Level and consumption measurement system based on sensor fusion

Abstract

The present application belongs to the technical field of measurement and sensing of process state of feeding on blast furnace channel, and particularly relates to a material level and consumption measurement system based on sensor fusion. The system receives radar level gauge, weight sensor, belt, gate, feeder, unloading trolley and data of material wanting, material slowing and bin position switching event, generates material state sequence according to unified time scale; in different stages, corresponding sensor credibility combination is called, combined with the conservation relation of material quantity into bin, storage capacity in bin and consumption quantity out of bin, joint estimation of material level, effective inventory, instantaneous consumption speed and cumulative consumption quantity is carried out; at the same time, semantic state flow is generated through material level quality mapping update, credibility rollback, bin switching transition identification and cross-bin conservation checking. The present application can reduce the influence of radar local echo, weighing impact and bin position switching time sequence dislocation on measurement result, and improve the bin position attribution consistency of material level, inventory and consumption data and control input reliability.

Description

Technical Field

[0001] This invention belongs to the field of measurement and blast furnace trough feeding process status sensing technology, specifically involving a material level and consumption measurement system based on sensor fusion. Background Technology

[0002] A blast furnace top-feeding system typically consists of belt conveyors, silo storage, unloading trolley distribution, gate or feeder discharge, and furnace top receiving. Material level and consumption are the fundamental measurement parameters for maintaining continuous feeding. Existing systems often use radar level gauges to obtain the material height in the silo, and weight sensors or bin weighing devices to acquire changes in material mass. The control system then generates commands for material demand, deceleration, and silo cut-off based on preset upper and lower limits, weight change curves, or manual confirmation. In this type of system, radar level gauges are usually installed on the top or side of the silo for non-contact distance measurement of the material level. The system then calculates the material level or inventory based on the silo's geometric parameters. Weight sensors reflect the material increase or decrease trend in the silo through weighing signals and calculate the consumption rate and cumulative consumption within a certain time window. Since the object being measured is in a state of alternating continuous conveying and intermittent unloading, the system needs to simultaneously receive signals such as belt operation, gate opening and closing, feeder start and stop, and unloading trolley position. However, conventional solutions often use these signals as alarm or interlock inputs, rather than as boundary conditions for material level and consumption measurement.

[0003] In existing material level measurement methods, radar level gauges can output continuous material level data. However, the raw materials and fuels inside the silo are not a stable plane. Affected by material particle size, moisture content, angle of repose, silo wall adhesion, material drop impact, and dust environment, radar echoes can easily correspond to localized material levels or abnormal reflection points. If the system directly converts radar ranging values ​​into inventory levels, it easily interprets material level tilting, material hanging, material collapse, or short-term echo fluctuations as actual inventory changes. Weight sensors can reflect the overall change in mass within the silo, but weighing signals are affected by mechanical vibration, belt conveyor impact, instantaneous unloading impact, and support structure hysteresis, resulting in short-term data jitter or lag. Conventional systems typically use filtering, smoothing, or manual verification to process radar and weighing data, then perform conversions according to fixed weights or empirical rules. While these methods can suppress some random fluctuations, it is difficult to determine whether a particular material level change corresponds to a material inflow or outflow event, or whether a particular weight change is due to actual material flow or equipment disturbance.

[0004] In existing consumption measurement and automatic feeding control methods, the system typically calculates consumption based on the weight drop in the silo, feeder running time, gate opening status, or furnace top receiving reports. It then issues a material request command based on the low material level limit and a slowdown command based on the high material level limit or process rhythm. Silo switching is executed when a preset condition is met in a particular silo. This control logic processes measurement, status judgment, and action execution in segments, lacking a unified temporal hierarchy among material level data, weight data, and equipment action data. During continuous feeding, there are time differences between belt feeding to the target silo, unloading trolley movement, gate status changes, and silo weight changes. During discharge, there are also response lags in feeder action, material outflow, and silo level drop. Existing systems often read data at fixed sampling periods and then correct deviations using reports or manual recording, resulting in unclear boundaries between replenishment, discharge, resting, and silo-cutting segments. Material in the transition section between silos is easily misclassified into the wrong silo.

[0005] The main technical problem with existing technologies is that the measurement of material level and consumption during the entire process of feeding material into the blast furnace bin lacks a unified fusion mechanism constrained by equipment event boundaries. Radar level data, weight sensor data, and status data from belts, gates, feeders, and unloading trolleys participate in judgment independently, making it impossible to determine the source and attribution of material changes within the same time series. Consequently, when the silo is in different stages such as replenishment, discharge, resting, or silo cutting, the system is still prone to using the same data processing logic, leading to the superposition of radar local material level errors, weighing impact errors, and equipment action delays. The essence of this problem is not the insufficient accuracy of a single sensor, but the lack of a closed-loop measurement relationship constrained by the amount of material entering the silo, the amount of material remaining in the silo, and the amount of material consumed upon discharge. This makes it difficult for the system to simultaneously provide reliable material level, inventory, instantaneous consumption rate, and cumulative consumption under complex operating conditions. Summary of the Invention

[0006] The purpose of this invention is to provide a material level and consumption measurement system based on sensor fusion, which can effectively solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The material level and consumption measurement system based on sensor fusion includes a status acquisition and processing part for receiving radar level gauge data, weight sensor data, belt running status, gate status, feeder status, unloading trolley positioning status, material demand event, material delay event and bin switching event throughout the entire process of feeding material on the blast furnace bin; The status acquisition and processing section generates a material status sequence according to a unified time scale; The material state sequence input stage identification and processing part divides the feeding process into the feeding stage, the discharging stage, the settling stage and the silo cutting stage according to the equipment event boundary; The fusion measurement and processing section calls different sensor confidence combinations according to the stage type, and performs joint estimation of material level, effective inventory, instantaneous consumption rate and cumulative consumption based on the conservation relationship between the amount of material entering the warehouse, the amount of material in the warehouse and the amount of material consumed when leaving the warehouse.

[0008] Preferably, the status acquisition and processing part includes a time scale tuning process, which uses the edge of equipment action, the inflection point of weight change, the inflection point of radar material level change and the bin switching event as composite alignment anchor points to establish event windows for data of different sampling periods. Within the same event window, radar level gauge data is converted into silo space occupancy status, weight sensor data is converted into material quality status, and belt running status, gate status, feeder status, and unloading trolley positioning status are converted into material flow direction constraint status. The material flow direction constraint status, along with material demand events and material delay events, are written into the material status sequence.

