Catalyst industry chain-oriented full-life-cycle tracing method and catalyst industry chain-oriented full-life-cycle tracing system

Abstract

The invention relates to the technical field of life cycle management, in particular to a catalyst industry chain-oriented full life cycle traceability method and system, and the method comprises the steps: obtaining the state attribute data of the same batch of recovered products, and carrying out the analysis to obtain an efficiency feature vector; inputting the efficiency feature vector into a preset quality evaluation model for matching, and identifying quality deviation data of the recycled product; the batch identity codes are analyzed, and resume data of the production full life cycle are obtained; a multi-dimensional data association model is constructed, quality deviation data is taken as a target variable, resume data is taken as an input feature, an influence weight coefficient and an attribution confidence index of each production control variable on the quality deviation are analyzed, and key control elements are screened out; and performing production regulation and control optimization operation based on the key control elements, calculating target parameter offset for correcting quality deviation, dynamically calculating a dynamic tolerance constraint range of a production control variable in combination with an attribution confidence index, and packaging into an execution management and control protocol.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of life cycle management, in particular to a full life cycle traceability method and system for a catalyst industry chain. BACKGROUND

[0002] As the core material of petroleum chemical industry, automobile exhaust treatment and new energy industry, catalysts cover multiple key links such as raw material preparation, precision manufacturing, industrial application and recycling in the industry chain. With the promotion of industrial digital transformation, the management mode of the catalyst industry chain is gradually transitioning from traditional physical management to digital full life cycle management. In the existing technical system, enterprise resource planning systems and manufacturing execution systems have been widely used in material control and parameter recording in the production process, and logistics tracking systems are mainly used to monitor the in-transit state and delivery progress of products. However, the existing catalyst industry chain management generally has the technical pain points of one-way data flow and cross-stage information island.

[0003] Current traceability technologies mainly focus on forward tracking, i.e. recording the flow of products from factory to user, mainly solving the problems of anti-counterfeiting and inventory management. Once the catalyst is put into use and eventually scrapped into the recycling link, the data chain of its full life cycle is often broken. As a reference for recording the performance of products under actual working conditions, waste catalysts contain a large amount of feedback data about product quality defects, endurance bottlenecks and potential hidden dangers of production processes. However, in the existing management process, the detection data at the recycling end is usually only used to evaluate the recycling value of residual precious metals, and cannot be digitally analyzed as a failure feature vector, nor can it be associated with production and manufacturing data a year ago or even earlier.

[0004] This lack of data closure leads to decision lag and inefficient resource allocation at the production management level. When unexpected failures occur in the same batch of products at the user end, the manufacturing end often lacks quantitative data-based attribution analysis means, making it difficult to accurately identify which production control variable fluctuation caused the quality deviation.

[0005] Therefore, a full life cycle traceability method and system for a catalyst industry chain are proposed. SUMMARY

[0006] The purpose of the present application is to provide a full life cycle traceability method and system for a catalyst industry chain, which performs production control optimization operations through key control elements, calculates target parameter offsets for correcting quality deviations, and dynamically calculates dynamic tolerance constraint ranges of production control variables in combination with attribution confidence indexes, encapsulates them as execution control protocols, and realizes full life cycle traceability of catalysts.

[0007] To achieve the above object, the present application provides a catalyst industry chain-oriented full life cycle traceability method, comprising: Obtaining state attribute data of the same batch of recycled products, analyzing and processing the state attribute data, and constructing a standardized performance feature vector; inputting the performance feature vector into a pre-set quality evaluation model for matching to identify the quality deviation data of the recycled products; Analyzing the batch identity code of the recycled products, obtaining the history data of the production full life cycle, and the history data covering the raw material batch information and the production control variables in the production process; A multi-dimensional data correlation model is constructed, the quality deviation data is used as the target variable, the history data is used as the input feature, the influence weight coefficient and the attribution confidence index of each production control variable on the quality deviation are analyzed, and the variable with the influence weight coefficient higher than the pre-set threshold is defined as the key control element; Based on the key control element, production control optimization operation is performed, the target parameter offset for correcting the quality deviation is calculated, and the dynamic tolerance constraint range of the production control variable is dynamically calculated combined with the attribution confidence index; the target parameter offset and the dynamic tolerance constraint range are encapsulated as an execution control protocol.

[0008] The state attribute data covers performance evaluation indicators in three aspects of microstructure, surface deposition and macro mechanics: The first evaluation index is the microstructure characterization index, including the specific surface area attenuation index, the pore volume distribution skewness index and the average pore size collapse index; The second evaluation index is the surface chemical deposition index, including the total amount index of carbon deposition on the catalyst surface, the carbon deposition graphitization degree index, and the deposition density index of exogenous toxic elements; The third evaluation index is the macro mechanical property index, including the average value index of particle lateral pressure strength, the wear index and the change rate index of bulk density.

[0009] The state attribute data is analyzed and processed to construct a standardized performance feature vector, including: Performing consistency verification of each state attribute data in the same batch of recycled products, identifying abnormal data points exceeding the compliance range by using a pre-set outlier determination logic; For the identified abnormal data points, data validity audit is performed combined with the sampling position label, if it is determined as a sampling error, it is excluded, if it is determined as a local working condition abnormality, it is retained and marked as an abnormal sub-sample, obtaining high-fidelity cleaned data; using a dimensionless mapping protocol, the cleaned data is mapped into a pre-set standardized evaluation interval to generate the performance feature vector of the recycled products.

[0010] The quality deviation data includes an abnormal type label and deviation feature data; the abnormal type label includes high-temperature sintering type failure, chemical poisoning type failure, pore channel blockage type failure and mechanical crushing type failure; the deviation feature data is a failure severity index, which is used to quantitatively describe the severity value of the failure type; The identification logic of the quality evaluation model is designed as follows: a business feature reference library containing various standard failure modes is constructed in advance, and the standard benchmark model corresponding to different failure types is stored in the library; in the matching process, the similarity evaluation rule is applied to calculate the matching correlation degree of the current input performance feature vector and the standard benchmark model in the business feature reference library; the abnormal type label and the deviation feature data are correspondingly output to obtain the quality deviation data.

[0011] The multi-dimensional data correlation model comprises: The reverse mapping alignment of the space-time data is performed; according to the logistics flow rule of the production line, the time lag amount of the product from the input of raw materials to the output of finished products is calculated, the batch identity code of the recovered product is accurately mapped back to the operation data slice when flowing through each production process, and the input feature matrix is established; The weight evaluation based on multi-dimensional attribution analysis is performed; the input feature matrix and the target variable are imported into the multi-dimensional data correlation analysis model for learning, the information contribution value of each production control variable in the model node splitting process is evaluated, and the influence weight coefficient is quantitatively obtained; The confidence evaluation based on repeated verification is performed; the data set is reconstructed and repeatedly analyzed multiple times by using a random sampling strategy, the frequency stability of the production control variable being identified as a key element in repeated evaluation is counted, and the frequency stability value is defined as an attribution confidence index.

[0012] The production regulation and optimization operation adopts an optimization design based on a reverse compensation strategy, comprising: Based on the data of the historical batches, an associated response model between the key control elements and the quality deviation data is fitted and constructed, representing the sensitivity relationship of the change of the production control variable to the quality deviation of the final product; A management objective function is defined in the associated response model, which aims to minimize the modulus of the quality deviation data; a multi-objective optimization strategy is used to search for an optimal control strategy within a feasible process window, calculate the numerical change amount of the key control elements that should occur to offset the currently identified quality deviation, and the numerical change amount is defined as a target parameter offset.

