A sintering machine digital twin collaborative optimization method and system based on bidirectional fusion

Abstract

The application discloses a sintering machine digital twin cooperative optimization method and system based on bidirectional fusion, relates to the technical field of sintering, and comprises the following steps: establishing a single-car time sequence history file, wherein the single-car time sequence history file contains single-car health data and single-car experienced working condition data identified by a time stamp and a spatial position under a unique identity; constructing a car digital twin body which is one-to-one mapped with a physical car; aligning and binding the single-car health data and the single-car experienced working condition data based on a space-time mapping rule to generate a single-car time sequence alignment file; calibrating and iterating the single-car time sequence alignment file and the digital twin body in real time; generating a cooperative optimization decision based on the digital twin body and the single-car time sequence alignment file; and responding to maintenance operations, process control operations and car replacement operations corresponding to the cooperative optimization decision. The method realizes intelligent fusion operation and maintenance of sintering machine equipment and processes.

Description

Technical Field

[0001] This invention relates to the field of sintering technology, and in particular to a digital twin collaborative optimization method and system for sintering machines based on bidirectional fusion. Background Technology

[0002] Sintering machines are key equipment in steel production, consisting of mobile clusters of hundreds of trolleys. Digital twin technology, as a crucial means of digitally mapping physical entities to information space, has shown immense potential in the industrial sector. However, its application in the complex process of sintering production faces a dual bottleneck: First, equipment and process data are fragmented. In existing technologies, equipment condition monitoring systems and production process control systems (DCS / PLC) typically operate independently. The equipment management system focuses only on the health status of components such as trolleys, sideboards, and grates, while the process control system focuses on optimizing parameters such as airflow, temperature, and machine speed to stabilize quality and output. The lack of an effective data bridge between the two creates a situation where the equipment is unaware of the process, and the process disregards the equipment. Second, operation and control decisions are isolated. Equipment maintenance decisions and process adjustment decisions are made separately by different departments based on different data sources, making collaborative optimization difficult. This leads to problems such as hidden waste of process potential, increased equipment wear and tear due to process operations, and slow response to anomalies.

[0003] Patent application number 2021112212538 discloses a fault identification and analysis method and system for sintering machine trolley side panels, which discloses fault identification and analysis of trolley side panels based on image processing; patent application number 2022101532467 discloses a diagnostic system and method for sintering machine trolley grate bars and a storage medium, which discloses fault identification and analysis of sintering machine trolley grate bars; patent application number 2021112212542 discloses a fault detection method and system for sintering machine trolley wheels, which discloses fault identification and analysis of sintering machine trolley wheels. In the prior art, local detection devices are used to detect local faults in each link, but these schemes are all isolated subsystems. The high-precision equipment status data generated by them cannot be deeply integrated with the process parameters of the entire sintering process, and cannot support the collaborative optimization decision-making between equipment maintenance and process control.

[0004] Therefore, there is an urgent need for an innovative method to break down equipment and process barriers, achieve data fusion and collaborative decision-making, and realize intelligent integrated operation and maintenance of sintering machine equipment and processes. Summary of the Invention

[0005] The main objective of this invention is to provide a digital twin collaborative optimization method and system for sintering machines based on bidirectional fusion, which aims to solve the technical problem that existing technologies only use local detection devices to detect local faults in various links, making it difficult to achieve intelligent operation and maintenance by integrating equipment and processes and making collaborative decisions.

[0006] To achieve the above objectives, this invention provides a sintering machine digital twin collaborative optimization method based on bidirectional fusion, comprising the following steps: S1. Assign a unique identifier to each physical trolley and establish a time-series history file for each trolley. The time-series history file for each trolley includes health data and working condition data of each trolley, which are marked by timestamp and spatial location identifiers under the unique identifier. S2. Construct a digital twin of the trolley that maps one-to-one with the physical trolley. The digital twin is a comprehensive virtual entity with physical information neural network and physical mechanism model as the core and geometric model as the form carrier. The physical mechanism model includes the overall structure model of the trolley, the geometric model, mechanical model and thermal model of each sub-component. Based on the spatiotemporal mapping rules, complete the alignment and binding of the health data of a single trolley and the working condition data experienced by a single trolley to generate a time-series alignment file of a single trolley. The time-series alignment file of a single trolley and the digital twin are bidirectionally linked in real time for calibration and iteration. S3 generates collaborative optimization decisions based on digital twins and time-series aligned files of individual vehicles. These collaborative optimization decisions include predictive maintenance decisions, adaptive process control decisions, and group detection decisions for latent equipment conditions. S4 responds to maintenance operations, process control operations, and trolley replacement operations corresponding to collaborative optimization decisions, collects operation execution records, status verification results, and updated single-vehicle health data and single-vehicle operating condition data, and updates single-vehicle time sequence history files, single-vehicle time sequence alignment files, and digital twins.

[0007] Furthermore, in step S1, the health data of a single vehicle includes sub-component health data, sub-component wear data, fault record data, maintenance record data, and overall vehicle health data. Sub-components include side panels, grate bars, and wheels; sub-component health data includes side panel health, grate bar health, and wheel health. Maintenance record data includes oil filling record data. Overall vehicle health data is obtained through a weighted fusion calculation based on the side panel health, grate bar health, and wheel health. The operating data of a single vehicle includes the bellows monitoring data during the operation of the physical vehicle, which includes bellows negative pressure data, temperature data, and valve opening data.

[0008] Furthermore, "aligning and binding single-vehicle health data with single-vehicle operational condition data based on spatiotemporal mapping rules, generating a single-vehicle time-series alignment file, and bidirectionally and in real-time associating the single-vehicle time-series alignment file with the digital twin for calibration and iteration" specifically includes: Each physical trolley is assigned a unique identifier. Based on the trolley's running speed and the position of each inspection station, the estimated time node for the physical trolley to pass through each inspection station is calculated. A data alignment time window is defined based on a preset dynamic buffer duration. The data alignment time window is used to define the effective time period for data collection when the corresponding physical trolley passes through the inspection station. Physical data of the corresponding physical trolley is collected by physical sensors on the inspection station. After calculation and processing, the health data of a single trolley is obtained. At the same time, process parameters of the corresponding area of ​​the inspection station that match the time of the physical data collection of the trolley are collected as the working condition data of a single trolley. The health data and working condition data of a single trolley are both associated with a unique identifier and a collection timestamp. Based on the data alignment time window, the single vehicle health collection data and single vehicle operating condition data corresponding to the same unique identity and whose collection timestamp is within the data alignment time window are filtered out. The two are then spatiotemporally aligned and fused to match the single vehicle health collection data with the single vehicle operating condition data at the corresponding time and location, resulting in the bound single vehicle health data and single vehicle operating condition data, and generating a single vehicle time-series alignment file. Establish a two-way real-time association between the time-series alignment file of a single vehicle and its digital twin for synchronous calibration and iteration; The data alignment time window is determined based on the sintering machine operating speed, sensor response delay, and data transmission delay.

[0009] Furthermore, the predictive maintenance decision-making for equipment in step S3 specifically includes: Based on the health time series data in the single vehicle time alignment file, linear regression or exponential smoothing method is used to predict the vehicle health value within a preset time period in the future. If the predicted vehicle health value is less than the warning threshold, a predictive maintenance command is triggered. According to the lubrication record and cumulative running mileage, the grease consumption time is calculated. When the grease consumption time reaches the preset lubrication interval time, a grease injection command is generated.