[0009] Preferably, the stage identification and processing part constructs an event boundary chain based on the material state sequence. The event boundary chain includes at least the unloading trolley arrival boundary, the belt allowable running boundary, the gate or feeder action boundary, the weight change boundary, the radar material level change boundary, and the bin switching boundary. For data segments between adjacent event boundaries, stage labels are generated according to the feeding direction, discharging direction, action duration, and silo affiliation. Data segments that simultaneously exhibit both feeding and discharging states are marked as coupled stages, allowing the fusion measurement and processing section to use a confidence combination independent of the static stage.

[0010] Preferably, the fusion measurement processing part includes a dynamic reliability allocation process. The dynamic reliability allocation process limits the target silo to the unloading trolley positioning state and belt running state in the feeding stage, limits the discharge path to the gate state and feeder state in the discharge stage, uses the lack of effective equipment action events as the state maintenance condition in the static stage, and uses the silo ownership relationship before and after the switch as the constraint condition in the silo switching stage. At different stages, credibility levels are assigned to radar level gauge data, weight sensor data, and equipment event data, and it is prohibited to directly use mutation data without event support as the basis for inventory updates.

[0011] Preferably, the fusion measurement processing part further includes a material level quality mapping update process, which extracts radar material level change, weight change, material category, silo identifier and working condition from the replenishment segment, discharge segment and static segment confirmed by the stage tag, and generates material level quality mapping relationship corresponding to different silos and different materials. Before updating the mapping relationship, data segments that do not satisfy the material conservation relationship, storage location relationship, or equipment event boundary relationship are removed, and the retained data segments are used as update samples for subsequent material location and inventory conversion.

[0012] Preferably, the dynamic credibility allocation process further includes a credibility rollback process, which compares the radar level change direction, weight change direction, and material flow direction constraint status within the same event window. When the direction of radar level change is inconsistent with the direction of weight change and there is no corresponding equipment event boundary, the corresponding data segment will be transferred to the anomaly candidate queue. When the weight change exhibits impact characteristics and the radar level change does not meet the response conditions corresponding to the stage label, the participation ratio of weight sensor data in the corresponding event window is reduced, and the equipment event data is retained as the basis for subsequent attribution.

[0013] Preferably, the material level quality mapping update process further includes a segmented mapping maintenance process, which divides the material level quality mapping relationship into a low segment, a middle segment and a high segment according to the silo space occupancy status, and maintains candidate mapping samples corresponding to the replenishment stage, the discharge stage and the settling stage in each segment respectively. Before entering the formal mapping relationship, the candidate mapping sample must simultaneously meet the following conditions: complete event boundary chain, unique silo ownership, closed loop between incoming material quantity and outgoing material consumption quantity, and no isolated echo abrupt change in radar level curve.

[0014] Preferably, the stage identification processing part further includes a silo switching transition identification process, which establishes a raw material silo tail-end status window and a target silo start-up status window after detecting a silo switching event. Within the tail section status window of the raw material silo, the tail material consumption is defined by the feeder status, gate status, and weight drop boundary. Within the starting status window of the target silo, the feeding is defined by the unloading trolley positioning status, belt running status, and radar level rise boundary. The overlapping data segments between the two status windows are marked as transitional segments awaiting confirmation.

[0015] Preferably, the transition segment to be confirmed is processed through a cross-warehouse conservation verification process, which combines the weight decrease of the raw material warehouse tail state window, the weight increase of the target material warehouse initial state window, the belt conveyor status, and the unloading trolley positioning status into a cross-warehouse constraint group. When there is a conflict in warehouse location ownership within the cross-warehouse constraint group, the transitional unconfirmed segment is segmented according to the order of the event boundary chain and written into the raw material warehouse consumption sequence and the target material warehouse inventory sequence respectively. Data segments that do not meet the segmentation conditions do not participate in the cumulative consumption update.

[0016] Preferably, the fusion measurement and processing part further includes a semantic state release process. The semantic state release process generates a semantic state stream of the entire process of feeding material on the blast furnace trough according to the semantic skeleton of equipment, material, workstation, working condition, event and rule, based on the material level, effective inventory, instantaneous consumption rate, cumulative consumption, stage label, sensor credibility, abnormal candidate queue and trough assignment result after the cross-warehouse conservation verification process. The semantic state stream is written into the knowledge center of intelligent perception as the input state for the automatic control of material demand, material delay and trough switching.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention forms a material state sequence by aligning radar level gauge data, weight sensor data, belt conveyor status, gate status, feeder status, unloading trolley positioning status, material demand events, material deceleration events, and bin switching events with a unified timescale. The material supply process is divided into replenishment, discharge, resting, and bin-cutting stages using equipment event boundaries. Joint estimation is then performed based on the conservation relationship between the amount of material entering the bin, the amount of material remaining in the bin, and the amount of material consumed upon discharge. Because level changes, weight changes, and equipment actions are included in the same event window, the system can limit material flow direction, bin affiliation, and stage attributes during measurement, avoiding the direct use of radar mutations or weighing shocks without event support as the basis for inventory updates. The radar level gauge provides bin space occupancy status, the weight sensor provides material quality status, and equipment events provide change boundaries and affiliation conditions. These three types of data are cross-checked under conservation constraints, ensuring consistency in the source, time boundaries, and bin affiliation of level, effective inventory, instantaneous consumption rate, and cumulative consumption. This processing method addresses the problem of the separation between level measurement and consumption measurement in existing technologies, so that the measurement results no longer depend solely on the instantaneous reading of a single sensor, but are jointly determined by sensor response, equipment events, and material conservation.

