Infrared drying method based on dynamic evolution digital twin modeling and twin cooperation

Abstract

The present application provides an infrared drying method based on dynamic evolution digital twin modeling and twin cooperation, the method comprising collecting equipment and material state, according to the partition physical connection rule and the material conveying alignment rule, first calculating the equipment radiation heat transfer capacity, then evolving the material temperature and moisture content and obtaining the predicted drying trajectory; under the control constraint constituted by the target and the execution boundary, generating and evaluating the candidate control instruction sequence, screening the optimal with the cost function and forming the normal baseline of working condition; through virtual-real alignment, deviation characteristics and adaptive threshold, positioning the fault partition and type, and online correcting the related model layer parameters; the method is used for predictive optimization control and early health diagnosis of the infrared drying process, which can ensure the outlet index and consider energy consumption at the same time and form a closed loop.

Description

Technical Field

[0001] This invention relates to the field of digital twin modeling technology in infrared drying, and particularly to an infrared drying method based on dynamic evolution digital twin modeling and twin collaboration. Background Technology

[0002] Infrared drying is used for dehumidification and heating of continuously conveyed materials. The equipment is arranged in zones along the conveyor belt, with the material receiving combined radiation and convection heat exchange under the influence of belt speed and airflow. Production must simultaneously meet the requirements of outlet moisture content, outlet temperature, and uniformity, while also considering energy consumption and zone coordination. There are physical relationships of adjacent influence and coverage between zones, and the heating sequence changes with belt speed. The process formula specifies the target range and allowable deviation. On-site data collection includes zone temperature, power, array status, airflow, belt speed, and material inlet and outlet temperatures and humidity. The zone coverage and material zones are marked using conveyor belt coordinates, and the process exhibits time-varying and spatial distribution characteristics.

[0003] Existing infrared drying control methods mostly employ zone setting and closed-loop regulation, allocating power according to outlet moisture content and temperature, and coordinating tuning with belt speed and air volume. Modeling often uses steady-state or quasi-steady-state radiation-convective heat transfer and empirical mapping to describe equipment and materials, reconstructing the distribution state using inlet and outlet measurements. Other methods utilize rolling prediction and digital simulation platforms, using zone signals to drive prediction and generating control commands based on process objectives and execution boundaries. During operation monitoring, health assessments are performed using rules or statistical characteristics of zone temperature, power, array status, air volume, and outlet volume, and settings are adjusted or maintained in case of anomalies. Material conveying typically estimates the heating time window based on belt speed, the order of zone action is determined by equipment layout, and control evaluation is ranked based on objective satisfaction and energy consumption estimation.

[0004] The above-mentioned scheme has shortcomings in the continuous drying scenario with multiple zones and varying belt speeds: First, it does not explicitly constrain the physical connection and influence transmission between adjacent zones in the model and prediction link; second, it lacks an alignment mechanism based on material conveying, making it difficult to maintain the consistent heating order of the same batch when the belt speed changes, resulting in unstable correspondence between control steps and material batches; third, the control assessment and health judgment lack virtual and real deviations and adaptive thresholds based on the normal operating baseline, making it difficult to locate faulty zones in a timely manner and implement localized corrections.

[0005] Therefore, an infrared drying method that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an infrared drying method based on dynamic evolution digital twin modeling and twin collaboration. The core technical problem to be solved by this application is: under the conditions of belt speed variation and partition coupling, to construct a set of integrated digital twin prediction, control and diagnosis methods that are constrained by the physical connection of adjacent partitions and can maintain the consistent heating order of the same batch of materials, so that control and diagnosis can be carried out under a unified alignment.

[0007] An infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to embodiments of the present invention includes:

[0008] S1. Obtain the zone temperature, zone power, emission array status, conveyor belt speed, air volume, and material temperature and moisture content of the infrared drying equipment. Calculate the material arrival sequence table of the zone based on the conveyor belt speed to form the current process status and actual control quantities of the equipment.

[0009] S2. Establish a dynamic evolution digital twin model based on the current state of the equipment process and the actual control variables. Write the adjacent order and coverage of the infrared partitions into the dynamic evolution digital twin model to form the physical connection rules of the partitions and limit the transferable influence between the partitions. Form the material conveying alignment rules according to the material partition arrival order table and limit the consistent heating order of materials under different belt speeds. Under the constraints of the partition physical connection rules and the material conveying alignment rules, the dynamic evolution digital twin model first calculates the change of the equipment's radiative heat transfer capacity, and then calculates the changes in material temperature and moisture content to obtain the predicted drying trajectory.

[0010] S3. Determine the outlet moisture content, outlet temperature and uniformity indicators based on the process formula, and form control constraints by combining power, belt speed and air volume execution boundaries, and generate candidate control command sequences under the control constraints;

[0011] S4. Input the candidate control command sequence into the dynamic evolution digital twin model, combine the partition physical connection rules and material conveying alignment rules to generate the corresponding predicted drying trajectory, calculate the target cost of the predicted drying trajectory under control constraints and screen to obtain the optimal control command, determine the predicted drying trajectory corresponding to the optimal control command as the normal baseline of the working condition and issue the optimal control command.

[0012] S5. Obtain the current state of the measured equipment process after the implementation of the optimal control command, align the current state of the measured equipment process with the normal working baseline according to the material conveying alignment rule, calculate the virtual and real deviation characteristics, and obtain the adaptive judgment threshold based on the predicted drying trajectory distribution.

[0013] S6. Based on the virtual-real deviation characteristics and adaptive judgment threshold, determine the early fault type and fault zone, generate health diagnosis results, and correct the parameters of the dynamic evolution digital twin model according to the health diagnosis results for the next control cycle.

[0014] Optionally, S1 is as follows:

[0015] The infrared drying equipment collects zone temperature, zone power, and transmitter array status according to zone number, and also collects air volume and conveyor belt speed to form zone temperature vector, zone power vector, transmitter array status vector, air volume and conveyor belt speed.

[0016] The material temperature and moisture content are collected at the inlet and outlet positions: material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content. These data are then combined with the zone temperature vector, zone power vector, transmitter array state vector, air volume, and conveyor belt speed to obtain the current process status of the equipment.

[0017] Based on the adjacent order and coverage of the infrared zones and the speed of the conveyor belt, the material is divided into material zones along the direction of the conveyor belt, corresponding to the number of zones. The order in which each material zone enters the coverage of each infrared zone is calculated, and a material zone arrival order table is generated.

[0018] The partition power vector, conveyor belt speed, and air volume are determined as the actual control variables, and the current state of the equipment process and the actual control variables are used as inputs to the dynamic evolution digital twin model.

[0019] Optionally, S2 is as follows:

[0020] The current state of the equipment process and the actual control quantity are used as inputs to the dynamic evolution digital twin model. Within the dynamic evolution digital twin model, an initial mapping is established to the material zone temperature vector and the material zone moisture content vector based on the material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content. This results in the material zone temperature vector and the material zone moisture content vector as the hidden state.

[0021] Based on the adjacent order and coverage of infrared partitions, write the partition physical connection rules into the dynamic evolution digital twin model, and configure the partition physical connection rules to limit the direction and range of the influence that can be transmitted between partitions and limit the information path to be transmitted only along the adjacent order of infrared partitions.

[0022] A partitioned input coding layer is established for the partitioned temperature vector, partitioned power vector, and transmitter array state vector by partition. The partitioned temperature, partitioned power, and transmitter array state of each infrared partition are fused and coded to output the partitioned coding vector.

[0023] Under the constraints of the physical connection rules of the partitions, an adjacent partition influence propagation layer is established. Only the partition coding vectors of adjacent partitions are weighted and summarized and nonlinearly mapped to obtain the partition coding vectors after the adjacent influence.

[0024] Establish a calculation layer for the equipment's radiative heat transfer capacity, fuse the partition coding vector after adjacent influences with the air volume, calculate the feature vector of the equipment's radiative heat transfer capacity for each partition, and thus complete the calculation of the changes in the equipment's radiative heat transfer capacity first.

