A multi-source sensor fusion-based silica gel desiccant production process intelligent management method

Abstract

The present application belongs to the technical field of production process control, and discloses a kind of silica gel desiccant production process intelligent management method based on multi-source sensor fusion, comprising, according to the time series translation alignment of monitoring data of different process links according to physical transmission lag time, construct the live phase change graph of current physical phase change stage, and deduce the real transmission path and variation result of fluctuation in production process based on live phase change graph, form state correlation result;Along the live phase change graph reverse tracking, based on live phase change edge positioning source abnormal node;Along the tracking path, update source abnormal node along the horizontal investigation of humidity counterattack crosstalk edge;According to the grading of edge attribute and investigation result, generate production process determination result set;Sensitivity decoupling matrix is constructed, and decoupled virtual control quantity is obtained;It is inversely transformed into actual process control parameter to generate production management control scheme;Realize the intelligent management of silica gel desiccant production process based on multi-source sensor fusion.

Description

Technical Field

[0001] This invention relates to the field of production process control technology, and more specifically, to an intelligent management method for the production process of silica gel desiccant based on multi-source sensor fusion. Background Technology

[0002] With the widespread application of desiccants in food packaging, pharmaceutical moisture protection, electronic component storage and transportation, and precision instrument protection, the production process of silica gel desiccants has increasingly higher requirements for material moisture stability, particle uniformity, drying adequacy, and batch consistency.

[0003] In detail, the production of silica gel desiccant usually involves multiple process steps such as raw material mixing, granulation, drying, cooling, sieving and temporary storage. These steps are highly continuous and coupled. If the production status cannot be perceived and coordinated in real time, it can easily lead to problems such as fluctuations in product moisture content, unstable particle performance, low energy utilization efficiency and untimely response to process switching.

[0004] However, existing silica gel desiccant production management methods mainly rely on single-point equipment monitoring, manual inspections, and experience-based parameter adjustments. These methods typically focus on independent control of a single production equipment or process variable, resulting in problems such as limited perception dimensions, delayed anomaly identification, difficulty in linking information across processes, and a lack of overall assessment of production status. This makes it difficult to achieve dynamic and refined management of the production process based on changes in raw materials, equipment status, and real-time fluctuations in process parameters. Furthermore, the limited perception dimensions lead to the fragmented setting of monitoring functions, and the lack of a unified data fusion and collaborative analysis mechanism for cross-process information results in incomplete judgment of production status and difficulty in locating the source of anomalies, affecting the production efficiency and product quality stability of silica gel desiccants.

[0005] Therefore, it is necessary to propose an intelligent management method for the silica gel desiccant production process based on multi-source sensor fusion. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent management method for silica gel desiccant production process based on multi-source sensor fusion, comprising: S1. Collect multi-source production monitoring data during the production process of the target silica gel desiccant; perform time-series shifting and alignment of monitoring data from different process stages based on physical transmission lag time to form a spatiotemporally aligned dataset of the same microscopic material batch. S2. Based on the spatiotemporal aligned dataset, construct the active phase transition diagram of the current physical phase transition stage, and deduce the actual transmission path and variation results of the wave in the production process based on the active phase transition diagram to form state correlation results; S3. When the state of a process node fluctuates, trace back along the active phase transition diagram and locate the source of the abnormal node based on the active phase transition edge; check the humidity backlash crosstalk edge laterally along the tracing path and update the source of the abnormal node; classify the level according to the edge attributes and the investigation results and generate a production process judgment result set. S4. Construct a sensitivity decoupling matrix based on the production process judgment result set and state association results to obtain the decoupled virtual control quantity; perform multi-objective optimization on the virtual control quantity and select the optimal virtual control solution; inversely transform it into actual process control parameters to generate a production management and control scheme.

[0007] Preferably, the spatiotemporally aligned dataset forming the same batch of microscopic materials includes: Multi-source production monitoring data includes temperature data, humidity data, pressure data, equipment operating status data, and material conveying status data; Calculate the physical transmission lag time between each process step based on the material conveying status data, and perform time series translation and alignment on the multi-source production monitoring data of different process steps based on the physical transmission lag time to obtain the initial spatiotemporal aligned dataset. Based on the thermodynamic mechanism of silica gel adsorption, deep mechanism features are extracted from the temperature and humidity data in the spatiotemporally aligned dataset. The deep mechanism features include the drying rate derivative, dew point approximation, and estimated values ​​of the moisture content gradient inside the particles. The deep mechanism features are merged with the initial spatiotemporal aligned dataset to form the final spatiotemporal aligned dataset output.

[0008] Preferably, constructing the active phase transition diagram of the current physical phase transition stage includes: The process nodes are obtained based on the process division of the target silica gel desiccant production process; Based on the drying rate derivative in the spatiotemporally aligned dataset, the current physical phase transition stage of the silica gel material in each process step is identified. For materials that point from upstream process nodes to downstream process nodes, when the material is identified as being in a constant-rate drying period, the edge attribute connecting the upstream and downstream process nodes is defined as a fluctuation absorption edge; when the material is identified as being in a falling-rate drying period, the edge attribute connecting the upstream and downstream process nodes is defined as a fluctuation locking amplification edge. Both the wave absorption edge and the wave locking amplification edge are marked as the active phase transition edge; Humidity backlash and crosstalk between process steps are dynamically generated based on workshop space layout and exhaust status. The active phase transition diagram of the current physical phase transition stage is constructed by process node, active phase transition edge, and humidity backlash crosstalk edge.

[0009] Preferably, the dynamic generation of humidity backlash crosstalk edges between process steps based on workshop space layout and exhaust status includes: Acquire humidity data from the exhaust port of the drying process, humidity data from the air intake area of ​​the downstream cooling or temporary storage process, and workshop exhaust pressure data in the spatiotemporally aligned dataset. When the humidity data at the exhaust port of the drying process is greater than the humidity data in the air intake area of ​​the downstream cooling process or temporary storage process, and the workshop exhaust pressure data is lower than the preset exhaust pressure threshold, a humidity backlash crosstalk edge is generated from the drying exhaust node to the downstream cooling or temporary storage node.