[0018] 2. This invention also constrains the subdivided measurement deviations under different working conditions through processing mechanisms such as dynamic credibility allocation, material level quality mapping update, credibility rollback, segmented mapping maintenance, silo switching transition identification, cross-silo conservation verification, and semantic state release. In the replenishment stage, the target silo is defined by the unloading trolley positioning status and belt running status; in the discharge stage, the discharge path is defined by the gate status and feeder status; in the static stage, the lack of effective equipment action events serves as the state maintenance condition; and in the silo switching stage, the silo affiliation before and after the switch serves as the constraint condition. This ensures that radar data, weighing data, and event data from different stages participate in the calculation according to different credibility levels. The material level quality mapping relationship is updated by data segments that have been confirmed by stage labels and satisfy the conservation relationship, reducing interference from material hanging, material collapse, isolated echoes, and impact weighing segments on subsequent conversions. Silo switching transition identification and cross-silo conservation verification separate the raw material silo tail-end consumption, the target silo initial feeding, and the transition unconfirmed segments, reducing the consumption statistics affiliation deviation during cross-silo switching. The material location, inventory, consumption, stage label, credibility, and warehouse location results processed as described above can be written into the knowledge center of intelligent sensing, providing structured input status for automatic control of material demand, material delay, and warehouse location switching. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall fusion measurement process of the sensor fusion-based level and consumption measurement system of the present invention. Figure 2 This is a flowchart of the time scale adjustment and stage identification process of the state acquisition and processing section of the present invention; Figure 3 This is a flowchart illustrating the dynamic credibility allocation, credibility rollback, and material level quality mapping update process of the present invention. Figure 4 This is a flowchart of the switching transition identification, cross-warehouse conservation verification, and semantic state publishing process of the present invention; Figure 5 This is a composite observation diagram of the present invention; Figure 6 This is the belt increment closure difference diagram of the present invention; Figure 7 This is a graph showing the instantaneous consumption rate of the present invention. Detailed Implementation

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

[0021] Please refer to Figure 1This embodiment provides a sensor fusion-based material level and consumption measurement system for measuring material level, inventory, and consumption throughout the entire blast furnace trough feeding process. The system uses radar level gauge data, weight sensor data, belt operation status, gate status, feeder status, unloading trolley positioning status, material demand events, material delay events, and silo switching events as input data. These input data are derived from existing measurement points, control points, and event points during the trough feeding process. The state acquisition and processing section performs unified time-stamping on the above data, forming a material state sequence that describes the state changes of the material in the silo, conveying path, and discharge path. The stage identification processing section divides the continuous feeding process into a replenishment stage, a discharge stage, a stationary stage, and a silo cutting stage based on equipment event boundaries. The fusion measurement processing section calls different sensor reliability combinations at different stages, using the conservation relationship between the amount of material entering the silo, the amount of material remaining in the silo, and the amount of material consumed upon discharge as constraints to measure material level, effective inventory, and instantaneous consumption rate. In this embodiment, radar level gauge data is not used as the sole basis for inventory calculation, but is first converted into silo space occupancy status. Weight sensor data is not used as the sole basis for consumption calculation, but is first converted into material quality status. Belt running status, gate status, feeder status, and unloading trolley positioning status are not used as separate alarm statuses, but are collectively converted into material flow direction constraint statuses. This forms a closed-loop measurement system jointly defined by measured values, action status, and event boundaries. Through the above configuration, this embodiment ensures that each change in level and weight can be limited to the corresponding replenishment, discharge, resting, or silo cutting stage, avoiding level jumps, weighing impact misjudgments, and consumption attribution errors caused by using the same measurement logic under different operating conditions. The advantage of this embodiment is that level, inventory, and consumption are no longer calculated from isolated measuring points, but are jointly defined by material conservation relationships, sensor responses, and equipment event boundaries. The measurement results have clear time boundaries and silo attribution.

[0022] In this embodiment, the status acquisition and processing part is a software processing structure that can be deployed on an edge computing node, a trough feeding control server, or a data processing server connected to the blast furnace feeding control system. The status acquisition and processing part sets a data description field for each type of input data. This data description field includes the data source, acquisition time, associated device, associated silo, associated material path, numerical status, action status, and event type. Radar level gauge data generates level observation records according to the associated silo; weight sensor data generates mass observation records according to the associated trough or silo; belt conveyor operation status generates conveying path records; gate status and feeder status generate discharge path records; unloading trolley positioning status generates infeed path records; and material demand events, material hold events, and silo switching events generate control event records. The status acquisition and processing part writes the above records into the same data buffer according to a unified timescale, and records with the same silo identifier, similar timestamps, and adjacent... The data organization of equipment path relationships is a material state sequence, which can be represented as a state tuple arranged in ascending order of time. The state tuple simultaneously includes material level observation, quality observation, feed path status, discharge path status, control event status, and data confidence flag. After reading the material state sequence, the stage identification and processing part does not directly perform stage segmentation according to a fixed time interval, but uses the equipment event boundary as the segmentation basis. The feeding stage is triggered by the event chain that the unloading trolley is positioned to the target silo and the belt is in an allowed running state. The discharge stage is triggered by the gate or feeder action event. The stationary stage is triggered by the absence of effective feed and discharge action events and the data is in a continuous observation state. The silo switching stage is triggered by the silo switching event and the change of the target silo before and after the switching. The advantage of this embodiment is that the organization of the measurement data is consistent with the actual action chain of feeding on the blast furnace trough, so that subsequent fusion measurement can select the data objects to participate in the calculation according to the stage attributes.

[0023] ; in, Indicates the first The material state tuple at each moment. Indicates the data collection time under a unified time scale. This indicates the level reading obtained from the radar level gauge. This represents the mass measurement obtained from the weight sensor. Indicates the belt running status. Indicates the gate status. Indicates the status of the feeder. This indicates the positioning status of the unloading trolley. This indicates the state of at least one of the following events: material demand event, material delay event, or warehouse switching event, for example, at a certain moment. Seconds, radar level observation Weight observation Belt running status Gate status Feeder status Positioning status of the unloading trolley Warehouse, material event status Then the tuple is identified as a candidate state tuple related to the feed, and if in the same event window thereafter... and If a change occurs simultaneously in the same direction as the material infeed, the stage identification and processing unit classifies the data segment into the replenishment stage. The fusion measurement and processing unit uses the stage label output by the stage identification and processing unit as a condition, invoking different confidence combinations for calculation. In the replenishment stage, the unloading trolley positioning status is used to define the target silo, the belt running status is used to define whether the material has an infeed path, radar level changes are used to characterize changes in silo space occupancy, and weight changes are used to characterize changes in the target silo's mass. In the discharge stage, the gate status and feeder status are used to define the discharge path, weight decrease changes are used to characterize the material discharge amount, and radar level decrease changes are used to characterize the silo space release status. In the static stage, the lack of a valid equipment action event is used as a condition for maintaining the state. If the radar level or weight changes without corresponding event support, the fusion measurement and processing unit does not directly update the valid inventory, but instead treats the data segment as static. As an anomaly candidate, during the silo switching phase, the silo switching event defines the time boundaries of the raw material silo and the target silo. The fusion measurement processing part calculates the tail-end consumption and the initial feeding based on the silo affiliation before and after the switching. The fusion measurement processing part uses the conservation relationship between the amount of material entering the silo, the amount of material in the silo, and the amount of material leaving the silo for joint estimation at each stage, so that the inventory status at a certain moment is determined by the inventory at the previous moment, the amount of material entering the silo at the current stage, and the amount of material leaving the silo at the current stage. In this embodiment, the material level observation and weight observation are not directly output values, but participate in the calculation as observation constraints under the conservation relationship. The advantage of this embodiment is that the same sensor data has different participation methods at different stages, which can reduce the interference of feeding impact, material discharge lag, static drift, and silo switching transition on the measurement results.