[0025] Based on the material zone arrival order table, a material conveying alignment rule is formed. In the material conveying alignment layer, the material zone temperature vector and the material zone moisture content vector are indexed and mapped to be aligned according to the material conveying alignment rule and the conveyor belt speed, so as to obtain the aligned material zone temperature vector and the aligned material zone moisture content vector.

[0026] A material drying evolution calculation layer is established, which integrates the characteristic vector of equipment radiative heat transfer capacity with the aligned material partition temperature vector and the aligned material partition moisture content vector. Under the information path defined by the partition physical connection rules, the changes of material temperature and moisture content at the next moment are calculated. The material partition temperature vector and material partition moisture content vector are updated through residual connection. Finally, the state prediction output layer outputs the predicted drying trajectory containing the material partition temperature vector sequence and the material partition moisture content vector sequence.

[0027] Optionally, the material conveying alignment layer, under the constraints of the material conveying alignment rules and the material partition arrival order table, uses an alignment function to perform index mapping and alignment of the material partition temperature vector and the material partition moisture content vector, obtaining the aligned material partition temperature vector and the aligned material partition moisture content vector, wherein the alignment function is specifically:

[0028] ;

[0029] in, This is the aligned material partition temperature vector. This is the aligned moisture content vector for the material partitions. For material zone temperature vectors, This represents the moisture content vector for each material zone. For the index mapping matrix, For the index mapping matrix, the first Line 1 Column elements, To arrive at the sequence in position The corresponding material partition number, The heating sequence position and its value is to , The material is assigned a partition number and its value is... to , This is for controlling the cycle number.

[0030] Optionally, S3 specifically refers to:

[0031] The target requirements for outlet moisture content, outlet temperature and uniformity are determined based on the process formula, and these target requirements are set as target constraints.

[0032] Based on the execution boundaries of the power, belt speed, and air volume of the infrared drying equipment, the range of values ​​for the partition power vector, conveyor belt speed, and air volume, as well as the variation limits of adjacent control cycles, are determined to form execution boundary constraints.

[0033] The target constraints and execution boundary constraints are combined to form control constraints, which consistently limit the partition power vector, conveyor belt speed, and air volume of each control step in the candidate control command sequence.

[0034] Multiple candidate control command sequences are generated under control constraints, and the control steps of the candidate control command sequences are established with the heating sequence of the materials according to the material zoning arrival order table, so that the candidate control command sequences under different conveyor belt speeds correspond to the zoning action order of the same material batch.

[0035] Optionally, S4 specifically refers to:

[0036] Using the current state of the equipment process as the initial state for prediction, the candidate control command sequence is written one by one into the actual control input of the dynamic evolution digital twin model, so that the partition power vector, conveyor belt speed and air volume in the candidate control command sequence drive the dynamic evolution digital twin model to make rolling predictions according to the control steps.

[0037] In each control step, the partition physical connection rules are invoked to limit the direction and range of the influence that can be transmitted between partitions, so that the dynamic evolution digital twin model only transmits the influence of the partition coding vector of the adjacent partition in the adjacent partition influence transmission layer, and generates the partition coding vector after the adjacent influence accordingly.

[0038] In each control step, the partition coding vector after adjacent influence is fused with the air volume, and the calculation of the equipment radiation heat transfer capacity calculation layer is performed to obtain the equipment radiation heat transfer capacity feature vector of each partition, thereby completing the prediction of equipment radiation heat transfer capacity change formed by partition power vector, transmitter array state vector and air volume.

[0039] In each control step, the material conveying alignment rule is invoked to match the conveyor belt speed in the candidate control command sequence with the material partition arrival order table. The material conveying alignment layer is executed to perform index mapping and alignment on the material partition temperature vector and the material partition moisture content vector to obtain the aligned material partition temperature vector and the aligned material partition moisture content vector.

[0040] The feature vector of the equipment's radiative heat transfer capacity, the aligned material partition temperature vector, and the aligned material partition moisture content vector are input into the material drying evolution calculation layer. Under the information path defined by the partition physical connection rules, the changes in material temperature and moisture content at the next moment are calculated and the hidden state is updated through residual connection. Then, the state prediction output layer outputs the predicted drying trajectory of the corresponding candidate control command sequence.

[0041] Under control constraints, progressive constraint verification is performed on candidate control command sequences and candidate control command sequences that do not meet the control constraints are eliminated. For the predicted drying trajectory that meets the control constraints, the target cost is calculated. The target cost is jointly determined by the degree of satisfaction of the outlet moisture content, outlet temperature and uniformity indicators, and the energy consumption related quantities determined by the partition power vector, conveyor belt speed and air volume.

[0042] The optimal control command is obtained by screening based on the target cost. The predicted drying trajectory corresponding to the optimal control command is determined as the normal operating baseline, and the optimal control command is sent to the infrared drying equipment for execution.

[0043] Optionally, the target cost for calculating the predicted drying trajectory that satisfies the control constraints is calculated using a cost function, wherein the cost function is specifically:

[0044] ;

[0045] in, Candidate control instruction sequence The target cost, Candidate control instruction sequence In control step Predicted material outlet moisture content, Candidate control instruction sequence In control step Predicted material outlet temperature, Candidate control instruction sequence In control step The uniformity index To meet the export moisture content target requirements, For the allowable deviation of moisture content at export, To meet the target temperature requirements for the outlet, For the allowable deviation of the outlet temperature, The upper limit of the uniformity index is allowed. , , , To preset weights, Candidate control instruction sequence In control step The partition power vector of the first One portion, Candidate control instruction sequence In control step air volume, Candidate control instruction sequence In control step conveyor belt speed, , , Preset conversion factors are used to convert zone power, air volume, and conveyor belt speed into energy-related quantities. To control the cycle duration, This is a preset reference energy consumption used for energy consumption item normalization. Predict the step size for candidate control command sequences. For the number of partitions, For partition numbering, This is the prediction step number.

[0046] Optional, S5 specifically includes:

[0047] After the optimal control command is issued and implemented, the current state of the infrared drying equipment process is collected. The current state of the infrared drying equipment process includes the zone temperature vector, zone power vector, emission array state vector, air volume, conveyor belt speed, material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content.

[0048] According to the material conveying alignment rules, the current state of the measured equipment process is aligned with the normal working condition baseline in terms of the material arrival order in the partition, the baseline alignment time corresponding to the measured conveyor belt speed is determined, and the corresponding predicted drying trajectory segment is extracted.

[0049] The predicted material temperature and predicted moisture content at the corresponding outlet position are extracted from the predicted drying trajectory segment, and the difference between the predicted material outlet temperature and the predicted material outlet moisture content is calculated to obtain the virtual-real deviation characteristics.

[0050] The predicted drying trajectory segments in the normal operating baseline are calculated and distributed according to the alignment time, and an adaptive judgment threshold is generated accordingly, so that the adaptive judgment threshold and the virtual-real deviation characteristics adopt the same alignment caliber.

[0051] Optionally, step S6 specifically includes:

[0052] Using the virtual-real deviation feature as input, the virtual-real deviation feature is compared with the adaptive judgment threshold for each partition to determine the set of abnormal partitions that exceed the adaptive judgment threshold;

[0053] Based on the correspondence between the abnormal partition set and the normal baseline of the operating condition, combined with the partition power vector, the transmitter array state vector, and the measured change direction of air volume, the early fault type is determined and the abnormal partition set is identified as the fault partition.

[0054] The early fault types are combined with fault partitions to generate health diagnosis results.

[0055] The calibration range is limited by the health diagnosis results. The current state of the aligned measured equipment process is used as the calibration target. The parameters of the equipment radiation heat transfer capacity calculation layer and the material drying evolution calculation layer corresponding to the fault zone in the dynamic evolution digital twin model are adjusted so that the adjusted dynamic evolution digital twin model can be used in the next control cycle.