[0010] Preferably, the process of extrapolating the actual transmission path and variation results of fluctuations in the production process based on the active phase transition diagram to form state correlation results includes: When the rate of change of temperature data, humidity data, or material conveying status data of any process node in the time-space aligned dataset exceeds the corresponding baseline fluctuation range, it is determined that the process node has experienced a status fluctuation. The propagation direction of state fluctuations is deduced along the positive edge of the living phase transition diagram; When the simulation path passes through the fluctuation absorption edge, the fluctuation transmission type of the current simulation path is marked as attenuation and resolution type; When the simulation path passes through the fluctuation locking amplification edge, the fluctuation transmission type of the current simulation path is marked as the distortion amplification type; When the simulation path passes through the humidity backlash crosstalk edge, the fluctuation transmission type of the current simulation path is marked as environmental back damp crosstalk type; Summarize the deduction paths and corresponding fluctuation transmission types to generate state correlation results.

[0011] Preferably, the step of tracing backwards along the living phase transition diagram and locating the source anomaly node based on the living phase transition edge includes: The process node experiencing state fluctuation is taken as the current traversal node; the upstream node of the current traversal node is obtained in reverse along the edge of the live phase transition graph, as well as the live phase transition edge connecting the current traversal node and the upstream node; if the live phase transition edge is a fluctuation absorption edge, the reverse traversal of the current traversal branch is terminated; if the live phase transition edge is a fluctuation locking amplification edge, the upstream node is recorded as the suspected fluctuation source node, and the upstream node is updated as the current traversal node, and the reverse traversal continues until the starting process node with no upstream node is reached; the finally recorded suspected fluctuation source node is determined as the source of the anomaly.

[0012] Preferably, the step of laterally investigating humidity backlash crosstalk along the tracking path and updating the source abnormal node includes: Obtain all process nodes traversed during the reverse tracing process as a check node set; traverse the check node set to check if there is a humidity backlash crosstalk edge input to any node in the check node set; if there is, determine the node that receives the humidity backlash crosstalk edge as the crosstalk receiving node; obtain the generation timestamp of the humidity backlash crosstalk edge and the timestamp of the state fluctuation of the crosstalk receiving node. When the generated timestamp is within a preset time window before the timestamp of the state fluctuation, the sending node of the humidity backlash crosstalk edge is updated to the source abnormal node.

[0013] Preferably, the step of generating a production process judgment result set based on edge attributes and investigation results includes: If the source of the abnormal operation is updated by the node that sends the humidity backlash crosstalk edge, the operating status will be classified as a bypass crosstalk type abnormal operating status; if the source of the abnormal operation is determined by the suspected fluctuation source node, the reverse traversal path from the source of the abnormal operation node to the process link node where the state fluctuation occurs will be obtained. If the reverse traversal path contains a wave-locking amplification edge, the running state will be classified as a forward amplification abnormal running state; if the reverse traversal path contains only wave-absorbing edges, the running state will be classified as a wave running state. The operation status classification results, source abnormal nodes, active phase transition edges, and humidity backlash crosstalk edges are combined to generate a production process judgment result set.

[0014] Preferably, the step of constructing a sensitivity decoupling matrix based on the production process judgment result set and state association results to obtain the decoupled virtual control quantity includes: Obtain the actual process control variables in the production process of the target silica gel desiccant, as well as the target control objects corresponding to the current production task; the actual process control variables include drying temperature, circulating air volume, and material residence time; the target control objects include moisture content, particle uniformity, and energy consumption. Based on the fluctuation propagation type in the state correlation results, obtain the state partial derivatives of the actual process control variables with respect to the target control object; construct a sensitivity decoupling matrix based on the state partial derivatives; and perform an inversion operation on the sensitivity decoupling matrix to obtain the virtual control quantity.

[0015] Preferably, the step of performing multi-objective optimization on the virtual control quantity and selecting the optimal virtual control solution; and inversely transforming it into actual process control parameters to generate a production management and control scheme includes: Based on the production process judgment result set, the objective function is to reduce the deviation between the target control object and the corresponding set benchmark value; using virtual control variables as decision variables, multi-objective non-dominated ranking optimization is performed to obtain the Pareto front solution set; based on the safety constraints of the current equipment operating status and the priority weight of the current production task, the optimal virtual control solution is selected from the Pareto front solution set; the optimal virtual control solution is inversely transformed through the sensitivity decoupling matrix to generate actual process control parameters; and a production management and control scheme is generated based on the actual process control parameters.

[0016] The technical effects and advantages of this invention, based on a multi-source sensor fusion-based intelligent management system for silica gel desiccant production process, are as follows: By calculating the physical transport lag time based on the material conveying status, the monitoring data of different process links are time-series shifted and aligned, effectively eliminating the problem of mismatched monitoring data timestamps caused by material spatial displacement, and restoring the true state correspondence of the same microscopic material batch in each process. At the same time, based on the thermodynamic mechanism of silica gel adsorption, deep features such as drying rate derivative, dew point approximation degree and moisture content gradient are extracted, breaking through the limitation of existing technologies that only rely on superficial temperature and humidity data, and enabling the perception of production status to penetrate to the essential level of physical phase change.

[0017] The living phase transition diagram breaks through the limitations of traditional static directed graphs that only represent physical pipeline connections. It makes explicit and graphical the system's self-stabilizing absorption and solidification amplification response characteristics to fluctuations, as well as the crosstalk paths in non-pipeline environments. This enables accurate deduction of the actual transmission path and variation results of fluctuations during the production process. Through reverse tracing and lateral investigation along the living phase transition diagram, the phase transition properties (absorption or amplification) of edges are used as logical criteria to automatically truncate irrelevant transmission branches and trace back to the source of the anomaly. By capturing hidden humidity backlash crosstalk edges through lateral investigation and updating the source according to the time window, this process achieves a leap from superficial anomaly alarms to physical causal localization. It can clearly distinguish between fluctuation operation status, downstream amplification anomalies, and bypass crosstalk anomalies, significantly improving the accuracy and timeliness of anomaly source localization.