[0010] Furthermore, the process adaptive control decision in step S3 specifically includes: When the health data of a single vehicle corresponding to a physical vehicle in a continuous area is all below the first threshold, it is determined that there is a regional common problem of air leakage or uneven air permeability in the continuous area, and an instruction is generated to reduce the valve opening of the continuous area and its adjacent air boxes proportionally. When the wear level data of a sub-component exceeds the second threshold, it is determined that there is a single critical problem with safety risk on the physical trolley corresponding to the wear level data of the sub-component, and a safety alarm and a shutdown maintenance instruction are generated. When the overall vehicle health data is less than the third threshold and the predicted remaining safe operating time is less than the remaining time until the next planned maintenance cycle, an immediate online replacement instruction is generated.

[0011] Furthermore, the decision-making process for detecting the latent state of equipment specifically includes: The spatiotemporal patterns of negative pressure and temperature curves of each bellows in the digital twin are continuously analyzed. When an abnormal pattern of a coordinated decrease in negative pressure and a shift of the high temperature zone in a continuous area is detected, it is identified as an abnormal area and a diagnosis is triggered. By tracing back the time-series alignment files of all physical trolleys that passed through the abnormal area, statistically significant common features in the equipment status are found through data mining algorithms. A group early warning report is generated based on statistically significant common characteristics. The group early warning report includes diagnostic conclusions, the scope of impact of hidden faults, energy efficiency impact assessment, and batch maintenance recommendations.

[0012] Furthermore, the operating condition data of a single vehicle also includes temperature data, load data, running time data, start-stop count data, environmental parameter data, cumulative running mileage data, sintering endpoint data, material layer thickness data, and ignition temperature data during the operation of the physical vehicle; the wear data of sub-components includes the wear of the side panels, the wear of the grate bars, and the wear of the wheels.

[0013] Furthermore, the acquisition of sub-component health data and sub-component wear data in the health data of a single vehicle includes the following steps: The image information of the sideboard of the physical trolley is obtained at the first preset position point, and the deformation and wear of the sideboard are obtained based on the image information. The image information of the grate bars of the physical trolley is obtained at the second preset position point, and the grate bar deformation, blockage area ratio and grate bar wear are obtained based on the grate bar image information; At the third preset position point, obtain the wheel image information of the physical trolley, and obtain the wheel deformation, wheel axle clearance and wheel wear based on the wheel image information; During the movement of the trolley, the status information of the air box corresponding to the physical trolley during sintering is obtained. The air box status information includes at least the air box negative pressure, air box temperature and valve opening data. The deformation of the guardrail, the deformation of the grate bars, the proportion of the blocked area, the deformation of the wheel, the wheel axle clearance, the wear of the guardrail, the wear of the grate bars, the wear of the wheel, the negative pressure of the air box, and the temperature of the air box are all single-dimensional data and are marked with identity information, timestamp information and physical location information respectively. The image information of the guardrail, the grate, and the wheel is processed to obtain dimensionless quantified values ​​of guardrail health, grate health, and wheel health.

[0014] Furthermore, the overall vehicle health data is obtained by weighted fusion calculation of the measured values ​​of the side panel health, the measured values ​​of the grate health, and the measured values ​​of the wheel health.

[0015] This invention also provides a sintering machine digital twin collaborative optimization system based on bidirectional fusion, comprising: The data access and preprocessing module is used to collect multi-source equipment data and process data, assign a unique identifier, timestamp identifier and spatial location identifier to each data, and transmit the collected data to the archive construction module after preprocessing. The file construction module is used to assign a unique identifier to each physical trolley, establish and update the time-series history file of a single trolley, store the health data and working condition data of a single trolley, synchronously associate the time-series alignment file of a single trolley, and use a distributed storage method to realize row-based and column-based storage of data, which facilitates data retrieval and analysis. The digital twin construction module is used to build a digital twin that maps one-to-one with the physical trolley. The digital twin is based on a physical information neural network, a physical mechanism model and a data-driven correction model as its core, and a geometric model as its form carrier. It realizes bidirectional real-time association, data synchronization and model calibration and iteration between the time-series alignment file of a single trolley, the time-series history file of a single trolley and the digital twin. The collaborative optimization instruction generation module is used to generate predictive maintenance instructions, adaptive process control instructions, and hidden status group detection instructions for equipment based on digital twin and single vehicle time-series aligned archive data. This module has built-in linear regression, exponential smoothing, and data mining algorithms to achieve health trend prediction, hidden fault root cause location, and generation of various control instructions. The closed-loop execution and monitoring module is used to respond to various operations corresponding to the collaborative optimization decision, monitor the operation execution process and status verification throughout the process, collect feedback data, update the single vehicle time sequence history file, single vehicle time sequence alignment file and digital twin model, and complete the closed-loop iteration. The operation and maintenance decision module is used to generate trolley health reports, group early warning reports and maintenance suggestions based on digital twin simulation results, single trolley time-series history archives, single trolley time-series alignment archive data and decision execution status, providing decision support for intelligent operation and maintenance of sintering machine trolleys throughout their entire life cycle.

[0016] Compared with existing technologies, the sintering machine digital twin collaborative optimization method based on bidirectional fusion provided by this invention has the following beneficial effects: This invention provides a bidirectional fusion-based digital twin collaborative optimization method for sintering machines. First, a unique identifier is assigned to each physical trolley, establishing a single-trolley time-series history file. This file includes health data and operational condition data of the single trolley, calibrated by the unique identifier, timestamp, and spatial location, enabling data traceability. Next, a digital twin is constructed, mapping one-to-one with the physical trolley. Using a physical information neural network and a physical mechanism model as its core, the health data and operational condition data of each trolley are aligned and bound through spatiotemporal mapping rules, generating a single-trolley time-series alignment file. Finally, spatiotemporal calibration is performed on the health data and operational condition data of each trolley to achieve time... The binding of nodes and spatial physical locations enables the alignment of a single vehicle's health data with its operational data to obtain a time-series aligned file, achieving deep integration of equipment and process data. Through the bidirectional association between the file and the digital twin, model calibration and iteration are completed. Based on the simulation capabilities of the digital twin and the fused data from the time-series aligned file, three types of collaborative decisions are generated: equipment maintenance, process control, and latent fault detection. Finally, through decision execution feedback, the dual files and the digital twin are updated, forming a complete technical link of data acquisition, fusion modeling, decision generation, and closed-loop iteration. This enables collaborative optimization of equipment and processes, and achieves intelligent integrated operation and maintenance of sintering machine equipment and processes. The solution of this invention achieves deep integration of equipment data and process data through spatiotemporal calibration and alignment binding of dual archives, building a data bridge between equipment and process. This enables collaborative optimization decision-making based on dual-dimensional considerations of equipment and process. Based on fused data and digital twin simulation, it links equipment maintenance decisions with process control decisions, avoiding the problems of hidden waste of process potential and process operation exacerbating equipment wear and tear, and improving the synergy of operation and maintenance and control. It also enhances the level of intelligent operation and maintenance by using fused data from time-series aligned archives to achieve equipment health trend prediction, hidden fault group detection, early warning of equipment anomalies, and reduction of sudden failure downtime. At the same time, it reduces operation and maintenance costs through online replacement and precise maintenance decisions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a bidirectional fusion-based digital twin collaborative optimization method for sintering machines according to one embodiment of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0023] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0024] Please refer to the appendix. Figure 1 This invention provides a digital twin collaborative optimization method for sintering machines based on bidirectional fusion. S1, assign a unique identity to each physical trolley and establish a single trolley time-series history file. The single trolley time-series history file includes the single trolley health data and single trolley working condition data marked by the unique identity under the timestamp and spatial location identifier. S2. Construct a digital twin of the trolley that maps one-to-one with the physical trolley. The digital twin is a comprehensive virtual entity with physical information neural network and physical mechanism model as the core and geometric model as the form carrier. The physical mechanism model includes the overall structure model of the trolley, the geometric model, mechanical model and thermal model of each sub-component. Based on the spatiotemporal mapping rules, complete the alignment and binding of the health data of a single trolley and the working condition data experienced by a single trolley to generate a time-series alignment file of a single trolley. The time-series alignment file of a single trolley and the digital twin are bidirectionally linked in real time for calibration and iteration. S3 generates collaborative optimization decisions based on digital twins and time-series aligned files of individual vehicles. These collaborative optimization decisions include predictive maintenance decisions, adaptive process control decisions, and group detection decisions for latent equipment conditions. S4 responds to maintenance operations, process control operations, and trolley replacement operations corresponding to collaborative optimization decisions, collects operation execution records, status verification results, and updated single-vehicle health data and single-vehicle operating condition data, and updates single-vehicle time sequence history files, single-vehicle time sequence alignment files, and digital twins.