[0013] The calculation of the dynamic tolerance constraint range of the production control variable comprises: identifying the attribution confidence index to distinguish high confidence intervals, medium confidence intervals and low confidence intervals; based on the confidence interval where the attribution confidence index is located, the data interval of the dynamic tolerance constraint range is dynamically adjusted.

[0014] A catalyst industry chain-oriented whole life cycle traceability system comprises: A data acquisition module acquires state attribute data of the same batch of recycled products, analyzes and processes the state attribute data, constructs a standardized performance feature vector, inputs the performance feature vector into a preset quality evaluation model for matching, and identifies quality deviation data of the recycled products. A data analysis module analyzes batch identity codes of the recycled products, and acquires life cycle data, which covers raw material batch information and production control variables in the production process. An element identification module constructs a multi-dimensional data correlation model, takes the quality deviation data as a target variable, takes the life cycle data as an input feature, analyzes the influence weight coefficient and the attribution confidence index of each production control variable on the quality deviation, and screens variables with an influence weight coefficient higher than a preset threshold value as key control elements. A production control module performs production control optimization operation based on the key control elements, calculates a target parameter offset for correcting the quality deviation, dynamically calculates a dynamic tolerance constraint range of the production control variable in combination with the attribution confidence index, and encapsulates the target parameter offset and the dynamic tolerance constraint range as an execution control protocol.

[0015] Compared with the prior art, the beneficial effects of the present application are: 1. The present application quantitatively characterizes the failure state of recycled products in the microstructure, surface chemistry and macro-mechanical level by constructing a standardized performance feature vector; the reverse mapping alignment technology can calculate the hysteresis based on the logistics flow rules, accurately trace the batch code of the recycled products to the production operation slice one year ago or even earlier; the manufacturing end can optimize the process based on the whole life cycle performance of the product under real working conditions; not only the information silos are eliminated, but also the production decisions are based on objective quantitative data, realizing the closed-loop iteration of product quality management.

[0016] 2. The present application accurately identifies and removes logistics damage outliers caused by transportation bumps or dampness by monitoring the cumulative vibration energy spectrum density and environmental humidity integral value in the logistics process, ensuring that the data input for analysis accurately reflects the production quality; in the attribution analysis stage, the chemical mechanism knowledge graph is used to verify the path of the key elements screened by the statistical model; this double check of "physics + data" ensures the scientificity and uniqueness of the traceability result, avoiding the misattribution of logistics problems to production processes or misleading process adjustment due to statistical coincidence.

[0017] 3. The application proposes a flexible control strategy based on confidence level grading, which not only calculates the target parameter offset for correcting deviation, but also introduces a core dimension of attribution confidence index. According to the index, the pipe behavior is divided into different grades. The mechanism of converting statistical probability into management strength generates a machine-readable execution control protocol, which not only gives the production system the ability of self-repairing, but also retains the necessary fault tolerance flexibility through the dynamic tolerance mechanism, achieving the best balance between quality improvement and safe operation of production. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A process schematic diagram of a full life cycle traceability method for a catalyst industry chain according to the application is shown in the figure. Figure 2 A process schematic diagram of an execution control protocol according to the application is shown in the figure. Figure 3 A structure schematic diagram of a full life cycle traceability system for a catalyst industry chain according to the application is shown in the figure. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0020] Embodiment one: The application proposes a full life cycle traceability method for a catalyst industry chain. The process of the method is shown in the figure. Figure 1 The process of executing the control protocol is shown in the figure, which includes: Figure 2 Obtaining the state attribute data of the same batch of recycled products, analyzing and processing the state attribute data, and constructing a standardized performance feature vector; inputting the performance feature vector into a pre-set quality evaluation model for matching to identify the quality deviation data of the recycled products; Analyzing the batch identity code of the recycled products to obtain the history data of the full life cycle of production, which covers the batch information of raw materials and the production control variables in the production process; Constructing a multi-dimensional data correlation model, taking the quality deviation data as the target variable and the history data as the input feature, analyzing the influence weight coefficient and attribution confidence index of each production control variable on the quality deviation; screening out the variables with an influence weight coefficient higher than a pre-set threshold value, and defining them as key control elements.

[0021] ​Based on the key control elements, production control optimization operations are performed, target parameter offsets for correcting quality deviations are calculated, and dynamic tolerance constraint ranges of production control variables are dynamically calculated in combination with attribution confidence indexes; and the target parameter offsets and the dynamic tolerance constraint ranges are encapsulated as an execution control protocol.

[0022] The state attribute data covers performance evaluation indexes in three aspects of microstructure, surface deposition and macro mechanics: The first evaluation index is a microstructure characterization index, including a specific surface area attenuation index, a pore volume distribution skewness index and an average pore size collapse index. The second evaluation index is a surface chemical deposition index, including a total amount index of carbon deposition on the surface of the catalyst, a graphitization degree index of the carbon deposition, and a deposition density index of exogenous toxic elements. The third evaluation index is a macro mechanical performance index, including an average value index of particle side pressure strength, a wear index and a change rate index of bulk density.

[0023] The microstructure characterization index is obtained by nitrogen adsorption and desorption experiments on the recovered particles. The specific surface area attenuation index is used to quantify the thermal stability of the carrier skeleton, and is calculated by comparing the difference between the specific surface area of the recovered product and the specific surface area of the fresh product. First, the recovered catalyst particle sample is subjected to vacuum degassing treatment, and the specific surface area value is determined by using the gas adsorption principle; at the same time, the fresh catalyst specific surface area value of the batch product at the time of leaving the factory is called from the historical database; the difference between the fresh catalyst specific surface area value and the recovered catalyst specific surface area value is calculated, and the difference is divided by the fresh catalyst specific surface area value, and the quotient obtained is the specific surface area attenuation index. The larger the index value, the more serious the thermal collapse or sintering degree of the carrier skeleton.

[0024] The pore volume distribution skewness index: the pore size inside the catalyst usually presents a normal distribution; this index is used to measure whether the pore distribution has been non-uniformly distorted; based on the nitrogen adsorption and desorption isotherm, the pore size distribution calculation model is used to obtain the pore size distribution curve, which shows the volume proportion of pores of different sizes. The statistical skewness value of the pore size distribution curve is calculated, specifically, the difference between the volume distribution probability density of each pore size point and the average pore size is cubed, and the skewness value is obtained after standardization; if the index deviates significantly from zero, it indicates that the pore structure inside the catalyst has been non-uniformly collapsed or blocked at a specific pore size; if the skewness is too large, it indicates that the pores of a specific pore size have been concentratedly collapsed.

[0025] Average pore size collapse index: used to characterize the openness of the gas transport channels; the average pore size values of fresh catalyst and recovered catalyst are obtained respectively. The difference between the average pore size value of fresh catalyst and the average pore size value of recovered catalyst is calculated, and the difference is divided by the average pore size value of fresh catalyst. This index is used to quantify the degree of shrinkage of the gas transport channels.

[0026] Surface chemical deposition index: mainly measured by high-frequency infrared carbon-sulfur analyzer, laser Raman spectrometer and X-ray fluorescence spectrometer, used to evaluate the coverage and poisoning of active sites; Total carbon deposition index: characterizes the amount of carbon deposits covering the active sites; using high-frequency infrared carbon-sulfur analyzer, the catalyst sample is burned at high temperature in an oxygen-rich environment, and the amount of carbon dioxide gas released is detected to determine the carbon content in the sample. Read the mass percentage value of carbon element output by the instrument, the higher the value, the more coke deposits covering the active sites, the smaller the effective reaction area of the catalyst.

