Abnormal traceability and parameter regulation and control system of premix production line

Abstract

The invention belongs to the field of production quality traceability, and particularly relates to an abnormity traceability and parameter regulation and control system for a premix production line, which comprises a monitoring module, a reverse mapping module and a forward adjustment module, a monitoring module can acquire whole-course fingerprint information of an abnormal production process in real time by constructing a whole-course tracking chain containing a process-control association relationship; the reverse mapping module performs reverse whole-course reasoning by using a hidden Markov algorithm, performs longitudinal local reasoning in combination with an index mapping relation, accurately locates an abnormal source, and generates a reverse abnormal information fingerprint chain; and the forward adjustment module performs forward processing control parameter adaptive adjustment from the abnormal starting node according to the fingerprint chain and a preset correction strategy until all indexes return to normal. According to the invention, closed-loop intelligent control from abnormity monitoring and source tracing to production self-adjustment is realized, and the quality consistency and tracing efficiency of premix production are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of production quality traceability, and in particular relates to an anomaly traceability and parameter control system for a premix production line. Background Technology

[0002] In the premix production field, end-to-end material traceability and precise management of each production step are crucial for product quality control. Traditional management methods rely heavily on manual recording or partial data collection from single devices. This lack of integration with various testing equipment, such as high-performance liquid chromatographs, infrared spectrophotometers, analytical balances, and vacuum drying ovens, makes it difficult to form a material traceability chain covering the entire process from raw materials to finished products. Furthermore, the absence of a systematic correlation mapping mechanism between the testing indicators of each production step and the corresponding production control parameters hinders the efficient and accurate use of test results to optimize the production process. Therefore, how to construct a traceability chain covering the entire premix production process, integrating indicators from multiple testing devices and parameters of each production step, and establishing a precise mapping relationship between test results and production step control to achieve efficient material traceability from raw materials to finished products and precise management of the production process has become an urgent problem to be solved. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an anomaly tracing and parameter control system for a premixed agent production line. This system includes a monitoring module, a reverse mapping module, and a forward adjustment module. By constructing a full-process tracking chain containing process-control correlations, the monitoring module can acquire real-time fingerprint information of the entire abnormal production process. The reverse mapping module uses a Hidden Markov Model (HMM) for reverse full-process reasoning and combines it with indicator mapping relationships for vertical local reasoning to accurately locate the root cause of the anomaly and generate a reverse anomaly information fingerprint chain. The forward adjustment module, based on this fingerprint chain and a preset correction strategy, adaptively adjusts the processing control parameters from the anomaly initiation node until all indicators return to normal. This invention achieves closed-loop intelligent control from anomaly monitoring and root cause tracing to production self-adjustment, significantly improving the quality consistency and traceability efficiency of premixed agent production.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An anomaly tracing and parameter control system for a premix production line includes: a monitoring module, a reverse mapping module, and a forward adjustment module;

[0006] The monitoring module is used to obtain the full fingerprint information of the abnormal production process based on the preset full-process tracking chain and the preset full-process monitoring indicator anomaly judgment threshold. The full-process fingerprint information includes the timestamp of each standard process step and the corresponding full-process monitoring indicator and local monitoring indicator, the mapping relationship between the full-process monitoring indicator and local monitoring indicator of each standard process step, and the mapping relationship between the full-process monitoring indicator and local monitoring indicator, the detection accuracy of each indicator and the processing control parameters of the corresponding standard process step.

[0007] The reverse mapping module is used to perform reverse full-process inference based on the full fingerprint information of the abnormal production process combined with the Hidden Markov Algorithm, and at the same time to perform vertical local inference using the mapping relationship between the full-process monitoring indicators and the local monitoring indicators to obtain a reverse anomaly information fingerprint chain; the reverse anomaly information fingerprint chain contains at least one abnormal full-process monitoring indicator or an abnormal local monitoring indicator.

[0008] The forward adjustment module is used to select at least one correction strategy from a preset abnormal processing correction strategy library based on the mapping relationship between the reverse abnormal information fingerprint chain, the full-process monitoring indicators, the local monitoring indicators, and the processing control parameters of the corresponding standard process steps. The forward processing control parameters are adjusted from the starting abnormal node in the reverse abnormal information fingerprint chain until all full-process monitoring indicators and local monitoring indicators in the reverse abnormal information fingerprint chain meet the corresponding indicator thresholds.

[0009] Specifically, the construction process of the end-to-end tracking chain includes:

[0010] By acquiring standard production process information of the target product and combining it with a hierarchical parsing algorithm, and taking the process flow implemented by each piece of equipment as the boundary, a global standard process step sequence and global standard process correlation relationship are obtained.

[0011] Based on standard process steps, and taking the production process parameters corresponding to the indivisible actions of the process flow implemented by each piece of equipment as the boundary, we obtain the local standard process sub-step sequence and the corresponding local correlation relationship for each standard process step.

[0012] The local association includes the priority order of the timestamps of the corresponding sub-steps and the image probability distribution of the degree of standardization of the current sub-step's execution on the degree of standardization of subsequent sub-steps, as well as the order of magnitude of the image sub-steps.

[0013] Specifically, the process of building the end-to-end tracking chain also includes:

[0014] Based on the global standard process step sequence and the corresponding global standard process correlation, the full-process monitoring indicators and corresponding cumulative relationships are obtained; the corresponding cumulative relationships include positive cumulative and negative cumulative.

[0015] The full-process monitoring indicators include at least the total quality inspection score and the probability of the production efficiency being affected throughout the process; the total quality inspection score is calculated by weighted average method based on the real-time quality inspection score of each standard process step and the contribution of the corresponding standard process step to the total quality inspection score corresponding to the standard production process information.

[0016] The production efficiency deviation is calculated by combining the deviation between the real-time production time achieved by the current standard process step and the corresponding preset standard production time with the deviation between the real-time production time achieved by the next standard process step and the corresponding preset standard production time, using a Bayesian network model. It is used to measure the probability distribution of the cumulative impact of the production efficiency of each standard process step on the production efficiency of the remaining standard process steps.

[0017] Based on the full-process monitoring indicators and their corresponding cumulative relationships, and the processing control nodes constructed with the labels of each processing equipment, combined with a loosely coupled algorithm, an initial full-process tracking chain is constructed.

[0018] Specifically, the process of building the end-to-end tracking chain also includes:

[0019] Based on the sequence of local standard process sub-steps, with each processing control node as the root node of the corresponding standard process step, and combining the local implementation priority relationship of the sequence of local standard process sub-steps corresponding to each standard process step and the probability distribution of the first influence of the standard degree of the execution action of each sub-step on the standard degree of the execution action of the next sub-step with a tree database, a forward process flow tree for each standard process step is obtained.

[0020] Based on the local monitoring indicators corresponding to each sub-step of the process flow tree for each standard process step, combined with the full-process monitoring indicators of each standard process step in the corresponding production period, the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step are obtained through factor analysis algorithm. The contribution of the detection accuracy of the corresponding local monitoring indicators to the deviation values ​​of the full-process monitoring indicators in the corresponding time period and to the deviation values ​​of the related sub-steps under the remaining standard process steps are also considered.

[0021] Specifically, the process of building the end-to-end tracking chain also includes:

[0022] Based on the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of the full monitoring indicators for the corresponding time period, a vertical internal derivation connection is constructed using the Hidden Markov algorithm.

[0023] The vertical internal derivation connection is embedded into the forward process flow tree of each standard process sub-step. At the same time, the cross-inference connection constructed by the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of the related sub-steps under the remaining standard process steps is embedded into the sub-steps with control relationships in the forward process flow tree under different standard process sub-steps.

[0024] Specifically, the process of building the end-to-end tracking chain also includes:

[0025] Based on the processing control parameters and standard processing time length corresponding to each sub-step in the forward process flow tree of each standard process step, a forward processing control tree for each standard process step is obtained. At the same time, the process-control connection is established by combining the standard degree of the execution action corresponding to each processing control parameter with the deviation value of the local monitoring index of the corresponding standard process sub-step.

[0026] The forward processing control tree of each standard process step under the initial full-process tracking chain is mapped to the corresponding forward process flow tree through the process-control connection to obtain the full-process tracking chain containing the process-control association tree;

[0027] The full-process tracking chain, which includes a process-control association tree, is combined with random forest and simulation algorithms. The positional deviation and corresponding timestamp deviation of the corresponding outlier points and their cumulative outlier points are used as the derivation loss function for simulation training to obtain a fully trained full-process tracking chain.

[0028] Specifically, the process of obtaining the reverse anomaly information fingerprint chain includes:

[0029] The total quality inspection score of each standard process step is compared with the preset threshold for the total quality inspection score, and the production efficiency deviation is compared with the preset threshold for the production efficiency deviation. The first standard process step with a total quality inspection score lower than the corresponding preset threshold or a production efficiency deviation higher than the corresponding preset threshold is selected to obtain the abnormal starting standard process step.

[0030] Based on the abnormal starting standard process step and the full-process monitoring indicators of all standard process steps before this step, as well as the timestamps of each standard process step, the total quality inspection score and production efficiency deviation of all steps are extracted. The two sets of indicator values ​​are arranged in order from back to front according to the timestamps of each standard process step and the corresponding total quality inspection score and production efficiency deviation, to obtain the observation sequence of the Hidden Markov Algorithm.

[0031] Specifically, the process of obtaining the reverse anomaly information fingerprint chain also includes:

[0032] Retrieve the pre-stored state transition probability matrix between each standard process step from the full tracking chain; the state transition probability matrix includes the positive cumulative impact probability of abnormal total quality inspection score between any adjacent steps, the negative cumulative impact probability of abnormal total quality inspection score, the positive cumulative impact probability of abnormal production efficiency deviation between adjacent steps, and the negative cumulative impact probability of abnormal production efficiency deviation.