[0025] This invention provides a bidirectional fusion-based digital twin collaborative optimization method for sintering machines. First, a unique identifier is assigned to each physical trolley, establishing a single-trolley time-series history file. This file includes health data and operational condition data of the single trolley, calibrated by the unique identifier, timestamp, and spatial location, enabling data traceability. Next, a digital twin is constructed, mapping one-to-one with the physical trolley. Using a physical information neural network and a physical mechanism model as its core, the health data and operational condition data of each trolley are aligned and bound through spatiotemporal mapping rules, generating a single-trolley time-series alignment file. Finally, spatiotemporal calibration is performed on the health data and operational condition data of each trolley to achieve time... The binding of nodes and spatial physical locations enables the alignment of a single vehicle's health data with its operational data to obtain a time-series aligned file, achieving deep integration of equipment and process data. Through the bidirectional association between the file and the digital twin, model calibration and iteration are completed. Based on the simulation capabilities of the digital twin and the fused data from the time-series aligned file, three types of collaborative decisions are generated: equipment maintenance, process control, and latent fault detection. Finally, through decision execution feedback, the dual files and the digital twin are updated, forming a complete technical link of data acquisition, fusion modeling, decision generation, and closed-loop iteration. This enables collaborative optimization of equipment and processes, and achieves intelligent integrated operation and maintenance of sintering machine equipment and processes. The solution of this invention achieves deep integration of equipment data and process data through spatiotemporal calibration and alignment binding of dual archives, building a data bridge between equipment and process. This enables collaborative optimization decision-making based on dual-dimensional considerations of equipment and process. Based on fused data and digital twin simulation, it links equipment maintenance decisions with process control decisions, avoiding the problems of hidden waste of process potential and process operation exacerbating equipment wear and tear, and improving the synergy of operation and maintenance and control. It also enhances the level of intelligent operation and maintenance by using fused data from time-series aligned archives to achieve equipment health trend prediction, hidden fault group detection, early warning of equipment anomalies, and reduction of sudden failure downtime. At the same time, it reduces operation and maintenance costs through online replacement and precise maintenance decisions.

[0026] Furthermore, in step S1, the health data of a single vehicle includes sub-component health data, sub-component wear data, fault record data, maintenance record data, and overall vehicle health data. Among them, sub-components include sideboards, grate bars, and wheels, and the sub-component health data includes the health of sideboards, grate bars, and wheels; the maintenance record data includes oiling record data; the overall vehicle health data is obtained by weighted fusion calculation based on the health of sideboards, grate bars, and wheels; the operating condition data experienced by a single vehicle includes the bellows monitoring data during the operation of the physical trolley, and the bellows monitoring data includes the bellows negative pressure data, temperature data, and valve opening data.

[0027] In a specific embodiment of the present invention, firstly, unified access and spatiotemporal preprocessing of all elements of data are achieved: specifically, multi-source heterogeneous data streams are received from the perception layer. These multi-source heterogeneous data streams include single-vehicle health data and single-vehicle operational condition data, with each data point appended with a globally unified high-precision timestamp and spatial location identifier. The multi-source heterogeneous data streams include: equipment identity and status data, vehicle identity ID and its timestamp T_id and spatial location identifier P_id at the identification point. The collection of single-vehicle health data includes: the obtained panel fault type (such as perforation, crack, fracture) and its quantified area and location information; the obtained maximum grate gap width value and blockage area value; the obtained unique identity ID of each wheel determined by RFID identification tags and its existence status; the obtained wheel tilt error e and tilt status; and the oiling record uploaded by the intelligent grease injection machine for a specific vehicle identity ID, including the oiling time T_oil and the oiling amount Q_oil. The operating data of a single machine includes process parameter data for the entire process, including the material layer thickness in the head area, the temperature of the ignition furnace and the material surface condition image; the exhaust gas temperature, negative pressure, oxygen content and valve opening of each air box in the main sintering process; the location and temperature used to determine the sintering endpoint (BTP); and the cross-sectional condition image of the tail area.

[0028] Furthermore, "aligning and binding single-vehicle health data with single-vehicle operational condition data based on spatiotemporal mapping rules, generating a single-vehicle time-series alignment file, and bidirectionally and in real-time associating the single-vehicle time-series alignment file with the digital twin for calibration and iteration" specifically includes: Each physical trolley is assigned a unique identifier. Based on the trolley's running speed and the position of each inspection station, the estimated time node for the physical trolley to pass through each inspection station is calculated. A data alignment time window is defined based on a preset dynamic buffer duration. The data alignment time window is used to define the effective time period for data collection when the corresponding physical trolley passes through the inspection station. Physical data of the corresponding physical trolley is collected by physical sensors on the inspection station. After calculation and processing, the health data of a single trolley is obtained. At the same time, process parameters of the corresponding area of ​​the inspection station that match the time of the physical data collection of the trolley are collected as the working condition data of a single trolley. The health data and working condition data of a single trolley are both associated with a unique identifier and a collection timestamp. Based on the data alignment time window, the single vehicle health collection data and single vehicle operating condition data corresponding to the same unique identity and whose collection timestamp is within the data alignment time window are filtered out. The two are then spatiotemporally aligned and fused to match the single vehicle health collection data with the single vehicle operating condition data at the corresponding time and location, resulting in the bound single vehicle health data and single vehicle operating condition data, and generating a single vehicle time-series alignment file. Establish a two-way real-time association between the time-series alignment file of a single vehicle and its digital twin for synchronous calibration and iteration; The data alignment time window is determined based on the sintering machine operating speed, sensor response delay, and data transmission delay.