[0027] Carbon graphitization degree index: characterizes the hardness of carbon deposition, the higher the index, the closer the carbon deposition is to the graphite structure, the more difficult the regeneration. Use laser Raman spectrometer to scan the surface carbon of the catalyst, obtain the Raman spectrum. Identify the defect peak representing disordered carbon structure and the graphite peak representing graphite carbon structure in the spectrum; calculate the ratio of the intensity peak value of the graphite peak to the intensity peak value of the defect peak. The higher the ratio, the greater the hardness and density of the deposited carbon, indicating that it is more difficult to remove carbon by conventional regeneration methods, and the irreversibility of deactivation is stronger.

[0028] Exogenous poison deposition density index: focuses on detecting sulfur, phosphorus, arsenic and heavy metal elements, quantifying their aggregation density per unit surface area, and judging whether there is chemical poisoning. Use X-ray fluorescence spectrometer to scan the elements on the surface of the catalyst particles, focusing on detecting elements such as sulfur, phosphorus, arsenic and heavy metals that are not components of the catalyst itself. Add up the mass percentage values of the detected toxic elements, or calculate the ratio of the content of a specific toxic element to the specific surface area value of the catalyst. This index is used to represent the density of adsorbed poisons per unit surface area, and to quantify the severity of chemical poisoning.

[0029] Measurement of average value index of particle side pressure strength: randomly select a certain number (e.g. 50) of complete catalyst particles, and use the particle strength tester to apply radial pressure one by one until the particle breaks, and record the critical pressure value of each particle. The calculation logic is: calculate the arithmetic mean of all recorded critical pressure values to obtain the average value of side pressure strength. In order to construct the index, the percentage decrease of the average value relative to the strength of fresh catalyst can be further calculated.

[0030] Wear index calculation: A certain weight of catalyst sample is placed in a standard drum wear tester and rotated at a specified speed and time. After the test, the sample is sieved using a standard aperture screen, and the fine powder weight is collected and weighed. The calculation logic is to calculate the quotient of the fine powder weight divided by the total sample weight before the test. This value directly reflects the anti-pulverization ability of the particles under fluid scouring.

[0031] Change rate index of bulk density calculation: Use a standard volume cylinder to freely accumulate the catalyst sample to the scale line, weigh the sample in the cylinder, and calculate the weight per unit volume to obtain the bulk density. The calculation logic is to calculate the difference between the bulk density value of the recovered catalyst and the bulk density value of the fresh catalyst, and then divide by the bulk density value of the fresh catalyst. If the index is positive and large, it usually means that the particles have been broken, and the fine particles have filled the interstitial space between the particles, resulting in a more compact bulk, increasing fluid resistance.

[0032] The present application overcomes the limitations of traditional methods that only focus on a single physical indicator by establishing a three-dimensional index system. In particular, the introduction of deep indicators such as distribution skewness and graphitization degree for solid particle groups can more accurately capture potential quality risks caused by minor fluctuations in production processes, providing a high-quality data foundation for subsequent precise attribution.

[0033] The state attribute data is analyzed and processed to construct a standardized performance feature vector, including: Conduct consistency verification of each state attribute data in the same batch of recovered products, and use the preset outlier determination logic to identify abnormal data points that exceed the compliance range; For the identified abnormal data points, combined with the sampling location label, data validity audit is conducted, if it is determined to be a sampling error, it is excluded, if it is determined to be a local working condition anomaly, it is retained and marked as an abnormal sub-sample, to obtain high-fidelity cleaned data; using dimensionless mapping protocol, the cleaned data is mapped into the preset standardized evaluation interval, generating the performance feature vector of the recovered product.

[0034] Consistency verification process: Given that there are often noise points in industrial big data, consistency verification is first performed. For multiple sample data within the same batch, calculate the dispersion degree. Using the preset outlier determination logic, such as the statistical rule based on quartiles, automatically identify abnormal data points that are significantly deviated from the normal range.

[0035] Data validity audit process: Instead of blindly removing outliers, secondary audit is conducted in combination with sampling location labels; Specifically including: Due to the segregation of solid particles in the transport barrel, fine particles tend to deposit at the bottom; If the abnormal data point is from the sample at the bottom of the barrel, and shows a high wear index, it is determined that this is a normal physical sedimentation phenomenon, and is removed or corrected; If the abnormal data point is from the middle and upper part of the barrel, it is determined to be a local working condition anomaly, and is retained and marked as a special subsample to prevent loss of critical fault information.

[0036] Regarding the specific design of the outlier determination logic, an exemplary is described in detail below taking the specific example of the specific surface area attenuation index: First step: Calculate the quartile index; Based on the specific surface area attenuation index data of all samples in the same batch, arrange in ascending order, calculate the first quartile (Q1, the 25th percentile) and the third quartile (Q3, the 75th percentile), and then calculate the interquartile range IQR = Q3-Q1.

[0037] Second step: Determine the outlier determination boundary; Set the lower boundary of outlier determination as Q1-kxIQR, and the upper boundary as Q3+kxIQR, where the coefficient k is 1.5, which is based on industry experience values combined with actual surface area attenuation data to dynamically adjust to prevent excessive removal. If the specific surface area attenuation index value of a sample falls outside the boundary range, it is determined to be an outlier.

[0038] Third step: Multi-dimensional outlier comprehensive determination; For the case where multiple indicators are outliers at the same time, the logic AND rule is used, and only when a sample exceeds the boundary in two or more key indicators at the same time will it be listed as an outlier. Single-index outliers usually reflect differences in working conditions at different sampling locations and should not be directly removed.

[0039] Fourth step: For other indicators; such as pore size, carbon deposition, poisons, etc., the same quartile-IQR method is applied, but the coefficient k value can be adjusted according to the physical meaning of the indicator. For example, for wear index outliers caused by transportation vibration, k=2.0 can be set to improve fault tolerance; For failure severity outliers caused by production process abnormalities, k=1.2 should be set to improve sensitivity.

[0040] Dimensionless mapping process: Adopt dimensionless mapping protocols, such as the range standardization method, to map the cleaned multi-dimensional data uniformly into the standardized evaluation interval of 0 to 1, eliminating physical unit differences, and generating a standardized performance feature vector.

[0041] Further, the outlier determination logic further comprises a pre-processing elimination mechanism based on state-logistics two-dimensional verification, aiming to exclude physical loss interference caused by non-production factors; simultaneously acquiring the macro mechanical performance index of the batch of recycled products in the state attribute data, and the logistics vibration monitoring data in the transportation process; presetting the vibration intensity threshold and the wear abnormality threshold; when the peak acceleration in the logistics vibration monitoring data is monitored to exceed the vibration intensity threshold, and the wear index in the corresponding macro mechanical performance index simultaneously exceeds the wear abnormality threshold, correlation determination is performed; if it is determined that there is strong positive correlation between the two, the sample is defined as a logistics damage type outlier, and is eliminated from the input data set of the attribution analysis, and only the samples not subjected to significant transportation vibration interference are reserved for subsequent analysis.

[0042] The macro mechanical performance index detected at the recycling end is read, and the wear index therein is focused on; the three-axis acceleration sensor data of the transportation vehicle is read through the logistics interface to obtain the maximum vibration acceleration in the whole transportation process.

[0043] The preset vibration intensity threshold is 2.5G (an empirical value, and particles are easy to break above this value); and the preset wear abnormality threshold is 1.5% (normal wear is usually lower than 1%).

[0044] If the wear index of a batch is 0.8% (normal), or the wear index is high but the maximum vibration acceleration value in the transportation process is only 0.5G (the road condition is stable), it is determined that the wear is caused by production quality problems, and is reserved.

