermodynamic equation, and industry experience. Each rule is encapsulated as an independent calculation unit, and the core formula is the dye uptake kinetic equation. Where E is the dye uptake rate, E eq To determine the equilibrium staining rate, k is the reaction rate constant, and t is the staining time.

[0060] It should be further explained that the dye uptake rate Where C0 is the initial concentration, C t The concentration at time t; the reaction rate constant. Where A is the pre-exponential factor, and in the disperse dye system, A = 1.2 * 10⁻⁶. 6 Through experiments, k was back-calculated at three different temperatures, and E was calculated. a Let R be the activation energy, R be the gas constant, and T be the absolute temperature. ,in The staining temperature is [value].

[0061] It should be further explained that the dynamic feasible domain is based on the material attributes (F) of the current order. type D type ), where F type For fiber type, D type For each dye category, the corresponding rule unit is invoked using a fiber-dye type matching algorithm. The specific steps include:

[0062] Step 1: Identify F type With D type ;

[0063] Step 2: Match the polyester-disperse dye calculation unit in the rule base and input the material properties;

[0064] Step 3: Set quality constraints and back-calculate the upper and lower limits of process parameters; for example, when the holding time t ins When the time is 30 min and E ≥ 85%, substituting into the kinetic equation yields k ≥ 0.065 min. -1 Substituting this into the Arrhenius formula, we get T ≥ 353 K. Combining this with the equipment's highest temperature, we determine the feasible temperature range. ;

[0065] Step 4: Repeat the above process to generate the dynamic feasible domain of all process parameters, and output the parameter-upper and lower limit list.

[0066] It should be noted that the data-driven prediction subunit adopts a long short-term memory network model based on a multi-head attention mechanism. Its attention weights are used to quantify the impact of different process stages in historical data packets on the final product quality. The specific quantification dimensions include the parameter fluctuation range, duration, and adaptability to material properties of the process stage.

[0067] It should be further explained that the model input layer consists of preprocessed feature vectors from historical data packets, with dimensions of (number of samples, time step, number of features). The time step is divided into heating, holding, and cooling stages according to the dyeing process, with 10 time points taken for each stage, 1 minute / point, for a total time step of 30. The number of features is 18, including 6 process parameters, 5 material properties, 4 sensor time-series data, and 3 environmental data. The output layer includes 3 core quality indicators: color difference ΔE, color fastness Q, and color accuracy Q. rub K / S value, with dimensions (number of samples, 3).

[0068] It should be further explained that the attention module quantifies the impact of different process stages, and the weight calculation formula is as follows: ,in Let be the attention weight at time step i, 0 ,all The sum is 1, ΔP i Let t be the parameter fluctuation amplitude at the i-th time step. i S represents the percentage of the duration of the process stage at time step i. i Let ω1, ω2, and ω3 be the parameters for the i-th time step – material fit. ω1, ω2, and ω3 are weight coefficients, each initially set to 1 / 3 and optimized through model training, ranging from 0 to 1. During model training, historical data packets are divided into training, validation, and test sets in a 7:2:1 ratio. The optimizer is Adam, with a learning rate of 0.001, a decay rate of 0.0001, and a loss function of mean squared error (MSE). The number of iterations is 100 rounds. Training stops when the MSE on the validation set shows no decrease for 5 consecutive rounds. The prediction errors on the test set must satisfy the following: ΔE prediction error ≤ 0.1, color fastness prediction error ≤ 0.2, and K / S prediction error ≤ 0.3. If these conditions are not met, the attention weight coefficients are readjusted and training resumes.

[0069] Parameter recommendation unit: For the current order data package, it generates multiple sets of process parameter recommendation schemes by combining the quality prediction model and multi-objective optimization strategy.

[0070] It should be noted that the multi-objective optimization strategies in the parameter recommendation unit include quality-first mode, energy-first mode, efficiency-first mode, and comprehensive optimization mode.

[0071] It should be further explained that the priority of the multi-objective optimization strategy is as follows: quality level > energy consumption requirement > production cycle; quality level is divided into high-end ΔE≤0.8, color fastness ≥4.5 grade, regular ΔE≤1.0, color fastness ≥4.0 grade, basic ΔE≤1.2, color fastness ≥3.5 grade; energy consumption requirement is divided into low energy consumption, energy consumption reduction ≥10% compared with historical orders of the same material, regular energy consumption, energy consumption fluctuation ≤±5%, and no energy consumption restriction; production cycle is divided into emergency, no more than 50 minutes, regular, 50-80 minutes, and relaxed, more than 80 minutes.