[0029] In an optional embodiment of the present invention, a deterministic data binding method based on spatiotemporal rules is used to achieve high-fidelity fusion; the system presets a virtual inspection pipeline containing multiple fixed stations; identity triggering, when trolley A-101 is identified at the identity recognition station P_id at time T_id, the system creates a data collection session for it; data alignment time window prediction, the system calculates the estimated time T_arrival for the trolley to reach each downstream station based on the fixed distance L from P_id to the downstream bar detection station P_img_bar, grate detection station P_img_grate, and wheel detection station P_wheel, and the real-time speed V of the sintering machine, and provides a dynamic preset dynamic buffer duration ΔT_buffer to form a data alignment time window [T_arrival - ΔT_buffer, T_arrival + ΔT_buffer]. State data binding involves using existing technologies to identify data collected within each workstation's time window and then deterministically binding the identification results to A-101. For example, P_img_bar is bound to the panel status; P_img_grate to the grate status; and P_wheel to the wheel ID and tilt status. Process data binding: After state binding, the system uses the trolley motion model to infer the precise coordinates (X, Y, Z) of A-101 in virtual space based on the binding timestamp. Subsequently, a spatial query is performed to bind the process parameters of the bellows directly below the coordinate point, as well as the key parameters of the process area to which the coordinate point belongs (e.g., under the ignition furnace, BTP determination area), to the twin of trolley A-101.

[0030] In an optional embodiment of the present invention, a structured, time-evolving single-car time-series history file is created for each trolley, realizing the dynamic generation and updating of the trolley's entire lifecycle history file. The file structure of the single-car time-series history file includes: an identity core, a unique ID, online time, and cumulative running mileage; an equipment health sub-file, recorded in the form of a time-series table, with each record including {timestamp, location, sideboard health H_b, grate health H_s, wheel status (present / tilted / normal), lubrication status}; a process history sub-file, recording the data of this sintering cycle in the form of a time-series table, including {timestamp, location, material layer thickness, ignition temperature, wind box temperature / negative pressure, BTP mark, ..., tail section score}; and a sub-component health quantification data stream is used to evaluate the health status of each sub-component, including sideboard health, grate health, wheel and axle health, etc. The overall health status of the trolley is obtained by multi-index fusion health calculation of key components, and the preliminary diagnostic conclusion of the whole machine can include the overall health status of the trolley. Specifically, the general formula for calculating the health of the railing is as follows: ,in To measure the actual deformation, This represents the maximum allowable deformation. If the calculated result is negative, it is set to 0. The health status of other components can be defined in a similar manner, such as the health status of the grate bars. wheel and axle health Etc; Grate health H_s: calculated by fusing the clogged area value η and the maximum gap width; Wheel health H_w is a composite state. If a wheel is detected to be missing, then H_w=0; if the tilt error e is detected to exceed the threshold, then H_w is mapped to the (0,1) interval according to the e value; otherwise, H_w=1.

[0031] Furthermore, the predictive maintenance decision-making for equipment in step S3 specifically includes: Based on the health time series data in the single vehicle time alignment file, linear regression or exponential smoothing method is used to predict the vehicle health value within a preset time period in the future. If the predicted vehicle health value is less than the warning threshold, a predictive maintenance command is triggered. According to the lubrication record and cumulative running mileage, the grease consumption time is calculated. When the grease consumption time reaches the preset lubrication interval time, a grease injection command is generated.

[0032] Furthermore, the process adaptive control decision in step S3 specifically includes: When the health data of a single vehicle corresponding to a physical vehicle in a continuous area is all below the first threshold, it is determined that there is a regional common problem of air leakage or uneven air permeability in the continuous area, and an instruction is generated to reduce the valve opening of the continuous area and its adjacent air boxes proportionally. When the wear level data of a sub-component exceeds the second threshold, it is determined that there is a single critical problem with safety risk on the physical trolley corresponding to the wear level data of the sub-component, and a safety alarm and a shutdown maintenance instruction are generated. When the overall vehicle health data is less than the third threshold and the predicted remaining safe operating time is less than the remaining time until the next planned maintenance cycle, an immediate online replacement instruction is generated.

[0033] Furthermore, the decision-making process for detecting the latent state of equipment specifically includes: The spatiotemporal patterns of negative pressure and temperature curves of each bellows in the digital twin are continuously analyzed. When an abnormal pattern of a coordinated decrease in negative pressure and a shift of the high temperature zone in a continuous area is detected, it is identified as an abnormal area and a diagnosis is triggered. By tracing back the time-series alignment files of all physical trolleys that passed through the abnormal area, statistically significant common features in the equipment status are found through data mining algorithms. A group early warning report is generated based on statistically significant common characteristics. The group early warning report includes diagnostic conclusions, the scope of impact of hidden faults, energy efficiency impact assessment, and batch maintenance recommendations.

[0034] In an optional embodiment of the present invention, collaborative optimization decision-making based on fused data specifically includes: Predictive maintenance decision-making for equipment: A trend forecasting method is employed, using time-series health data from the trolley's history archive to predict the health trend over a future period using linear regression or exponential smoothing. If the predicted value falls below a warning threshold, a predictive maintenance work order is triggered. Lubrication decision-making method: Based on lubrication records and cumulative mileage, grease consumption data is calculated. When a preset lubrication interval is reached, a grease injection work order is automatically generated and sent to the intelligent grease injection machine.

[0035] Adaptive Process Control Decision: This invention distinguishes between two typical abnormal equipment operating conditions and adopts different strategies: **Regional Widespread Problems:** When the system detects that the health of the sideboards or grates of a trolley in a continuous area is generally low, it is determined that there is a risk of air leakage or uneven air permeability in that area. The response is for the decision engine to generate instructions to proportionally reduce the valve opening of the problematic area and its adjacent air boxes, thereby distributing more airflow to the healthy areas of the equipment to maintain airflow balance and BTP stability across the entire sintering section. **Single-Point Severe Problems:** When a trolley section is detected to have a missing wheel (H_w=0) or severe tilting, it is determined that the trolley poses a significant safety risk. The response is for the decision engine to immediately generate a high-priority safety alarm and a shutdown maintenance work order, and can, depending on the situation, interlock the sintering machine main control system to execute an emergency or planned shutdown to prevent the accident from escalating. Online replacement decision for faulty trolleys: When the system determines that the health of a trolley (e.g., the combined health of the sideboards, grate bars, and wheels) is consistently below the scrap threshold and predicts that it cannot safely operate until the next planned maintenance cycle, the decision engine will perform the following operations: Generate a highest-priority instruction for immediate online replacement, which will be simultaneously sent to the online replacement device and the sintering machine main control system. The system will utilize a preset maintenance window or coordinate a very short process stabilization period, under conditions where the sintering machine is not shut down or only running at a very low speed, to trigger and complete the automatic disassembly of the faulty trolley and the installation of a new trolley. After replacement, the system automatically updates the trolley queue in the digital twin and verifies the operating status of the new trolley.