[0033] The total quality inspection score and production efficiency deviation value of each step in the observation sequence are input into the back-inference model constructed by the Hidden Markov Algorithm. With the goal of maximizing the matching degree between the abnormal index values ​​in the observation sequence and the cumulative influence probability in the state transition probability matrix, the hidden state of each preceding standard process step is derived one by one, starting from the abnormal initial standard process step. The specific hidden state type and corresponding timestamp of each standard process step are recorded. The derivation stops when the cumulative sum of the deviations between the total quality inspection score and the production efficiency deviation is 0. The root node with the first deviation of 0 is taken as the starting deviation node, and the specific hidden state and corresponding timestamp of each standard process step are obtained. The hidden states are: the initial state of abnormal total quality inspection score, the state of abnormal transmission of abnormal total quality inspection score, the initial state of abnormal production efficiency deviation, the state of abnormal transmission of abnormal production efficiency deviation, and the normal production state.

[0034] Specifically, the process of obtaining the reverse anomaly information fingerprint chain also includes:

[0035] Based on the specific hidden state type, corresponding timestamp, and starting deviation node of each standard process step obtained through derivation, the reverse state sequence representing the entire dimensional anomaly propagation path is obtained by arranging the specific hidden state type-timestamp in the order of the timestamps of each standard process step from back to front.

[0036] Based on the reverse state sequence representing the entire dimensional anomaly propagation path, the first influence probability distribution under each root node on the path, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node are extracted.

[0037] Based on the cumulative deviation value of the full-process monitoring indicators of each root node and its forward root node in the reverse state sequence, combined with the first influence probability distribution under each node, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node, the abnormal deviation index value of each local detection indicator under each node is obtained by inversion.

[0038] Specifically, the process of obtaining the reverse anomaly information fingerprint chain also includes:

[0039] Based on the abnormal deviation index value of each local detection index under each node, combined with the process-control connection, the standard deviation value of the corresponding processing process execution action and the deviation value of the standard processing time length of the corresponding execution action are obtained by inversion.

[0040] Based on the reverse state sequence of the full-dimensional anomaly propagation path, the corresponding equipment label information, processing technology execution priority, and the standard deviation value of the corresponding processing technology execution action and the deviation value of the standard processing time length of the corresponding execution action are combined with the graph algorithm to construct the reverse anomaly information fingerprint chain.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This invention addresses the shortcomings of existing technologies by constructing a deep-integrated end-to-end tracking chain of process and control parameters, enabling precise and multi-dimensional perception and tracing of production anomalies. The monitoring module can capture end-to-end fingerprint information, including the accuracy of indicator detection, in real time, providing a highly reliable data foundation for analysis. The reverse mapping module innovatively combines hidden Markov models for reverse end-to-end reasoning and vertical local reasoning, enabling it to penetrate complex process relationships, accurately locate the root cause of anomalies and their transmission paths, and generate a reverse anomaly information fingerprint chain, completely changing the limitation of traditional traceability systems that can only record data. Finally, the forward adjustment module, based on the fingerprint chain, achieves adaptive adjustment of forward parameters from the anomaly initiation point, forming an intelligent closed-loop control of monitoring-diagnosis-repair, thereby significantly improving product quality consistency, production efficiency, and system autonomy, and effectively reducing quality risks and production costs. Attached Figure Description

[0043] Figure 1 This is a block diagram of an anomaly tracing and parameter control system for a premix production line according to Embodiment 1 of the present invention;

[0044] Figure 2 This is a diagram of the end-to-end tracking chain architecture of Embodiment 1 of the present invention; Detailed Implementation

[0045] Example 1

[0046] In premix production, ensuring product quality requires traceability of the entire process, monitoring indicators, and processing parameters. However, existing management methods often involve fragmented data collection, failing to establish a comprehensive tracking system that integrates indicator detection accuracy and sub-step correlations. This makes it difficult to efficiently pinpoint the root cause after anomalies occur. The specific technical problem is that when adjusting processing control parameters forward from the starting point of the reverse anomaly fingerprint chain, it's impossible to determine a reasonable adjustment range based on the precise mapping relationship between local monitoring indicator deviations and processing control parameters, combined with the probability distribution of influence between sub-steps. This easily leads to over- or under-adjustment of parameters, resulting in low correction efficiency and difficulty in quickly achieving target compliance. Please refer to [link / reference]. Figure 1 The present invention provides an embodiment of an anomaly tracing and parameter control system for a premix production line, comprising: a monitoring module, used to obtain full-process fingerprint information of an abnormal production process based on a preset full-process tracking chain and a preset full-process monitoring indicator anomaly discrimination threshold; the full-process fingerprint information includes a timestamp of each standard process step and the corresponding full-process monitoring indicator and local monitoring indicator, the mapping relationship between the full-process monitoring indicator and local monitoring indicator of each standard process step, and the mapping relationship between the full-process monitoring indicator and local monitoring indicator, the detection accuracy of each indicator, and the processing control parameters of the corresponding standard process step;

[0047] The reverse mapping module is used to perform reverse full-process inference based on the full fingerprint information of the abnormal production process combined with the Hidden Markov Algorithm, and at the same time to perform vertical local inference using the mapping relationship between the full-process monitoring indicators and the local monitoring indicators to obtain a reverse anomaly information fingerprint chain; the reverse anomaly information fingerprint chain contains at least one abnormal full-process monitoring indicator or an abnormal local monitoring indicator.

[0048] The forward adjustment module is used to select at least one correction strategy from a preset abnormal processing correction strategy library based on the mapping relationship between the reverse anomaly information fingerprint chain, the full-process monitoring indicators, the local monitoring indicators, and the processing control parameters of the corresponding standard process steps. The forward processing control parameters are adjusted from the starting abnormal node in the reverse anomaly information fingerprint chain until all full-process monitoring indicators and local monitoring indicators in the reverse anomaly information fingerprint chain meet the corresponding indicator thresholds.

[0049] It should be further explained that the construction process of the full-process tracking chain in this embodiment includes:

[0050] By acquiring standard production process information of the target product and combining it with a hierarchical parsing algorithm, and taking the process flow implemented by each piece of equipment as the boundary, a global standard process step sequence and global standard process correlation relationship are obtained.

[0051] For example, to clearly illustrate the stratified analysis process, the production of a compound vitamin premix is ​​used as an example. First, its standard production process information is obtained, which covers the following six core stages: raw material preparation, raw material crushing, material mixing, isothermal reaction, vacuum drying, and finished product quality testing. In the raw material preparation stage, vitamin B12 raw material and wheat bran carrier excipients are added to remove impurities. In the raw material crushing stage, the impurity-removed raw materials are crushed to obtain powder of a specified fineness. In the material mixing stage, the crushed raw materials and excipients are mixed to achieve uniform material distribution. In the isothermal reaction stage, the mixture undergoes a chemical reaction under constant conditions to improve product stability. In the vacuum drying stage, the reacted materials are treated to reduce moisture content. In the finished product quality testing stage, the vitamin content of the dried finished product is determined to ensure compliance with quality standards.

[0052] Subsequently, a hierarchical analytical algorithm was applied to divide the standard process steps into three levels: the process stage level, the equipment function level, and the operation step level. Using the process flow implemented by each piece of equipment as the decomposition boundary, the specific decomposition is as follows: The vibrating screen corresponds to the raw material preparation process, and its operation steps include equipment preheating, adding vitamin B12 raw materials and wheat bran adjuvants, setting the screening frequency, performing the screening operation, and collecting qualified raw materials; the universal pulverizer corresponds to the raw material pulverization process, and its operation steps cover equipment startup, feeding, setting the pulverization speed, performing pulverization, and collecting the pulverized raw materials; the double helix mixer corresponds to the material mixing process. The operating steps include cleaning the mixer, feeding materials, setting the mixing speed, executing the mixing process, and collecting the uniformly mixed materials; the constant temperature and humidity chamber corresponds to the constant temperature reaction process, and the operating steps are equipment preheating, feeding materials, maintaining reaction conditions, executing the reaction process, and collecting the semi-finished product; the vacuum drying oven corresponds to the vacuum drying process, and the operating steps include vacuuming, feeding materials, setting the drying temperature, executing the drying process, and collecting the dried product; the high performance liquid chromatograph corresponds to the finished product quality inspection process, and the operating steps are sampling, preparing the test solution, injecting the sample for analysis, detecting the vitamin B12 content, and determining whether the results meet the standards.

[0053] Finally, based on the chronological order of process execution, the operation steps corresponding to each of the above-mentioned equipment are integrated to form a global standard process step sequence for compound vitamin premix. At the same time, according to the material transfer logic between steps, such as the qualified raw materials produced by the vibrating screen being transported to the universal pulverizer, and the output of the universal pulverizer being transported to the twin-helix mixer, and combined with the synergistic effect between equipment and the interdependence of process parameters, the global standard process correlation is established.

[0054] Based on standard process steps, and taking the production process parameters corresponding to the indivisible actions of the process flow implemented by each piece of equipment as the boundary, we obtain the local standard process sub-step sequence and the corresponding local correlation relationship for each standard process step.

[0055] For example, in order to more clearly illustrate the local decomposition process of each standard process step here, each standard process step is further decomposed into a unit based on the basic actions that cannot be further decomposed in the process flow of each piece of equipment, and based on the key production process parameters corresponding to the actions.

[0056] Taking the material mixing steps performed by a double helix mixer as an example, the following steps are explained: First, it is clear that the non-decomposable actions involved in this step include mixer cleaning, raw material feeding, mixing speed setting, mixing process timing, and material unloading. Each action is associated with specific process parameters. For example, the cleaning action involves the amount and duration of cleaning medium, the feeding action involves the amount of raw materials and auxiliary materials fed, the speed setting and timing actions determine the mixing intensity and time, and the unloading action involves the opening degree and control duration of the unloading valve.

[0057] Based on the above action and parameter limits, the standard process step of material mixing is decomposed into a sequence of local standard process sub-steps, which are as follows: cleaning the mixer, adding vitamin raw materials and carrier excipients, setting the speed of the twin-helix mixer, controlling the mixing process time, and unloading the mixed materials according to the set parameters.

[0058] Simultaneously, the local correlations between each sub-step are determined; the timestamp priority order is fixed as follows: cleaning before addition, addition before speed setting, speed setting before mixing timing, and mixing timing before unloading; in addition, the influence of the execution standard of the preceding sub-step on the subsequent sub-step is defined. For example, if the mixer is not thoroughly cleaned, it will significantly increase the risk of contamination of the subsequently added material; if the mixing speed setting is insufficient, it will directly reduce the possibility of the material uniformity meeting the standard in the subsequent mixing timing stage, and the scope of the subsequent sub-steps affected by its influence is clarified; the local correlations include the image probability distribution of the timetamp priority order of the corresponding sub-steps and the image order of the sub-steps on the execution standard of the current sub-step.