[0072] In the Quality First mode, when the quality requirement is high, there are no mandatory requirements for efficiency / energy consumption; in the Energy First mode, when the order requires low energy consumption, quality is sufficient; in the Efficiency First mode, when the order requires an urgent production cycle, quality is sufficient; in the Comprehensive Optimization mode, there is no clear priority, and quality, energy consumption, and efficiency must be balanced.

[0073] It should be specifically noted that the parameter recommendation unit is configured with order attribute-optimization mode matching logic. By parsing the order attributes in the current order data packet, it automatically selects the optimization mode that matches the order priority. For the selected optimization mode, the weighting coefficient of each optimization objective is calculated using the analytic hierarchy process. Based on the weighting coefficient, multiple objectives are solved in a weighted manner to generate a process parameter recommendation scheme.

[0074] It should be further explained that the weighting coefficients for each optimization objective were calculated using the analytic hierarchy process (AHP). First, for the selected optimization model, three dyeing process experts were invited to score the importance of the three objectives—quality (Q), energy consumption (E), and efficiency (P)—using a 1-9 scale, where 1 = equally important and 9 = extremely important, thus generating a judgment matrix. ,in Then, the largest eigenvalue of the judgment matrix is ​​found using the eigenvalue method. The corresponding eigenvectors, after normalization, become the weighting coefficients ω. Q ω E ω P The consistency check is satisfied.

[0075] It should be further explained that the process parameter recommendation scheme is generated by first constructing an optimization objective function, using weighted quality, energy consumption, and efficiency as objectives, and then searching for the optimal parameters within the dynamic feasible region. The objective function is: , where ΔE max E is the maximum permissible color difference value. act The actual energy consumption is determined by the heating power P. heat and heat preservation time t ins calculate, E max P represents the highest energy consumption for similar orders in history. act For actual efficiency, Pmax To maximize the efficiency of similar historical orders, a genetic algorithm was used for optimization, with parameters set as follows: population size 50, crossover probability 0.8, mutation probability 0.05, and number of iterations 50. Finally, five parameter schemes satisfying the objective function J≥0.85 were generated. Each scheme includes complete process parameters and predicted quality indicators, and is recommended in descending order of J value.

[0076] Feedback Learning Unit: By comparing the actual quality indicators of products produced based on the recommended process parameters with the prediction results of the quality prediction model, the source of deviation is analyzed, and the quality prediction model is updated and optimized online.

[0077] It should be noted that the feedback learning unit includes a deviation source analysis module, which is used to distinguish deviation types; deviation types include random noise, process execution fluctuations, and systemic concepts.

[0078] It should be further explained that deviation analysis first requires calculating the deviation value for each quality indicator. Q act Q represents the actual quality index value. pred For the model's predicted values, finally determine the type of bias:

[0079] Random noise: All ΔQ ≤ preset threshold, color difference threshold 0.1, color fastness threshold 0.2, K / S threshold 0.3, and no fluctuation in the same direction for 3 consecutive batches;

[0080] Process execution fluctuations: at least one ΔQ > threshold, and abnormal equipment status data;

[0081] Systematic concept drift: at least one ΔQ > threshold in three consecutive batches, and is related to material batch differences or environmental mutations.

[0082] Further explanation is needed: if incremental learning is triggered by random noise, the process parameters and actual quality data of the current order are added to the training set, accounting for 5%, and the output layer parameters of the model are retrained. The attention layer and LSTM layer are frozen, and the process is iterated for 10 rounds. If local reconstruction is triggered by process execution fluctuations, data from periods of abnormal equipment status are extracted, the weight parameters related to equipment status and quality in the model are adjusted, and the local network is retrained for 20 rounds. If a systemic concept drift triggers an early warning, an alarm is sent to the system backend to prompt staff to investigate material batches or environmental issues. After the problem is resolved, the normal data from the 10 most recent batches are added to the historical data package, and the hybrid modeling process in step two is re-executed to update the dynamic feasible region and the quality prediction model.