[0036] Equipment latent condition group detection based on process parameter reverse diagnosis: This aims to solve the systemic performance degradation problem caused by minor collective deterioration of the trolley, which cannot be detected by a single sensor. Its core process includes: The system continuously analyzes the spatiotemporal patterns of negative pressure and temperature curves of each air box in the digital twin for process anomaly detection. When an anomaly pattern such as "decreased negative pressure coordination and shift of high temperature zone" is detected in a continuous area, a diagnosis is triggered. The system also performs group equipment file backtracking and correlation analysis, tracing the entire lifecycle of all trolleys passing through the anomaly area and comparing it with the files of trolleys in normal areas using data mining algorithms to find statistically significant common features in equipment status. Finally, the system locates and decides on hidden fault root causes, identifying hidden root causes that are difficult to detect with a single sensor, such as slight deformation of the grate panels or inconsistent deterioration of grate health. It then generates a group early warning report to guide planned and batch maintenance. Based on this, the system generates a group early warning and decision report, such as: "Diagnosis: There is a risk of systemic air leakage in the 5-9 wind box area due to slight deformation of the side panels (average health decrease of 8%), which is expected to affect the overall energy efficiency by about 2%. Decision recommendation: For this batch of 35 vehicles, arrange a special inspection and repair of the side panel sealing during the next planned shutdown window."

[0037] Furthermore, the operating condition data of a single vehicle also includes temperature data, load data, running time data, start-stop count data, environmental parameter data, cumulative running mileage data, sintering endpoint data, material layer thickness data, and ignition temperature data during the operation of the physical vehicle; the wear data of sub-components includes the wear of the side panels, the wear of the grate bars, and the wear of the wheels.

[0038] Furthermore, the acquisition of sub-component health data and sub-component wear data in the health data of a single vehicle includes the following steps: The image information of the sideboard of the physical trolley is obtained at the first preset position point, and the deformation and wear of the sideboard are obtained based on the image information. The image information of the grate bars of the physical trolley is obtained at the second preset position point, and the grate bar deformation, blockage area ratio and grate bar wear are obtained based on the grate bar image information; At the third preset position point, obtain the wheel image information of the physical trolley, and obtain the wheel deformation, wheel axle clearance and wheel wear based on the wheel image information; During the movement of the trolley, the status information of the air box corresponding to the physical trolley during sintering is obtained. The air box status information includes at least the air box negative pressure, air box temperature and valve opening data. The deformation of the guardrail, the deformation of the grate bars, the proportion of the blocked area, the deformation of the wheel, the wheel axle clearance, the wear of the guardrail, the wear of the grate bars, the wear of the wheel, the negative pressure of the air box, and the temperature of the air box are all single-dimensional data and are marked with identity information, timestamp information and physical location information respectively. The image information of the guardrail, the image information of the grate, and the image information of the wheel are processed to obtain dimensionless quantified values ​​of guardrail health, grate health, and wheel health. Furthermore, the overall vehicle health data is obtained by weighted fusion calculation of the measured values ​​of the side panel health, the measured values ​​of the grate health, and the measured values ​​of the wheel health.

[0039] This invention provides a specific collaborative optimization method for sintering machine digital twins based on bidirectional fusion, applicable to sintering machine clusters consisting of hundreds of sintering machines. The core is to use time node constraints to align the health data of a single machine with the operating condition data experienced by a single machine in its time-series history file to form a time-series aligned file for a single machine, and then collaborate with the digital twin to achieve data fusion and collaborative decision-making between equipment and process. The specific forms of the single-vehicle time-series archive and the single-vehicle time-series alignment archive are as follows: The single-vehicle time-series archive adopts distributed storage (row + column), with a unique identifier and timestamp as dual indexes, and stores the original health data, original operating condition data and all update records of the entire life cycle of a single vehicle in chronological order. It is the original data archive without spatiotemporal alignment. The single-vehicle time-series alignment archive also uses a unique identifier and timestamp as indexes, and stores the fused data after alignment and binding according to spatiotemporal mapping rules. That is, each data contains health data and operating condition data at the corresponding time and location. It is the fused and aligned data archive. The two are linked by a unique identifier. The single-vehicle time-series alignment archive originates from the original data of the single-vehicle time-series archive and is updated synchronously with the update of the single-vehicle time-series archive. At the same time, the bidirectional association of digital twins realizes model calibration.

[0040] In an optional embodiment of the present invention, taking the data related to the unique identifier TC-001 on March 5, 2026 as an example, the presentation format of a single vehicle time-series history file (original data not aligned, example of a single set of original data, multiple such sets sorted by timestamp to form a complete file) can be: {Timestamp: 2026-03-05 08:45:00, Location: No. 3 windbox area, Material layer thickness: 850mm, Ignition temperature: 1150℃, Windbox temperature: 370℃, Windbox negative pressure: 13kPa, BTP mark: Not reached, Sidewall health: 0.68, Grate health: 0.73, Wheel health: 0.64, Sidewall wear: 0.8mm, Grate blockage area ratio: 12%, Wheel axle clearance: 0.3mm, Oil injection record: 2026-03-01, Cumulative mileage: 800km, Overall vehicle health: 0.66, Tail section score: 82 points, Equipment identification: TC-001}; {Timestamp: 2026-03-05 08:50:00, Location: Area 4 (Blowbox), Material layer thickness: 850mm, Ignition temperature: 1150℃, Blowbox temperature: 375℃, Blowbox negative pressure: 12.5kPa, BTP mark: Not reached, Sidewall health: 0.67, Grate health: 0.72, Wheel health: 0.63, Sidewall wear: 0.8mm, Grate blockage area ratio: 12%, Wheel axle clearance: 0.3mm, Oil filling record: 2026-03-01, Cumulative mileage: 800km, Overall vehicle health: 0.65, Tail section score: Not generated, Equipment identification: TC-001; {Timestamp: 2026-03-05} 09:05:00, Location: 5# airbox inspection station, Material layer thickness: 850mm, Ignition temperature: 1150℃, Airbox temperature: 380℃, Airbox negative pressure: 12kPa, BTP mark: Not reached, Sideboard health: 0.65, Grate health: 0.72, Wheel health: 0.63, Sideboard wear: 0.8mm, Grate blockage area ratio: 12%, Wheel axle clearance: 0.3mm, Oil filling record: 2026-03-01, Cumulative mileage: 800km, Overall vehicle health: 0.65, Tail section score: Not generated, Equipment identification: TC-001, etc. The single vehicle time-series history file, that is, the health data and working condition data of the same vehicle at different timestamps and different locations, are stored independently without a clear binding relationship.A single vehicle time-series aligned archive (fusion alignment, example of a single fused data set; multiple such sets sorted by timestamps form a complete archive) can be: {timestamp: 2026-03-05} 09:00:00 (time after alignment), Location: 5# airbox inspection station (position after alignment), Material layer thickness: 850mm, Ignition temperature: 1150℃, Airbox temperature: 380℃, Airbox negative pressure: 12kPa, BTP mark: Not reached, Side panel health: 0.6, Grate bar health: 0.7, Wheel health: 0.6, Side panel wear: 0.8mm, Grate bar blockage area ratio: 12%, Wheel axle clearance: 0.3mm, Overall vehicle health: 0.65, Airbox valve opening: 60%, Tail section score: 82 points, Equipment identification: TC-001, Data alignment window: 8:55-9:55, Alignment status: Aligned; {Timestamp: 2026-03-05} 09:10:00 (time after alignment), Location: Area 6 (after alignment), Material layer thickness: 850mm, Ignition temperature: 1150℃, Air box temperature: 385℃, Air box negative pressure: 11.8kPa, BTP mark: Not reached, Sideboard health: 0.59, Grate health: 0.69, Wheel health: 0.62, Sideboard wear: 0.8mm, Grate blockage area ratio: 13%, Wheel axle clearance: 0.3mm, Overall vehicle health: 0.64, Air box valve opening: 60% Tail section score: 82 points, equipment identification: TC-001, data alignment window: 9:05-10:05, alignment status: aligned, etc. The characteristics of the single vehicle time sequence alignment archive are spatiotemporal alignment fusion, that is, each set is bound to health data and operating condition data at the same time and location, supplemented with alignment-related identifiers (alignment window, alignment status). The data items are consistent with the single vehicle time sequence history archive, but achieve precise matching of health data and operating condition data, which can be directly used for digital twin calibration and collaborative decision generation.