[0018] To address the highly coupled multivariate characteristics of the silica gel drying process, a sensitivity decoupling matrix is ​​constructed based on state partial derivatives to obtain virtual control variables. This eliminates the cross-coupling effects of drying temperature, circulating air volume, and other variables on moisture content, particle uniformity, and energy consumption. Furthermore, by combining equipment safety constraints and production task priorities, a multi-objective non-dominated ranking optimization is performed on the virtual control variables. The optimal solution is selected from the Pareto front and inversely transformed into actual process parameters. This method effectively avoids process oscillations and quality sacrifices caused by traditional univariate regulation, achieving global coordinated control of quality assurance and energy consumption optimization. Attached Figure Description

[0019] Figure 1This is a schematic diagram of the intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to the present invention; Figure 2 This is a schematic diagram of the structure of the active phase transition diagram in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, please refer to Figure 1 and Figure 2 In this embodiment of the invention, a method for intelligent management of silica gel desiccant production process based on multi-source sensor fusion includes: S1. Collect multi-source production monitoring data during the production process of the target silica gel desiccant; perform time-series shifting and alignment of monitoring data from different process stages based on physical transmission lag time to form a spatiotemporally aligned dataset of the same microscopic material batch. S2. Based on the spatiotemporal aligned dataset, construct the active phase transition diagram of the current physical phase transition stage, and deduce the actual transmission path and variation results of the wave in the production process based on the active phase transition diagram to form state correlation results; S3. When the state of a process node fluctuates, trace back along the active phase transition diagram and locate the source of the abnormal node based on the active phase transition edge; check the humidity backlash crosstalk edge laterally along the tracing path and update the source of the abnormal node; classify the level according to the edge attributes and the investigation results and generate a production process judgment result set. S4. Construct a sensitivity decoupling matrix based on the production process judgment result set and state association results to obtain the decoupled virtual control quantity; perform multi-objective optimization on the virtual control quantity and select the optimal virtual control solution; inversely transform it into actual process control parameters to generate a production management and control scheme.

[0022] The spatiotemporally aligned dataset that forms the same batch of microscopic materials includes: Multi-source production monitoring data includes temperature data, humidity data, pressure data, equipment operating status data, and material conveying status data; specifically, a multi-source sensor network is deployed at key nodes in each process step of the target silica gel desiccant production process for real-time monitoring. Temperature data is collected by temperature sensors placed inside the mixing tank, granulator die head, air inlets and outlets of each temperature zone of the drying oven, cooling section air chamber and temporary storage bin. Humidity data is collected by placing humidity-sensitive capacitive sensors at the granulation outlet, the dryer exhaust vent, the cooling section air inlet, and the workshop environment. Pressure data is collected by installing absolute pressure sensors inside the drying oven, in the inlet and outlet pipes of the cooling fan, and at key nodes of the dust removal pipe. The equipment operating status data is collected by accessing the data interface of the control PLC of each production equipment, including the granulator motor frequency, drying fan speed, opening degree of each air valve, and digital signals for equipment start and stop. Material conveying status data is obtained by deploying belt scales or loss-in-weight scales at key locations on the material conveyor belt to acquire instantaneous material flow rate, by deploying speed sensors to acquire real-time conveyor belt speed, and by deploying rotary paddle level gauges to acquire material level status in temporary storage bins.

[0023] Based on material conveying status data, the physical transmission lag time between each process step is calculated. Then, based on this physical transmission lag time, multi-source production monitoring data from different process steps are time-series shifted and aligned to obtain an initial spatiotemporally aligned dataset (represented as a multi-dimensional time series matrix, where each row corresponds to an aligned timestamp, and each column corresponds to the shifted and aligned monitoring values ​​of each sensor in different process steps). Specifically… For adjacent process steps, obtain the physical distance L of the material conveyor belt between the two steps, and extract the real-time speed v of the current conveyor belt from the material conveying status data; calculate the physical transmission lag time Δt=L / v required for the material to be transferred from the previous process node to the next process node. If there is a temporary storage bin between the steps, calculate the temporary storage time based on the feed flow rate into the temporary storage bin and the current material level, and include it in the physical transmission lag time. The timestamp of the core process step (such as the drying process) is selected as the reference time axis. For monitoring data upstream of the core process, the timestamp is added to the corresponding physical transmission lag time, so that the data time axis is shifted backward. For monitoring data downstream of the core process, the timestamp is subtracted from the corresponding physical transmission lag time, so that the data time axis is shifted forward. In this way, the monitoring data generated by different processes for the same micro-material batch are aligned to the same time index.

[0024] Based on the thermodynamic mechanism of silica gel adsorption, deep mechanistic features are extracted from the temperature and humidity data in the spatiotemporally aligned dataset. These features include the drying rate derivative, dew point approximation, and estimated internal moisture content gradient of the particles. It should be noted that the drying rate derivative refers to the amount of moisture evaporation per unit time calculated by extracting the temperature and humidity data at the air inlet of the drying process from the spatiotemporally aligned dataset and combining it with the drying air volume. The time series of this drying rate is then differentially calculated to obtain the drying rate derivative. This drying rate derivative reflects the kinetic transformation characteristics of the drying stage; it tends to zero during the constant-rate drying period and is negative during the falling-rate drying period.

[0025] Dew point approximation refers to the extraction of temperature and humidity data from the air intake area of ​​the cooling or temporary storage process from the spatiotemporally aligned dataset, and the calculation of the current environment's dew point temperature based on the thermodynamic temperature-humidity mapping relationship; at the same time, the material surface temperature data of this area is obtained, and the difference between the material surface temperature and the dew point temperature is calculated as the dew point approximation; the closer this value is to zero, the more likely the current environment's water vapor is to condense and become damp on the material surface.

[0026] The estimated internal moisture content gradient of the particles refers to the temperature and humidity data extracted from the spatiotemporally aligned dataset during the drying and falling period, combined with the characteristic radius of the silica particles, and the equilibrium moisture content calculated using the silica adsorption isotherm model at the current temperature and humidity. Based on the simplified spherical coordinate model of Fick's diffusion law, combined with the current drying rate and equilibrium moisture content, the difference in moisture content between the center and the surface of the silica particles is estimated as the estimated internal moisture content gradient of the particles. This value reflects the resistance to moisture migration and the degree of crusting inside the particles.

[0027] The deep mechanism features are merged with the initial spatiotemporal aligned dataset to form the final spatiotemporal aligned dataset output.

[0028] It should be explained that the drying rate derivative, dew point approximation, and estimated particle internal moisture content gradient values ​​extracted in the above steps are added as new data dimension columns according to their corresponding calculation source timestamps and concatenated into the multidimensional time series matrix of the initial spatiotemporal aligned dataset. That is, after each row of the original matrix for the original monitoring data such as temperature, humidity, and pressure at each aligned timestamp, the deep mechanism feature values ​​corresponding to that time are extended and added; this is output as the final spatiotemporal aligned dataset.

[0029] The construction of the active phase transition diagram of the current physical phase transition stage includes: The process nodes are obtained by dividing the production process of the target silica gel desiccant. It should be explained that, based on the actual process flow of the target silica gel desiccant production process, each independent processing unit with a clear physical boundary on the production line is divided into corresponding processes, specifically including raw material mixing process, granulation process, drying process, cooling process, sieving process, and temporary storage process. Each divided process is abstracted as a vertex in a graph model and assigned a unique node identifier (e.g., Node_Mixing, Node_Granulation, Node_Drying, Node_Cooling, Node_Sieving, Node_Temporary Storage), thereby obtaining the complete set of process nodes that constitute the active phase transition diagram.