[0083] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0084] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for dyeing process parameter optimization and recommendation based on big data analysis, characterized in that, The application relates to a dyeing process parameter optimization system, which comprises the following units: a data interface unit for collecting and processing multi-source heterogeneous data to generate historical data packages and current order data packages; the multi-source heterogeneous data comprises process parameters, material attributes, sensor time series data, product quality indicators, environmental data and equipment state data; the data interface unit adds process-quality-energy consumption association tags to the multi-source data, divides a 1-minute time window based on a dyeing process stage, binds process parameters and sensor time series data in the window with quality indicators and energy consumption data at the end of the window, and forms a time-parameter-quality-energy consumption four-dimensional data block; a hybrid modeling unit for establishing a mapping relationship between process parameters and dyeing results according to the historical data packages provided by the data interface unit; the hybrid modeling unit comprises: a mechanism constraint subunit for determining a dynamic feasible region of process parameters according to material attributes and dyeing process principles; and a data-driven prediction subunit for constructing a quality prediction model according to the historical data packages under the constraint of the dynamic feasible region; the data-driven prediction subunit adopts a long short-term memory network model based on a multi-head attention mechanism, and an attention weight module of the model is configured to quantify the influence degree of different process stages in the historical data packages on the final product quality, and specific quantification dimensions include parameter fluctuation amplitude, duration and adaptation degree of the process stage to the material attributes; a parameter recommendation unit for generating multiple sets of process parameter recommendation schemes for the current order data package in combination with the quality prediction model and a multi-objective optimization strategy; a feedback learning unit for analyzing the deviation source by comparing the actual quality indicators of the products produced based on the process parameter recommendation schemes with the prediction results of the quality prediction model, and performing online updating and optimization on the quality prediction model; the feedback learning unit comprises a deviation source analysis module for distinguishing deviation types; the deviation types include: random noise: the deviation value between the actual quality indicators and the prediction results is within a preset threshold value, and there is no continuous fluctuation trend; process execution fluctuation: the deviation value exceeds the preset threshold value, and is directly related to abnormal equipment state data; systematic concept drift: the deviation value of three or more batches exceeds the preset threshold value, and is related to batch differences of material attributes or mutations of environmental data; if the deviation type is random noise, incremental learning of the quality prediction model is triggered; if the deviation type is process execution fluctuation, local reconstruction of the quality prediction model is triggered; and if the deviation type is systematic concept drift, an early warning signal is triggered, and full model updating is triggered after the problem is solved.

2. The big data analytics based dyeing process parameter optimization and recommendation system as claimed in claim 1 wherein: the process parameters include dyeing temperature, holding time, dye liquor pH value, dye concentration, bath ratio and stirring rate; the material attributes include fiber types, dye categories, yarn counts, fabric organization densities and pretreatment effect indicators; the sensor time series data includes time series data output by real-time temperature sensors, pH sensors, liquid level sensors and dye concentration sensors, the collection frequency is 1 minute / time, and the collection duration is synchronized with the dyeing process cycle. The product quality indicators include color fastness, color difference value, fabric hand rating, and Lab color space parameters and K / S dyeing depth values periodically collected by an online spectrometer during the production process; The environmental data include water quality hardness value, water quality pH value, water quality conductivity, and production environment temperature and humidity of the production water; The equipment state data include circulating pump operating frequency, valve opening, heating pipe actual power, and nozzle working pressure.

3. The big data analytics based dyeing process parameter optimization and recommendation system as claimed in claim 1 wherein: The dyeing process principles in the mechanism constraint subunit include: According to the fiber type and dye category in the input material attributes, a predefined physical and chemical rule library is called to calculate and output the dynamic feasible region of the process parameters.

4. The big data analytics based dyeing process parameter optimization and recommendation system as claimed in claim 3, wherein: The physical and chemical rule library is constructed based on dye-fiber binding kinetics theory, dye liquor diffusion thermodynamic equation, and industry-verified process empirical data.

5. The big data analytics based dyeing process parameter optimization and recommendation system as claimed in claim 3, wherein: Each rule in the physical and chemical rule library is encapsulated as an independent calculation unit; the mechanism constraint subunit is configured to be able to dynamically call and combine different rule calculation units through a preset fiber-dye type matching algorithm according to different material attributes, so as to generate a customized feasible region for a specific production order.

6. The big data analytics based dyeing process parameter optimization and recommendation system as claimed in claim 1 wherein: The multi-objective optimization strategy in the parameter recommendation unit includes quality priority mode, energy consumption priority mode, efficiency priority mode, and comprehensive optimization mode; The parameter recommendation unit is configured with order attribute-optimization mode matching logic, which automatically selects an optimization mode matching the order priority by analyzing the order attributes in the current order data packet, calculates the weighting coefficients of each optimization objective using the analytic hierarchy process for the selected optimization mode, and generates a process parameter recommendation scheme by weighted solving of the multi-objectives based on the weighting coefficients.

Images