[0045] The wear index of a batch is as high as 2.0%, and the logistics data of the same period shows that a severe bump of 3.0G has occurred, such as sudden braking or falling. It is determined that the high wear is directly caused by transportation vibration and belongs to a logistics damage type outlier.

[0046] The present application effectively eliminates the noise interference of the logistics link by combining the transportation vibration condition, ensures that the wear data fed back to the production end are all real internal quality defects, thereby ensuring the accuracy of subsequent adjustment of production control variables and avoiding misoperation; especially for the friction and wear phenomenon unique to solid particles, calibration and elimination are realized, and the precision of data cleaning and the accuracy of attribution analysis are significantly improved.

[0047] The quality deviation data comprises an abnormal type label and deviation characteristic data; the abnormal type label comprises high-temperature sintering type failure, chemical poisoning type failure, pore blockage type failure and mechanical crushing type failure; and the deviation characteristic data is a failure severity index, which is used to quantitatively describe the severity value of the failure type; The identification logic of the quality evaluation model is designed to: pre-construct a business feature reference library containing various types of standard failure modes, and the library stores standard benchmark models corresponding to different failure types; in the matching process, the similarity evaluation rule is applied to calculate the matching correlation degree of the current input performance feature vector and the standard benchmark model in the business feature reference library; the abnormal type label and the deviation feature data are correspondingly output to obtain the quality deviation data.

[0048] The quality evaluation model in the embodiment adopts a center point matching architecture based on a feature space, which is based on a geometric space measurement logic with strong interpretability; the model contains two core components: Business feature reference library: a high-dimensional database storing standard benchmark models, each benchmark model representing a typical failure state recognized by the industry, specifically including high-temperature sintering failure, chemical poisoning failure, pore blockage failure, and mechanical crushing failure.

[0049] Similarity calculation engine: a vector operation unit based on Euclidean distance or cosine similarity, used to calculate the spatial position relationship between the input vector and the benchmark model.

[0050] The construction of the quality evaluation model is an iterative process based on historical data clustering; wherein, the standard benchmark model of the quality evaluation model is constructed based on the clustering method of the feature space. Each benchmark model is represented as a D-dimensional vector, where D is equal to the dimension of the standardized performance feature vector, and in the embodiment, D=9, corresponding to 3 microstructure indicators, 3 surface chemical indicators, and 3 mechanical performance indicators. The basic composition of the benchmark model is: the center point of the benchmark feature vector and the spatial distribution range of its multi-dimensional features. The model uses the K-means clustering algorithm for iterative optimization, and the clustering number K is initially set to 4, corresponding to four standard failure types. When the ratio of the sum of intra-class distance squares to inter-class distance converges, the iteration stops. The clustering convergence condition is defined as: the sum of Euclidean distances of clustering center movement in two consecutive iterations is less than 1×10 -6 , or the maximum number of iterations is reached, which is set to 500.

[0051] Step one (sample accumulation): collect the state attribute data of the past accumulated spent catalyst batches and their corresponding expert determination results, such as artificially determined failure types.

[0052] Step two (feature space mapping): convert these historical data into standardized performance feature vectors and map them into a multi-dimensional feature space.

[0053] Step three (cluster center extraction): using the logic of K-means clustering algorithm, the dense sample points in the space are clustered into clusters. For example, all samples judged by artificial as high-temperature sintering failure are clustered into a class, and the arithmetic mean of all feature vectors of the samples in the class is calculated. The average vector is defined as the standard feature vector center point of high-temperature sintering failure.

[0054] Step four (reference model solidification): each center point calculated above is solidified and stored in the service feature reference library as a benchmark for subsequent matching.

[0055] Specifically, when a new recycling batch performance feature vector is received, the following identification logic is executed: Distance measurement: the Euclidean distance between the input vector and each standard feature vector center point in the reference library is calculated respectively; the specific calculation method is: the difference between the numerical value of each dimension of the input vector and the numerical value of the corresponding dimension of the center point vector is calculated, the square sum of the difference is calculated, and finally the square root of the sum is calculated.

[0056] Correlation degree calculation: for each failure dimension, the spatial distance between the input vector and the reference model is calculated respectively by applying similarity evaluation rules such as Euclidean distance calculation.

[0057] Severity reverse mapping: the calculated spatial distance is converted into a failure severity index. Reverse mapping logic or normalization function is used; the smaller the distance value, the more typical the feature in that dimension, and the higher the converted severity index; the farther the distance, the lower the severity index.

[0058] The quality deviation data is constructed as a multi-dimensional feature vector, and each dimension represents the severity of a specific failure mode; if the catalyst quality is perfect and there is no deviation, the vector should be a zero vector in theory.

[0059] Further, the history data also covers environmental logistics factors of the product in the storage and transportation stage; the environmental logistics factors include cumulative vibration energy spectrum density in the transportation process, environmental humidity integral value and temperature change cycle number; when constructing a multi-dimensional data correlation model, the environmental logistics factors and production control variables are used as input features together to identify non-productive quality deviations caused by improper logistics links.

[0060] Cumulative vibration energy spectrum density: used to quantify the total cumulative fatigue damage of the catalyst particles caused by the vibration energy in a specific frequency range, i.e. the mechanical resonance sensitive frequency band of the catalyst particles, during transportation; this parameter is different from the ordinary average acceleration, and focuses on evaluating the energy accumulation that causes resonance breakage.

[0061] Calculation method: First, collect the three-axis acceleration data during transportation at a set high sampling rate; second, perform fast Fourier transform on the acceleration data to convert it into a frequency domain signal to obtain the power spectral density; then, only the power spectral density values in the catalyst vulnerable sensitive frequency band are intercepted for frequency domain integration to obtain the instantaneous vibration intensity (square of the root mean square acceleration); finally, the instantaneous vibration intensity is time-integrated on the entire transportation time axis to calculate the cumulative vibration fatigue damage index of the whole process.

[0062] Environmental humidity integral value: used to represent the cumulative damage effect of water molecules invading the pores through capillary condensation during the storage and transportation of the catalyst under high humidity environment; this parameter not only considers the peak value of humidity, but also focuses on the cumulative effect of high humidity duration and temperature synergy.

[0063] Calculation method: First, set a critical moisture absorption threshold according to the physical properties of the catalyst material, such as a relative humidity of 65%; second, continuously collect real-time relative humidity data and temperature data; third, select the time period when the relative humidity value exceeds the critical moisture absorption threshold; finally, integrate the humidity difference value that exceeds the threshold with respect to time, and introduce the diffusion coefficient based on real-time temperature as a correction factor during the calculation process, to obtain the final cumulative integral value.

[0064] Temperature change cycle number: used to represent the frequency of temperature fluctuations (i.e. thermal shock), reflecting the number of microscopic thermal stress shocks caused by the difference in thermal expansion coefficients between the catalyst carrier and the active component.

[0065] Calculation method: First, collect real-time temperature data during the entire transportation period and calculate the derivative of temperature with respect to time, i.e. the temperature change rate; second, set a thermal shock threshold (e.g. temperature change of more than ±5 degrees Celsius per minute); finally, apply the rainflow counting method to count the number of complete hysteresis loops whose temperature change rate exceeds the above thermal shock threshold, which is the number of temperature change cycles.

[0066] By reading the Internet of Things sensor tag data attached to the catalyst packaging barrel, the accelerometer and hygrometer records during transportation are obtained. In the attribution analysis, it is found that the mechanical crushing type failure of a batch is not related to the production parameters, but has a high correlation of 0.9 with the cumulative vibration energy spectrum density of the transportation segment, because it passed through a bumpy road section and the packaging buffer was insufficient, so the output control protocol is not to adjust the production pressure, but to issue a logistics instruction to upgrade the packaging buffer material.