[0030] Based on the drying rate derivative in the spatiotemporally aligned dataset, the current physical phase transition stage of the silica gel material in each process step is identified; specifically... From the final spatiotemporally aligned dataset, the drying rate derivative corresponding to the drying process at the current moment is extracted. Based on the drying kinetics of silica gel, a derivative judgment threshold is set: if the absolute value of the drying rate derivative at the current moment is less than or equal to the derivative judgment threshold, it indicates that the current drying rate is stable, and the vaporization rate of surface moisture and the outward diffusion rate of internal moisture are in balance, thus identifying the current physical phase change stage of the silica gel material as the constant-rate drying period; if the drying rate derivative at the current moment is negative and its absolute value is greater than the derivative judgment threshold, it indicates that the drying rate decreases significantly over time, the outward diffusion rate of internal moisture is lower than the surface vaporization rate, and the drying process is controlled by internal diffusion resistance, thus identifying the current physical phase change stage of the silica gel material as the falling-rate drying period.

[0031] For materials where upstream process nodes point to downstream process nodes, when the material is identified as being in a constant-rate drying period, the edge connecting these upstream and downstream process nodes is defined as a fluctuation-absorbing edge; when the material is identified as being in a falling-rate drying period, the edge connecting these upstream and downstream process nodes is defined as a fluctuation-locking amplification edge. Specifically... Traverse all upstream and downstream process nodes in the active phase transition diagram that have the actual material transport direction; Query the current physical phase transition stage of the material flowing between the upstream and downstream process nodes; If the query identifies that the material is in a constant-rate drying period, since the material surface remains wet during this stage, the hot air medium has the buffering capacity to receive and evaporate fluctuations. The small moisture fluctuations from the upstream input will be absorbed and smoothed out by the hot air when they are transmitted to the downstream, generating a directed edge from the upstream process node to the downstream process node. The attribute parameter field of the directed edge is assigned the value of fluctuation absorption edge, which is used to characterize the transmission characteristics of the path to attenuate and dissolve fluctuations from the upstream input. If the query identifies that the material is in the slow-drying stage, the drying resistance increases sharply because the material surface is crusted and the water inside is difficult to escape. The small moisture fluctuations input from upstream cannot be buffered when they are transmitted to downstream. Instead, they are locked inside the particles and amplified, causing deformation or cracking. A directed edge is generated from the upstream process node to the downstream process node, and the attribute parameter field of the directed edge is assigned the value of fluctuation locking and amplification edge. This is used to characterize the transmission characteristics of the path to the fluctuations input from upstream, which have the characteristics of solidification and amplification.

[0032] Both the wave absorption edge and the wave locking amplification edge are marked as the active phase transition edge; Humidity backlash and crosstalk between process steps are dynamically generated based on workshop space layout and exhaust status. The active phase transition diagram of the current physical phase transition stage is constructed by process node, active phase transition edge, and humidity backlash crosstalk edge.

[0033] The dynamic generation of humidity backlash and crosstalk edges between process steps based on workshop space layout and exhaust status specifically includes: Acquire humidity data from the exhaust port of the drying process, humidity data from the air intake area of ​​the downstream cooling or temporary storage process, and workshop exhaust pressure data in the spatiotemporally aligned dataset. When the humidity data at the exhaust port of the drying process is greater than the humidity data in the air intake area of ​​the downstream cooling process or temporary storage process (used to characterize that the humidity of the drying exhaust gas exceeds the self-purification capacity of the workshop environment), and the workshop exhaust pressure data is lower than the preset exhaust pressure threshold (used to characterize that the air intake area of ​​the downstream cooling process or temporary storage process is not in an independent isolation state), a humidity backlash crosstalk edge is generated from the drying exhaust node to the downstream cooling or temporary storage node.

[0034] It needs to be explained that, The drying process refers to the process equipment unit (such as mesh belt drying oven or fluidized bed drying bed) on the target silica gel desiccant production line that heats and dehydrates the granulated wet silica gel particles. The exhaust port of the drying process refers to the physical pipeline outlet used to discharge the high-temperature exhaust gas carrying a large amount of evaporated moisture in the drying process. The downstream cooling process or temporary storage process air intake area refers to the cold air inlet of cooling equipment (such as fluidized bed cooler) used to cool down the high-temperature silica gel after drying, immediately after the drying process, or the internal environmental space of the temporary storage silo used to temporarily store silica gel to be screened. A workshop refers to the physical space of a closed or semi-closed production plant that houses the aforementioned process equipment units.

[0035] It should also be noted that the preset exhaust pressure threshold is based on the physical exhaust capacity of the workshop exhaust system and the calibration of the exhaust gas overflow critical point: according to the design parameters of the workshop exhaust duct and the historical normal operation records, the minimum critical pressure value that the exhaust main duct needs to maintain when the workshop exhaust system effectively extracts the high humidity exhaust gas from the drying process to the outside without overflowing and spreading into the workshop is obtained, and this minimum critical pressure value is set as the preset exhaust pressure threshold.

[0036] The deduction of the actual transmission path and variation results of fluctuations in the production process based on the active phase transition diagram forms state correlation results, specifically including: When the rate of change of temperature, humidity, or material conveying status data at any process node in the spatiotemporal aligned dataset exceeds the corresponding baseline fluctuation range, it is determined that a status fluctuation has occurred at that process node; specifically... From the final spatiotemporally aligned dataset, extract the temperature, humidity, or material conveying status data of any process node at the current moment and its historical adjacent moments; calculate the difference between the data sample value at the current moment and the data sample value at the previous moment, and divide the difference by the time interval between adjacent moments to obtain the first-order difference value of the data at the current moment, and use the first-order difference value as the rate of change of the data corresponding to the process node; It should be noted that the baseline fluctuation range refers to the allowable fluctuation range of various monitoring data at each process node due to inherent equipment disturbances and measurement noise, when the target silica gel desiccant production process is in a stable operating state. Acquisition method: Extract the spatiotemporal aligned dataset of historical normal production periods, and statistically analyze the probability distribution of the rate of change of temperature data, humidity data and material conveying status data of each process node, obtain the mean and standard deviation of the rate of change, and define the standard deviation range of the mean fluctuation by a set multiple based on the set confidence coefficient (such as the 3 sigma principle) as the corresponding benchmark fluctuation range.