[0056] The beneficial effects of this invention are:

[0057] (1) This proposal presents an improved dynamic evolution digital twin modeling method and a twin collaborative control diagnostic technique. Compared to the approach of handling the influence of partitions through experience or free propagation and approximating the heating time window with belt speed, this method incorporates partition physical connection rules into the model, limiting the influence to be transmitted only along the adjacent order and coverage of infrared partitions; at the same time, it constructs material conveying alignment rules based on the material partition arrival order table, and uses index mapping to maintain the consistency of the heating order of the same batch. The model follows the calculation link of "equipment first, material later", first calculating the combined radiation-convection heat transfer capacity of the equipment, and then updating the temperature and moisture content of the material partitions with residuals. After the improvement, under the conditions of belt speed change and partition coupling, the consistency of prediction and control steps in the partition dimension can be guaranteed, reducing the evaluation bias caused by heating misalignment.

[0058] (2) This proposal puts forward a novel integrated control-assessment and health diagnosis technology. Candidate control command sequences are generated under unified control constraints formed by process objectives and execution boundaries. The predicted drying trajectory is obtained through rolling prediction by a twin model under the same alignment caliber. A cost function containing outlet moisture content, outlet temperature, uniformity, and energy consumption-related quantities is used for screening to form a normal operating baseline. After implementation, deviation characteristics are obtained through virtual-real alignment, and an adaptive judgment threshold is constructed based on the baseline prediction dispersion to locate fault zones and types. Only the parameters of the equipment radiative heat transfer capacity calculation layer and the material drying evolution calculation layer in the fault zone are corrected online, ensuring that the model maintains a consistent response with the field without changing the parameters of non-fault zones, supporting zone-level diagnosis and correction.

[0059] (3) This proposal puts forward a digital twin prediction-control-diagnosis integrated method and technology for infrared drying processes. Taking the physical connection of zones and the alignment of material conveying as the core, it integrates modeling, control evaluation, baseline generation, virtual-real deviation judgment and local parameter correction into the same calculation link. In continuous production lines, belt speed variation and zone coupling scenarios, this integrated method enables control and diagnosis to be carried out under a unified alignment, enhances the accessibility of process constraints and operational stability, and provides a basis for zone-level maintenance and adjustment. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0061] Figure 1 This is a flowchart of an infrared drying method based on dynamic evolution digital twin modeling and twin collaboration proposed in this invention;

[0062] Figure 2 This is a flowchart illustrating the acquisition of the current state and actual control quantities of the equipment process in an infrared drying method based on dynamic evolution digital twin modeling and twin collaboration proposed in this invention.

[0063] Figure 3 This is a flowchart illustrating the modeling and calculation process of a dynamic evolution digital twin model for an infrared drying method based on dynamic evolution digital twin modeling and twin collaboration proposed in this invention.

[0064] Figure 4 Schematic diagram of a drum infrared dryer Figure 1 ;

[0065] Figure 5 Schematic diagram of a drum infrared dryer Figure 2 ;

[0066] Figure 6 Schematic diagram of a drum infrared dryer Figure 3 . Detailed Implementation

[0067] In Example 1, reference Figures 1 to 3 An infrared drying method based on dynamic evolution digital twin modeling and twin collaboration includes:

[0068] S1. Obtain the zone temperature, zone power, emission array status, conveyor belt speed, air volume, and material temperature and moisture content of the infrared drying equipment. Calculate the material arrival sequence table of the zone based on the conveyor belt speed to form the current process status and actual control quantities of the equipment.

[0069] S2. Establish a dynamic evolution digital twin model based on the current state of the equipment process and the actual control variables. Write the adjacent order and coverage of the infrared partitions into the dynamic evolution digital twin model to form the physical connection rules of the partitions and limit the transferable influence between the partitions. Form the material conveying alignment rules according to the material partition arrival order table and limit the consistent heating order of materials under different belt speeds. Under the constraints of the partition physical connection rules and the material conveying alignment rules, the dynamic evolution digital twin model first calculates the change of the equipment's radiative heat transfer capacity, and then calculates the changes in material temperature and moisture content to obtain the predicted drying trajectory.

[0070] S3. Determine the outlet moisture content, outlet temperature and uniformity indicators based on the process formula, and form control constraints by combining power, belt speed and air volume execution boundaries, and generate candidate control command sequences under the control constraints;

[0071] S4. Input the candidate control command sequence into the dynamic evolution digital twin model, combine the partition physical connection rules and material conveying alignment rules to generate the corresponding predicted drying trajectory, calculate the target cost of the predicted drying trajectory under control constraints and screen to obtain the optimal control command, determine the predicted drying trajectory corresponding to the optimal control command as the normal baseline of the working condition and issue the optimal control command.

[0072] S5. Obtain the current state of the measured equipment process after the implementation of the optimal control command, align the current state of the measured equipment process with the normal working baseline according to the material conveying alignment rule, calculate the virtual and real deviation characteristics, and obtain the adaptive judgment threshold based on the predicted drying trajectory distribution.

[0073] S6. Based on the virtual-real deviation characteristics and adaptive judgment threshold, determine the early fault type and fault zone, generate health diagnosis results, and correct the parameters of the dynamic evolution digital twin model according to the health diagnosis results for the next control cycle.

[0074] In this embodiment, step S1 specifically includes:

[0075] In one embodiment, the infrared drying equipment is divided along the conveyor belt direction into sections. One infrared zone The number of partitions is specified, and the partition numbers are in the same order as the adjacent infrared partitions. Each infrared partition has a corresponding coverage area, which is calibrated using conveyor belt coordinates. Each infrared zone is given its starting coordinates. and termination coordinates ,in For partition number and Values to This is used to perform index mapping alignment in the material delivery alignment layer, which is then used to generate a material partition arrival order table and support the dynamic evolution digital twin model.

[0076] Control cycle is In the Each control cycle collects data on zone temperature, zone power, transmitter array status, airflow, and conveyor belt speed, among which... To control the cycle number, for the first... Each infrared zone collects zone temperature. Zone power Transmit array status ,Will The temperature vectors of the zones are arranged in order of their zone numbers. ,Will The partition power vector is formed according to the partition number order. ,Will The transmit array state vector is composed according to the partition number order. ,in , , All dimensions are And the vector of the first The components correspond to the first... The data collected includes the zone temperature, zone power, and transmitter array status for each infrared zone, as well as airflow data. The conveyor belt speed was collected as follows: To ensure the availability of the partitioned structure constraint calculation link, , , The component order is consistent with the partition number, so that the subsequent adjacent partition influence propagation layer can carry out influence propagation according to the adjacent relationship defined by the partition physical connection rules;

[0077] The material inlet temperature was collected at both the inlet and outlet locations along the material conveying direction. Moisture content of material inlet Material outlet temperature Moisture content of material outlet ,Will , , , , , , , , The current state of the device process is obtained by combining fixed fields in a specific order. ,in, As the current state input of the device process in the dynamically evolving digital twin model, it drives the partition input coding layer to fuse and encode the partition temperature, partition power, and emission array state of each infrared partition, and supports the internal model based on... , , , Establish the initial hidden states of the material zone temperature vector and the material zone moisture content vector;

[0078] Based on the adjacent order and coverage of infrared zones, and conveyor belt speed The material is divided along the conveyor belt into sections corresponding to the number of partitions. Each material partition is assigned a unique partition number, with each partition number corresponding to a specific partition number. The length of each material partition along the conveyor belt is equal to the length of the corresponding infrared partition's coverage area along the conveyor belt. The material inlet section is used as the origin of the conveyor belt coordinate system. The position of each material partition in the conveyor belt coordinate system is described by its leading edge and trailing edge positions. Within each control cycle, the leading edge and trailing edge positions of the material partitions are synchronously advanced along the conveyor belt direction, with the displacement measured by the conveyor belt speed. With control cycle The product of these factors is then used to calculate the order in which material zones enter the infrared zone coverage area: for each material zone, the order of adjacent infrared zones is used to determine whether the leading edge of the material zone has entered the starting coordinates of the corresponding infrared zone. Upon entry, the correspondence between the material zone and the infrared zone is recorded, and then it is determined whether the leading edge of the material zone has crossed the termination coordinate. To proceed to the next infrared zone, the order in which each material zone enters within this control cycle is recorded as a material zone arrival sequence table. , This is used to form material conveying alignment rules, so that the dynamic evolution digital twin model can maintain the consistency of the partitioning order of the same material batch through index mapping at different conveyor belt speeds;

[0079] partition power vector Conveyor belt speed Air volume Determined as the actual control quantity and the current state of the equipment process. With actual control quantity As input to the dynamic evolution digital twin model, the dynamic evolution digital twin model can simultaneously obtain the partition temperature vector, partition power vector, and emission array state vector corresponding to the partition structure information, and the material partition arrival order table corresponding to the material conveying information in subsequent steps through the above method. This supports the continuous reference of the partition physical connection rules and the material conveying alignment rules within the same computing link and outputs the predicted drying trajectory.