[0067] By introducing the cumulative vibration energy spectrum density, the environmental humidity integral value and the number of temperature change cycles and other refined parameters, the hole collapse caused by the production process and the hole collapse caused by moisture during transportation can be distinguished; The physical crushing at the recycling end is often mistakenly attributed to the insufficient binder strength at the production end, leading to incorrect process adjustment, such as excessive addition of binder leading to reduced activity, and the present application can effectively eliminate such non-productive noise, protecting the stability of the production process; The definition of the whole life cycle is improved, preventing the quality problems caused by logistics transportation from being mistakenly attributed to the production and manufacturing link, avoiding the misadjustment of the production process, and improving the responsibility definition accuracy of traceability.

[0068] The multi-dimensional data correlation model comprises: performing reverse mapping alignment of space-time data; according to the logistics flow rule of the production line, calculating the time lag amount of the product from the input of raw materials to the output of finished products, accurately mapping the batch identity code of the recycled product back to the operation data slice when flowing through each production process, and establishing an input feature matrix; Performing weight evaluation based on multi-dimensional attribution analysis; importing the input feature matrix and the target variable into the multi-dimensional data correlation analysis model for learning, and quantitatively obtaining the influence weight coefficient by evaluating the information contribution value of each production control variable in the model node splitting process; Performing confidence evaluation based on repeated verification; using a random sampling strategy to reconstruct and repeatedly analyze the data set multiple times, and statistically analyzing the frequency stability of the production control variables being identified as key elements in repeated evaluation, and defining the frequency stability value as the attribution confidence index.

[0069] The reverse mapping alignment of space-time data is a key step in data preparation before model training, aiming to eliminate the time misalignment between production and recycling; Lag time calculation: for each production process, such as mixing, extrusion, drying, and calcination, according to the time distribution of the historical production process, the average residence time of the material in the process is calculated, i.e. the time lag amount; Slice mapping: analyze the batch identity code of the recycled product to obtain its output time point; then, subtract the residence time of each subsequent process in turn, and reversely deduce the specific time stamp of the batch material at each time, such as raw material mixing, extrusion molding, drying, and calcination; Feature matrix establishment: grab the process parameters corresponding to the above specific time stamp, such as temperature, pressure, and flow, to establish an accurate micro-space-time correspondence relationship and form an input feature matrix; The gradient boosting decision tree (GBDT) architecture based on ensemble learning is adopted, which is composed of decision trees and can handle nonlinear relationships and automatically evaluate the importance of features.

[0070] Training process: The input feature matrix (production control variables) is imported as input, and the quality deviation data (such as the severity value of each failure mode) is imported as the target variable to learn the model.

[0071] Weight calculation: During the model training process, whenever a production control variable is selected as a decision tree splitting node, the reduction of squared error or the improvement of purity brought by the splitting is calculated as the information contribution value. The improvement brought by this variable in all decision trees is accumulated and divided by the sum of all variable improvement values to quantify the influence weight coefficient of the variable, which reflects the contribution degree of the production control variable to the final quality deviation.

[0072] Confidence evaluation based on repeated verification: In order to ensure the reliability of the attribution result, confidence evaluation based on repeated verification is performed, using Bootstrap resampling strategy.

[0073] Resampling setting: Set the resampling number to 100 times, randomly sample 80% of the samples from the original data set with replacement each time, and construct a new training subset.

[0074] Repeated training: Train 100 independent correlation models on these 100 subsets respectively.

[0075] Frequency statistics: For the variables identified as key control elements, statistics of how many times their ranking exceeds the preset threshold in the 100 training are counted.

[0076] Index definition: The frequency stability of production control variables identified as key elements in repeated evaluation is counted. For example, if a variable ranks in the top 3 in 95 training, its frequency stability value is high, and it is defined as the attribution confidence index, such as 95%. This indicates that the causal relationship is extremely stable and not accidental noise.

[0077] The present application solves the problems of space-time misalignment and pseudo-correlation commonly encountered in industrial big data analysis through time alignment at the physical level and repeated verification at the statistical level; ensures that the key control elements found are not only mathematically related, but also physically real causal, providing a solid scientific basis for subsequent process adjustment.

[0078] Further, for complex production processes, since most catalyst production involves mixing, kneading, extrusion or precipitation processes, these processes have a clear residence time distribution in chemical engineering principles. The material in the reactor or mixer is not ideal piston flow, but there is back mixing. This means that the finished product particles produced at a certain time are actually the result of the mixing of raw materials and process parameters input over a period of time (may be several minutes or even several hours) in the past. It may be difficult to accurately reflect the production process using data slicing.

[0079] Therefore, the construction of the input feature matrix can also be designed as follows: according to the logistics flow rule of the production line and the average residence time distribution of each process, the time lag interval of the product from the raw material input to the finished product output is calculated, the batch identity code of the recycled product is mapped back to the time period weighted operation data when flowing through each production process, and the input feature matrix is established; that is, the relatively stable operation process is regarded as a data point; for example, the production control variables of the mixing process are clustered to identify several relatively stable time periods as data points.

[0080] Further, the multi-dimensional data correlation model can also fuse a pre-constructed chemical mechanism knowledge graph; the chemical mechanism knowledge graph includes entity nodes representing production control variables, result nodes representing quality deviation data, and relationship edges representing chemical reaction principles; when calculating the influence weight coefficient, a mechanism path verification step is performed: using a graph traversal algorithm to search for a connected path from a high-weight production control variable node to a quality deviation data node in the chemical mechanism knowledge graph; According to the search results of the connected path, a weight correction strategy is performed: If there is no connected path, it is determined that the statistical correlation between the production control variable and the quality deviation data is pseudo-correlation, and a damping penalty factor is introduced to attenuate the influence weight coefficient; If there is a connected path, it is determined to be a mechanism strong correlation, and a gain enhancement factor is introduced to enhance the attribution confidence index.

[0081] The chemical mechanism knowledge graph is defined as a structured graph database (such as constructed based on Neo4j) for storing expert knowledge and chemical reaction principles in the field of catalyst production.

[0082] The node design includes parameter entities, intermediate state entities, and result entities. Parameter entities: including specific process parameters, such as calcination section temperature, nitric acid solution concentration, kneader speed, etc.; intermediate state entities: including micro physical and chemical changes, such as grain growth, pore collapse, active component migration, and carbon precursor generation; result entities: quality deviation data, such as specific surface area attenuation, poisoning type failure, and insufficient side pressure strength.

[0083] The relationship edge design includes: defining the causal or promoting relationship between nodes, for example: high calcination temperature leads to grain growth, and grain growth leads to specific surface area attenuation; each edge can be attached with attributes, such as threshold conditions for reaction occurrence, for example, temperature > 500°C activates this path; when the multi-dimensional data correlation model identifies that a variable X has high statistical correlation with a quality deviation Y, graph verification is started with X as the starting point and Y as the terminal point, and a depth-first search algorithm is applied to search for one or more connected paths in the knowledge graph; if a path is found, it is determined that there is mechanism support; Pseudo-correlation rejection, i.e., damping penalty mechanism: when a non-process type characteristic variable (such as a batch produced on Tuesday afternoon) is monitored to have a strong statistical correlation with a product quality deviation (such as high product wear rate), but in the chemical mechanism knowledge graph, the timestamp node cannot be connected to the wear rate node through any known chemical reaction path or physical action mechanism; the operation performed: determine that the statistical correlation belongs to a statistical coincidence without any physical and chemical mechanism support, i.e., pseudo-correlation, introduce a damping penalty factor to attenuate the original influence weight of the variable, which is specifically manifested as reducing the weight value of the variable by a predetermined proportion, so that the modified weight value is significantly reduced; after this operation, the priority of the variable is degraded, and it is no longer included as a key control element in the subsequent management and control system.