[0059] Based on the global standard process step sequence and the corresponding global standard process correlation, the full-process monitoring indicators and their corresponding cumulative relationships are obtained; the corresponding cumulative relationships include positive and negative cumulative relationships; for example, taking the global standard process step sequence of premix production and the local correlation between each step as an example, the full-process monitoring indicators are selected as the total quality inspection score and the deviation of production efficiency, and the specific manifestations of positive and negative cumulative relationships are as follows:

[0060] Positive accumulation refers to the phenomenon where local monitoring indicators in preceding standard process steps are better than preset standards. Through local correlation, this positively promotes the local monitoring indicators in subsequent steps, thus accumulating the overall monitoring indicators towards a better direction. For example, if the local monitoring indicators in the raw material screening step meet the standard better, based on local correlation, it will improve the compliance level of the local monitoring indicators in the subsequent raw material crushing step. The optimization of the indicators in the raw material crushing step will further improve the local monitoring indicators in the material mixing step. The local indicators in each step develop in sequence towards improvement, ultimately accumulating the total quality inspection score to a higher level and the production efficiency deviation accumulating in a direction more conducive to production progress.

[0061] Negative accumulation refers to a situation where the failure of local monitoring indicators in a preceding standard process step to meet the preset standard negatively impacts the local monitoring indicators in subsequent steps through local correlation, causing the overall monitoring indicators to accumulate towards a worse state. For example, if the local monitoring indicators in the raw material screening step fail to meet the standard, based on local correlation, it will lower the compliance level of the local monitoring indicators in the subsequent raw material crushing step. The deficiencies in the raw material crushing step indicators will further lower the local monitoring indicators in the material mixing step. The failure of the mixing step indicators will also affect the local monitoring indicators in the isothermal reaction step. The local indicators in each step progressively worsen, ultimately causing the total quality inspection score to accumulate to a lower level, and the production efficiency deviation to accumulate in a direction unfavorable to the production schedule.

[0062] The full-process monitoring indicators include at least the total quality inspection score and the probability of the production efficiency being affected throughout the process; the total quality inspection score is calculated using a weighted average method based on the real-time quality inspection score of each standard process step combined with the contribution of the corresponding standard process step to the total quality inspection score corresponding to the standard production process information; for example, the process of obtaining the total quality inspection score in this embodiment includes:

[0063] Step 1: Determine the overall standard process sequence for premix production. This sequence consists of six core steps: raw material screening, raw material crushing, material mixing, isothermal reaction, vacuum drying, and finished product quality inspection. The execution order of each step in the production process must be clearly defined, and the specific role of each step in shaping the final quality of the premix product must be explained.

[0064] Step 2: Based on the impact of each standard process step on the final product quality of the premix, quantitatively determine its contribution to the total quality inspection score. Among these, the material mixing step, which directly determines the uniformity of the material, is assigned the highest contribution; the isothermal reaction step affects the stability of the product components, with the second highest contribution; the raw material pulverization step indirectly affects subsequent processes by influencing the fineness of the raw materials, with the next lowest contribution; the vacuum drying step mainly controls the product's moisture content, with the lowest contribution; the raw material screening step concerns the initial impurity content of the raw materials, with a relatively low contribution; and the finished product quality inspection step, as the verification step for the final result, has the lowest contribution. The sum of the contributions of all standard process steps is 100% to ensure the integrity of the weighted calculation system.

[0065] Step 3: After each standard process step is completed, real-time quality testing and scoring are immediately performed based on the specific quality inspection indicators for that step. Specifically, the raw material screening step is scored based on whether the impurity removal rate meets the preset standard; the raw material crushing step is scored based on whether the fineness of the obtained raw materials meets the process requirements; the material mixing step is scored based on whether the uniformity of the mixed material meets the standard; the isothermal reaction step is scored based on whether the stability of the reaction product components is qualified; the vacuum drying step is scored based on whether the moisture content of the dried material reaches the specified range; and the finished product quality inspection step is scored based on whether the effective ingredient content of the final product meets the quality standards. All real-time quality inspection scores for each step are quantified using a percentage system.

[0066] Step 4: For each completed standard process step that has obtained a real-time quality inspection score, its score is weighted by its pre-determined contribution level. Specifically, the real-time quality inspection score of a step is multiplied by its corresponding contribution level; the result is the weighted quality score for that step. This calculation rule applies to every standard process step in the sequence.

[0067] Step 5: Sum the weighted quality scores of all standard process steps calculated in Step 4. The final arithmetic sum is the total quality test score that characterizes the overall quality level of the batch of premixed agent production. This total score can comprehensively and intuitively reflect the degree of compliance of the entire production process execution results with the quality standards.

[0068] The production efficiency deviation is calculated using a Bayesian network model, combining the deviation between the real-time production time of the current standard process step and the corresponding preset standard production time, along with the deviation between the real-time production time of the next standard process step and the corresponding preset standard production time. It measures the probability distribution of the cumulative impact of the production efficiency of each standard process step on the production efficiency of the remaining standard process steps. For example, to better illustrate the production efficiency deviation, it is specifically as follows:

[0069] Step 1: In the global standard process sequence for premix production, select the current standard process step whose efficiency impact needs to be evaluated, and the next standard process step that immediately follows it and has a direct material transfer and timing dependency. Obtain the preset standard production time of the current step and the preset standard production time of the next step as benchmark reference values.

[0070] Step 2: In actual production operation, when the current standard process step is started, timing begins, and its real-time production duration is recorded upon completion. Similarly, after the next standard process step is completed, its complete real-time production duration is recorded.

[0071] Step 3: Subtract the preset standard production time from the recorded real-time production time of the current step to obtain the production time deviation value for the current step. Similarly, subtract the preset standard production time from the recorded real-time production time of the next step to obtain the production time deviation value for the next step. Positive values ​​indicate production delays, and negative values ​​indicate production ahead of schedule.

[0072] Step 4: Construct a Bayesian network inference model. This model contains two input nodes, corresponding to the production time deviation of the current step and the production time deviation of the next step, respectively. The output node of the model is defined as the probability distribution of the cumulative influence of the production efficiency of the current step on the production efficiency of all remaining standard process steps (i.e., all steps in the sequence following the next step). Internally, based on historical production data, a conditional probability table is established to describe the probabilistic dependencies between the states of the input nodes and the states of the output nodes.

[0073] Step 5: Perform probabilistic inference and output the efficiency deviation. The production time deviation between the current step and the next step calculated in Step 3 is used as evidence input into the Bayesian network model constructed in Step 4. Through the model's probabilistic inference mechanism, the cumulative impact probability distribution of the current step's production efficiency on the production efficiency of each subsequent remaining standard process step is calculated. This probability distribution is the defined production efficiency deviation of the current standard process step, used to quantitatively assess the potential ripple effect of its efficiency anomaly on the downstream production process. Based on the full-process monitoring indicators and their corresponding cumulative relationships, and the processing control nodes constructed with each processing equipment label, combined with a loosely coupled algorithm, an initial full-process tracking chain is constructed.

[0074] Based on the sequence of local standard process sub-steps, with each processing control node as the root node of the corresponding standard process step, and combining the local implementation priority relationship of the sequence of local standard process sub-steps corresponding to each standard process step and the probability distribution of the first influence of the standard degree of the execution action of each sub-step on the standard degree of the execution action of the next sub-step with a tree database, a forward process flow tree for each standard process step is obtained. It should be further noted that the standard process sub-step here corresponds to a specific action of each device when implementing the current standard process step.

[0075] It should be further explained that each root node in this embodiment corresponds one-to-one with a processing device, and the standard process step corresponding to each root node corresponds to the specific production process parameters of the processing device; for example, the exemplary construction process of the forward process flow tree for each standard process step in this embodiment includes:

[0076] Step 1: Determine the target standard process step in premix production. Taking the isothermal reaction standard process step as an example, and considering the core processing requirement of achieving a constant temperature and humidity reaction of materials in this step, the unique corresponding production equipment is identified as the isothermal and humidity chamber. This is then designated as the root node of the forward process flow tree for the isothermal reaction standard process step. Based on the core control functions of the isothermal and humidity chamber, three secondary control attributes—temperature control, humidity control, and timing control—are set under the root node. Each secondary control attribute corresponds to a type of core control parameter of the isothermal and humidity chamber, used to assist in classifying subsequent sub-steps related to this type of control.

[0077] Step 2: Obtain the sequence of local standard process sub-steps corresponding to the standard process steps of the isothermal reaction. All sub-steps in this sequence are independent processing actions that the isothermal and humidity chamber can perform. Specifically, these include: starting the isothermal and humidity chamber and performing preheating (operation of the equipment's own heating system); injecting the mixed material into the internal cavity of the isothermal and humidity chamber (operation of the equipment's feeding mechanism); adjusting the temperature parameters of the isothermal and humidity chamber (operation of the equipment's temperature control system); adjusting the humidity parameters of the isothermal and humidity chamber (operation of the equipment's humidity control system); maintaining the set temperature and humidity in the isothermal and humidity chamber and timing (operation of the equipment's temperature and humidity maintenance system and timing system); and stopping the isothermal and humidity chamber and performing unloading (operation of the equipment's unloading mechanism). It is clear that the processing actions of each sub-step depend on the isothermal and humidity chamber for completion, and that executing them sequentially constitutes a complete isothermal reaction process.

[0078] Step 3: Analyze the local implementation priority relationships of the standard process sub-step sequence, and determine the execution order of the sub-steps based on the actual processing logic of the temperature and humidity chamber: First, perform the temperature and humidity chamber preheating action; then, perform the material injection action; next, perform the temperature setting and humidity setting actions; then, perform the heat preservation and timing action; and finally, perform the reaction termination action. Ensure that subsequent sub-nodes strictly follow the processing logic priority of this equipment.