[0037] The propagation direction of state fluctuations is deduced along the positive edge of the living phase transition diagram; When the simulation path passes through the fluctuation absorption edge, the fluctuation transmission type of the current simulation path is marked as attenuation and resolution type; When the simulation path passes through the fluctuation locking amplification edge, the fluctuation transmission type of the current simulation path is marked as the distortion amplification type; When the simulation path passes through the humidity backlash crosstalk edge, the fluctuation transmission type of the current simulation path is marked as environmental back damp crosstalk type; Summarize the deduction paths and corresponding fluctuation transmission types to generate state correlation results.

[0038] Specifically The process node where the state fluctuation is determined is taken as the current simulation starting node, and an empty historical simulation path is initialized. In the currently constructed active phase transition diagram, find all positive outgoing edges that start from the current simulation starting node and point to the downstream node; among them, positive outgoing edges include active phase transition edges (fluctuation absorption edges or fluctuation locking amplification edges) and humidity backlash crosstalk edges emitted by the current simulation starting node as the drying exhaust node. If a positive outgoing edge exists, the state fluctuation is propagated along the positive outgoing edge to the corresponding downstream node, generating a deduction path that includes the current deduction starting node, the positive outgoing edge, and the downstream node. Identify and mark the attribute type of the forward outgoing edge along the deduction path: if the forward outgoing edge is a wave absorption edge, mark the wave conduction type field of the deduction path as attenuation and resolution type; if the forward outgoing edge is a wave locking and amplification edge, mark the wave conduction type field of the deduction path as distortion and amplification type; if the forward outgoing edge is a humidity backlash crosstalk edge, mark the wave conduction type field of the deduction path as environmental humidity crosstalk type. Update the downstream nodes of the simulation path to the new current simulation starting node. Repeat the above operation until there are no positive outgoing edges at the current starting node of the deduction, and complete the full deduction of the state fluctuation along the positive propagation direction of the living phase transition diagram; All the deduction paths generated by the deduction and their corresponding wave propagation types are summarized. The set of deduction paths and the set of wave propagation types are combined to generate a state association result that includes physical phase transition logic and environmental crosstalk logic.

[0039] The reverse tracing along the active phase transition diagram, based on the active phase transition edge to locate the source anomaly node, includes: The process node experiencing state fluctuation is taken as the current traversal node; the upstream node of the current traversal node is obtained in reverse along the edge of the live phase transition graph, as well as the live phase transition edge connecting the current traversal node and the upstream node; if the live phase transition edge is a fluctuation absorption edge, the reverse traversal of the current traversal branch is terminated; if the live phase transition edge is a fluctuation locking amplification edge, the upstream node is recorded as the suspected fluctuation source node, and the upstream node is updated as the current traversal node, and the reverse traversal continues until the starting process node with no upstream node is reached; the finally recorded suspected fluctuation source node is determined as the source of the anomaly.

[0040] It should be clearly stated that when obtaining upstream nodes in reverse along the edges of the active phase transition diagram, tracing is only performed along the active phase transition edges (i.e., the fluctuation absorption edge and the fluctuation locking amplification edge), and not along the humidity backlash crosstalk edge.

[0041] In addition, when the current traversal node has multiple incoming edges (i.e., corresponding to multiple upstream process nodes sending materials), independent traversal branches need to be generated along each incoming edge and executed in parallel. If a branch is terminated when it passes through a fluctuation absorption edge, it does not affect other branches from continuing to execute the reverse traversal. Only when all parallel branches are terminated or have traversed to the starting process node, is the entire reverse tracing process considered to be over.

[0042] The theoretical basis for this treatment also needs further explanation, as follows: The fluctuation absorption edge corresponds to the constant-rate drying period of silica gel material. During this physical phase change stage, the material surface remains fully wetted. The vaporization rate of moisture on the surface is determined by the external hot air conditions. The rate at which moisture inside the material migrates to the surface is sufficient to replenish the moisture loss on the surface. At this time, the system itself has a strong heat and mass transfer buffer capacity (self-stabilizing capability). Even if a small fluctuation in moisture or temperature is input from the upstream process, the excess moisture will be quickly vaporized and carried away by the sufficient hot air medium when it reaches this process. The fluctuation is absorbed and smoothed out by this process and will not continue to be transmitted downstream. Therefore, if a state fluctuation occurs in the downstream process, and the edge connecting the upstream and downstream has the attribute of fluctuation absorption edge, according to the principle of heat and mass balance, the abnormality in the downstream cannot be caused by the downstream small fluctuation being transmitted downstream (because it has been absorbed in the middle). Thus, the upstream node is cleared of suspicion in physical logic, and there is no need to trace upwards, terminating the branch traversal. The fluctuation-locked amplification edge corresponds to the slow-drying period of silica gel material. During this physical phase transition stage, insufficient moisture on the material surface leads to localized drying crust formation, drastically increasing resistance to moisture migration. The drying process is controlled by internal diffusion. At this point, the system loses its buffering capacity and is in a critically sensitive state. If a small moisture fluctuation is input from the upstream process, the surface crust will block the rapid adjustment of moisture, locking the small fluctuation inside the particle. This results in a local increase in vapor pressure or stress concentration, leading to defects such as particle cracking and deformation. The originally small fluctuation is solidified and significantly amplified, then transmitted downstream. Therefore, if a state fluctuation occurs in the downstream process, and the edge connecting the upstream and downstream processes is a fluctuation-locked amplification edge, according to the principle of internal diffusion dynamics, the severe anomaly downstream is highly likely caused by the distortion of the small fluctuation upstream through this amplification stage. Therefore, the upstream node is highly suspicious and must be recorded as the suspected fluctuation source node. The process should continue to trace upstream along this path until the true source of the initial small fluctuation is found.

[0043] The process of laterally investigating humidity backlash crosstalk along the tracking path and updating the source abnormal node includes: Obtain all process nodes traversed during the reverse tracing process as a check node set; traverse the check node set to check for the existence of a humidity backlash crosstalk edge input to any node in the check node set; if it exists, determine the node receiving the humidity backlash crosstalk edge as the crosstalk receiving node; obtain the generation timestamp of the humidity backlash crosstalk edge (referring to the system time when the conditions of the drying exhaust port humidity being greater than the downstream air inlet humidity and the workshop exhaust pressure being lower than the preset exhaust pressure threshold are first met when the system dynamically instantiates and generates the humidity backlash crosstalk edge during the construction of the active phase transition diagram; this timestamp represents the moment when the physical path of environmental crosstalk is established and begins to take effect). The system time when the rate of change of the monitoring data of the crosstalk receiving node (i.e., the cooling or temporary storage node) first exceeds the corresponding baseline fluctuation range during the simulation of the propagation of the state fluctuation is determined. This time stamp represents the moment when a visible anomaly occurs in the downstream node. When the generated time stamp is within a preset time window before the state fluctuation occurrence time stamp, the node that sends the humidity backlash crosstalk edge is updated to the source abnormal node.