[0084] Mechanism confirmation, i.e., gain enhancement mechanism: when it is found that a certain process parameter (for example, steam partial pressure) has a correlation with a specific quality defect (for example, framework dealumination), and a clear connection path is retrieved in the chemical mechanism knowledge graph (for example, the path is shown as: the increase of steam partial pressure leads to the intensification of hydrothermal aging, and then induces the phenomenon of framework dealumination); the operation performed: determine that the correlation is a real physical and chemical causal relationship, introduce a gain enhancement factor to amplify the original influence weight of the variable, which is specifically manifested as increasing the weight value of the variable by a predetermined amplification multiple, thereby significantly improving the confidence value of the attribution result, and ensuring that the key control element is preferentially encapsulated into the execution control protocol.

[0085] In complex chemical production, there are a large number of accidental statistical coincidences, and a pure statistical model is prone to false positives. The present application uses graph reasoning as a filter to automatically identify and reject pseudo-correlated variables that lack physical and chemical mechanism support, preventing the issuance of incorrect process adjustment instructions and avoiding misoperation of the production line. When the sample data is small or there are noise points, the statistical model is prone to distortion. At this time, the prior expert knowledge provided by the knowledge graph can be used as a stable reference system to strengthen the weight of the correct path through the gain enhancement factor, ensuring that correct traceability judgments consistent with chemical principles can be made under small sample conditions.

[0086] The production regulation and optimization operation adopts an optimization design based on a reverse compensation strategy, including: Based on the data of historical batches, an associated response model between the key control elements and the quality deviation data is fitted and constructed to represent the sensitivity relationship of the change of the production control variable to the final product quality deviation; A management objective function is defined in the associated response model, which aims to minimize the modulus of the quality deviation data; a multi-objective optimization strategy is used to search for the optimal control strategy within the feasible process window, and the numerical change amount that should occur in the key control elements to offset the currently identified quality deviation is calculated, which is defined as the target parameter offset.

[0087] Based on the big data of historical batches, a second-order response surface regression method is used to fit and construct the associated response model between the key control elements and the quality deviation data; the model is a polynomial regression model, which not only contains linear terms (variables themselves), but also contains quadratic terms (squares of variables) and interaction terms (products of variables), which represents how the small changes of production control variables will lead to changes in the final product quality deviation, that is, the sensitivity relationship.

[0088] Management objective function definition: define a management objective function, the mathematical meaning of which aims to minimize the modulus of the quality deviation data. The quality deviation data is a multi-dimensional vector, and the modulus represents the comprehensive distance of the vector from the perfect zero point in the multi-dimensional space, that is, the comprehensive quality loss value, which aims to find a set of production parameters to minimize the output value of the function.

[0089] Multi-objective optimization strategy: sequential quadratic programming algorithm or genetic algorithm is used for optimization.

[0090] Constraint condition: the optimization process is strictly limited within the feasible process window. For example, the upper limit of the search of the baking temperature cannot exceed the highest temperature that the equipment can tolerate, and the lower limit cannot be lower than the starting temperature of the reaction.

[0091] Iterative search: the algorithm continuously tries to adjust the numerical value of the key control elements within the constraint range, inputs the response model to predict the deviation modulus, until a parameter combination that minimizes the modulus is found.

[0092] Offset calculation: calculate the difference between the optimal parameter combination and the parameter setting value currently used on the production line, calculate the numerical change that the key control elements should undergo in order to offset the current identified quality deviation, and define the target parameter offset.

[0093] The present application realizes the leap from qualitative adjustment to quantitative and accurate compensation, no longer relying on manual experience to estimate the adjustment range, but accurately calculating the optimal solution that can minimize the quality loss through mathematical model; making the adjustment of production process more scientific and accurate, which can effectively suppress the recurrence of similar failure modes in the next batch production.

[0094] The calculation of the dynamic tolerance constraint range of the production control variable includes: identifying the attribution confidence index, distinguishing the high confidence interval, the medium confidence interval and the low confidence interval; based on the confidence interval where the attribution confidence index is located, dynamically adjusting the data interval of the dynamic tolerance constraint range.

[0095] According to the output attribution confidence index, the dynamic tolerance constraint range of the production control variable is dynamically adjusted.

[0096] High confidence interval (e.g. greater than or equal to 90%): extremely high confidence, must be strongly corrected. Tolerance calculation: dynamic tolerance constraint range narrowed to the limit of equipment capability or a smaller multiple of historical standard tolerance (e.g. 30% of standard tolerance). This means that the production equipment must be extremely precise to perform the new parameters.

[0097] Medium confidence interval (e.g. 60% to 90%): strong suspicion of abnormality, but with uncertainty, leaving moderate flexibility. Tolerance calculation: dynamic tolerance constraint range maintains historical standard tolerance (e.g. 100% of standard tolerance).

[0098] Low confidence interval (e.g. less than or equal to 60%): insufficient evidence, avoid excessive intervention to cause production shock. Tolerance calculation: dynamic tolerance constraint range is expanded to a larger multiple of historical standard tolerance (e.g. 1.2 to 1.5 times), or only recommended values are issued without locking modification authority.

[0099] Finally, the calculated target parameter offset is combined with the dynamic tolerance constraint range to generate a machine-readable execution control protocol. The protocol is directly issued to the production manufacturing system through a data interface, automatically updating the control logic of the production workflow.

[0100] The present application converts statistical confidence into management control intensity, establishing an adaptive risk management mechanism; it can both solve problems with lightning speed when the cause is clear, and maintain production stably when the cause is in doubt, achieving the best balance between quality improvement and production safety, and minimizing the risk of misoperation caused by data noise.

[0101] Embodiment two: The present application proposes a full life cycle traceability system for the catalyst industry chain, the structure of the system is as shown in Figure 3 The system includes a data acquisition module, a data analysis module, an element identification module, and a production control module. It is deployed on a cloud server and interconnected with a laboratory management system, a manufacturing execution system, and a logistics tracking system through a data interface.

[0102] The data acquisition module acquires state attribute data of the same batch of recycled products, analyzes and processes the state attribute data, and constructs a standardized performance feature vector. The performance feature vector is input into a pre-set quality evaluation model for matching to identify the quality deviation data of the recycled products. This module is responsible for interpreting the current situation from waste, and is the starting point of the entire traceability process.

[0103] Multi-dimensional index collection: When the recycled catalyst enters the processing center, the data acquisition module first retrieves the multi-dimensional state attribute data of the batch from the laboratory management system. The collected data covers three levels: Microstructure characterization indices, including specific surface area decay index, pore volume distribution skewness index, and average pore size collapse index; Surface chemical deposition indices, including total carbon deposition on the catalyst surface index, carbon graphitization degree index, and exogenous toxic element deposition density index; Macroscopic mechanical performance indices, including average particle side pressure strength index, wear index, and bulk density change rate index.

[0104] Data cleaning and feature construction: The module performs consistency checking and finds that the wear index of two samples is abnormally high. The module automatically calls the sampling location label and logistics data for auditing and finds that the two samples are taken from the bottom of the packaging barrel and no severe vibration is recorded during transportation. Therefore, it is determined that the local working condition is abnormal rather than sampling error, and it is retained and marked as an abnormal sub-sample. Subsequently, the range standardization method is used to map all cleaned data to the interval of 0 to 1, and a standardized performance feature vector is constructed.