[0079] Step 4: Analyze the probability distribution of the first influence of the standardization of the corresponding action in each sub-step on the standardization of the corresponding action in the next sub-step. This analysis is based on the correlation within the processing environment of the constant temperature and humidity chamber. For example, in the preheating action, the degree to which the preheating temperature reaches the preset standard affects the probability of achieving the target temperature setting action, because the internal temperature field of the equipment needs to be preheated to near the target temperature to more accurately adjust to the set value. In the material injection action, the degree to which the injected material reaches the preset standard affects the probability of achieving the target humidity setting action, because the material quantity affects the balance and regulation of humidity within the equipment. For all adjacent sub-steps, based on the interaction logic of the equipment's own processing actions, clarify the probability distribution of the first influence between the two.

[0080] Step 5: Initialize the tree-structured database. Create new tree-structured entries in the database system, and enter the constant temperature and humidity chamber identified in Step 1 as the root node. Configure a unique equipment identifier for the root node and associate it with the corresponding standard process step name. Create three secondary category fields under the root node, corresponding to the three secondary control attributes: temperature control, humidity control, and timing control. Each secondary control attribute field records the corresponding equipment control parameter type, ensuring that the secondary control attributes and the root node form a clear hierarchical relationship.

[0081] Step 6: Assign the local standard process sub-steps obtained in Step 2 to the corresponding levels of the tree database according to the priority relationships determined in Step 3. Specifically, the preheating action and temperature setting action of the constant temperature and humidity chamber, both related to equipment temperature control, are assigned to the temperature control secondary attribute field under the root node; the humidity setting action, related to equipment humidity control, is assigned to the humidity control secondary attribute field under the root node; the heat preservation timing action, related to equipment timing control, is assigned to the timing control secondary attribute field under the root node; the material injection action and reaction termination action, being basic equipment operation actions and not specific to any particular control function, are directly assigned to the constant temperature and humidity chamber root node, at the same level as the secondary control attribute field. During the assignment process, the logical matching between the sub-steps and the assignment levels must be verified to ensure that all sub-steps belong to the constant temperature and humidity chamber root node and that the hierarchy conforms to the equipment processing function classification.

[0082] Step 7: Use the first influence probability distribution determined in Step 4 as the association attribute between child nodes, and enter it into the association field of the corresponding adjacent child nodes in the tree database. Specifically, find adjacent sub-step records with a sequential order in the database, such as the preheating action sub-node of the constant temperature and humidity chamber and the temperature setting action sub-node, and enter the influence rule based on the equipment processing logic in the association field of the two. For all other adjacent sub-steps, enter the corresponding influence rule in the same way to ensure that each influence rule is consistent with the interaction logic of the constant temperature and humidity chamber's own processing action and is accurately associated with the corresponding adjacent child node.

[0083] Step 8: Using the hierarchical construction function of the tree database, first establish a hierarchical association between the root node of the constant temperature and humidity chamber and its three subordinate secondary control attribute fields. Then, establish a subordinate association between each sub-step and its corresponding root node or secondary control attribute field. Finally, associate the influence rules between adjacent child nodes with the corresponding child node records to form a complete hierarchical structure of root node (constant temperature and humidity chamber) - secondary control attribute - sub-step (equipment processing action) - influence rules between child nodes.

[0084] Based on this structure, a forward process flow tree of the standard process steps for isothermal reaction is generated, ensuring that the tree structure fully conforms to the principle that one standard process step corresponds to one production equipment root node, and that the equipment completes the implementation process of all sub-steps of the step, with clear and traceable logic at each level.

[0085] Based on the local monitoring indicators corresponding to each sub-step of the process flow tree for each standard process step, combined with the full-process monitoring indicators of each standard process step in the corresponding production period, the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step are obtained through factor analysis algorithm. The contribution of the detection accuracy of the corresponding local monitoring indicators to the deviation values ​​of the full-process monitoring indicators in the corresponding time period and to the deviation values ​​of the related sub-steps under the remaining standard process steps are also considered.

[0086] Step 1: Based on the process flow tree of each standard process step, collect the local monitoring index data corresponding to the sub-nodes constructed by all standard process sub-steps under this step, specifically including the actual detection value, preset standard value and detection accuracy of each local monitoring index; at the same time, collect the actual detection value and preset standard value of the full-process monitoring index of this standard process step under the corresponding production period, to ensure that the collected data covers the complete information of the local monitoring of sub-steps and the full-process monitoring of steps, and that the data source is consistent with the logical association between the sub-nodes and steps in the process flow tree.

[0087] Step 2: First, for each local monitoring indicator corresponding to a standard process sub-step, calculate the difference between the actual measured value and the preset standard value to obtain the deviation value of each local monitoring indicator; then, for the full-process monitoring indicator corresponding to the standard process step, calculate the difference between the actual measured value and the preset standard value to obtain the deviation value of the full-process monitoring indicator for the corresponding production period, ensuring that the deviation value calculation logic is consistent and can reflect the direction and degree of deviation between the actual value and the standard value.

[0088] Step 3: Process the calculated deviation values ​​of local monitoring indicators and full-process monitoring indicators using the specified standardization method to eliminate the impact of differences in the dimensions of different indicators on subsequent analysis, so that the standardized data has a unified statistical analysis basis; at the same time, convert the detection accuracy of local monitoring indicators into corresponding standardized weight coefficients, wherein the higher the detection accuracy of the local monitoring indicator, the larger the corresponding standardized weight coefficient, so as to reflect the degree of influence of detection accuracy on the deviation value analysis results.

[0089] Step 4: Based on the standardized deviation values ​​of local monitoring indicators and the deviation values ​​of the whole process monitoring indicators, calculate the correlation coefficients between the deviation values ​​of local monitoring indicators and the correlation coefficients between the deviation values ​​of local monitoring indicators and the deviation values ​​of the whole process monitoring indicators, and integrate them to form a first correlation matrix, which is used to reflect the degree of correlation between local monitoring indicators and the whole process monitoring indicators. At the same time, according to the logical relationship between the current standard process step and the remaining standard process steps in the process flow tree, determine the related sub-steps under the remaining standard process steps, collect the deviation values ​​of local monitoring indicators of these related sub-steps and perform standardization processing, calculate the correlation coefficients between the deviation values ​​of local monitoring indicators of the current step and the deviation values ​​of local monitoring indicators of the remaining related sub-steps, and integrate them to form a second correlation matrix, which is used to reflect the degree of correlation between the local monitoring indicators of the current step and the local monitoring indicators of the remaining related sub-steps.

[0090] Step 5: Perform eigenvalue decomposition on the constructed first correlation matrix and second correlation matrix respectively to obtain the eigenvalues ​​corresponding to each matrix; based on the statistical significance of the eigenvalues, select the vectors corresponding to the eigenvalues ​​that meet the preset conditions as common factors, and determine the number of common factors corresponding to the first correlation matrix and the second correlation matrix respectively, to ensure that the selected common factors can effectively reflect the main information of the original data and avoid information omission or redundancy.

[0091] Step 6: Use principal component analysis to extract common factors from the first and second correlation matrices respectively. For the first correlation matrix, extract the corresponding number of common factors through principal component analysis to obtain the initial factor loading matrix between the local monitoring index deviation values ​​and the common factors. For the second correlation matrix, similarly extract the corresponding number of common factors through principal component analysis to obtain the initial factor loading matrix between the local monitoring index deviation values ​​and the common factors, ensuring that the extracted common factors can accurately reflect the correlation characteristics in the original correlation matrix.

[0092] Step 7: Perform maximum variance rotation on the two initial factor loading matrices obtained after extracting common factors; by adjusting the direction of the factor axes, make each local monitoring indicator deviation value show a higher loading on a few common factors and a lower loading on other common factors, thereby improving the actual interpretability of the common factors and ensuring that the rotated common factors can more clearly correspond to the specific monitoring indicator influence dimensions.

[0093] Step 8: Standardize the weighted coefficients of the local monitoring indicators obtained in Step 3, and multiply them by the loading coefficients of the corresponding local monitoring indicators in the rotated first factor loading matrix and second factor loading matrix in Step 7, respectively. Through this operation, the influence of detection accuracy is incorporated into the factor loading, resulting in the weighted first factor loading matrix and the weighted second factor loading matrix, ensuring that the subsequent contribution calculation can reflect the influence of detection accuracy on the results.

[0094] Step 9: For the weighted first factor loading matrix, calculate the proportion of the loading coefficient corresponding to the deviation value of each local monitoring index according to the direction of the common factor column. Summate the proportions of the loading coefficients of all local monitoring indices under the same common factor column, and normalize the summation result to obtain the contribution of each local monitoring index deviation value to the deviation value of the monitoring index of the entire time period corresponding to the standard process step, combined with the corresponding detection accuracy. For the weighted second factor loading matrix, first calculate the proportion of the loading coefficient corresponding to the deviation value of each local monitoring index according to the direction of the common factor column. Then, according to the logical association weights of the current standard process sub-step and the related sub-steps under the remaining standard process steps in the process flow tree, multiply each load coefficient proportion with the corresponding logical association weight. Summate the multiplication result and normalize it to obtain the contribution of each local monitoring index deviation value to the deviation value of the related sub-steps under the remaining standard process steps, combined with the corresponding detection accuracy.

[0095] Step 10: Systematically organize the two types of contribution values ​​calculated in Step 9 to form a contribution analysis report for each standard process step. The report clearly identifies the local monitoring indicators corresponding to all standard process sub-steps under this step, and marks the contribution of each local monitoring indicator to the deviation value of the overall monitoring indicator and the contribution to the deviation value of the remaining related sub-steps, ensuring that the contribution results are clearly presented and can be directly used for subsequent process optimization and anomaly analysis.

[0096] Based on the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of the full monitoring indicators for the corresponding time period, a vertical internal derivation connection is constructed using the Hidden Markov algorithm.

[0097] It should be further explained that the process of constructing the vertical internal derivation connection using the Hidden Markov algorithm in this embodiment includes:

[0098] Step 1: Determine the target standard process step for which the vertical internal derivation connection needs to be constructed. Collect the local monitoring index deviation data corresponding to all standard process sub-steps under this step. The data sources include real-time deviation data of the current production cycle and deviation data of multiple historical production cycles. At the same time, collect the full-process monitoring index deviation data of this step under the corresponding production period to ensure that the local monitoring index deviation data and the full-process monitoring index deviation data maintain a strict correspondence in the time dimension.