[0044] The preset time window is calibrated based on the physical diffusion time of the high-humidity airflow in the workshop and the response lag time of the sensors in the downstream process. The specific steps are as follows: Obtain the maximum physical space distance between the exhaust port of the drying process and the air intake area of ​​the downstream cooling process or temporary storage process. Combined with the normal minimum diffusion velocity of the airflow in the workshop environment, calculate the maximum physical diffusion time required for the high-humidity exhaust gas to cause back-moistening and spread in the workshop space. The sensing delay time of the downstream process humidity sensor to the sudden change in environmental humidity, and the cumulative response time required for the material surface to adsorb environmental water vapor until the monitoring data shows obvious fluctuations are obtained. The sum of the two is taken as the crosstalk response lag time. Add the maximum physical diffusion time to the crosstalk response lag time, and set the total time span obtained as the length of the preset time window; This preset time window represents the maximum physical time limit required from the triggering of crosstalk by the humidity rebound (i.e., the generation of the crosstalk source) to the inevitable state fluctuation of the downstream node due to the absorption of moisture.

[0045] The process of classifying and grading based on edge attributes and investigation results generates a set of production process judgment results, including: Before performing the classification, the execution record identifier of the horizontal investigation step needs to be read: if the horizontal investigation successfully matches a humidity backlash crosstalk edge that meets the time window conditions and performs the operation of updating the sending node to the source abnormal node, then the judgment branch of "the source abnormal node is obtained by updating the sending node of the humidity backlash crosstalk edge" is triggered; otherwise, the judgment branch of "the source abnormal node is determined by the suspected fluctuation source node" is triggered.

[0046] If the source of the abnormal operation is updated by the node that sends the humidity backlash crosstalk edge, the operating status will be classified as a bypass crosstalk type abnormal operating status; if the source of the abnormal operation is determined by the suspected fluctuation source node, the reverse traversal path from the source of the abnormal operation node to the process link node where the state fluctuation occurs will be obtained. If the reverse traversal path contains a wave-locking amplification edge, the running state will be classified as a forward amplification abnormal running state; if the reverse traversal path contains only wave-absorbing edges, the running state will be classified as a wave running state. It should be further explained that after entering the judgment branch determined by the suspected fluctuation source node, it is necessary to obtain the set of all active phase transition edges on the reverse traversal path; traverse the attribute field of each edge in the set: if there is at least one active phase transition edge in the set with the attribute of fluctuation locking amplification edge, then regardless of whether there is a fluctuation absorption edge in the path at the same time, the classification of "downstream amplification abnormal operation state" will be directly triggered (because as long as the fluctuation undergoes locking amplification once, it will cause downstream explicit anomalies, and the amplification effect will cover the absorption effect); the classification of "fluid operation state" will only be triggered when the attribute of all active phase transition edges in the set is fluctuation absorption edge.

[0047] The operational status classification results, source anomaly nodes, active phase transition edges, and humidity backlash crosstalk edges are combined to generate a production process judgment result set. The specific data structure of the final result set needs to be explained. The operational status classification results are used as the primary key, and the identifiers of source anomaly nodes, the sequence of active phase transition edges traversed along the reverse traversal path (sorted by tracing level), and the implicit environmental bypass edges identified through horizontal investigation (i.e., the identifiers of humidity backlash crosstalk edges that trigger crosstalk; if they do not exist, they are empty) are used as attribute values. These are encapsulated into a structured data dictionary or data object, which serves as the final production process judgment result set.

[0048] The step of constructing a sensitivity decoupling matrix based on the production process judgment result set and state association results, and obtaining the decoupled virtual control quantity, includes: Obtain the actual process control variables in the production process of the target silica gel desiccant, as well as the target control objects corresponding to the current production task; the actual process control variables include drying temperature, circulating air volume, and material residence time; the target control objects include moisture content, particle uniformity, and energy consumption. It should be further explained that by connecting to the DCS control system of the target silica gel desiccant production line, the list of process parameters and current set values ​​that are currently in automatic adjustment mode and can be directly issued commands are read; from this list, three core adjustable parameters that directly participate in the drying heat and mass transfer process are extracted, namely, the set value of the heating medium temperature of the drying process as the drying temperature, the air volume corresponding to the frequency of the frequency converter of the circulating fan as the circulating air volume, and the material residence time in the drying zone derived from the running speed of the mesh belt conveyor as the material residence time. The above three items are used as actual process control variables.

[0049] By analyzing the production task orders and corresponding process specification documents issued by the current production scheduling system, the terminal quality indicators and production constraint indicators that must be guaranteed for the current batch of silica gel desiccant are extracted. Specifically, these include the final output moisture content specified in the product standard as the moisture content, the particle size distribution standard deviation required by the screening process as the particle uniformity, and the unit output energy consumption limit approved by the current production task as the energy consumption. The above three items are taken as the target control objects.