[0105] Quality deviation identification: The above performance feature vector is input into the pre-set quality evaluation model. The model internally stores a business feature reference library, including standard benchmark models of high-temperature sintering, chemical poisoning, etc.

[0106] Matching process: By calculating the Euclidean distance between the input vector and each benchmark model, it is found that the batch characteristics are closest to the "high-temperature sintering failure" benchmark model, with a high degree of similarity; wherein the abnormal type label includes high-temperature sintering failure, chemical poisoning failure, pore blockage failure, and mechanical crushing failure.

[0107] Output result: The module identifies quality deviation data, which contains the abnormal type label of high-temperature sintering failure, and the deviation feature data is 0.85, indicating that the sintering condition is serious.

[0108] Data analysis module, analyzing the batch identity code of the recycled product, obtaining the life cycle data of the production, which covers the batch information of raw materials and production control variables in the production process; This module is responsible for cross-time and space data alignment, establishing a connection between the failure at the recycling end and the parameters at the production end.

[0109] Life cycle history acquisition: The module analyzes the batch identity code of the recycled product and traces back to its production record. The obtained life cycle data includes batch information of raw materials, such as alumina carrier origin, and production control variables in the production process, such as temperature, pressure, and feed rate in each temperature zone.

[0110] Time-space reverse mapping alignment: In order to solve the hysteresis problem in the production process, the module performs reverse mapping alignment operation.

[0111] Lag calculation: According to the logistics flow rule of the production line, the system calculates the average time lag from the granulation process to the finished product packaging, which is four hours.

[0112] Precise slicing: The module deduces the output time point of the recycled product forward, accurately locates the specific time window when the batch material flows through the three-stage roasting furnace, and captures the operation data slice in the window, such as three-stage roasting temperature, heating rate, and ventilation volume, establishing an input feature matrix.

[0113] Factor identification module, build a multi-dimensional data correlation model, take the quality deviation data as the target variable, take the history data as the input feature, analyze the influence weight coefficient and attribution confidence index of each production control variable on the quality deviation; Screen out variables with influence weight coefficient higher than the preset threshold, defined as key control factors; Multi-dimensional correlation analysis: The module constructs a multi-dimensional data correlation model, taking the quality deviation data as the target variable and the above input feature matrix as the input feature; The composition and feature engineering of the exemplary input feature matrix include: The production control variable feature set includes but is not limited to the following, a total of 18 numerical features; Mixing process: mixing time, mixing speed, mixing temperature; Extrusion process: extrusion pressure, extrusion temperature, extrusion yield; Drying process: drying temperature, drying time, drying humidity; Roasting process: one-stage roasting temperature, two-stage roasting temperature, three-stage roasting temperature, heating rate, ventilation volume; Cooling process: cooling rate, cooling final temperature.

[0114] For raw material batch information, such as alumina origin, use one-hot encoding to convert to categorical features; If the number of original suppliers is 5, generate 5 binary features, where the product of a certain supplier corresponds to the feature value 1, and the rest are 0.

[0115] The environmental logistics factor feature set includes: cumulative vibration energy spectrum density, environmental humidity integral value, and temperature change cycle number, a total of 3 features.

[0116] After space-time mapping, the final input feature matrix dimension is (sample number x 26 feature dimensions).

[0117] Multi-output design of target variable: The quality deviation data is a 4-dimensional vector corresponding to the severity index of four failure modes [high temperature sintering, chemical poisoning, pore blockage, mechanical crushing].

[0118] The multi-output GBDT model architecture is adopted, which contains four independent decision tree sets that share the integrated learning hyperparameters, learning the relationship between each failure severity and production control variable. This design not only preserves the indirect association between different failure modes, but also allows for fine-grained learning of specific features for each failure mode.

[0119] GBDT model hyperparameter settings: decision tree maximum depth: max_depth = 5, to prevent overfitting while retaining sufficient nonlinear fitting capability; learning rate: learning_rate = 0.1; subsample ratio: subsample = 0.8, randomly sample 80% of the samples each iteration; feature subsample ratio: colsample_bytree = 0.8, randomly sample 80% of the features each split; minimum leaf sample size: min_child_weight = 5, to prevent overfitting; L1 regularization parameter: reg_alpha = 0.1; L2 regularization parameter: reg_lambda = 1.0; number of iterations: use early stopping strategy, monitor the root mean square error on the validation set; stop training when there is no improvement in the validation set RMSE for 20 consecutive iterations, with a maximum of 500 iterations.

[0120] Division of training data and multicollinearity processing, the historical sample set is divided into three parts in chronological order; Training set: the earliest 70% of the samples, used for model training; Validation set: the middle 15% of the samples, used for parameter tuning and early stopping determination; Test set: the latest 15% of the samples, used for independent evaluation of model generalization performance.

[0121] In the feature engineering stage, the Pearson correlation coefficient matrix between all input features is calculated. For feature pairs with an absolute correlation coefficient greater than 0.9, strong multicollinearity is determined, and one of the features is deleted, retaining the one with stronger correlation with the target variable.

[0122] After GBDT training, the contribution value of each feature in the model is quantified by the loss reduction amount brought by the feature during iteration, and the contribution values of all features are normalized to obtain the influence weight coefficient.

[0123] Training and weight calculation: use the gradient boosting decision tree algorithm for training to evaluate the information contribution value of each variable in node splitting. The results show that the influence weight coefficient of the average temperature of the three-stage roasting furnace is the highest; Mechanism verification and confidence evaluation: to prevent statistical misjudgment, the module performs double verification; Mechanism path verification: call the chemical mechanism knowledge graph, search and find that there is a clear connection path between the too high roasting temperature and the specific surface area attenuation (sintering), confirming that it is a strong mechanism correlation, and giving a gain enhancement factor to increase its weight; Repeated verification: 100 times of resampling analysis are performed on the data by using a random sampling strategy; the results show that the average temperature of the three-stage roasting furnace is identified as a key element in 95 analyses, and therefore the attribution confidence index is determined to be 95%; Key element locking: finally, the module screens out variables with an impact weight coefficient higher than a preset threshold, and formally defines the average temperature of the three-stage roasting furnace as the key control element leading to the high-temperature sintering failure.

[0124] Production control module, based on the key control element, performs production control optimization operation, calculates the target parameter offset for correcting the quality deviation, and dynamically calculates the dynamic tolerance constraint range of the production control variable combined with the attribution confidence index; the target parameter offset and the dynamic tolerance constraint range are packaged as an execution control protocol.

[0125] This module is responsible for generating a strategy closed-loop control and converting the analysis results into machine executable instructions.

[0126] Reverse compensation optimization: the module is based on historical data to fit and build an associated response model between the average temperature of the three-stage roasting furnace and the sintering severity; in the model, a management objective function aimed at minimizing the quality deviation module length is defined; The associated response model includes constant terms, first-order linear terms, quadratic terms, and interaction terms; in order to prevent overfitting caused by too complex model, stepwise regression method is used to select significant interaction terms. The specific process is as follows: The initial model only contains first-order linear terms and quadratic terms, with a total of 26+26=52 coefficients; based on the significance judgment (p-value<0.05) of F test, interaction terms are gradually included; after each interaction term is included, the adjusted R² is recalculated, and if the R² growth is less than 2%, the inclusion is stopped, and R² represents the proportion of the variance (variation) of the response variable (i.e. failure mode) that the model can explain to the total variance; in this way, the interaction terms can be compressed from the theoretical 325 to usually about 15-20, which not only retains important nonlinear and coupling relationships, but also avoids parameter explosion; For multiple output situations (four failure modes), four independent response surface models are established, allowing different failure modes to have different sensitivities to the same production parameter; The least squares method is used for model fitting, and the sample size should be no less than 10 times the number of coefficients; if the number of available samples is insufficient, ridge regression or Lasso regression is used to introduce a regularization term to improve the stability of the model.