[0080] In this embodiment, step S2 specifically includes:

[0081] In one embodiment, the dynamically evolving digital twin model is based on the current state of the device process. and actual control quantity As input, where Includes partitioned temperature vector Partition power vector Transmit array state vector Air volume Conveyor belt speed Material inlet temperature Moisture content of material inlet Material outlet temperature Moisture content of material outlet , Depend on , , Composition, material arrival order table is as follows , For the number of partitions, For control cycle number;

[0082] A hidden state is established within the dynamic evolution digital twin model, which includes the material zone temperature vector. Moisture content vector of material zones , and All dimensions are And the first The component corresponds to the first Each material partition, the initial mapping adopts a one-dimensional distribution reconstruction method with inlet and outlet measurement constraints: [The remaining text appears to be incomplete and requires further context.] The first component is assigned the value ,Will The Component assignment value The intermediate components are linearly interpolated and assigned values ​​from the inlet to the outlet according to the partition number. The first component is assigned the value ,Will The Component assignment value The intermediate components are linearly interpolated from the inlet to the outlet according to the partition number, so that the hidden state is consistent with the infrared partition structure in the partition dimension and can be directly updated in a rolling manner.

[0083] Based on the adjacent order and coverage of infrared partitions, physical connection rules for the partitions are written into the dynamically evolving digital twin model. The physical connection rules for the partitions adopt an adjacent constraint table. express, Each partition is listed by partition number. The transitive influence of a set of adjacent partitions, which is composed of partitions... The model is composed of adjacent infrared partitions with contiguous coverage areas, and the influence transmission is limited to the direction of the conveyor belt, from the upstream partition to the downstream partition. Subsequent aggregation of adjacent information and injection of adjacent influences are only allowed to reference. Given a set of adjacent partitions, the relationship between partitions is determined by the device partition structure;

[0084] Establish a partitioned input encoding layer for the first... Infrared partition construction partition input , By the The partition temperature, partition power, and transmit array status of each partition are concatenated in a fixed order. The partition input coding layer uses fully connected neurons. Perform fusion coding and output partition coding vectors. , Used to characterize the Each partition can be used for a comprehensive state representation of the subsequent impact transmission within the current control cycle;

[0085] Establish an adjacent partition influence propagation layer for each partition. According to the adjacent constraint table Read the adjacent partition set, only for partitions within the set. Perform restricted weighted aggregation, which is performed dimension-wise by partition: for each adjacent partition Take the corresponding partition encoding vector The adjacent summation vector is obtained by multiplying the summation vector by the trainable weight coefficients corresponding to the adjacent directions. This adjacent summation vector is then input into a fully connected neuron and mapped through a non-linear activation function to obtain the partitioned encoding vector after the adjacent influence. This allows the transmission of thermal effects between zones to occur along information pathways strictly defined by the physical connection rules of the zones;

[0086] Establish a layer for calculating the radiation heat transfer capacity of the equipment, and encode the partitioned vectors after the influence of adjacent elements. With air volume The fusion process employs fully connected neurons for nonlinear mapping to obtain the feature vector of the device's radiative heat transfer capacity. , Used to characterize the The radiation and convection heat transfer capacity of each zone in the current control cycle is determined by the zone power, the status of the emission array, and the air volume, thus completing the calculation link of equipment-side capacity changes before the material drying evolution calculation.

[0087] According to the material arrival order table Form material conveying alignment rules, and perform index mapping alignment on the hidden state in the material conveying alignment layer. Represented as the arrival order sequence arranged according to the order of heating. ,in Indicates the first Within the first control cycle, the first The material zone number that enters the infrared zone coverage area, and the Based on conveyor belt speed Based on the determination of the material position and coverage area driven by the material conveying system, the material conveying alignment layer constructs an index mapping matrix. and to and Rearrange the materials to obtain the aligned temperature vectors for each material partition. Aligned material partition moisture content vector Its calculation follows the alignment function, specifically as follows:

[0088] ;

[0089] in, This is the aligned material partition temperature vector. This is the aligned moisture content vector for the material partitions. For material zone temperature vectors, This represents the moisture content vector for each material zone. For the index mapping matrix, For the index mapping matrix, the first Line 1 Column elements, To arrive at the sequence in position The corresponding material partition number, The heating sequence position and its value is to , The material is assigned a partition number and its value is... to , For control cycle number;

[0090] Establish a material drying evolution calculation layer and incorporate the characteristic vector of equipment radiative heat transfer capacity. With Alignment , The process involves merging the data and calculating the changes at the next time step within the information path defined by the partition physical connection rules. Specifically, this involves: for each partition... Construct the material evolution input vector, and and , The corresponding number in the middle The components of each material partition are concatenated in a fixed order and input into a fully connected neuron to obtain temperature increment components and moisture content increment components. These temperature increment components and moisture content increment components are numerical outputs that can be directly added to update the hidden state. Then, based on the adjacent constraint table... Only the constrained influence corresponding to the upstream adjacent partition is introduced, and this constrained influence is used together with the incremental component to form the partition. The final increment is obtained by using residual connections to connect the final increment component with the current... , The corresponding components are added together to form the updated result. and Then, the state prediction output layer outputs the predicted drying trajectory, which at least includes the material zone temperature vector sequence. With material zone moisture content vector sequence ,in The prediction step number is given, and the dimension of the vector at each time step in the sequence is 1. .

[0091] In this embodiment, step S3 specifically includes:

[0092] In one embodiment, step S3 is used to construct a unified sequence of control constraints and candidate control commands before control evaluation, so that the candidate control command sequence can be called by the subsequent dynamic evolution digital twin model according to the partitioning action order of the same material batch. Let the control cycle number be... The number of partitions is The prediction step size of the candidate control command sequence is ,in The number of control steps contained in the candidate control instruction sequence;

[0093] Based on the process formulation, the target requirements are read and transformed into target constraints that can be directly used for constraint determination. The target requirement for outlet moisture content is expressed as follows: and its permissible deviation The target temperature requirement for the outlet is expressed as and its permissible deviation The uniformity index target requirement is expressed as: and its allowed upper limit The uniformity index is characterized by the degree of dispersion of the moisture content of each material zone at the outlet position in the zone dimension. The degree of dispersion is obtained by taking the maximum and minimum values ​​of the moisture content vector of the material zone corresponding to the outlet position and calculating the difference. The outlet moisture content, outlet temperature, and uniformity index are respectively limited within the range of the target requirements and the allowable deviation or allowable upper limit to form target constraints, which are used to determine the degree of target satisfaction of the predicted drying trajectory in the subsequent process.

[0094] Based on the execution boundaries of the infrared drying equipment's power, belt speed, and air volume, execution boundary constraints are formed, and the power vector of the zones is defined. Each partition component is given a lower power limit With power limit ,in The partition number has a value of to And give power variation limits for adjacent control cycles. It is used to limit the range of power variation between adjacent control steps and to set a lower limit for the conveyor belt speed. With belt speed limit And give the belt speed variation limit for adjacent control cycles. Set a lower limit for the air volume. With upper limit of air volume And give limits on airflow variation in adjacent control cycles. Boundary constraints are used to ensure that the subsequently generated candidate control command sequence varies within the executable range of the device and is consistent with the actual control input dimension of the dynamically evolving digital twin model.

[0095] The target constraints and execution boundary constraints are combined to form control constraints. The control constraints consist of two parts. One part limits the range of values ​​and changes of the partition power vector, conveyor belt speed and air volume of the candidate control command sequence in each control step. The other part limits the range of the predicted drying trajectory generated by the dynamic evolution digital twin model of the candidate control command sequence in terms of outlet moisture content, outlet temperature and uniformity index. The two parts of the control constraints are configured together as a unified screening criterion so that the elimination of candidate control command sequences and the calculation of target cost in subsequent steps adopt the same constraint caliber.