[0099] Step 2: Based on the fluctuation range of the deviation values ​​of local monitoring indicators, define the set of hidden states of the model. This set includes three states: normal state, mildly abnormal state, and severely abnormal state. The normal state corresponds to a deviation value within the preset normal threshold range. The mildly abnormal state corresponds to a deviation value exceeding the normal threshold but not reaching the severe threshold. The severely abnormal state corresponds to a deviation value exceeding the severe threshold. Based on the fluctuation range of the deviation values ​​of the entire monitoring indicators, define the set of observed states of the model. The observed states also include three states: normal state, mildly abnormal state, and severely abnormal state. The threshold division adopts the same threshold interval division principle as the hidden states.

[0100] Step 3: Retrieve the contribution data of the deviation values ​​of each local monitoring index to the deviation values ​​of the whole process monitoring index under each target standard process step, sort each local monitoring index according to the size of the contribution, normalize the contribution values, and use the processing results as the initial weights of the corresponding local monitoring index deviation state in the hidden Markov model.

[0101] Step 4: Construct the hidden state transition probability matrix. Based on historical local monitoring indicator deviation data, statistically analyze the transition frequencies between each hidden state. The statistics include the frequency of transitions from a normal state to a slightly anomalous state, the frequency of transitions from a slightly anomalous state to a severely anomalous state, and all other possible state transition scenarios. Using the initial weights determined in Step 3, the statistically obtained transition frequencies are weighted and corrected, with higher-weighted local monitoring indicators receiving larger correction coefficients. The corrected transition frequencies are then converted into probability values, forming the hidden state transition probability matrix of the Hidden Markov Model.

[0102] Step 5: Based on the correspondence between historical local monitoring index deviation data and whole-process monitoring index deviation data, count the frequency of occurrence of each hidden state corresponding to each observation state. The count includes all possible correspondences such as the frequency of whole-process monitoring index being in a slightly abnormal state when the local monitoring index is in a slightly abnormal state, and the frequency of whole-process monitoring index being in a severely abnormal state when the local monitoring index is in a severely abnormal state. Similarly, combined with the initial weights determined in Step 3, the count is weighted and corrected, and the corrected frequency is converted into probability values ​​to form the emission probability matrix of the Hidden Markov Model.

[0103] Step 6: Use the historical local monitoring indicator deviation state sequence as the hidden state training sequence for the model input, and the historical full-process monitoring indicator deviation state sequence as the observed state training sequence for the model input. Substitute these two sets of sequences into the initially constructed state transition probability matrix and emission probability matrix, and iteratively optimize the matrix parameters using the Baum-Welch algorithm. The optimization process aims to minimize the error between the model's predicted observed state and the actual observed state. Iteration stops when the error value is lower than a preset threshold. At this point, the trained Hidden Markov Model is obtained.

[0104] Step 7: Select local monitoring indicator deviation data from a new production cycle that has not participated in training, convert the data into a hidden state sequence, and input it into the trained Hidden Markov Model. The model predicts the observed state sequence of the entire monitoring indicator deviation. The predicted observed state sequence is compared with the actual state sequence of the entire monitoring indicator deviation for the production cycle, and the prediction accuracy is calculated. If the accuracy is higher than the preset qualified threshold, the model is validated. If the standard is not met, return to step 6 to iteratively optimize the model parameters again.

[0105] Step 8: Using the validated Hidden Markov Model as the core, define the input and output rules for the derivation connection. The input rule specifies that the input is the sequence of hidden states transformed from the real-time collected local monitoring index deviation values ​​under the target standard process step. The output rule specifies that the output is the observed state of the full-process monitoring index deviation predicted by the model and the corresponding probability value. Simultaneously, a reverse derivation rule is defined. This rule specifies that the input is the sequence of observed state of the full-process monitoring index deviation, and the output is the most likely hidden state sequence of local monitoring index deviations and the contribution ratio of each local index. Through the definition of the input, output, and reverse derivation rules, a vertical internal derivation connection between the local monitoring index deviations and the full-process monitoring index deviations under the target standard process step is finally formed.

[0106] The vertical internal derivation connection is embedded into the forward process flow tree of each standard process sub-step. Simultaneously, the cross-inference connection, constructed by the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of related sub-steps under the remaining standard process steps, is embedded into the sub-steps with control relationships within the forward process flow tree under different standard process sub-steps. To better illustrate the process of this embodiment, an exemplary process is as follows:

[0107] Step 1: Determine the set of forward process flow trees to be embedded and connected. Taking the forward process flow trees of the raw material crushing standard process steps and the isothermal reaction standard process steps for premix production as examples, clarify the identifiers, hierarchical relationships, and existing secondary control attributes of each sub-step within the two flow trees. These secondary control attributes include the fineness control attributes of the raw material crushing tree and the temperature control attributes of the isothermal reaction tree.

[0108] Step 2: For the forward process flow tree of the isothermal reaction standard process step, retrieve the vertical internal derivation connections already constructed for that step. Add a vertical derivation rule subfield to the attribute fields of each sub-step node within the tree, and enter the input / output logic of the vertical internal derivation connections. Simultaneously, add a vertical derivation traceability field under the whole-process monitoring indicator association node of this flow tree, associating it with the sub-step node with the highest contribution, forming a hierarchical association of sub-step-vertical connection-whole-process indicator. The vertical internal derivation connections retrieved here are used to correlate the quantitative relationship between the local monitoring indicator deviation of the sub-step and the whole-process monitoring indicator deviation of the corresponding standard process step.

[0109] Step 3: Based on the contribution data of the local monitoring index deviation values ​​of the raw material grinding fineness control sub-step to the local monitoring index deviation values ​​of the temperature setting sub-step under the isothermal reaction step, clarify the control correlation between the two sub-steps. Using this contribution as the core, formulate cross-reasoning rules. The rules include inputting the grinding fineness control deviation state and, combined with the contribution value, deriving the possible states and correlation probabilities of the output temperature setting deviation, forming a cross-reasoning connection from raw material grinding – grinding fineness control to isothermal reaction – temperature setting. The contribution data here reflects the degree of influence of grinding fineness deviation on subsequent temperature setting deviation.

[0110] Step 4: Add a cross-deduction association node field to the attribute of the fineness control sub-step node in the forward process flow tree of the raw material crushing standard process steps, and enter the unique identifier of the temperature setting sub-step in the isothermal reaction process flow tree and the identifier of its respective process tree. Simultaneously, add a reverse cross-deduction source node field to the attribute of the temperature setting sub-step node in the isothermal reaction process tree, and enter the unique identifier of the fineness control sub-step in the raw material crushing process tree. Enter the cross-deduction rule fields of the two sub-step nodes according to the cross-deduction rule definition in Step 3, ensuring that the rule content is consistent with the contribution logic, thus achieving bidirectional cross-connection between related sub-steps in different process trees. This bidirectional connection ensures bidirectional traceability for anomaly tracing.

[0111] Step 5: Traverse the two forward process flow trees of raw material crushing and isothermal reaction, checking whether the vertical internal derivation connections are associated with all relevant sub-steps and full-process monitoring index nodes within the isothermal reaction tree, and whether the cross-reasoning connections accurately associate the control-related sub-steps within the two trees. By simulating the input of the crushing fineness control sub-step deviation state, verify whether the temperature setting sub-step deviation state can be derived through cross-reasoning connections. Simultaneously, by simulating the input of the temperature setting sub-step deviation state, verify whether the vertical internal derivation connections can derive the full-process monitoring index deviation state of the isothermal reaction, ensuring that the embedded connections can normally realize the reasoning function and do not conflict with the original hierarchical logic of the process tree. This verification process is a key step in ensuring system reliability.

[0112] Based on the processing control parameters and standard processing time length corresponding to each sub-step in the forward process flow tree of each standard process step, a forward processing control tree for each standard process step is obtained. At the same time, the process-control connection is established by combining the standard degree of the execution action corresponding to each processing control parameter with the deviation value of the local monitoring index of the corresponding standard process sub-step.

[0113] The forward processing control tree of each standard process step under the initial full-process tracking chain is mapped to the corresponding forward process flow tree through the process-control connection to obtain the full-process tracking chain containing the process-control association tree;

[0114] The full-process tracking chain, which includes a process-control association tree, is combined with random forest and simulation algorithms. The positional deviation and corresponding timestamp deviation of the corresponding outlier points and their cumulative outlier points are used as the derivation loss function for simulation training to obtain a fully trained full-process tracking chain.

[0115] In this embodiment, to better illustrate the end-to-end tracking chain, such as Figure 2 As shown, A, B, ..., N are the root nodes corresponding to the entire standard process flow. A1, A2, A3, ..., Am are the sub-step nodes constructed from the m standard process sub-steps corresponding to the root node A. The bidirectional arrows connecting A1, A2, A3, ..., Am form a forward process flow tree for each standard process step. The upward arrows represent vertical internal derivation connections. A11, A22, A33, ..., Amm are the control parameter nodes corresponding to each sub-step node. The connection between A1 and A11 is a process-control connection. The corresponding structure from root node B to root node N is the same as the corresponding structure from root node A, so it will not be elaborated further here. A1 to B2 or Am to B3 are cross-reasoning connections between different root nodes. From left to right, it is the control logic flow corresponding to the associated sub-steps. From right to left, it is the reverse derivation flow.

[0116] It should be further explained that the process of obtaining the reverse anomaly information fingerprint chain in this embodiment includes:

[0117] The total quality inspection score of each standard process step is compared with the preset threshold for the total quality inspection score, and the production efficiency deviation is compared with the preset threshold for the production efficiency deviation. The first standard process step with a total quality inspection score lower than the corresponding preset threshold or a production efficiency deviation higher than the corresponding preset threshold is selected to obtain the abnormal starting standard process step.

[0118] Based on the abnormal starting standard process step and the full-process monitoring indicators of all standard process steps before this step, as well as the timestamps of each standard process step, the total quality inspection score and production efficiency deviation of all steps are extracted. The two sets of indicator values ​​are arranged in order from back to front according to the timestamps of each standard process step and the corresponding total quality inspection score and production efficiency deviation, to obtain the observation sequence of the Hidden Markov Algorithm.