[0127] Offset calculation: Using a multi-objective optimization strategy to search for the optimal solution, it is calculated that in order to offset the current sintering deviation, the set temperature should be reduced by 8 degrees Celsius, that is, the target parameter offset is -8 degrees Celsius. Dynamic tolerance calculation: The module reads that the attribution confidence index output by the feature recognition module is 95%, and determines that it is in the high confidence interval; Tolerance narrowing: The system executes a strong correction strategy, dynamically adjusting the dynamic tolerance constraint range. This involves narrowing the historical standard tolerance of the temperature variable (…). The tolerance (in degrees Celsius) is narrowed to 30% of the equipment's limit capability, meaning the new dynamic tolerance is... Celsius, to ensure absolute precision in temperature control; Protocol encapsulation and delivery: The module will adjust the target parameter offset (down by 5 degrees Celsius) and the dynamic tolerance constraint range ( The temperature (in degrees Celsius) is encapsulated into a standardized execution control protocol. This protocol is automatically pushed to the manufacturing execution system via an interface. In the next batch of production, the temperature control program of the calcining furnace will automatically apply this more stringent process standard, thereby achieving closed-loop quality management throughout the entire life cycle.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A catalyst industry chain-oriented full life cycle traceability method, characterized in that, The method comprises the following steps: acquiring state attribute data of the same batch of recycled products, analyzing the state attribute data, and constructing a standardized performance feature vector; inputting the performance feature vector into a pre-set quality evaluation model for matching to identify quality deviation data of the recycled products; analyzing the batch identity code of the recycled products to obtain production life cycle history data, which covers raw material batch information and production control variables in the production process; constructing a multi-dimensional data correlation model, taking the quality deviation data as the target variable and the history data as the input feature, analyzing the influence weight coefficient and the attribution confidence index of each production control variable on the quality deviation, and screening out variables with an influence weight coefficient higher than a pre-set threshold value as key control elements; based on the key control elements, performing production control optimization operation, calculating a target parameter offset for correcting the quality deviation, and dynamically calculating a dynamic tolerance constraint range of the production control variable in combination with the attribution confidence index; and encapsulating the target parameter offset and the dynamic tolerance constraint range as an execution control protocol.

2. The full life cycle tracing method for the catalyst industry chain according to claim 1, wherein: the state attribute data covers performance evaluation indexes in three aspects of microstructure, surface deposition and macro mechanics: the first evaluation index is a microstructure characterization index, including a specific surface area attenuation index, a pore volume distribution skewness index and an average pore size collapse index; the second evaluation index is a surface chemical deposition index, including a total amount index of carbon deposition on the catalyst surface, a carbon deposition graphitization degree index and a deposition density index of exogenous toxic elements; the third evaluation index is a macro mechanical property index, including an average value index of particle lateral pressure strength, a wear index and a change rate index of bulk density.

3. The full life cycle tracing method for the catalyst industry chain according to claim 1, wherein: the state attribute data is analyzed and processed to construct a standardized performance feature vector, including: performing consistency verification of each state attribute data in the same batch of recycled products, and identifying abnormal data points beyond the compliance range by using a pre-set outlier judgment logic; for the identified abnormal data points, data validity audit is performed in combination with a sampling position label, if it is determined as a sampling error, it is excluded, if it is determined as a local working condition abnormality, it is retained and marked as an abnormal sub-sample, to obtain high-fidelity cleaned data; and using a dimensionless mapping protocol, the cleaned data is mapped into a pre-set standardized evaluation interval to generate a performance feature vector of the recycled products.

4. The full life cycle tracing method for the catalyst industry chain according to claim 1, wherein: the quality deviation data includes an abnormal type label and deviation feature data; the abnormal type label includes high-temperature sintering failure, chemical poisoning failure, pore blockage failure and mechanical crushing failure; and the deviation feature data is a failure severity index for quantitatively describing the severity value of the failure type. The identification logic of the quality evaluation model is designed to: pre-construct a business feature reference library containing various types of standard failure modes, and store standard benchmark models corresponding to different failure types in the library; in the matching process, the similarity evaluation rule is applied to calculate the matching correlation degree of the current input performance feature vector and the standard benchmark model in the business feature reference library; the abnormal type label and the deviation feature data are correspondingly output, and the quality deviation data is obtained.

5. The full life cycle tracing method for the catalyst industry chain according to claim 1, characterized in that: The multi-dimensional data correlation model comprises: Performing reverse mapping alignment of space-time data; calculating the time lag of products from raw material input to finished product output according to the logistics flow rule of the production line, accurately mapping the batch identity code of the recycled product back to the operation data slice when flowing through each production process, and establishing an input feature matrix; Performing weight evaluation based on multi-dimensional attribution analysis; inputting the input feature matrix and the target variable into the multi-dimensional data correlation analysis model for learning, quantifying the influence weight coefficient by evaluating the information contribution value of each production control variable in the model node splitting process; Performing confidence evaluation based on repeated verification; using a random sampling strategy to reconstruct and repeatedly analyze the data set multiple times, and statistically analyzing the frequency stability of the production control variables identified as key elements in repeated evaluation, and defining the frequency stability value as the attribution confidence index.

6. The full life cycle tracing method for the catalyst industry chain according to claim 1, characterized in that: The production control optimization operation adopts an optimization design based on a reverse compensation strategy, comprising: Based on the data of historical batches, an associated response model between the key control elements and the quality deviation data is fitted and constructed, representing the sensitivity relationship between the change of the production control variable and the final product quality deviation; In the associated response model, a management objective function is defined, which aims to minimize the modulus of the quality deviation data; an optimal control strategy is searched within the feasible process window using a multi-objective optimization strategy, and the numerical change amount of the key control elements that should occur to offset the currently identified quality deviation is calculated, and the numerical change amount is defined as the target parameter offset.

7. The full life cycle tracing method for the catalyst industry chain according to claim 1, characterized in that: The calculation of the dynamic tolerance constraint range of the production control variable comprises: identifying the attribution confidence index to distinguish high confidence interval, medium confidence interval and low confidence interval; based on the confidence interval where the attribution confidence index is located, the data interval of the dynamic tolerance constraint range is dynamically adjusted.

8. A catalyst industry chain-oriented whole life cycle traceability system, characterized in that, Comprise: The data acquisition module is used for acquiring the state attribute data of the same batch of recycled products, analyzing and processing the state attribute data, constructing a standardized performance feature vector, inputting the performance feature vector into the pre-set quality evaluation model for matching, and identifying the quality deviation data of the recycled products; The data analysis module is used for analyzing the batch identity code of the recycled products, obtaining the history data of the production full life cycle, and the history data covers the raw material batch information and the production control variables in the production process; An element identification module is configured to construct a multi-dimensional data correlation model, take the quality deviation data as a target variable, take the history data as an input feature, analyze an influence weight coefficient and an attribution confidence index of each production control variable on the quality deviation, and screen out variables with an influence weight coefficient higher than a preset threshold value, and define the variables as key control elements. A production control module is configured to perform production control optimization operation based on the key control elements, calculate a target parameter offset for correcting the quality deviation, and dynamically calculate a dynamic tolerance constraint range of the production control variable in combination with the attribution confidence index, and encapsulate the target parameter offset and the dynamic tolerance constraint range as an execution control protocol.

Images