[0041] This invention provides a specific solution for handling single-car malfunctions in the No. 3 sintering machine (containing 200 trolleys, numbered TC-001 to TC-200) of a steel plant, including: Step 1: Construct a timeline history file for a single vehicle: Each vehicle is assigned a unique identifier (e.g., TC-001). Through the system's file construction module, a single-vehicle time-series archive for TC-001 is created (distributed storage, row-based storage of time series, column-based storage of different data types, dual index: TC-001 + timestamp). Various types of raw data are collected and labeled, and stored in this archive. Some key data include: single-vehicle health data (raw, unaligned): sideboard health 0.65 (dimensionless 0-1, quantized through image processing), grate health 0.72, wheel health 0.63, sideboard wear 0.8mm, grate wear 0.5mm, wheel wear 1.2mm, oil filling record (last oil filling date March 1, 2026, cumulative mileage 800km), and fault records (no major faults, only 2). On February 28, 2026, a minor deformation warning was issued for the sideboard. The overall vehicle health score was 0.65 (weighted fusion calculation, weights: sideboard 0.3, grate 0.4, wheels 0.3). The operating data of a single vehicle (original data not aligned): Data collected between 8:40 and 9:10 on March 5, 2026, showing the operating data of TC-001 passing through air boxes 1#-6#, including the negative pressure data of air box 5# (corresponding to the testing station) as 12kPa, the temperature data as 380℃, and the valve opening as 60%, and the data of other air boxes. All original data are marked with a unique identifier (TC-001), a timestamp (accurate to the minute), and a spatial location identifier (such as "air box 5# testing station" or "air box 1# area"), and are all stored in the single vehicle time-series archive of TC-001 to form a complete original data sequence.

[0042] Step 2: Digital twin construction and generation of time-series aligned files for individual vehicles: Digital Twin Construction: Through the digital twin construction module, digital twins are constructed that are mapped one-to-one with 200 trolleys. Among them, the digital twin of TC-001 contains a physical information neural network (PINN) and a physical mechanism model (overall structure model of the trolley, geometric model of the sideboard / grate / wheel, mechanical model and thermal model). The geometric shape is restored to the physical trolley 1:1. It can receive data in real time and perform simulation calculations. Spatiotemporal alignment and alignment file generation: Based on spatiotemporal mapping rules, the original data in the single-vehicle time-series archive of TC-001 is aligned and bound to generate a single-vehicle time-series aligned archive (same storage mode as the history archive, dual index: TC-001 + timestamp): Calculate the estimated time and data alignment time window: Given that the sintering machine's operating speed is 2.2 m / min, and the spatial coordinates of the detection station corresponding to the #5 wind box are 15 m, combined with the real-time position of TC-001, the estimated time for it to pass through this detection station is calculated to be 9:00 AM on March 5, 2026; Combining sensor response delay and data transmission delay h, a dynamic buffer duration is set to define the data alignment time window; Data filtering and alignment fusion: From the single-vehicle time sequence history file of TC-001, filter out those with the identity identifier TC-001 and whose collection timestamp is within... The original health data and original operating condition data within the data alignment time window are spatiotemporally aligned and fused. The sideboard health is 0.6, the grate health is 0.7, the wheel health is 0.6, and the sideboard wear is 0.8mm. This corresponds to a negative pressure of 12kPa in the No. 5 air box, a temperature of 380℃, and a valve opening of 60%. This fused data is stored in the single-vehicle time-series alignment file of TC-001 to form a complete time-location-health-operating condition binding data. At the same time, the data in other time windows are processed in the same way to generate a complete single-vehicle time-series alignment file. The dual archives (single vehicle time-series history archive and single vehicle time-series alignment archive) are bidirectionally linked with the digital twin: a bidirectional real-time link is established between the single vehicle time-series alignment archive of TC-001 and the digital twin. Through the fused data in the alignment archive (such as the health of the sideboard and the temperature of the bellows), the mechanical model of the digital twin is calibrated (correcting the simulation error of the sideboard deformation so that the deviation between the simulated deformation and the actual wear is controlled within ±0.1mm) and the thermal model (correcting the simulation error of the vehicle temperature conduction), completing the first iteration of the digital twin. At the same time, the simulation data of the digital twin (such as the predicted sideboard deformation trend) is synchronously fed back to the time-series alignment archive and supplemented into the corresponding time series.

[0043] Step 3: Generate collaborative optimization decisions based on the digital twin and the single vehicle time-series alignment file: Based on the single-vehicle time-series alignment archive data and digital twin simulation results from TC-001, three types of collaborative optimization decisions are generated, all of which are associated with specific data in the alignment archive: Predictive maintenance decision-making: Extract the health time series data of the past 7 days (February 26, 2026 - March 5, 2026) from the time series alignment file (the health of the guardrail drops from 0.75 to 0.6). Use linear regression (built-in algorithm) to predict that the health of the guardrail will drop to 0.45 in the next 7 days (March 6 - March 12), which is lower than the warning threshold (0.5), triggering a predictive maintenance instruction. The instruction associates the wear data (0.8mm) and time series in the alignment file, specifying the maintenance content as "guardrail deformation correction and wear repair welding", with a planned maintenance time before March 10. At the same time, based on the oiling record (March 1) and cumulative mileage (800km) associated in the alignment file, use the built-in algorithm to calculate the grease consumption time, determine that the preset lubrication interval (1000km / time) has been reached, generate a grease injection instruction, specifying the grease injection time as March 6 and the grease injection volume as 500ml.

[0044] Adaptive process control decision: Through digital twin simulation, combined with the binding data of TC-001 corresponding to the No. 5 wind box in the timing alignment file (negative pressure 12kPa, temperature 380℃, and plate health 0.6), and compared with the normal operating standard of the sintering machine (wind box negative pressure 16-18kPa, temperature 390-410℃), it was determined that the No. 5 wind box has a slight risk of air leakage, and the air leakage will aggravate the wear of the plate (the alignment file shows that when the negative pressure is below 14kPa, the wear rate of the plate increases by 20%). Therefore, a process control instruction is generated: adjust the valve opening of the No. 5 wind box from 60% to 50% to reduce air leakage, and balance process stability (ensuring that the sintering temperature meets the standard) and equipment protection (reducing plate wear).

[0045] Device latent state group detection decision: This embodiment is a single point problem and this decision has not yet been triggered.