[0024] ; in, Indicates the first The effective inventory in the warehouse at each time point is obtained through fusion estimation. This represents the effective inventory in the warehouse obtained through fusion estimation at the previous moment. Indicates the first The amount of incoming materials in the event window at any given time. Indicates the first Outbound consumption within the event window at any given time. The operator represents the correction amount consisting of measurement residuals. and These represent increases and decreases in inventory, respectively, such as the effective inventory at the previous moment. Material quantity entering the warehouse during the replenishment stage No warehouse exit action within the same event window and Measurement residual correction amount ,but If it is in the material discharge stage and Outbound consumption Measurement residual correction amount Then it can be calculated using the same formula. .

[0025] In a preferred embodiment, reference Figure 2 The status acquisition and processing section includes a time-scale calibration process. This process addresses timing misalignment issues caused by different sampling periods, transmission delays, and event recording granularities from multiple data sources. Specifically, the time-scale calibration process does not use a single system timestamp as the sole alignment basis. Instead, it sets equipment action edges, weight change inflection points, radar level change inflection points, and bin switching events as composite alignment anchors. Equipment action edges include changes in the conveyor belt from stopped to running or from running to stopped; changes in the gate from closed to open or from open to closed; and changes in the feeder from stopped to running or from running to stopped. Weight change inflection points include curve inflection points where the weight curve changes from stable to rising, from stable to falling, or from rising or falling to stable. Radar level change inflection points include... The material level curve transitions from stable to rising, from stable to falling, and from rising or falling back to stable at the curve inflection points. Bin switching events include records of events where the original target bin stops supplying material and the new target bin starts supplying material. The time-scale tuning process establishes an event window based on the aforementioned composite alignment anchor points. Within the event window, radar level gauge data is converted into bin space occupancy status, weight sensor data is converted into material mass status, and belt conveyor status, gate status, feeder status, and unloading trolley positioning status are converted into material flow direction constraint status. This material flow direction constraint status, along with material demand events and material hold-up events, is written into the material status sequence. The advantage of this embodiment is that data alignment no longer depends on a single acquisition moment but is jointly defined by physical actions, measurement inflection points, and control events, allowing cross-system data to enter the same measurement window. The organization of status fields within the event window is shown in Table 1.

[0026] Table 1. Organization of Status Fields within the Event Window ; In this embodiment, after the fields listed in Table 1 are written into the same material status sequence, the status acquisition and processing part also performs an integrity check on the event window. If an event window only has radar level changes but no weight changes, equipment action edges, or control event records, the window type is marked as an observation window to be confirmed. If an event window only has equipment action edges but no radar level changes or weight changes, the window type is marked as an action non-response window. If an event window contains both infeed constraint fields and outfeed constraint fields, the window type is marked as a coupled observation window. The integrity check result does not replace subsequent measurement calculations, but is used as input for the fusion measurement processing part to allocate confidence and select conservation relationships. Specifically, the level observation field in the observation window to be confirmed does not directly update the inventory, the event field in the action non-response window is used for subsequent anomaly attribution, and the infeed and outfeed quantities in the coupled observation window are modeled separately to prevent the same weight change from being interpreted as both warehousing and outfeing. The advantage of this embodiment is that the time-scale tuning process has already marked the availability of the data window before entering the fusion calculation, which can reduce the impact of missing data, clock offset, and missed action acquisition on subsequent calculations.

[0027] ; in, This indicates the time deviation between two alignment anchor points. This indicates an anchor point moment in an event such as the edge of equipment operation, the inflection point of weight change, the inflection point of radar material level change, or a bin switching event. This indicates another anchor point moment used for comparison, such as the moment of belt start-up. Seconds, the moment when the weight increases at the inflection point seconds, then The seconds indicate that the belt start occurs before the inflection point of the weight increase. If the radar level rises at the inflection point... Seconds, the moment when the weight increases at the inflection point seconds, then The second indicates that the radar level response lags behind the relative weight inflection point by 0.5 seconds. The timescale tuning process is based on multiple... The value determines the event window boundary and performs state merging.

[0028] Furthermore, the stage identification and processing section constructs an event boundary chain based on the material state sequence. This event boundary chain includes at least the unloading trolley arrival boundary, the belt allowable operation boundary, the gate or feeder action boundary, the weight change boundary, the radar level change boundary, and the bin switching boundary. When constructing the event boundary chain, the stage identification and processing section sorts each boundary according to its chronological order and records the corresponding equipment object, bin object, and material path object for each boundary. If the unloading trolley arrival boundary, the belt allowable operation boundary, the weight increase boundary, and the radar level increase boundary appear in a reasonable order within the same event window, the stage identification and processing section classifies the data segment between adjacent boundaries as a replenishment stage. If the gate or feeder action boundary, the weight decrease boundary, and the radar level change boundary appear in a reasonable order within the same event window, the stage identification and processing section classifies the data segment between adjacent boundaries as a replenishment stage. When the material level drop boundary appears within the same event window, the stage identification and processing part classifies the data segment between adjacent boundaries as the discharge stage. If there is no belt operation, no change in the unloading trolley target, no gate or feeder action within the adjacent event window, and only stable material level and stable weight exist, it is classified as the stationary stage. If the bin switching boundary is within the event boundary chain, the stage identification and processing part classifies the data segment before and after the switching as the bin cutting stage. If the material replenishment related boundary and the material discharge related boundary overlap within the same time range, the stage identification and processing part marks the corresponding data segment as the coupling stage and distinguishes between the coupling stage and the stationary stage. The advantage of this embodiment is that the stage label is formed by the event boundary chain, avoiding the judgment of working conditions based on fixed duration or single sensor trend.