[0119] Retrieve the pre-stored state transition probability matrix between each standard process step from the full tracking chain; the state transition probability matrix includes the positive cumulative impact probability of abnormal total quality inspection score between any adjacent steps, the negative cumulative impact probability of abnormal total quality inspection score, the positive cumulative impact probability of abnormal production efficiency deviation between adjacent steps, and the negative cumulative impact probability of abnormal production efficiency deviation.

[0120] The total quality inspection score and production efficiency deviation value of each step in the observation sequence are input into the back-inference model constructed by the Hidden Markov Algorithm. With the goal of maximizing the matching degree between the abnormal index values ​​in the observation sequence and the cumulative influence probability in the state transition probability matrix, the hidden state of each preceding standard process step is derived one by one, starting from the abnormal initial standard process step. The specific hidden state type and corresponding timestamp of each standard process step are recorded. The derivation stops when the cumulative sum of the deviations between the total quality inspection score and the production efficiency deviation is 0. The root node with the first deviation of 0 is taken as the starting deviation node, and the specific hidden state and corresponding timestamp of each standard process step are obtained. The hidden states are: the initial state of abnormal total quality inspection score, the state of abnormal transmission of abnormal total quality inspection score, the initial state of abnormal production efficiency deviation, the state of abnormal transmission of abnormal production efficiency deviation, and the normal production state.

[0121] Based on the specific hidden state type, corresponding timestamp, and starting deviation node of each standard process step obtained through derivation, the reverse state sequence representing the entire dimensional anomaly propagation path is obtained by arranging the specific hidden state type-timestamp in the order of the timestamps of each standard process step from back to front.

[0122] Based on the reverse state sequence representing the entire dimensional anomaly propagation path, the first influence probability distribution under each root node on the path, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node are extracted.

[0123] Based on the cumulative deviation value of the full-process monitoring index of each root node and its forward root node in the reverse state sequence, combined with the first influence probability distribution under each node, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node, the abnormal deviation index value of each local detection index under each node is obtained by inversion.

[0124] It should be further explained that the process of obtaining the cumulative deviation value of the full-process monitoring index of each root node and its preceding root node in this embodiment includes:

[0125] To obtain the cumulative deviation value of the full-process monitoring indicators for each root node and its forward root node in the reverse state sequence, the reverse state sequence must be used as a basis. This sequence includes the specific hidden state of the standard process step corresponding to each root node, the corresponding timestamp, and the starting deviation node, i.e., the first root node with a deviation value of 0. First, retrieve the full-process monitoring indicators recorded in real time during the production process for the standard process step corresponding to each node, including the total quality inspection score and the production efficiency deviation. At the same time, extract the preset standard values ​​for each type of indicator. By calculating the difference between the real-time monitoring value and the preset standard value, the basic deviation value of the full-process monitoring indicator for each node is obtained.

[0126] Secondly, by combining the hidden state types of each root node in the reverse state sequence, including the initial state of abnormal total quality inspection score, the state of abnormal total quality inspection score transmission, the initial state of abnormal production efficiency deviation, the state of abnormal production efficiency deviation, and the normal production state, the attribute characteristics of the deviation of the monitoring indicators throughout the entire process of each root node are clarified. At the same time, the pre-stored state transition probability matrix between each standard process step is retrieved. This matrix covers the positive and negative cumulative impact probabilities of abnormal total quality inspection score and abnormal production efficiency deviation between adjacent steps. Based on this, the cumulative direction of the deviation of the monitoring indicators throughout the entire process between the current root node and the previous root node is determined. Positive accumulation refers to the positive promoting effect of the monitoring indicators of the preceding standard process step being better than the preset standard on the subsequent steps. Negative accumulation refers to the negative impact of the monitoring indicators of the preceding standard process step not meeting the preset standard on the subsequent steps. At the same time, the specific cumulative impact probability under the corresponding accumulation direction is determined.

[0127] Finally, using the cumulative deviation value of the entire monitoring index of the initial deviation node as a benchmark (this benchmark value is set to 0), iterative calculations are performed in reverse state sequence from back to front according to the timestamps of each root node: for each root node, its own basic deviation value of the entire monitoring index is superimposed with the cumulative deviation value passed from the forward root node to this node. In the positive accumulation scenario, the positive influence value is superimposed, and in the negative accumulation scenario, the negative influence value is superimposed. Through the above iterative calculation method, the cumulative deviation value of the entire monitoring index of each root node and its forward root node is finally obtained. This value comprehensively reflects the total influence of the entire monitoring index deviation of the forward root node passed through the accumulation relationship from the initial deviation node to the current node, and the superposition effect of the current node's own basic deviation value.

[0128] Based on the abnormal deviation index value of each local detection index under each node, combined with the process-control connection, the standard deviation value of the corresponding processing process execution action and the deviation value of the standard processing time length of the corresponding execution action are obtained by inversion.

[0129] Based on the reverse state sequence of the full-dimensional anomaly propagation path, the corresponding equipment label information, processing technology execution priority, and the standard deviation value of the corresponding processing technology execution action and the deviation value of the standard processing time length of the corresponding execution action in the local monitoring indicators are combined with the graph algorithm to construct the reverse anomaly information fingerprint chain.

[0130] It should be further explained that the process of adjusting the forward processing control parameters from the starting abnormal node in the reverse abnormal information fingerprint chain in this embodiment includes:

[0131] Step 1: Identification of the first deviation root node and extraction of fingerprint chain information. Based on the reverse anomaly information fingerprint chain (this chain extends backward from the starting node of the over-threshold anomaly to the first node without global or local deviation, covering all root nodes from the first deviation to the over-threshold anomaly, and each root node is associated with the forward processing control tree of the corresponding standard process step), first locate the "first deviation root node" in the chain. This node is the source of the entire deviation transmission chain. The deviations of subsequent nodes are directly or indirectly caused by the deviation of this node. It is determined as the starting point of the correction process. Next, key information is extracted from the first deviation root node and all subsequent root nodes in the chain (up to the abnormal node exceeding the threshold), including the standard process step identifier corresponding to each root node, the local standard process sub-step identifier, the local monitoring index type of the deviation or abnormal state, the full-process monitoring index type of the deviation or abnormal state, and the timestamp of the deviation or abnormality. At the same time, the forward processing control tree of the standard process step associated with each root node is extracted (the tree contains core control information such as specific processing control parameters and standard processing time length). The information collection of the starting correction node and all root nodes to be corrected in the chain is completed, and finally, the complete process association information of the first deviation root node and the list of nodes to be corrected in the chain are formed.

[0132] Step 2: Retrieve the process-control mapping relationship of the first deviation root node. Based on the process association information of the first deviation root node and the entire tracking chain, match according to the hierarchical logic of standard process step identifier → corresponding root node → forward machining control tree associated with the root node to accurately locate the standard process step to which the first deviation root node belongs and the forward machining control tree bound to it. From the forward machining control tree, further retrieve the machining control parameters (including the current running value and preset standard value) of the local standard process sub-step corresponding to the root node, as well as the mapping relationship between these machining control parameters and local monitoring indicators and overall monitoring indicators. This mapping relationship needs to clearly define the specific impact logic of parameter adjustment on indicator changes (e.g., an increase in a certain parameter will directly cause an improvement in a certain local indicator). At the same time, retrieve the process-control mapping relationship of all subsequent root nodes to be corrected in the chain in advance to prepare data for subsequent forward sequence correction, ensuring that all retrieved mapping relationships are consistent with the forward machining control tree parameters of the corresponding root node, and finally obtain the complete process-control mapping relationship of the first deviation root node and subsequent nodes in the chain.

[0133] Step 3: Initial Correction Strategy Matching for the First Deviation Root Node. Matching is performed based on the local and full-process deviation monitoring index types of the first deviation root node, combined with a pre-defined abnormal processing correction strategy library. The matching process must adhere to the core principle of "correspondence between the deviation type and the parameter type within the forward processing control tree of the first deviation root node"—for example, if the deviation index is temperature-related, strategies containing temperature parameter adjustment rules within the forward processing control tree should be prioritized. The selected adjustment rules must clearly define the adjustment type (e.g., speed adjustment, temperature adjustment, timing adjustment), adjustment direction (e.g., parameter increase, parameter decrease, time extension), and adjustment range (the range setting should refer to the standard values ​​of parameters within the forward processing control tree and the equipment's operating characteristics) for the parameters in the forward processing control tree of that root node. Multiple matched correction strategies are prioritized according to "efficiency of parameter adjustment in improving deviation," and the strategy with the highest improvement efficiency and the strongest adaptability to the parameters in the forward processing control tree is selected as the initial correction strategy for the first deviation root node.

[0134] Step 4: Initial Adjustment of Processing Control Parameters at the First Deviation Root Node. Based on the process-control mapping relationship of the first deviation root node, the initial correction strategy, and the current local and overall monitoring index deviation values ​​of this node, firstly, extract the specific processing control parameters consistent with the parameter type adjusted by the initial correction strategy from the forward processing control tree associated with the first deviation root node. Calculate the median value of the adjustment range determined in the initial correction strategy as the initial adjustment range—during the calculation, the safe operating range of this parameter within the forward processing control tree must be referenced simultaneously to ensure that the initial adjustment range does not exceed the equipment's permissible safety parameter boundaries. Adjust the corresponding processing control parameters within the forward processing control tree according to the calculated initial adjustment range to generate the adjusted processing control parameters. Simultaneously, record the parameter values ​​before and after adjustment, the adjustment time, and the adjustment range in detail, and link them to the parameter change record of the forward processing control tree for this root node to ensure the entire parameter adjustment process is traceable.

[0135] Step 5: Monitoring the effect of the first deviation root node adjustment. Based on the adjusted processing control parameters, drive the production equipment corresponding to the first deviation root node to execute those parameters. Set a reasonable real-time monitoring frequency according to the processing time of the local standard process sub-step corresponding to the first deviation root node. Increase the monitoring frequency appropriately for shorter processing times and decrease it appropriately for longer processing times. The core principle is to ensure at least three valid data acquisitions are completed during the full execution of this sub-step. Collect the local monitoring indicators of this local standard process sub-step and the full-process monitoring indicators of its corresponding standard process step in real time according to the set monitoring frequency. During the acquisition process, the parameter operating status of the forward processing control tree of the first deviation root node (such as whether the parameters are in a stable operating state and whether there are abnormal fluctuations) needs to be recorded simultaneously. Abnormal fluctuation data caused by equipment momentary failures or sensor errors are discarded, and finally, valid real-time monitoring indicators reflecting the true adjustment effect are retained.