[0096] Multiple sets of candidate control command sequences are generated under control constraints, based on the partitioned power vector in the current state of the equipment process. Conveyor belt speed Air volume As the initial control variable, a candidate control instruction sequence is generated. ,in Candidate number, Indicates from the first The control step to the first A sequence of partitioned power vectors for each control step. For conveyor belt speed sequence, For the air volume sequence, candidate generation is achieved through a stepwise discrete expansion method: for each control step, for each partition... Select the partition power candidate as the power of the previous control step, and increase the power of the previous control step. Power reduction in the previous control step Three discrete values ​​are available, and out-of-bounds values ​​are truncated to... to Within the range, the candidate conveyor belt speed is the belt speed of the previous control step, or the belt speed of the previous control step increased. The previous control step reduced the belt speed. Three discrete values ​​and truncation to Within the range, the air volume candidate is selected as the air volume of the previous control step, and the air volume of the previous control step is increased. The previous control step reduced the air volume. Three discrete values ​​and truncation to Within the range, multiple sets of discrete values ​​are obtained by recursively combining them according to the control steps. It retains only the candidate control instruction sequence that satisfies the execution boundary constraints, and then determines the order of arrival based on the material partitioning. Establish a correspondence between the control steps of the candidate control command sequence and the material heating sequence: This includes the conveyor belt speed sequence within the candidate control command sequence. As the input for material positioning, the displacement is calculated step by step and the position of the material partition in the conveyor belt coordinates is updated. Then, based on the coverage range of the infrared partitions, the order in which the material partitions enter the coverage range of each infrared partition is determined, resulting in a candidate arrival order table. ,by The material conveying alignment and calling order of candidate control command sequences in the dynamic evolution digital twin model is determined, so that the candidate control command sequences at different conveyor belt speeds correspond to the partition action order of the same material batch.

[0097] In this embodiment, step S4 specifically includes:

[0098] In one embodiment, step S4 is used to perform comparable rolling prediction and screening of multiple sets of candidate control command sequences within the same dynamic evolution digital twin model, and to determine the predicted drying trajectory corresponding to the optimal control command obtained by screening as the normal operating baseline. Let the current control cycle number be... The number of partitions is The candidate control instruction sequence number is The candidate control command sequence prediction step size is The control cycle duration is The current state of the equipment process is This includes partition temperature vectors. Partition power vector Transmit array state vector Air volume Conveyor belt speed Material inlet temperature Moisture content of material inlet Material outlet temperature Moisture content of material outlet The physical connection rules for partitions use an adjacency constraint table. This indicates that the material arrival order table for each zone is as follows: Control constraints are formed by a combination of objective constraints and execution boundary constraints;

[0099] by As the initial state for prediction, the candidate control command sequence Write the actual control inputs one by one into the dynamic evolution digital twin model, and... The input dynamic evolution digital twin model is used to initialize the hidden state, which includes the material zone temperature vector. Moisture content vector of material zones , and Depend on , , , The initial mapping is completed within the model, so that each candidate control command sequence is predicted in a rolling manner starting from the same current state of the device process and the same hidden state starting point.

[0100] In each control step The system invokes partition physical join rules, using the adjacent constraint table. By limiting the direction and range of influence that can be transmitted between partitions, the dynamic evolution digital twin model ensures that the influence transmission layer in the adjacent partitions only performs influence transmission on the partition coding vectors of adjacent partitions. Specifically, the partition input coding layer first constructs partition input and outputs partition coding vectors based on the partition temperature vector, partition power vector, and transmitter array state vector of the control step, and then... Only the partition coding vectors of the upstream adjacent partitions and the current partition are summarized and nonlinear mapping is completed to obtain the partition coding vector after the adjacent influence, thereby ensuring that the thermal influence between partitions propagates along the adjacent order of infrared partitions in the predicted drying trajectory;

[0101] In each control step The partition coding vector after the adjacent influence is combined with the control step air volume. The system integrates and executes the equipment radiation heat transfer capacity calculation layer to obtain the equipment radiation heat transfer capacity feature vector of each zone. The equipment radiation heat transfer capacity feature vector is used to characterize the equipment-side heat transfer capacity change determined by the zone power vector, the emission array state vector, and the air volume, and serves as the equipment-side input of the subsequent material drying evolution calculation layer, so that the dynamic evolution digital twin model can carry out predictions in the order of equipment first and then material.

[0102] In each control step The material conveying alignment rules are invoked to align the conveyor belt speed in the candidate control command sequence. Establish a correspondence with the material arrival sequence table to determine the material heating sequence for this control step, specifically: based on... The material is driven to advance to the designated zone position, and the entry order is determined based on the infrared zone coverage area. A candidate arrival order table corresponding to this control step is obtained, and the material is aligned in the material conveying alignment layer according to the candidate arrival order table. and Perform index mapping alignment to obtain the aligned material partition temperature vector and the aligned material partition moisture content vector, so that the candidate control command sequences under different conveyor belt speeds are comparable under the partition action order caliber of the same material batch;

[0103] The radiative heat transfer capacity feature vector of the equipment, the aligned material partition temperature vector, and the aligned material partition moisture content vector are input into the material drying evolution calculation layer. Under the information path defined by the partition physical connection rules, the changes in material temperature and moisture content at the next moment are calculated, and updated through residual connections. and get and The state prediction output layer predicts the drying trajectory based on the updated hidden state output, and extracts the predicted material outlet temperature corresponding to the outlet position from the predicted drying trajectory. Predicted material outlet moisture content Meanwhile, the difference between the maximum and minimum values ​​of the predicted material moisture content vector corresponding to the export location along the partition dimension is used as a uniformity index. ;

[0104] Under control constraints, progressive constraint verification is performed on candidate control command sequences, and candidate control command sequences that do not meet the control constraints are eliminated. Progressive constraint verification includes verifying the range of values ​​for the partition power vector, conveyor belt speed, and air volume for each control step, as well as verifying the variation restrictions between adjacent control steps. Target constraint verification is also performed on the corresponding predicted drying trajectory. For candidate control command sequences that meet the control constraints, the target cost is calculated. The target cost adopts a unified step-by-step cumulative calculation link, combining the target satisfaction degree of outlet moisture content, outlet temperature, and uniformity indicators with energy consumption-related quantities on the same scale, and calculating according to the cost function, specifically:

[0105] ;

[0106] in, Candidate control instruction sequence The target cost, Candidate control instruction sequence In control step Predicted material outlet moisture content, Candidate control instruction sequence In control step Predicted material outlet temperature, Candidate control instruction sequence In control step The uniformity index To meet the export moisture content target requirements, For the allowable deviation of moisture content at export, To meet the target temperature requirements for the outlet, For the allowable deviation of the outlet temperature, The upper limit of the uniformity index is allowed. , , , To preset weights, Candidate control instruction sequence In control step The partition power vector of the first One portion, Candidate control instruction sequence In control step air volume, Candidate control instruction sequence In control step conveyor belt speed, , , Preset conversion factors are used to convert zone power, air volume, and conveyor belt speed into energy-related quantities. To control the cycle duration, This is a preset reference energy consumption used for energy consumption item normalization. Predict the step size for candidate control command sequences. For the number of partitions, For partition numbering, This refers to the prediction step number;

[0107] The optimal control command is selected from all candidate control command sequences that satisfy the control constraints based on the target cost, and the predicted drying trajectory corresponding to the optimal control command is determined as the normal operating baseline. The normal operating baseline is then compared with... The optimal control command, the physical connection rules of the partition, and the material conveying alignment rules are all in one correspondence, so that the subsequent steps can align the current state of the measured equipment process with the normal operating baseline under the same alignment caliber and calculate the virtual and real deviation characteristics. At the same time, the optimal control command is sent to the infrared drying equipment for execution to enter the next control cycle.