[0029] In this embodiment, the stage identification processing unit also generates stage labels for data segments between adjacent event boundaries. The stage labels include the feeding direction, discharging direction, action duration, and silo affiliation. The feeding direction is determined by the unloading trolley positioning status and belt running status; the discharging direction is determined by the gate status and feeder status; the action duration is determined by continuous recording between event boundaries; and the silo affiliation is determined by the mapping relationship between equipment objects and silo objects. For data segments that simultaneously exhibit both feeding and discharging states, the stage identification processing unit does not use net weight change as the sole criterion for judgment, but instead retains the feeding direction and... The dual state of the discharge direction is marked as the coupling stage. The fusion measurement and processing part adopts a confidence combination for the coupling stage that is independent of the static stage. Specifically, in the coupling stage, the weight change is broken down into the inbound contribution and the outbound contribution. The radar level change is used to verify the change direction of the silo space occupancy status. The equipment event boundary is used to limit the time attribution of inbound and outbound. In the static stage, if the weight change and level change lack event support, they will not be used as the basis for inventory change. The advantage of this embodiment is that the complex working condition of simultaneous replenishment and discharge can be identified separately, avoiding the misunderstanding of net weight change as unidirectional consumption or unidirectional feeding.

[0030]

[0031] in, Indicates by the first The event boundary and the first A data segment defined by an event boundary. Represents the first in the event boundary chain At a boundary moment, Indicates the next boundary time immediately following, symbol This represents a half-open interval that includes the start boundary but not the end boundary, such as the unloading trolley's arrival boundary. Seconds, belt stop boundary Seconds, then data fragment The process covers all material status tuples from the unloading trolley's arrival to the belt's stop. If there is both a weight increase and a radar level increase within the segment, the segment is marked as a replenishment stage. If there is both feeder operation and weight decrease within the segment, the segment is marked as a coupling stage and both material flow directions are retained.

[0032] In a preferred embodiment, reference Figure 3 The integrated measurement and processing section includes a dynamic reliability allocation process. This process differentiates the participation levels of radar level gauge data, weight sensor data, and equipment event data at different stages. During the replenishment stage, the dynamic reliability allocation process defines the target silo based on the unloading trolley's positioning status and the belt's running status. If the target silo matches the silo associated with the radar level gauge and the weight change direction is increasing, the participation of radar level gauge data and weight sensor data in the inbound quantity estimation is increased. If the unloading trolley's positioning status is unstable or the belt's running status is interrupted, the corresponding event window is transferred to pending confirmation processing. During the discharge stage, the dynamic reliability allocation process defines the target silo based on the gate status and feeder status. If the weight change direction is downward and the radar level trend is consistent with the discharge direction, then the weight sensor data is used as the main observation constraint for the outflow consumption, and the radar level gauge data is used as the spatial state verification. In the static stage, the dynamic reliability allocation process uses the lack of effective equipment action events as the state maintenance condition. If a short-term change occurs, the corresponding data will not be directly updated to the inventory. In the silo switching stage, the dynamic reliability allocation process uses the silo ownership relationship before and after the switch as the constraint condition, and the sensor data of the raw material silo and the target silo are respectively assigned to the corresponding calculation window. The advantage of this embodiment is that the reliability is not determined by a fixed sensor type, but by the working condition stage, event boundary and material flow direction.

[0033] In this embodiment, the dynamic reliability allocation process can use a normalized reliability vector to represent the participation ratio of various types of data. The reliability vector includes at least radar level reliability, weight reliability, and equipment event reliability. The dynamic reliability allocation process updates the reliability vector based on stage labels, sensor change direction, equipment action integrity, and conservation residuals. Stage labels determine the initial reliability combination, sensor change direction determines whether it conforms to the material change direction of the current stage, equipment action integrity determines whether the event support is valid, and conservation residuals determine whether the current observation is consistent with the inventory closure relationship. When a certain type of data is inconsistent with the stage label or conflicts with the conservation relationship, The corresponding confidence level decreases, but the reduced data is still retained in the material state sequence for anomaly attribution and is not deleted from the original record. When a certain type of data is consistent with the stage label and closes with the conservation relationship, the corresponding confidence level is maintained or enters the sample update queue. In this embodiment, it is prohibited to directly use mutation data without event support as the basis for inventory updates. The mutation data includes isolated radar echo changes during the static stage, weight jumps without equipment action, and storage location abrupt changes outside the storage boundary. The advantage of this embodiment is that the system can perform hierarchical processing on the data participating in the fusion calculation while retaining the original observation records, so that measurement calculation and anomaly tracking are separated. The confidence participation rules under different stages are shown in Table 2.

[0034] Table 2 Credibility Participation Rules at Different Stages

[0035] ; in, Indicates the first The conserved residual at a given moment This represents the merged inventory calculated based on the amount of materials received, the amount of materials still in the warehouse, and the amount of materials consumed upon leaving the warehouse. This represents the observed inventory calculated from radar level-mass mapping or weight observations. (Operator) This indicates taking the absolute value; for example, calculating the merged inventory according to the conservation principle at a certain replenishment stage. The observed inventory was obtained by converting the material level and quality mapping relationship. ,but If the device events in the same window are complete and the residuals are within the observation range allowed in the corresponding stage, the relevant data fragments can be used as candidate samples to enter the mapping update process. If the residuals come from a sudden change without event support in the static stage, the corresponding observation data will only enter the abnormal candidate queue.