[0136] Step 6: Determine the effect of the first deviation root node adjustment. Compare the adjusted effective real-time monitoring indicators with the preset local monitoring indicator thresholds and the anomaly detection thresholds for the entire monitoring indicator. First, determine whether the local monitoring indicators have reached the "complete elimination of deviation" standard, i.e., the indicator value falls within the normal range without deviation. If the local monitoring indicators meet the standard, further determine whether the entire monitoring indicators have improved synchronously to avoid situations where local indicators meet the standard but the entire indicator still has deviation. For local or entire indicators that are in a "critical state close to the zero-deviation threshold but not completely met," a secondary data collection and confirmation step is required. After the sub-step is completed, the monitoring indicators are collected again to verify whether the indicators continue to change towards the standard. If both local and entire indicators meet the standard, the parameter stability must be verified through the positive processing control tree of the first deviation root node, i.e., confirm that the adjusted parameters have run continuously for a complete processing cycle without fluctuation. Finally, determine that the parameter adjustment of the first deviation root node is effective and the deviation has been completely eliminated.

[0137] Step 7: Iterative optimization of the first deviation root node correction strategy. If the adjustment effect is determined to be substandard, first calculate the deviation value between the real-time monitoring index and the no-deviation threshold to clarify the gap between the current index and the compliant state. From the process-control mapping relationship of the first deviation root node, retrieve the influence coefficient of the adjustment parameter on the monitoring index within the forward machining control tree. This coefficient reflects the specific degree of improvement of the index by each adjustment of the parameter by a certain amount. Based on the deviation value and the influence coefficient, determine the correlation ratio between the deviation value and the parameter adjustment range. That is, the larger the deviation value, the higher the expansion ratio of the parameter adjustment range, ensuring that the adjustment range can accurately match the deviation improvement needs. Update the adjustment range of the initial correction strategy based on this correlation ratio. During the update process, the maximum allowable adjustment range of the parameter within the forward machining control tree must be strictly followed to avoid the adjustment range exceeding the safe operating boundary of the equipment. Readjust the machining control parameters of the first deviation root node according to the updated correction strategy, and repeat steps 5-6 until the deviation of the first deviation root node is completely eliminated.

[0138] Step 8: Determine the propagation of deviations in subsequent root nodes within the chain. After the deviation of the first deviation root node is eliminated, the subsequent root nodes (arranged in the forward time sequence of the production process) in the reverse anomaly information fingerprint chain are used for determination, combined with the cross-inference connection information in the full-process tracking chain. First, the influence probability distribution between the first deviation root node and each subsequent root node is retrieved from the cross-inference connection information to clarify the possible influence of the first deviation root node on subsequent nodes. Next, it is analyzed whether there is a logical propagation relationship between the deviation type of the subsequent root node and the deviation type of the first deviation root node. For example, if the first deviation root node is a temperature deviation and the subsequent root node is a humidity deviation, and from the process logic, temperature changes directly affect humidity stability, then it is determined that there is a deviation propagation logic between the two. If the influence probability reaches the preset judgment standard and the deviation type has a clear propagation logic, then it is determined that the deviation of the subsequent root node is caused by the propagation of the first deviation root node; if the influence probability does not meet the standard or the deviation type has no propagation logic, then it is determined that the deviation of the subsequent root node is an independent deviation, and a separate correction strategy needs to be formulated.

[0139] Step 9: Formulate a derivative correction strategy for the root node of the transitive deviation. If the deviation of a subsequent root node is caused by transitivity, first, retrieve the forward processing control tree associated with that subsequent root node and extract the processing control parameters directly related to the deviation index from the tree, ensuring that the extracted parameters are the core control factors affecting the deviation of that node. Based on the probability distribution of the influence of the first deviation root node and the subsequent root node, calculate the probability ratio of influence, that is, the proportion of the probability of the first deviation root node influencing the subsequent node to the total probability of all deviation influencing factors of the subsequent node. Based on this probability ratio, combined with the correction strategy of the first deviation root node, determine the adjustment range of the processing control parameters of the subsequent root node. The higher the probability ratio, the closer the adjustment range is to the correction range of the first deviation root node, ensuring that the adjustment intensity matches the degree of influence of the deviation transitivity. At the same time, refer to the sensitivity characteristics of the parameter in the forward processing control tree of the subsequent root node (such as the sensitivity of the parameter to changes in the index) to optimize the adjustment direction, avoiding over-adjustment due to high parameter sensitivity, and finally generate the derivative correction strategy for the root node of the transitive deviation.

[0140] Step 10: Adjustment and Verification of Transitivity Deviation Root Node Parameters. Based on the derived correction strategy and the current deviation index value of subsequent root nodes, extract the processing control parameters to be adjusted from the forward processing control tree of this node, and adjust the parameters according to the adjustment range determined by the derived correction strategy. Drive the production equipment corresponding to this subsequent root node to execute the adjusted parameters. Refer to the monitoring logic in Step 5 to collect the local monitoring index of the local standard process sub-step corresponding to this node and the full-process monitoring index of the standard process step to which it belongs, to ensure that the collected data truly reflects the adjustment effect. Compare the collected monitoring index with the no-deviation threshold to determine whether the deviation has been eliminated; at the same time, additionally verify whether the adjustment of the node parameters has an improvement effect on the deviation of the subsequent root nodes (if they exist) in the reverse anomaly information fingerprint chain, to avoid new deviation transmission caused by the adjustment of this node. If the deviation is not eliminated, repeat the strategy optimization logic in Step 7, update the derived correction strategy and readjust the parameters; if the deviation is eliminated, proceed to the next subsequent root node in forward time sequence, repeat Steps 8-10, until the deviations of all transitivity deviation root nodes in the chain are completely eliminated.

[0141] The material traceability and management system for premixed agent production lines provided in this application addresses the core pain points of traditional management methods in premixed agent production, such as scattered data, inefficient anomaly location, and blind parameter adjustments. Through multi-module collaboration and key technological innovation, it forms a closed-loop control capability across the entire process. In particular, from the perspective of data integration and traceability, the system constructs a full-process tracking chain using a hierarchical parsing algorithm. It breaks down the standard production process information of the target product into a global standard process step sequence and a local standard process sub-step sequence according to the corresponding process flow of the equipment. Simultaneously, it integrates the mapping relationships between local monitoring indicators of each sub-step, overall monitoring indicators, indicator detection accuracy, and processing control parameters, thoroughly solving the problems of scattered data collection and lack of systematic correlation in traditional management. This integration is not a simple data accumulation, but rather uses factor analysis algorithms to calculate the contribution of the deviation of local monitoring indicators in each sub-step to the deviation of overall monitoring indicators, as well as its contribution to the deviation of remaining related sub-steps. This ensures that each data point has a clear impact logic and traceability path, providing accurate data support for subsequent anomaly analysis and parameter adjustment, and avoiding traceability gaps caused by data fragmentation. Regarding the accuracy of anomaly localization, the system's reverse mapping module, based on the full-process fingerprint information of the abnormal production process, combines hidden Markov algorithms for reverse full-process inference. Simultaneously, it utilizes the mapping relationship between full-process and local monitoring indicators to conduct vertical local inference. The constructed reverse anomaly information fingerprint chain can trace from the starting node of the threshold-exceeding anomaly to the first root node where a deviation occurred, clearly presenting the complete transmission path from the occurrence of the deviation to the anomaly exceeding the standard. Traditional methods often only detect threshold-exceeding anomalies but struggle to locate the source of the deviation and intermediate transmission nodes, resulting in a wide investigation scope and low efficiency. This system, through state transition probability matrices and hidden state derivation, can not only pinpoint the first deviation root node but also clearly identify the anomaly type (initial anomaly or transmitted anomaly) of each node, transforming anomaly localization from fuzzy investigation to precise source tracing, significantly shortening anomaly localization time and reducing production losses caused by blind shutdowns for investigation. Regarding the effectiveness and efficiency of parameter adjustment, the system's forward adjustment module uses the first deviation root node in the reverse anomaly information fingerprint chain as the correction starting point, combining the process-control mapping relationship in the full-process tracking chain with a pre-set anomaly processing correction strategy library to formulate an initial correction strategy. Unlike traditional adjustments that rely on experience to set the range and are prone to over- or under-adjustment, this system determines the adjustment range by mapping the parameters and indicators in the forward processing control tree of the first deviation root node, combined with the probability distribution of the influence between sub-steps. The first adjustment is based on scientific evidence. If the indicator does not meet the standard, the adjustment range can be dynamically optimized based on the deviation value and the influence coefficient to ensure that the parameter adjustment accurately matches the deviation improvement needs.Meanwhile, for transitive deviation nodes, the system formulates derivative correction strategies by cross-referencing information and the probability of impact, avoiding the recurrence of anomalies caused by only correcting the root node of the initial deviation while ignoring subsequent transit nodes. This achieves full-chain correction from the source of the deviation to all related nodes, enabling monitoring indicators to quickly reach the target, improving correction efficiency, and reducing product quality fluctuations caused by repeated parameter adjustments. From the perspective of improving overall production efficiency, the above-mentioned data integration, precise positioning, and effective adjustment form a synergistic closed loop: the systematic data support provided by the full-process tracking chain enables the reverse mapping module to achieve precise anomaly tracing; the precise tracing results provide clear correction targets for the positive adjustment module; and effective adjustment ensures that production indicators quickly return to normal. The combined effect of these three factors transforms the premix production process from "passively responding to anomalies" to "proactive prevention and precise control." On the one hand, by reducing the time for anomaly investigation and parameter adjustment, production efficiency is improved, and downtime and material loss are reduced; on the other hand, by precisely controlling the monitoring indicators of each process step, product quality substandardities caused by deviation transmission are avoided, improving product quality stability. Furthermore, the system trains and optimizes the entire tracking chain through random forest and simulation algorithms, enabling the system to continuously learn. As production data accumulates, the accuracy of anomaly location and parameter adjustment can be further improved, forming a virtuous cycle of production control and providing strong support for premixed agent manufacturers to achieve high-quality, high-efficiency, and low-cost production goals.