[0108] In this embodiment, step S5 specifically includes:

[0109] In one embodiment, step S5 is used to align the measured current state of the infrared drying equipment with the normal operating baseline under the same material zone arrival order caliber after the optimal control command is issued and implemented, and to generate virtual-real deviation features and adaptive judgment thresholds from the aligned predicted drying trajectory segments, assuming the number of zones is... The control cycle number is The candidate number corresponding to the optimal control command is The normal baseline of the operating condition corresponding to the optimal control command includes the predicted drying trajectory and a candidate arrival sequence table. The candidate arrival sequence table is denoted by control step as follows: ,in This refers to the prediction step number;

[0110] The current state of the measured equipment is collected in the next control cycle after the optimal control command is issued and implemented. superscript This indicates that the measured values ​​were obtained by collecting temperature vectors for each zone of the infrared drying equipment according to the zone number. Partition power vector Transmit array state vector And collect air volume With conveyor belt speed The material inlet temperature is collected as The moisture content of the material at the inlet was collected as follows: The material outlet temperature is collected as follows: The moisture content of the material at the outlet was collected as follows: The above collected data are combined in order of the current status field of the device process. This is used for subsequent alignment and deviation calculations;

[0111] According to the material conveying alignment rules Aligning with the normal operating baseline in terms of material arrival order in the partitions, firstly, based on the measured conveyor belt speed... The system drives the material partitions to advance to their positions and determines the order in which material partitions enter the coverage area of ​​each infrared partition based on the infrared partition coverage area, generating a measured material partition arrival order table. Subsequently, a search was conducted in the normal operating baseline for... The corresponding baseline alignment time, the baseline alignment time is used express, To meet The prediction step number is selected as the smallest prediction step number when multiple prediction step numbers satisfy the conditions. When no prediction step number meets the conditions, for each candidate prediction step number... Calculate the number of consistent components, where the number of consistent components is... and When comparing the arrival order of each item, the counts of equal items are used, and the prediction step number with the largest number of consistent components is selected as the prediction step number. In determining Then, the predicted drying trajectory from the normal operating baseline is extracted. Start, continuous The predicted drying trajectory segment for each control step, where Preset segment length;

[0112] The predicted material outlet temperature and predicted material outlet moisture content are extracted from the predicted drying trajectory segment, and the difference between these values ​​and the measured outlet quantity is calculated to obtain the virtual-actual deviation characteristic. This is then used to align the predicted drying trajectory segment at the initial control step. The predicted material outlet temperature at the location is denoted as The predicted moisture content of the material at the outlet is denoted as The temperature difference was calculated. The difference in moisture content was calculated to obtain ,Will and Composition of virtual and real deviation characteristics , As the input for subsequent decisions, and using the same alignment as the adaptive decision threshold;

[0113] The predicted drying trajectory segments in the normal operating baseline are scattered according to the alignment time, and an adaptive judgment threshold is generated accordingly. The predicted material outlet temperature sequence at the outlet position is extracted for each control step within the predicted drying trajectory segment. Compared with the predicted material outlet moisture content series ,in The sequence number within the fragment and its value is... to Calculate the mean and standard deviation for each of the two sequences, where the mean is calculated using the average of the two sequences. Sum of the sequence samples and divide by The standard deviation is obtained by squaring the difference between each sample and the mean, and then... Sum of the squares and divide by The result is then squared to obtain the temperature adaptive judgment threshold. The standard deviation of the outlet temperature series is multiplied by a preset amplification factor. The standard deviation of the export moisture content series is multiplied by a preset amplification factor to obtain the adaptive moisture content judgment threshold. ,Will and Composition of adaptive decision threshold ,make Characteristics of virtual and real deviation Arrival sequence table of the same measured material zone The determined alignment time allows the adaptive judgment threshold and the virtual-real deviation characteristics to be directly used for subsequent deviation judgment under the same material conveying alignment caliber.

[0114] In this embodiment, step S6 specifically includes:

[0115] In one embodiment, step S6 is used to obtain the virtual-real deviation characteristics. With adaptive decision threshold Then, abnormal partitions are located, early fault types are determined, and online correction is performed on the dynamic evolution digital twin model within a limited correction range. Let the number of partitions be... The control cycle number is The partition number is And the value is to The normal baseline alignment time under operating conditions is The candidate number corresponding to the optimal control command is ;

[0116] Based on the characteristics of virtual and real deviation Using this as input, the virtual-to-real deviation features are compared with an adaptive decision threshold for each partition, identifying the set of abnormal partitions that exceed the adaptive decision threshold. Among them, adaptive judgment threshold Temperature adaptive threshold determination Adaptive determination threshold for moisture content To obtain the virtual-real deviation characteristics that can be compared partition by partition, at the alignment time... Extract the predicted material zone temperature vector from the predicted drying trajectory segment of the normal operating baseline. With the predicted material zone moisture content vector The measured material inlet temperature With material outlet temperature Construct the measured material zone temperature vector using linear interpolation based on zone number. The interpolation rule is: Let , ,right Assign values ​​by linear interpolation between the inlet and outlet according to the zone number, and assign the measured material inlet moisture content. With the moisture content of the material outlet Construct the measured material zone moisture content vector in the same manner. For each partition Calculate temperature deviation components component with moisture content deviation and will and Compare, and The comparison will be performed, and if any comparison result exceeds the corresponding threshold, the partition will be determined. Write to abnormal partition set , will partition all and The virtual-real deviation feature is constructed by splicing the components in the order of the partition numbers. ,make It can be directly used for partition-level anomaly location;

[0117] Based on the abnormal partition set By analyzing the partition correspondence in the normal operating baseline, combined with the partition power vector, transmitter array state vector, and measured airflow change direction, the early fault type is determined, and the abnormal partition set is identified as the fault partition. ,in, , , The measured zone power vector, measured transmitter array state vector, and measured airflow, stored from the previous control cycle, change direction of the measured zone power vector from... The measured change direction of the transmit array state vector is determined partition by partition. The direction of measured air volume change was determined zone by zone. Let the power deviation tolerance be... The air volume deviates from the tolerance. For each abnormal partition Read the optimal control command in the control step partition power components With air volume And calculate the execution deviation. and Define "inadequate heating direction" as and The "insufficient drying direction" is defined as When partition Satisfy the direction of insufficient heating and At that time, the early fault type was determined to be insufficient partition power execution. Satisfy the direction of insufficient heating and and At that time, the early fault type is determined to be an abnormal state of the transmit array, when the partition Satisfying the direction of insufficient drying and At that time, the early fault type was determined to be insufficient airflow. Assigned value This is used to define the partition range for subsequent corrections;

[0118] Early fault types and fault partitions Combine to generate health diagnosis results ,Will Construct a data structure containing "early fault type identifier" and "fault partition number set" to enable... It can be directly used to limit the open update range of parameters in dynamically evolving digital twin models;

[0119] Based on health diagnosis results By limiting the correction range and using the current state of the aligned measured equipment process as the correction target, adjustments are made to the calculation layer parameters of the equipment radiative heat transfer capacity and the material drying evolution calculation layer parameters corresponding to the fault partition in the dynamic evolution digital twin model. The equipment radiative heat transfer capacity calculation layer parameters are denoted as... The parameters of the material drying evolution calculation layer are denoted as... The current state of the aligned measured equipment is recorded as follows: , Depend on , , , , , , , , The interpolated , Together they form a system, organized according to the partitioned order of material conveying alignment rules, so that... and , The partition numbers are matched one by one. During calibration, the parameters corresponding to the non-faulty partitions are frozen, and only the parameters corresponding to the non-faulty partitions are open. The matters involved Parameter subset and parameter subset, With optimal control commands in control steps The partition power vector, conveyor belt speed, and air volume are written into the dynamic evolution digital twin model and a forward calculation is performed to obtain the predicted material partition temperature vector at the alignment moment. With the predicted material zone moisture content vector Using the correction error on the faulty partition as the correction target: for each calculate and The scalar correction loss is obtained by summing the squares of the above errors along the fault partition dimension, and a preset learning rate is used. Update steps with preset Perform iterative adjustments: In each iteration, keep the input unchanged, recalculate the correction loss, obtain the gradient of the open parameter subset through backpropagation, update the open parameter subset according to gradient descent, keep the other parameters unchanged, and use the updated dynamic evolution digital twin model for rolling prediction and candidate control command sequence evaluation in the next control cycle.