[0036] Furthermore, the integrated measurement and processing section also includes a level-mass mapping update process. This process addresses conversion errors caused by nonlinear changes in the relationship between level and mass under different materials, silos, and operating conditions. Specifically, the level-mass mapping update process extracts radar level changes, weight changes, material categories, silo identifiers, and operating conditions from replenishment, discharge, and stationary segments confirmed by stage tags, forming candidate mapping samples. The replenishment segment provides the correspondence between level increase and mass increase, the discharge segment provides the correspondence between level decrease and mass decrease, and the stationary segment provides a stable reference relationship under conditions of no inflow and no outflow. Before candidate mapping samples enter the mapping relationship update, they need to be checked for material conservation relationship, warehouse location relationship and equipment event boundary relationship. If a data segment has a non-closed relationship between the amount of material entering the warehouse and the increase in weight, non-unique warehouse location, missing equipment event boundary or isolated abrupt change in radar curve, the data segment is removed and does not participate in the mapping relationship update. In this embodiment, the mapping relationship does not use a fixed conversion coefficient, but maintains the corresponding relationship according to warehouse identification, material category and working condition. The advantage of this embodiment is that the conversion relationship comes from the material change segment confirmed by the event boundary, which can reduce the continuous impact of dust echo, hanging material, collapse material and weighing impact on subsequent inventory conversion.

[0037] In this embodiment, the material level quality mapping update process maintains candidate mapping samples in groups. Samples from the same silo, the same material category, and the same operating condition are grouped into the same mapping group. The mapping group records the material level change, weight change, stage label, conservation residual, and data source window. If the replenishment segment and the discharge segment form mapping samples in opposite directions within the same material level range, the material level quality mapping update process performs a consistency check on the two. When there is a large residual between the replenishment mapping and the discharge mapping within the same material level segment, the corresponding sample group does not immediately replace the original mapping relationship, but waits for subsequent static segments or complete segments. The event chain segments are confirmed. If the candidate samples in multiple consecutive event windows all meet the conditions of conservation closure and unique warehouse location, the candidate mapping samples are merged into the formal mapping relationship. The formal mapping relationship is used to convert radar level observations into observed inventory and participate in fusion measurement processing together with weight observations. In this embodiment, the sample source event window can also be retained in the mapping group so that the corresponding replenishment, discharge or static segment can be traced back when an anomaly occurs. The advantage of this embodiment is that the mapping update does not depend on the single observation result, but on multi-segment data with stage labels and conservation closure relationship.

[0038] ; in, Indicates the first The comprehensive observations used for fusion measurement at each moment, This indicates the observed space occupied by the silo corresponding to the radar level gauge. This indicates the mass measurement corresponding to the weight sensor. This indicates the material flow constraint quantity corresponding to the equipment event data. This represents the radar level reliability coefficient. This represents the weight reliability coefficient. This represents the reliability coefficient of a device event. This indicates a weighted composite operation. For example, during the replenishment stage, when the radar level change is in the same direction as the weight increase and the unloading trolley is positioned at the target bin, the following can be taken: , , If a weighing impact occurs in the same window and the radar level changes continuously, it can be adjusted to... , , ,when , , At that time, the calculated Used for subsequent inventory estimations rather than directly as the final inventory output. (Reference) Figure 5 The horizontal axis represents time, displaying the trends of composite observations, inventory levels, and input components according to state time. The vertical axis represents tons, showing the composite observation mass and inventory mass, in tons. Figure 5 The solid line curve depicts the time-series change sequence of the main indicator in the historical sampling window, while the gray dashed or dotted lines represent auxiliary sequences, threshold lines, or closed reference lines. The star-shaped marker corresponds to the current sampling window. Figure 5 The calculation is based on a composite observation obtained by combining material level height, belt weight, and ultrasonic distance according to coefficients, and compared with inventory levels. The calculation relationship is displayed mathematically in the expression block below. The current output covers 240 sampling windows, and the overall trend remains stable, with only minor fluctuations in some local windows; the main indicator ranges from 115.9678 to 177.2837, with a mean of 148.0105. In a preferred embodiment, the dynamic confidence allocation process further includes a confidence rollback process. This rollback process compares the radar level change direction, weight change direction, and material flow constraint status within the same event window. The radar level change direction is determined by the difference between the current window's level observation and the previous stable window's level observation. The weight change direction is determined by the difference between the current window's weight observation and the previous stable window's weight observation. The material flow constraint status is jointly determined by the belt conveyor running status, unloading trolley positioning status, gate status, and feeder status. When the radar level change direction and weight change direction are inconsistent and a corresponding equipment event boundary is missing, [further details can be added]. The confidence rollback process transfers the corresponding data segment to the anomaly candidate queue and does not use the data segment to update the effective inventory and material level quality mapping relationship. When the weight change has impact characteristics and the radar material level change does not reach the response condition corresponding to the stage label, the confidence rollback process reduces the participation ratio of weight sensor data in the corresponding event window and retains the equipment event data as the basis for subsequent attribution. The impact characteristics can be represented by short-term weight surge, short-term weight drop, weight curve peak, or change pattern that highly coincides with the belt start-stop boundary. The advantage of this embodiment is that the sensor anomaly processing is bound to the event window, avoiding the coarse rejection of the entire segment of operating data.

[0039] In this embodiment, the anomaly candidate queue records anomaly fragments according to the anomaly source and attribution conditions. The anomaly sources include radar level direction anomalies, weight direction anomalies, missing material flow direction constraints, missing equipment event boundaries, and conservation residual anomalies. The attribution conditions include the corresponding event window, corresponding silo, corresponding material path, corresponding stage label, and confidence vector before participating in the calculation. The confidence rollback process uses the anomaly candidate queue as a bypass input for subsequent fusion measurement processing. When the subsequent event window can fill in the missing equipment action boundary or form a conservation closure relationship, the fragments in the anomaly candidate queue can be re-marked as delayed confirmation fragments. If the subsequent event window still cannot form event support or conservation closure relationship, the fragments remain in the anomaly candidate state and do not enter the level quality mapping update process. In this embodiment, the original change curve of the weight sensor is retained during the weighing impact processing. The original observation record is not deleted, but its participation ratio is reduced in the comprehensive observation calculation. The specific processing method is to adjust the confidence coefficient and record the reason for the confidence change in the semantic state flow. The advantage of this embodiment is that the abnormal data is not simply discarded, but can participate in the attribution when the subsequent event is completed, while not affecting the closure of the current inventory and consumption calculation.