[0142] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. An anomaly tracing and parameter control system for a premix production line, characterized in that, include: Monitoring module, reverse mapping module, and forward adjustment module; The monitoring module is used to obtain the full fingerprint information of the abnormal production process based on the preset full-process tracking chain and the preset full-process monitoring indicator anomaly judgment threshold. The full-process fingerprint information includes the timestamp of each standard process step and the corresponding full-process monitoring indicator and local monitoring indicator, the mapping relationship between the full-process monitoring indicator and local monitoring indicator of each standard process step, and the mapping relationship between the full-process monitoring indicator and local monitoring indicator, the detection accuracy of each indicator and the processing control parameters of the corresponding standard process step. The reverse mapping module is used to perform reverse full-process inference based on the full fingerprint information of the abnormal production process combined with the Hidden Markov Algorithm, and at the same time to perform vertical local inference using the mapping relationship between the full-process monitoring indicators and the local monitoring indicators to obtain a reverse anomaly information fingerprint chain; the reverse anomaly information fingerprint chain contains at least one abnormal full-process monitoring indicator or an abnormal local monitoring indicator. The forward adjustment module is used to select at least one correction strategy from a preset abnormal processing correction strategy library based on the mapping relationship between the reverse abnormal information fingerprint chain, the full-process monitoring indicators, the local monitoring indicators, and the processing control parameters of the corresponding standard process steps. The forward processing control parameters are adjusted from the starting abnormal node in the reverse abnormal information fingerprint chain until all full-process monitoring indicators and local monitoring indicators in the reverse abnormal information fingerprint chain meet the corresponding indicator thresholds.

2. The anomaly tracing and parameter control system for a premix production line as described in claim 1, characterized in that, The process of constructing the end-to-end tracking chain includes: By acquiring standard production process information of the target product and combining it with a hierarchical parsing algorithm, and taking the process flow implemented by each piece of equipment as the boundary, a global standard process step sequence and global standard process correlation relationship are obtained. Based on standard process steps, and taking the production process parameters corresponding to the indivisible actions of the process flow implemented by each piece of equipment as the boundary, we obtain the local standard process sub-step sequence and the corresponding local correlation relationship for each standard process step. The local association includes the priority order of the timestamps of the corresponding sub-steps and the image probability distribution of the degree of standardization of the current sub-step's execution on the degree of standardization of subsequent sub-steps, as well as the order of magnitude of the image sub-steps.

3. The anomaly tracing and parameter control system for a premix production line as described in claim 2, characterized in that, The process of constructing the end-to-end tracking chain also includes: Based on the global standard process step sequence and the corresponding global standard process correlation, the full-process monitoring indicators and corresponding cumulative relationships are obtained; the corresponding cumulative relationships include positive cumulative and negative cumulative. The full-process monitoring indicators include at least the total quality inspection score, the probability of the impact on production efficiency throughout the process, and the deviation of production efficiency. The total quality inspection score is calculated by weighted average method based on the real-time quality inspection score of each standard process step and the contribution of the corresponding standard process step to the total quality inspection score corresponding to the standard production process information. The production efficiency deviation is calculated by combining the deviation between the real-time production time achieved by the current standard process step and the corresponding preset standard production time with the deviation between the real-time production time achieved by the next standard process step and the corresponding preset standard production time, using a Bayesian network model. It is used to measure the probability distribution of the cumulative impact of the production efficiency of each standard process step on the production efficiency of the remaining standard process steps. Based on the full-process monitoring indicators and their corresponding cumulative relationships, and the processing control nodes constructed with the labels of each processing equipment, combined with a loosely coupled algorithm, an initial full-process tracking chain is constructed.

4. The anomaly tracing and parameter control system for a premix production line as described in claim 3, characterized in that, The process of building the end-to-end tracking chain also includes: Based on the sequence of local standard process sub-steps, with each processing control node as the root node of the corresponding standard process step, and combining the local implementation priority relationship of the sequence of local standard process sub-steps corresponding to each standard process step and the probability distribution of the first influence of the standard degree of the execution action of each sub-step on the standard degree of the execution action of the next sub-step with a tree database, a forward process flow tree for each standard process step is obtained. Based on the local monitoring indicators corresponding to each sub-step of the process flow tree for each standard process step, combined with the full-process monitoring indicators of each standard process step in the corresponding production period, the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step are obtained through factor analysis algorithm. The contribution of the detection accuracy of the corresponding local monitoring indicators to the deviation values ​​of the full-process monitoring indicators in the corresponding time period and to the deviation values ​​of the related sub-steps under the remaining standard process steps are also considered.

5. The anomaly tracing and parameter control system for a premix production line as described in claim 4, characterized in that, The process of building the end-to-end tracking chain also includes: Based on the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of the full monitoring indicators for the corresponding time period, a vertical internal derivation connection is constructed using the Hidden Markov algorithm. The vertical internal derivation connection is embedded into the forward process flow tree of each standard process sub-step. At the same time, the cross-inference connection constructed by the contribution of the deviation values ​​of the local monitoring indicators corresponding to all standard process sub-steps under each standard process step to the deviation values ​​of the related sub-steps under the remaining standard process steps is embedded into the sub-steps with control relationships in the forward process flow tree under different standard process sub-steps.

6. The anomaly tracing and parameter control system for a premix production line as described in claim 5, characterized in that, The process of building the end-to-end tracking chain also includes: Based on the processing control parameters and standard processing time length corresponding to each sub-step in the forward process flow tree of each standard process step, a forward processing control tree for each standard process step is obtained. At the same time, the process-control connection is established by combining the standard degree of the execution action corresponding to each processing control parameter with the deviation value of the local monitoring index of the corresponding standard process sub-step. The forward processing control tree of each standard process step under the initial full-process tracking chain is mapped to the corresponding forward process flow tree through the process-control connection to obtain the full-process tracking chain containing the process-control association tree; The full-process tracking chain, which includes a process-control association tree, is combined with random forest and simulation algorithms. The positional deviation and corresponding timestamp deviation of the corresponding outlier points and their cumulative outlier points are used as the derivation loss function for simulation training to obtain a fully trained full-process tracking chain.

7. The anomaly tracing and parameter control system for a premix production line as described in claim 6, characterized in that, The process of obtaining the reverse anomaly information fingerprint chain includes: The total quality inspection score of each standard process step is compared with the preset threshold for the total quality inspection score, and the production efficiency deviation is compared with the preset threshold for the production efficiency deviation. The first standard process step with a total quality inspection score lower than the corresponding preset threshold or a production efficiency deviation higher than the corresponding preset threshold is selected to obtain the abnormal starting standard process step. Based on the abnormal starting standard process step and the full-process monitoring indicators of all standard process steps before this step, as well as the timestamps of each standard process step, the total quality inspection score and production efficiency deviation of all steps are extracted. The two sets of indicator values ​​are arranged in order from back to front according to the timestamps of each standard process step and the corresponding total quality inspection score and production efficiency deviation, to obtain the observation sequence of the Hidden Markov Algorithm.

8. The anomaly tracing and parameter control system for a premix production line as described in claim 7, characterized in that, The process of obtaining the reverse anomaly information fingerprint chain also includes: Retrieve the pre-stored state transition probability matrix between each standard process step from the full tracking chain; the state transition probability matrix includes the positive cumulative impact probability of abnormal total quality inspection score between any adjacent steps, the negative cumulative impact probability of abnormal total quality inspection score, the positive cumulative impact probability of abnormal production efficiency deviation between adjacent steps, and the negative cumulative impact probability of abnormal production efficiency deviation. The total quality inspection score and production efficiency deviation value of each step in the observation sequence are input into the back-inference model constructed by the Hidden Markov Algorithm. With the goal of maximizing the matching degree between the abnormal index values ​​in the observation sequence and the cumulative influence probability in the state transition probability matrix, the hidden state of each preceding standard process step is derived one by one, starting from the abnormal initial standard process step. The specific hidden state type and corresponding timestamp of each standard process step are recorded. The derivation stops when the cumulative sum of the deviations between the total quality inspection score and the production efficiency deviation is 0. The root node with the first deviation of 0 is taken as the starting deviation node, and the specific hidden state and corresponding timestamp of each standard process step are obtained. The hidden states are: the initial state of abnormal total quality inspection score, the state of abnormal transmission of abnormal total quality inspection score, the initial state of abnormal production efficiency deviation, the state of abnormal transmission of abnormal production efficiency deviation, and the normal production state.

9. The anomaly tracing and parameter control system for a premix production line as described in claim 8, characterized in that, The process of obtaining the reverse anomaly information fingerprint chain also includes: Based on the specific hidden state type, corresponding timestamp, and starting deviation node of each standard process step obtained through derivation, the reverse state sequence representing the entire dimensional anomaly propagation path is obtained by arranging the specific hidden state type-timestamp in the order of the timestamps of each standard process step from back to front. Based on the reverse state sequence representing the entire dimensional anomaly propagation path, the first influence probability distribution under each root node on the path, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node are extracted. Based on the cumulative deviation value of the full-process monitoring indicators of each root node and its forward root node in the reverse state sequence, combined with the first influence probability distribution under each node, the contribution coefficient corresponding to the vertical internal derivation connection, and the contribution coefficient corresponding to the cross-inference connection information under the forward root node, the abnormal deviation index value of each local detection indicator under each node is obtained by inversion.

10. The anomaly tracing and parameter control system for a premix production line as described in claim 9, characterized in that, The process of obtaining the reverse anomaly information fingerprint chain also includes: Based on the abnormal deviation index value of each local detection index under each node, combined with the process-control connection, the standard deviation value of the corresponding processing process execution action and the deviation value of the standard processing time length of the corresponding execution action are obtained by inversion. Based on the reverse state sequence of the full-dimensional anomaly propagation path, the corresponding equipment label information, processing technology execution priority, and the standard deviation value of the corresponding processing technology execution action and the deviation value of the standard processing time length of the corresponding execution action are combined with the graph algorithm to construct the reverse anomaly information fingerprint chain.

Images