[0046] Step 4, Closed-loop execution and dual-file / digital twin updates: Decision Execution: Following the generated instructions, the grease injection operation of TC-001 was completed on March 6th; the side panel repair (correction of deformation and welding of worn parts) was completed on March 8th; simultaneously, the opening of the No. 5 air box valve was adjusted to 50%. The closed-loop execution and monitoring module monitored the entire execution process and recorded execution details (such as repair time 1.5h, grease injection volume 500ml, valve opening 50% after adjustment). Feedback Data Collection: After the repair, updated data of TC-001 was collected: side panel health 0.82, grate health 0.71, wheel health 0.62, side panel wear 0.1mm; No. 5 air box operating data: negative pressure 14kPa, temperature 390℃, valve opening 50%; all data were marked with identification, timestamp (March 8th 10:00), and spatial location identifier as feedback data. Dual Archives and Digital Twin Update: The above feedback data is updated to the single-vehicle time-series archive of TC-001 (supplementing maintenance records, grease injection records, and updated original health and operating condition data); at the same time, the feedback data is spatiotemporally aligned, and the single-vehicle time-series alignment archive is updated (adding fused data at 10:00 on March 8: side panel health 0.82 + 5# air box negative pressure 14kPa, etc.); finally, the updated alignment archive data is synchronized to the digital twin, the model parameters are calibrated, the model is completed in its second iteration, and closed-loop optimization is achieved.

[0047] The beneficial effects of the solution of the present invention include: This invention overcomes the challenge of information silos, achieving unified understanding and causal traceability of cross-domain data. In traditional systems, equipment status and process parameters belong to different systems and are difficult to correlate. This invention, by constructing a high-fidelity digital twin, creates for the first time a unified data model (history archive) for each mobile trolley, spanning its entire lifecycle and integrating all elements of "mechanical-electrical-engineering-process". This completely breaks down data barriers, enabling the system to accurately trace the root causes of process anomalies (e.g., abnormal negative pressure in the bellows below a trolley due to deformation of its side panel) or quantitatively assess the impact of process operations on equipment wear and tear, providing an unprecedented data foundation for accurate decision-making. This invention represents a fundamental shift in operation and maintenance from passive response to proactive prediction and collaborative optimization. It transcends the limitations of traditional discrete detection systems that merely issue alarms. Through data fusion and trend analysis using digital twins, it achieves state-based predictive maintenance. More importantly, it innovatively uses real-time equipment health status as a dynamic constraint for process optimization, enabling collaborative intelligent decision-making between equipment management and process control. For example, when a local trolley experiences a widespread decline in health, the system can automatically adjust airflow distribution to maintain overall production stability. This significantly improves equipment lifespan and reduces energy consumption without affecting output or quality, maximizing overall benefits.

[0048] By constructing a closed-loop intelligence, the system is endowed with the ability to continuously evolve. This invention forms a complete closed loop of "perception-fusion-decision-execution-verification". The effect data after each decision execution is fed back to the digital twin for optimization and calibration of the model and algorithm. This makes the system no longer a static program, but an "intelligent agent" with self-learning and self-optimization capabilities. Its decision accuracy and intelligence level continue to improve over time.

[0049] By making tacit knowledge explicit, this invention forms reusable core digital assets. It materializes the personal experience of operations and maintenance experts into analytical models, diagnostic rules, and control strategies within a digital twin. This allows best practices to be accumulated, solidified, and replicated on a large scale, effectively solving the problem of enterprises' reliance on the experience of key personnel and providing a solid core digital asset for enterprises' intelligent transformation.

[0050] This invention enables "group health checks" and proactive early warning of latent equipment faults. It surpasses the limitations of traditional monitoring technologies in perceiving individual components and obvious faults, creatively utilizing the spatiotemporal patterns of process big data to diagnose subtle, common latent degradation states existing within a group of production vehicles. This makes it possible to perform precise and proactive maintenance in the early stages of systemic performance degradation (such as a slow increase in air leakage rate), thereby improving the reliability and economy of system operation from the source.

[0051] This invention provides a sintering machine digital twin collaborative optimization system based on bidirectional fusion, comprising: The data access and preprocessing module is used to collect multi-source equipment data and process data, assign a unique identifier, timestamp identifier and spatial location identifier to each data, and transmit the collected data to the archive construction module after preprocessing. The file construction module is used to assign a unique identifier to each physical trolley, establish and update the time-series history file of a single trolley, store the health data and working condition data of a single trolley, and synchronously associate the time-series alignment file of a single trolley. The digital twin construction module is used to build a digital twin that maps one-to-one with the physical trolley. The digital twin uses a physical information neural network (PINN), a physical mechanism model and a data-driven correction model as its core, and a geometric model as its form carrier to realize bidirectional real-time association, data synchronization and model calibration and iteration between the time-series alignment file of a single trolley, the time-series history file of a single trolley and the digital twin. The collaborative optimization instruction module is used to generate predictive maintenance instructions, adaptive process control instructions, and hidden status group detection instructions for equipment based on digital twin and single vehicle time-series aligned archive data. This module has built-in linear regression, exponential smoothing, and data mining algorithms to achieve health trend prediction, hidden fault root cause location, and generation of various control instructions. The closed-loop execution and monitoring module is used to respond to various operations corresponding to the collaborative optimization decision, monitor the operation execution process and status verification throughout the process, collect feedback data, update the single vehicle time sequence history file, single vehicle time sequence alignment file and digital twin model, and complete the closed-loop iteration. The operation and maintenance decision module is used to generate trolley health reports, group early warning reports and maintenance suggestions based on digital twin simulation results, single trolley time-series history archives, single trolley time-series alignment archive data and decision execution status, providing decision support for intelligent operation and maintenance of sintering machine trolleys throughout their entire life cycle.

[0052] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A collaborative optimization method for a sintering machine based on bidirectional fusion digital twins, characterized in that, Includes the following steps: S1. Assign a unique identifier to each physical trolley and establish a single trolley time-series history file. The single trolley time-series history file includes single trolley health data and single trolley working condition data marked by a timestamp and a spatial location identifier under the unique identifier. S2, construct a digital twin of the trolley that maps one-to-one with the physical trolley. The digital twin is a comprehensive virtual entity with a physical information neural network and a physical mechanism model as its core and a geometric model as its form carrier. Based on the spatiotemporal mapping rules, complete the alignment and binding of the health data of the single trolley and the working condition data experienced by the single trolley to generate a time-series alignment file for the single trolley. The time-series alignment file of the single trolley and the digital twin are bidirectionally associated in real time for calibration and iteration. S3. Based on the digital twin and the time-series alignment file of the single vehicle, a collaborative optimization decision is generated. The collaborative optimization decision includes predictive maintenance decision, adaptive process control decision, and group detection decision of hidden equipment conditions. S4, respond to the maintenance operation, process control operation and trolley replacement operation corresponding to the collaborative optimization decision, collect operation execution records, status verification results and updated single trolley health data and single trolley experience working condition data, and update the single trolley time sequence history file, the single trolley time sequence alignment file and the digital twin.

2. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 1, characterized in that, In step S1, the single vehicle health data includes sub-component health data, sub-component wear data, fault record data, maintenance record data, and overall vehicle health data. Sub-components include side panels, grate bars, and wheels; sub-component health data includes side panel health, grate bar health, and wheel health. The maintenance record data includes oil filling record data. The overall vehicle health data is obtained through a weighted fusion calculation based on the side panel health, grate bar health, and wheel health. The operating condition data experienced by a single vehicle includes the bellows monitoring data during the operation of the physical vehicle, which includes bellows negative pressure data, temperature data, and valve opening data.

3. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, "Based on spatiotemporal mapping rules, the alignment and binding of the single vehicle's health data and its operational data are completed, generating a single vehicle time-series alignment file. The single vehicle time-series alignment file is then bidirectionally and in real-time associated with the digital twin for calibration and iteration." Specifically, this includes: Each physical trolley is assigned a unique identifier. Based on the trolley's running speed and the position of each inspection station, the estimated time node for the physical trolley to pass through each inspection station is calculated. A data alignment time window is defined based on a preset dynamic buffer duration. The data alignment time window is used to define the effective time period for data collection when the corresponding physical trolley passes through the inspection station. Physical data of the corresponding physical trolley is collected using physical sensors on the inspection station. After calculation and processing, the health data of a single trolley is obtained. At the same time, process parameters of the area corresponding to the inspection station that match the time of the physical data collection of the trolley are collected as the working condition data of the single trolley. The health data of the single trolley and the working condition data of the single trolley are both associated with a unique identifier and a collection timestamp. Based on the data alignment time window, the single vehicle health collection data and single vehicle operating condition data corresponding to the same unique identity and whose collection timestamp is within the data alignment time window are filtered out. The two are then spatiotemporally aligned and fused to match the single vehicle health collection data with the single vehicle operating condition data at the corresponding time and location, resulting in the bound single vehicle health data and single vehicle operating condition data, and generating a single vehicle time-series alignment file. Establish a bidirectional real-time association between the single vehicle time-series alignment file and the digital twin for synchronous calibration and iteration; The data alignment time window is determined based on the sintering machine operating speed, sensor response delay, and data transmission delay.

4. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, The predictive maintenance decision-making for equipment mentioned in step S3 specifically includes: Based on the health time series data in the single vehicle time alignment file, linear regression or exponential smoothing is used to predict the vehicle health value within a preset time period in the future. If the predicted vehicle health value is less than the warning threshold, a predictive maintenance command is triggered. Based on the lubrication record and cumulative running mileage, the grease consumption time is calculated. When the grease consumption time reaches the preset lubrication interval time, a grease injection command is generated.

5. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, The process adaptive control decision mentioned in step S3 specifically includes: When the health data of the single vehicle corresponding to the physical vehicle in a certain continuous area is all lower than the first threshold, it is determined that there is a regional common problem of air leakage or uneven air permeability in the continuous area, and an instruction is generated to reduce the valve opening of the continuous area and its adjacent air boxes proportionally. When the wear level data of the sub-component exceeds the second threshold, it is determined that the physical trolley corresponding to the wear level data of the sub-component has a single serious problem with safety risk, and a safety alarm and a shutdown maintenance command are generated. When the vehicle health data is less than the third threshold and the predicted remaining safe operating time is less than the remaining time until the next planned maintenance cycle, an immediate online replacement instruction is generated.

6. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, The decision-making process for detecting the latent state of equipment includes: The spatiotemporal patterns of negative pressure and temperature curves of each air box in the digital twin are continuously analyzed. When an abnormal pattern of a coordinated decrease in negative pressure and a shift of the high temperature zone in a certain continuous area is detected, it is identified as an abnormal area and a diagnosis is triggered. By tracing back the time-series alignment files of all physical trolleys that passed through the abnormal area, statistically significant common features in the equipment status are found through data mining algorithms. A group early warning report is generated based on the statistically significant common characteristics. The group early warning report includes diagnostic conclusions, the scope of impact of hidden faults, energy efficiency impact assessment, and batch maintenance recommendations.

7. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, The operating condition data of a single vehicle also includes temperature data, load data, running time data, start-stop count data, environmental parameter data, cumulative running mileage data, sintering endpoint data, material layer thickness data, and ignition temperature data during the operation of the physical vehicle; the wear data of the sub-components includes the wear of the side panels, the wear of the grate bars, and the wear of the wheels.

8. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 2, characterized in that, The acquisition of sub-component health data and sub-component wear data in the single vehicle health data includes the following steps: Image information of the sideboard of the physical trolley is obtained at a first preset position point, and the deformation and wear of the sideboard are obtained based on the image information of the sideboard. At the second preset location point, obtain the grate bar image information of the physical trolley, and based on the grate bar image information, obtain the grate bar deformation amount, blockage area ratio and grate bar wear amount; At the third preset position point, obtain the wheel image information of the physical trolley, and obtain the wheel deformation, wheel axle clearance and wheel wear based on the wheel image information; During the movement of the trolley, the status information of the air box corresponding to the physical trolley during sintering is obtained. The air box status information includes at least the air box negative pressure, air box temperature and valve opening data. The deformation of the guardrail, the deformation of the grate, the proportion of the blocked area, the deformation of the wheel, the wheel axle clearance, the wear of the guardrail, the wear of the grate, the wear of the wheel, the negative pressure of the air box, and the temperature of the air box are all single-dimensional data and are respectively marked with identity information, timestamp information and physical location information; The image information of the guardrail, the image information of the grate, and the image information of the wheel are processed to obtain dimensionless quantified values ​​of guardrail health, grate health, and wheel health.

9. The sintering machine digital twin collaborative optimization method based on bidirectional fusion according to claim 8, characterized in that, The overall vehicle health data is obtained by weighted fusion calculation of the quantified health values ​​of the sideboards, the quantified health values ​​of the grate strips, and the quantified health values ​​of the wheels.

10. A sintering machine digital twin collaborative optimization system based on bidirectional fusion, characterized in that, include: The data access and preprocessing module is used to collect multi-source equipment data and process data, assign a unique identifier, timestamp identifier and spatial location identifier to each data, and transmit the collected data to the archive construction module after preprocessing. The file construction module is used to assign a unique identifier to each physical trolley, establish and update the time-series history file of a single trolley, and store the health data and working condition data of a single trolley. The digital twin construction module is used to construct a digital twin that maps one-to-one with the physical trolley. The digital twin is based on a physical information neural network, a physical mechanism model, and a data-driven correction model as its core, and uses a geometric model as its form carrier to realize bidirectional real-time association, data synchronization, and model calibration and iteration between the single trolley time-series alignment file, the single trolley time-series history file and the digital twin. The collaborative optimization instruction generation module is used to generate predictive maintenance instructions, adaptive process control instructions, and hidden status group detection instructions for equipment based on digital twin and single vehicle time-series aligned archive data. This module has built-in linear regression, exponential smoothing, and data mining algorithms to achieve health trend prediction, hidden fault root cause location, and generation of various control instructions. The closed-loop execution and monitoring module is used to respond to various operations corresponding to the collaborative optimization decision, monitor the operation execution process and status verification throughout the process, collect feedback data, update the single vehicle time sequence history file, single vehicle time sequence alignment file and digital twin model, and complete the closed-loop iteration. The operation and maintenance decision module is used to generate trolley health reports, group early warning reports and maintenance suggestions based on digital twin simulation results, single trolley time-series history archives, single trolley time-series alignment archive data and decision execution status, providing decision support for intelligent operation and maintenance of sintering machine trolleys throughout their entire life cycle.

Images