[0040] Furthermore, the material level quality mapping update process also includes a segmented mapping maintenance process. This process divides the material level quality mapping relationship into low, medium, and high segments based on the silo space occupancy status. These segments are not set by hardware parameters but are determined based on the distribution range of silo space occupancy status in historical operational data and the observation density areas corresponding to stage labels. Within each segment, candidate mapping samples corresponding to the replenishment, discharge, and resting stages are maintained. The low segment describes the material level quality correspondence when the silo is close to a low inventory state, while the medium segment describes the normal supply state. The lower segment describes the material level quality correspondence, while the higher segment describes the material level quality correspondence when approaching a high inventory state. Before entering the formal mapping relationship, candidate mapping samples must simultaneously meet the following conditions: complete event boundary chain, unique silo ownership, closed loop between incoming and outgoing material consumption, and no isolated echo abrupt changes in the radar level curve. If a candidate mapping sample only meets some of the conditions, it is retained in the candidate sample area and does not participate in the formal mapping update. The advantage of this embodiment is that the material level quality mapping relationship can adapt to the nonlinear changes of different inventory ranges, while restricting abnormal segments from entering the formal mapping relationship through stage samples and complete event chains. The sample admission conditions in segmented mapping maintenance are shown in Table 3.

[0041] Table 3 Sample Admission Conditions in Segmented Mapping Maintenance

[0042] In this embodiment, the segmented mapping maintenance method listed in Table 3 is also used to handle the mutual verification between samples at different stages. The samples in the replenishment stage reflect the relationship between the increase in material level and the increase in mass; the samples in the discharge stage reflect the relationship between the decrease in material level and the decrease in mass; and the samples in the settling stage reflect the stability of material level and mass under conditions of no material flow. The segmented mapping maintenance process performs joint checks on the three types of samples within the same material level segment. If the slope directions corresponding to the replenishment sample and the discharge sample are consistent and the settling sample remains stable, the mapping relationship of the material level segment is retained and can be used for subsequent inventory conversion. If the settling sample shows that the radar material level is isolated... If the vertical echo changes abruptly, but the weight sensor data does not show a corresponding change, the corresponding radar sample will not enter the formal mapping. If the discharge sample shows a decrease in weight but the radar material level does not change accordingly, and the feeder status and gate status are intact, then the sample is retained as a possible candidate segment for material hanging in the bin, without directly correcting the mapping relationship. This embodiment uses segmented mapping maintenance to form a closed loop between the mapping relationship update and the stage label, the conservation relationship, and the abnormal candidate queue. The advantage of this embodiment is that the material level quality conversion relationship can be maintained separately according to the material level range and the working condition stage, reducing the dependence of a single linear conversion relationship on the complex material form in the bin.

[0043]

[0044] in, Indicates the first Material level quality mapping coefficient for each material level segment Indicates belonging to the first A sample set that is segmented by material location and meets the admission criteria. Indicates the first Weight change in each admission sample Indicates the first For example, if there are three incoming material replenishment samples in the median segment, with weight changes of 20, 18, and 22 respectively, and level changes of 0.40, 0.36, and 0.44 respectively, then... This indicates that the mass change corresponding to a unit change in material level within this median range can be used to participate in the observation inventory conversion according to the mapping coefficient. If the event boundary of a candidate sample is incomplete, it will not be included in the set. .

[0045] In a preferred embodiment, reference Figure 4 The stage identification and processing section also includes a silo switching transition identification process. This process handles the attribution relationships between the raw material silo tail-end consumption, the target silo initial feeding, and the transition segment during silo switching. Specifically, after detecting a silo switching event, the silo switching transition identification process establishes a raw material silo tail-end status window and a target silo initial status window. The raw material silo tail-end status window starts from the most recent discharge-related boundary before the silo switching event and ends at the tail-end response end boundary after the silo switching event. The target silo initial status window starts from the feeding preparation boundary or switching event boundary before the silo switching event and ends at the target silo... The boundary for forming a stable feeding or discharging response is terminated. Within the tail section status window of the raw material silo, the tail material consumption is defined by the feeder status, gate status, and weight descent boundary. Within the starting status window of the target silo, the feeding is defined by the unloading trolley positioning status, belt running status, and radar level rise boundary. If there is time overlap between the two status windows or the data segment cannot be directly assigned, the silo cutting transition identification process marks the overlapping data segment as a transition segment to be confirmed. The advantage of this embodiment is that the silo cutting period is not simply divided into two fixed time periods, but is defined by the actual event boundaries of the raw material silo and the target silo.

[0046] In this embodiment, the transition identification process for switching silos also sets dual-silo status identifiers for transition segments to be confirmed. These dual-silo status identifiers include a raw material silo consumption candidate identifier, a target silo infeed candidate identifier, a shared conveyor path identifier, and a segmentation identifier. The raw material silo consumption candidate identifier is triggered by the feeder status, gate status, and weight drop boundary. The target silo infeed candidate identifier is triggered by the unloading trolley positioning status, belt running status, and radar level rise boundary. The shared conveyor path identifier is triggered by an event window where the same belt running status simultaneously affects the silos before and after the switch. The segmentation identifier is used to record segments that cannot be temporarily... For data segments belonging to a single silo, the silo-cutting transition identification process writes the above-mentioned identifiers into the material status sequence and inputs them into the cross-silo conservation verification process. In specific processing, the weight decrease in the end status window of the raw material silo is first matched with the discharge path event, and the weight increase and radar level rise in the beginning status window of the target silo are first matched with the infeed path event. The data in the overlapping area of ​​the two windows is further processed by the cross-silo constraint group. The advantage of this embodiment is that the silo-cutting transition segment is retained as an independent processing object, avoiding the deviation of cumulative consumption caused by directly classifying it into the raw material silo or the target silo.

[0047] Furthermore, the transitional unconfirmed segment is processed through a cross-warehouse conservation verification process. This process combines the weight decrease in the raw material silo's tail-end state window, the weight increase in the target silo's initial state window, the belt conveyor status, and the unloading trolley's positioning status into a cross-warehouse constraint group. The weight decrease in the raw material silo's tail-end is used to characterize the silo's tail-end consumption before the switchover; the initial weight increase in the target silo is used to characterize the change in material input after the switchover; the belt conveyor status is used to limit whether the material is in the conveying path; and the unloading trolley's positioning status is used to limit the target silo location of the conveyed material. When there is a silo location ownership conflict in the cross-warehouse constraint group, the cross-warehouse conservation verification process follows... The sequence of events in the boundary chain divides the transitional segments to be confirmed. The first segment after division is written into the raw material warehouse consumption sequence, and the second segment after division is written into the target material warehouse inventory sequence. Data segments that do not meet the division conditions do not participate in the cumulative consumption update. The warehouse location conflict includes situations where the unloading trolley is positioned pointing to the target material warehouse but the weight decrease still occurs at the raw material warehouse window, the belt conveyor is in continuous status but the radar level rises after the switch to the target material warehouse, and the gate or feeder tail section action overlaps with the infeed path event. The advantage of this embodiment is that the material in the warehouse transition segment is handled through the cross-warehouse conservation relationship, and the consumption sequence and inventory sequence have clear division boundaries.