[0120] in, This is a characteristic of virtual-real deviation. To adaptively determine the threshold, The temperature adaptive determination threshold is used. The moisture content adaptive determination threshold is set. For the abnormal partition set, For fault partitioning, For health diagnosis results, This is the candidate number corresponding to the optimal control command. This is the time when the baseline is aligned under normal operating conditions. For the predicted material partition temperature vector at the alignment time, To align the predicted material zone moisture content vector at the time of alignment. This is the measured temperature vector of the material zone obtained through interpolation. This is the measured moisture content vector of the material zone obtained through interpolation. For power deviation tolerance, For air volume deviation from tolerance, Calculate the layer parameters for the equipment's radiative heat transfer capacity. Parameters for calculating the material drying evolution. The current state of the aligned measured equipment process. For learning rate, To update the step count.

[0121] Example 1:

[0122] This embodiment applies the infrared drying method based on dynamic evolution digital twin modeling and twin collaboration to a drum infrared dryer, such as... Figure 4-6 As shown, the equipment consists of a rotating cylinder, an infrared emitting array arranged in sections around the outer periphery of the cylinder, temperature measuring points in each section, power execution units for each section, a ventilation system, inlet and outlet ports, and a drive mechanism. The drying section is divided into several infrared zones along the cylinder's axial direction, each zone corresponding to a fixed axial coverage area. The zone numbers increase sequentially from the inlet to the outlet along the material's axial direction. The drive mechanism provides the cylinder's rotational speed, and the ventilation system provides adjustable airflow.

[0123] The control cycle is a fixed duration. Within each control cycle, the zone temperature, zone power, and emission array status of each infrared zone are collected. Simultaneously, the cylinder rotation speed, airflow, and material inlet temperature, inlet moisture content, outlet temperature, and outlet moisture content are also collected to form the current state of the equipment process and the actual control quantities. Since the material in this equipment tumbles and rolls within the cylinder due to rotation, and is axially propelled by the tilt angle, lifting plates, and loading rate, this embodiment replaces "conveyor belt speed" with "equivalent axial conveying speed." This speed is calculated by a calibration model based on parameters such as cylinder rotation speed, equipment tilt angle, feed rate, and material moisture content. Based on the equivalent axial conveying speed and the coverage area of ​​each zone, the material is divided into material zones along the axial direction, corresponding to the number of zones. The axial position of the material zones is updated in each control cycle, and the order in which the material zones enter the coverage area of ​​each infrared zone is calculated to generate a material zone arrival order table.

[0124] Based on the above inputs, a dynamic evolutionary digital twin model is established: the adjacent order and coverage of the partitions are written into the model to form the physical connection rules of the partitions, limiting the influence to be transmitted only along the axial adjacent partitions; according to the arrival order table, material conveying alignment rules are formed, and under different rotation speeds and different equivalent conveying speeds, the heating order of the same batch of materials is kept consistent through index mapping. The model follows the "equipment first, material second" link, first integrating partition temperature, partition power, array status, and air volume to obtain the radiative heat transfer capacity characteristics of each partition, and then combining the aligned material partition temperature and moisture content latent state to obtain the predicted drying trajectory through rolling evolution.

[0125] Based on the process formula, the outlet moisture content, outlet temperature, and uniformity targets are set, and control constraints are formed by combining the execution boundaries of zone power, drum rotation speed, and air volume. Under these constraints, multiple sets of candidate control command sequences are generated. These candidate sequences are input into a twin model to obtain the corresponding predicted drying trajectory. The optimal control command is selected according to a cost function that includes outlet index deviation and energy consumption, and the trajectory corresponding to the optimal command is issued as the normal operating baseline for execution. After execution, the measured status is collected, aligned with the baseline according to the material conveying alignment rules, and the virtual-real deviation characteristics are calculated. An adaptive threshold is formed based on the baseline trajectory dispersion. When the deviation exceeds the threshold, the fault zone and type are located by combining the directional characteristics of zone power deviation, array status abnormality, or air volume abnormality. Only the parameters of the radiation heat transfer capacity calculation layer and material evolution layer corresponding to the fault zone are corrected online, so that the corrected model can be used for the next control cycle, realizing the predictive optimization control and early health diagnosis closed loop of the drum infrared dryer.

[0126] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An infrared drying method based on dynamic evolution digital twin modeling and twin collaboration, characterized in that, include: S1. Obtain the zone temperature, zone power, emission array status, conveyor belt speed, air volume, and material temperature and moisture content of the infrared drying equipment. Calculate the material arrival sequence table of the zone based on the conveyor belt speed to form the current process status and actual control quantities of the equipment. S2. Establish a dynamic evolution digital twin model based on the current state of the equipment process and the actual control variables. Write the adjacent order and coverage of the infrared partitions into the dynamic evolution digital twin model to form the physical connection rules of the partitions and limit the transferable influence between the partitions. Form the material conveying alignment rules according to the material partition arrival order table and limit the consistent heating order of materials under different belt speeds. Under the constraints of the partition physical connection rules and the material conveying alignment rules, the dynamic evolution digital twin model first calculates the change of the equipment's radiative heat transfer capacity, and then calculates the changes in material temperature and moisture content to obtain the predicted drying trajectory. S3. Determine the outlet moisture content, outlet temperature and uniformity indicators based on the process formula, and form control constraints by combining power, belt speed and air volume execution boundaries, and generate candidate control command sequences under the control constraints; S4. Input the candidate control command sequence into the dynamic evolution digital twin model, combine the partition physical connection rules and material conveying alignment rules to generate the corresponding predicted drying trajectory, calculate the target cost of the predicted drying trajectory under control constraints and screen to obtain the optimal control command, determine the predicted drying trajectory corresponding to the optimal control command as the normal baseline of the working condition and issue the optimal control command. S5. Obtain the current state of the measured equipment process after the implementation of the optimal control command, align the current state of the measured equipment process with the normal working baseline according to the material conveying alignment rule, calculate the virtual and real deviation characteristics, and obtain the adaptive judgment threshold based on the predicted drying trajectory distribution. S6. Based on the virtual-real deviation characteristics and adaptive judgment threshold, determine the early fault type and fault zone, generate health diagnosis results, and correct the parameters of the dynamic evolution digital twin model according to the health diagnosis results for the next control cycle.

2. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, S1 specifically refers to: The infrared drying equipment collects zone temperature, zone power, and transmitter array status according to zone number, and also collects air volume and conveyor belt speed to form zone temperature vector, zone power vector, transmitter array status vector, air volume and conveyor belt speed. The material temperature and moisture content are collected at the inlet and outlet positions: material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content. These data are then combined with the zone temperature vector, zone power vector, transmitter array state vector, air volume, and conveyor belt speed to obtain the current process status of the equipment. Based on the adjacent order and coverage of the infrared zones and the speed of the conveyor belt, the material is divided into material zones along the direction of the conveyor belt, corresponding to the number of zones. The order in which each material zone enters the coverage of each infrared zone is calculated, and a material zone arrival order table is generated. The partition power vector, conveyor belt speed, and air volume are determined as the actual control variables, and the current state of the equipment process and the actual control variables are used as inputs to the dynamic evolution digital twin model.

3. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, S2 specifically refers to: The current state of the equipment process and the actual control quantity are used as inputs to the dynamic evolution digital twin model. Within the dynamic evolution digital twin model, an initial mapping is established to the material zone temperature vector and the material zone moisture content vector based on the material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content. This results in the material zone temperature vector and the material zone moisture content vector as the hidden state. Based on the adjacent order and coverage of infrared partitions, write the partition physical connection rules into the dynamic evolution digital twin model, and configure the partition physical connection rules to limit the direction and range of the influence that can be transmitted between partitions and limit the information path to be transmitted only along the adjacent order of infrared partitions. A partitioned input coding layer is established for the partitioned temperature vector, partitioned power vector, and transmitter array state vector according to the partition. The partitioned temperature, partitioned power, and transmitter array state of each infrared partition are fused and coded to output the partitioned coding vector. Under the constraints of the physical connection rules of the partitions, an adjacent partition influence propagation layer is established. Only the partition coding vectors of adjacent partitions are weighted and summarized and nonlinearly mapped to obtain the partition coding vectors after the adjacent influence. Establish a calculation layer for the equipment's radiative heat transfer capacity, fuse the partition coding vector after adjacent influences with the air volume, calculate the feature vector of the equipment's radiative heat transfer capacity for each partition, and thus complete the calculation of the changes in the equipment's radiative heat transfer capacity first. Based on the material zone arrival order table, a material conveying alignment rule is formed. In the material conveying alignment layer, the material zone temperature vector and the material zone moisture content vector are indexed and mapped to align according to the material conveying alignment rule and the conveyor belt speed, so as to obtain the aligned material zone temperature vector and the aligned material zone moisture content vector. A material drying evolution calculation layer is established, which integrates the characteristic vector of equipment radiative heat transfer capacity with the aligned material partition temperature vector and the aligned material partition moisture content vector. Under the information path defined by the partition physical connection rules, the changes of material temperature and moisture content at the next moment are calculated. The material partition temperature vector and material partition moisture content vector are updated through residual connection. Finally, the state prediction output layer outputs the predicted drying trajectory containing the material partition temperature vector sequence and the material partition moisture content vector sequence.

4. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 3, characterized in that, The material conveying alignment layer, under the constraints of the material conveying alignment rules and the material partition arrival order table, uses an alignment function to perform index mapping and alignment of the material partition temperature vector and the material partition moisture content vector, obtaining the aligned material partition temperature vector and the aligned material partition moisture content vector. The alignment function is specifically as follows: ； in, This is the aligned material partition temperature vector. This is the aligned moisture content vector for the material partitions. For material zone temperature vectors, This represents the moisture content vector for each material zone. For the index mapping matrix, For the index mapping matrix, the first Line number Column elements, To arrive at the sequence in position The corresponding material partition number, The heating sequence position and its value is to , The material is assigned a partition number and its value is... to , This is for controlling the cycle number.

5. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, S3 specifically refers to: The target requirements for outlet moisture content, outlet temperature and uniformity are determined based on the process formula, and these target requirements are set as target constraints. Based on the execution boundaries of the power, belt speed, and air volume of the infrared drying equipment, the range of values ​​for the partition power vector, conveyor belt speed, and air volume, as well as the variation limits of adjacent control cycles, are determined to form execution boundary constraints. The target constraints and execution boundary constraints are combined to form control constraints, which consistently limit the partition power vector, conveyor belt speed, and air volume of each control step in the candidate control command sequence. Multiple candidate control command sequences are generated under control constraints, and the control steps of the candidate control command sequences are established with the heating sequence of the materials according to the material zoning arrival order table, so that the candidate control command sequences under different conveyor belt speeds correspond to the zoning action order of the same material batch.

6. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, S4 specifically refers to: Using the current state of the equipment process as the initial state for prediction, the candidate control command sequence is written one by one into the actual control input of the dynamic evolution digital twin model, so that the partition power vector, conveyor belt speed and air volume in the candidate control command sequence drive the dynamic evolution digital twin model to make rolling predictions according to the control steps. In each control step, the partition physical connection rules are invoked to limit the direction and range of the influence that can be transmitted between partitions, so that the dynamic evolution digital twin model only transmits the influence on the partition coding vector of the adjacent partition in the adjacent partition influence transmission layer, and generates the partition coding vector after the adjacent influence accordingly. In each control step, the partition coding vector after adjacent influence is fused with the air volume, and the calculation of the equipment radiation heat transfer capacity calculation layer is performed to obtain the equipment radiation heat transfer capacity feature vector of each partition, thereby completing the prediction of equipment radiation heat transfer capacity change formed by partition power vector, transmitter array state vector and air volume. In each control step, the material conveying alignment rule is invoked to match the conveyor belt speed in the candidate control command sequence with the material partition arrival order table. The material conveying alignment layer is executed to perform index mapping and alignment on the material partition temperature vector and the material partition moisture content vector to obtain the aligned material partition temperature vector and the aligned material partition moisture content vector. The feature vector of the equipment's radiative heat transfer capacity, the aligned material partition temperature vector, and the aligned material partition moisture content vector are input into the material drying evolution calculation layer. Under the information path defined by the partition physical connection rules, the changes in material temperature and moisture content at the next moment are calculated and the hidden state is updated through residual connection. Then, the state prediction output layer outputs the predicted drying trajectory of the corresponding candidate control command sequence. Under control constraints, progressive constraint verification is performed on candidate control command sequences and candidate control command sequences that do not meet the control constraints are eliminated. For the predicted drying trajectory that meets the control constraints, the target cost is calculated. The target cost is jointly determined by the degree of satisfaction of the outlet moisture content, outlet temperature and uniformity indicators, and the energy consumption related quantities determined by the partition power vector, conveyor belt speed and air volume. The optimal control command is obtained by screening based on the target cost. The predicted drying trajectory corresponding to the optimal control command is determined as the normal operating baseline, and the optimal control command is sent to the infrared drying equipment for execution.

7. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 6, characterized in that, The target cost for calculating the predicted drying trajectory that satisfies the control constraints is calculated using a cost function, which is specifically as follows: ； in, Candidate control instruction sequence The target cost, Candidate control instruction sequence In control step Predicted material outlet moisture content, Candidate control instruction sequence In control step Predicted material outlet temperature, Candidate control instruction sequence In control step The uniformity index To meet the export moisture content target requirements, For the allowable deviation of moisture content at export, To meet the target temperature requirements for the outlet, For the allowable deviation of the outlet temperature, The upper limit of the uniformity index is allowed. , , , To preset weights, Candidate control instruction sequence In control step The partition power vector of the first One portion, Candidate control instruction sequence In control step air volume, Candidate control instruction sequence In control step conveyor belt speed, , , Preset conversion factors are used to convert zone power, air volume, and conveyor belt speed into energy-related quantities. To control the cycle duration, This is a preset reference energy consumption used for energy consumption item normalization. Predict the step size for candidate control command sequences. For the number of partitions, For partition numbering, This is the prediction step number.

8. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, S5 specifically refers to: After the optimal control command is issued and implemented, the current state of the infrared drying equipment process is collected. The current state of the infrared drying equipment process includes the zone temperature vector, zone power vector, emission array state vector, air volume, conveyor belt speed, material inlet temperature, material inlet moisture content, material outlet temperature, and material outlet moisture content. According to the material conveying alignment rules, the current state of the measured equipment process is aligned with the normal working condition baseline in terms of the material arrival order in the partition, the baseline alignment time corresponding to the measured conveyor belt speed is determined, and the corresponding predicted drying trajectory segment is extracted. The predicted material temperature and predicted moisture content at the corresponding outlet position are extracted from the predicted drying trajectory segment, and the difference between the predicted material outlet temperature and the predicted material outlet moisture content is calculated to obtain the virtual-real deviation characteristics. The predicted drying trajectory segments in the normal operating baseline are calculated and distributed according to the alignment time, and an adaptive judgment threshold is generated accordingly, so that the adaptive judgment threshold and the virtual-real deviation characteristics adopt the same alignment caliber.

9. The infrared drying method based on dynamic evolution digital twin modeling and twin collaboration according to claim 1, characterized in that, Step S6 is as follows: Using the virtual-real deviation feature as input, the virtual-real deviation feature is compared with the adaptive judgment threshold for each partition to determine the set of abnormal partitions that exceed the adaptive judgment threshold; Based on the correspondence between the abnormal partition set and the normal baseline of the operating condition, combined with the partition power vector, the transmitter array state vector, and the measured change direction of air volume, the early fault type is determined and the abnormal partition set is identified as the fault partition. The early fault types are combined with fault partitions to generate health diagnosis results. The calibration range is limited by the health diagnosis results. The current state of the aligned measured equipment process is used as the calibration target. The parameters of the equipment radiation heat transfer capacity calculation layer and the material drying evolution calculation layer corresponding to the fault zone in the dynamic evolution digital twin model are adjusted so that the adjusted dynamic evolution digital twin model can be used in the next control cycle.

Images