[0050] Based on the fluctuation propagation type in the state correlation results, obtain the state partial derivatives of the actual process control variables with respect to the target controlled object; construct a sensitivity decoupling matrix based on the state partial derivatives; invert the sensitivity decoupling matrix to obtain the virtual control quantity; the virtual control quantity is used to adjust a single target controlled object and suppress its coupling influence on other target controlled objects. Specifically, Extract the current wave propagation type (attenuation type, distortion amplification type, or environmental backflow crosstalk type) from the state correlation results. Since different wave conduction types correspond to different phase change stages and heat and mass transfer boundary conditions of silica gel materials, the pre-stored historical steady-state operation data cluster corresponding to the current working condition range or the pre-built silica gel drying mechanism simulation model is called according to the wave conduction type as the reference environment for subsequent partial derivative calculations. In the matched current operating condition data cluster or simulation model, a small disturbance of a preset unit is applied to each actual process control variable (drying temperature, circulating air volume, material residence time) in sequence; the steady-state output changes of each target control object (moisture content, particle uniformity, energy consumption) before and after the disturbance are collected respectively; the ratio of the change of the corresponding target control object to the disturbance of the actual process control variable is calculated, and the state partial derivative of the actual process control variable with respect to the target control object is obtained. All combinations are traversed to complete the full state partial derivative calculation. Using the target control object (moisture content, particle uniformity, energy consumption) as the row index of the matrix and the actual process control variables (drying temperature, circulating air volume, material residence time) as the column index of the matrix, the obtained partial derivatives of each state are filled into the corresponding row and column intersection positions to generate a sensitivity decoupling matrix; each element of this sensitivity decoupling matrix quantitatively characterizes the degree of instantaneous coupling influence of a specific actual process control variable on a specific target control object. The invertibility of the sensitivity decoupling matrix is ​​determined by calculating its determinant. If it is invertible, the sensitivity decoupling matrix is ​​inverted by using the algebraic cofactor method or elementary row transformation method to obtain its inverse matrix, which serves as the decoupling compensation matrix. This inverse matrix contains the cancellation logic for the original cross-coupling channels of the system. The desired adjustment increment vector of the target controlled object is multiplied with the inverse matrix to generate a virtual control vector. The virtual control vector contains three independent virtual control components. According to the properties of matrix operations, due to the introduction of inverse matrix compensation, each virtual control component only changes its unique corresponding target controlled object, while the effect on the other two target controlled objects is canceled out and becomes zero. This achieves the decoupling of the strongly coupled drying system and obtains independently adjustable virtual control quantities.

[0051] The process of performing multi-objective optimization on the virtual control quantity and selecting the optimal virtual control solution; then inversely transforming it into actual process control parameters to generate a production management and control scheme includes: Based on the production process judgment result set, the objective function is to reduce the deviation between the target control object and the corresponding set benchmark value; using virtual control variables as decision variables, multi-objective non-dominated ranking optimization is performed to obtain the Pareto front solution set; based on the safety constraints of the current equipment operating status and the priority weight of the current production task, the optimal virtual control solution is selected from the Pareto front solution set; the optimal virtual control solution is inversely transformed through the sensitivity decoupling matrix to generate actual process control parameters; and a production management and control scheme is generated based on the actual process control parameters.

[0052] The specific execution steps are as follows: Based on the production process judgment result set, extract the current operating status classification result and the location of the source anomaly node; obtain the actual operating values ​​of the target control objects (moisture content, particle uniformity, energy consumption) in the current spatiotemporal aligned dataset, as well as the set benchmark values ​​corresponding to the target control objects in the current production task; calculate the deviation between the actual operating values ​​and the set benchmark values ​​of each target control object; construct a multi-objective optimization objective function vector with the goal of reducing the deviation, where each component corresponds to the deviation minimization objective of a target control object; and dynamically adjust the constraint penalty terms of the objective function according to the operating status classification result. Using virtual control variables as decision variables, the boundary values ​​of the virtual control variables are set. A non-dominated sorting genetic algorithm is used to randomly initialize the virtual control variable population within the boundary values. The fitness value of each virtual control variable individual in the population is calculated by substituting it into the objective function vector. Based on the fitness value, fast non-dominated sorting and crowding calculation are performed, and the population is iteratively updated through selection, crossover, and mutation operations. When the number of iterations reaches the preset maximum number of generations, the set of all individuals in the population with a non-dominated level of 1 is output as the Pareto front solution set. Each solution in the Pareto front solution set represents a compromise allocation scheme for virtual control variables. The physical allowable limit values ​​of actual process control variables (drying temperature, circulating air volume, material residence time) are read from the parameter configuration table of the equipment control system and converted into safety constraint boundaries of the virtual control space. Priority configuration information is extracted from the process specification file of the current production task (for example, when the production task is for high-end pharmaceutical moisture-proof packaging, high priority weights of moisture content and particle uniformity are extracted, and when it is for conventional storage, high priority weights of energy consumption are extracted), and priority weight coefficient vectors corresponding to each target control object are generated. For each solution in the Pareto front solution set, first check whether the solution satisfies the safety constraint boundary and eliminate solutions that exceed the safety constraint; for the remaining feasible solutions, multiply the corresponding target control object deviation vector with the priority weight coefficient vector to calculate the comprehensive weighted deviation evaluation value; select the feasible solution with the smallest comprehensive weighted deviation evaluation value as the optimal virtual control solution.

[0053] The vector obtained from the optimal virtual control solution is multiplied with the constructed sensitivity decoupling matrix to complete the inverse mapping from the decoupled virtual space to the physical coupling space, and the corresponding actual process control parameter increment vector is output. This increment vector is added to the current actual process control variable running value to obtain the final drying temperature setpoint, circulating air volume setpoint, and material residence time setpoint.

[0054] The final drying temperature setpoint, circulating air volume setpoint, and material residence time setpoint are encapsulated into a structured control instruction data package, along with anomaly classification tags and source abnormality node identifiers extracted from the production process judgment result set. This generates a production management control scheme and sends it to the production line control system interface for execution.

[0055] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent management of silica gel desiccant production process based on multi-source sensor fusion, characterized in that, include: S1. Collect multi-source production monitoring data during the production process of the target silica gel desiccant; Based on the physical transport lag time, the monitoring data of different process links are time-series shifted and aligned to form a spatiotemporally aligned dataset of the same microscopic material batch. S2. Based on the spatiotemporal aligned dataset, construct the active phase transition diagram of the current physical phase transition stage, and deduce the actual transmission path and variation results of the wave in the production process based on the active phase transition diagram to form state correlation results; S3. When the state of a process node fluctuates, trace back along the active phase transition diagram and locate the source of the abnormal node based on the active phase transition edge. The source of the humidity backlash is updated by horizontally investigating the humidity backlash and crosstalk edges along the tracking path; the level is determined based on the edge attributes and investigation results, and a set of production process judgment results is generated. S4. Construct a sensitivity decoupling matrix based on the production process judgment result set and state association results, and obtain the decoupled virtual control quantity; Perform multi-objective optimization on the virtual control quantity and select the optimal virtual control solution; The parameters are then inversely transformed into actual process control parameters to generate production management and control schemes.

2. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 1, characterized in that, The spatiotemporally aligned dataset that forms the same batch of microscopic materials includes: Multi-source production monitoring data includes temperature data, humidity data, pressure data, equipment operating status data, and material conveying status data; Calculate the physical transmission lag time between each process step based on the material conveying status data, and perform time series translation and alignment on the multi-source production monitoring data of different process steps based on the physical transmission lag time to obtain the initial spatiotemporal aligned dataset. Based on the thermodynamic mechanism of silica gel adsorption, deep mechanism features are extracted from the temperature and humidity data in the spatiotemporally aligned dataset. The deep mechanism features include the drying rate derivative, dew point approximation, and estimated values ​​of the moisture content gradient inside the particles. The deep mechanism features are merged with the initial spatiotemporal aligned dataset to form the final spatiotemporal aligned dataset output.

3. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 2, characterized in that, The construction of the active phase transition diagram of the current physical phase transition stage includes: The process nodes are obtained based on the process division of the target silica gel desiccant production process; Based on the drying rate derivative in the spatiotemporally aligned dataset, the current physical phase transition stage of the silica gel material in each process step is identified. For materials that point from upstream process nodes to downstream process nodes, when the material is identified as being in a constant-rate drying period, the edge attribute connecting the upstream and downstream process nodes is defined as a fluctuation absorption edge; when the material is identified as being in a falling-rate drying period, the edge attribute connecting the upstream and downstream process nodes is defined as a fluctuation locking amplification edge. Both the wave absorption edge and the wave locking amplification edge are marked as the active phase transition edge; Humidity backlash and crosstalk between process steps are dynamically generated based on workshop space layout and exhaust status. The active phase transition diagram of the current physical phase transition stage is constructed by process node, active phase transition edge, and humidity backlash crosstalk edge.

4. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 3, characterized in that, The method of dynamically generating humidity backlash and crosstalk edges between process steps based on workshop space layout and exhaust status includes: Acquire humidity data from the exhaust port of the drying process, humidity data from the air intake area of ​​the downstream cooling or temporary storage process, and workshop exhaust pressure data in the spatiotemporally aligned dataset. When the humidity data at the exhaust port of the drying process is greater than the humidity data in the air intake area of ​​the downstream cooling process or temporary storage process, and the workshop exhaust pressure data is lower than the preset exhaust pressure threshold, a humidity backlash crosstalk edge is generated from the drying exhaust node to the downstream cooling or temporary storage node.

5. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 4, characterized in that, The deduction of the actual transmission path and variation results of fluctuations in the production process based on the active phase transition diagram forms state correlation results, including: When the rate of change of temperature data, humidity data, or material conveying status data of any process node in the time-space aligned dataset exceeds the corresponding baseline fluctuation range, it is determined that the process node has experienced a status fluctuation. The propagation direction of state fluctuations is deduced along the positive edge of the living phase transition diagram; When the simulation path passes through the fluctuation absorption edge, the fluctuation transmission type of the current simulation path is marked as attenuation and resolution type; When the simulation path passes through the fluctuation locking amplification edge, the fluctuation transmission type of the current simulation path is marked as the distortion amplification type; When the simulation path passes through the humidity backlash crosstalk edge, the fluctuation transmission type of the current simulation path is marked as environmental back damp crosstalk type; Summarize the deduction paths and corresponding fluctuation transmission types to generate state correlation results.

6. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 5, characterized in that, The reverse tracing along the active phase transition diagram, based on the active phase transition edge to locate the source anomaly node, includes: The process node experiencing state fluctuation is taken as the current traversal node; the upstream node of the current traversal node is obtained in reverse along the edge of the live phase transition graph, as well as the live phase transition edge connecting the current traversal node and the upstream node; if the live phase transition edge is a fluctuation absorption edge, the reverse traversal of the current traversal branch is terminated; if the live phase transition edge is a fluctuation locking amplification edge, the upstream node is recorded as the suspected fluctuation source node, and the upstream node is updated as the current traversal node, and the reverse traversal continues until the starting process node with no upstream node is reached; the finally recorded suspected fluctuation source node is determined as the source of the anomaly.

7. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 6, characterized in that, The process of laterally investigating humidity backlash crosstalk along the tracking path and updating the source abnormal node includes: Obtain all process nodes traversed during the reverse tracing process as a check node set; traverse the check node set to check if there is a humidity backlash crosstalk edge input to any node in the check node set; if there is, determine the node that receives the humidity backlash crosstalk edge as the crosstalk receiving node; obtain the generation timestamp of the humidity backlash crosstalk edge and the timestamp of the state fluctuation of the crosstalk receiving node. When the generated timestamp is within a preset time window before the timestamp of the state fluctuation, the sending node of the humidity backlash crosstalk edge is updated to the source abnormal node.

8. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 7, characterized in that, The process of classifying and grading based on edge attributes and investigation results generates a set of production process judgment results, including: If the source of the abnormal operation is updated by the node that sends the humidity backlash crosstalk edge, the operating status will be classified as a bypass crosstalk type abnormal operating status; if the source of the abnormal operation is determined by the suspected fluctuation source node, the reverse traversal path from the source of the abnormal operation node to the process link node where the state fluctuation occurs will be obtained. If the reverse traversal path contains a wave-locking amplification edge, the running state will be classified as a forward amplification abnormal running state; if the reverse traversal path contains only wave-absorbing edges, the running state will be classified as a wave running state. The operation status classification results, source abnormal nodes, active phase transition edges, and humidity backlash crosstalk edges are combined to generate a production process judgment result set.

9. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 8, characterized in that, The step of constructing a sensitivity decoupling matrix based on the production process judgment result set and state association results, and obtaining the decoupled virtual control quantity, includes: Obtain the actual process control variables in the production process of the target silica gel desiccant, as well as the target control objects corresponding to the current production task; the actual process control variables include drying temperature, circulating air volume, and material residence time; the target control objects include moisture content, particle uniformity, and energy consumption. Based on the fluctuation propagation type in the state correlation results, obtain the state partial derivatives of the actual process control variables with respect to the target control object; construct a sensitivity decoupling matrix based on the state partial derivatives; and perform an inversion operation on the sensitivity decoupling matrix to obtain the virtual control quantity.

10. The intelligent management method for silica gel desiccant production process based on multi-source sensor fusion according to claim 9, characterized in that, The process involves multi-objective optimization of the virtual control quantity and selection of the optimal virtual control solution. This is inversely transformed into actual process control parameters to generate a production management and control scheme, including: Based on the production process judgment result set, the objective function is to reduce the deviation between the target control object and the corresponding set benchmark value; using virtual control variables as decision variables, multi-objective non-dominated ranking optimization is performed to obtain the Pareto front solution set; based on the safety constraints of the current equipment operating status and the priority weight of the current production task, the optimal virtual control solution is selected from the Pareto front solution set; the optimal virtual control solution is inversely transformed through the sensitivity decoupling matrix to generate actual process control parameters; and a production management and control scheme is generated based on the actual process control parameters.

Images