[0048]

[0049] in, Indicates the first Cross-position conservation check value for each position transition window This indicates the amount of weight loss within the status window at the end of the raw material silo. This indicates the amount of weight increase within the initial state window of the target silo. This represents the in-transit material quantity observation item defined by the belt conveyor status and the unloading trolley positioning status. (Operator) This indicates that the changes in position size before and after the switch will be combined. (Operator) This indicates the deduction of in-transit material quantity observation items, such as the weight decrease at the end of the raw material storage section. Increase in the initial weight of the target silo Observation item for material quantity in transit ,but This indicates that a closed loop has been formed between the raw material silo tail-end consumption, the target silo initial feed, and the materials in transit within the switching window. If the event cannot be closed and the event boundary chain is missing, the corresponding transitional unconfirmed segment will not be included in the cumulative consumption update. (See reference) Figure 6 , Figure 6 The horizontal axis represents time, displaying belt closure amount and inventory residuals according to the time of the status. The vertical axis represents tons, showing closure amount, closure error, and residuals, all in tons. Figure 6 The solid line curve depicts the time-series change sequence of the main indicator in the historical sampling window, while the gray dashed or dotted lines represent auxiliary sequences, threshold lines, or closed reference lines. The star-shaped marker corresponds to the current sampling window. Figure 6 The calculation is based on combining the previous belt increment, the current belt increment, and the loss term into the same balance equation to determine the closure deviation on the conveyor side. The calculation relationship is displayed mathematically in the expression block below. The current output covers 240 sampling windows, and the overall trend remains stable, with only minor fluctuations in certain windows; the main indicator ranges from -21.4414 to 15.6766, with a mean of -0.3417. In this embodiment, the cross-warehouse conservation verification process also includes a data segment processing mechanism that does not meet the segmentation conditions. Data segments that do not meet the segmentation conditions include segments lacking the switching status of the unloading trolley, segments lacking the feeding status of the belt conveyor, segments lacking the consumption segment indicating the weight decrease boundary at the tail end of the raw material silo, segments lacking the feeding segment indicating the rising radar level boundary of the target silo, and segments where the cross-warehouse conservation verification values ​​cannot form a closed loop. The cross-warehouse conservation verification process writes the above segments into a transitional pending confirmation queue and records their corresponding silo switching events, raw material silo identifiers, target silo identifiers, missing boundary types, and other relevant information in the semantic state stream. If a new event window fills in the missing boundary during the cumulative consumption update, the segment can re-enter the cross-warehouse conservation verification process. If the missing boundary cannot be filled in, it is retained as a transitional segment awaiting confirmation. In this embodiment, the cross-warehouse conservation verification process does not modify the original radar level gauge data and weight sensor data. Instead, it controls the impact on effective inventory and cumulative consumption by segmentation, deferred participation, and attribution marking. The advantage of this embodiment is that the data during the warehouse switching period will not cause continuous consumption statistics errors due to a one-time attribution judgment, and each segment that does not participate in the cumulative update has a traceable event boundary cause.

[0050] In a preferred embodiment, the fusion measurement processing section further includes a semantic state publishing process. This process takes the material level, effective inventory, instantaneous consumption rate, cumulative consumption, stage label, sensor reliability, anomaly candidate queue, and warehouse location attribution results processed by the cross-warehouse conservation verification process, and generates a semantic state flow for the entire blast furnace trough feeding process based on a semantic skeleton of equipment, materials, workstations, operating conditions, events, and rules. Equipment semantics include the states of radar level gauges, weight sensors, belt conveyors, gates, feeders, and unloading trolleys; material semantics include material category and material flow direction; and workstation semantics... The semantics include the material silo, conveying path, and furnace top receiving position; the working condition semantics include material replenishment, material discharge, settling, silo cutting, and coupling stages; the event semantics include material demand, material respite, silo switching, and equipment action boundaries; and the rule semantics include material conservation relationships, silo attribution relationships, event boundary chain integrity, and mapping sample admission conditions. The semantic state release process writes the above semantic state stream into the knowledge center of intelligent perception as the input state for automatic control of material demand, material respite, and silo switching. The advantage of this embodiment is that the fused measurement results not only include numerical values, but also the corresponding stage, credibility, attribution relationship, and rule verification status.

[0051] In this embodiment, the semantic state publishing process uses a unified record format to output state streams. Each state stream record includes at least the state time, the associated silo, the associated material path, the material level value, the effective inventory, the instantaneous consumption rate, the cumulative consumption, the stage label, the radar material level reliability, the weight reliability, the equipment event reliability, the anomaly identifier, the silo ownership identifier, and the rule verification identifier. The state stream records undergo a consistency check before being written into the intelligent sensing knowledge center. If two conflicting stage labels exist in the same silo at the same time, the state record with a complete event boundary chain is retained, and the other record is written to the conflict record area. If the silo ownership result is consistent with the unloading... If the vehicle's positioning status is inconsistent, the corresponding record will be marked as a warehouse location conflict. If the cumulative consumption update depends on a data segment that has not passed the cross-warehouse conservation check, the semantic state publishing process will prevent the record from entering the automatic control input state. The knowledge hub can establish query indexes and constraint relationships according to the semantic skeleton of equipment, materials, workstations, working conditions, events, and rules, so that the status read by the automatic control of material demand, material delay, and warehouse switching has computable source information. The advantage of this embodiment is that the input status used by the material supply control is processed by measurement fusion, stage identification, conservation check, and semantic consistency, which can reduce the risk of control actions driven by erroneous status.

[0052]

[0053] in, Indicates the first Each moment corresponds to the instantaneous consumption rate within the event window. This indicates the effective inventory at the previous moment. This indicates the current effective inventory. This indicates the quantity of incoming materials in the current event window. This represents the time interval between the current moment and the previous moment. Fractional operations represent the net decrease in inventory per unit time, adjusted for the amount of goods received, and the outbound rate. For example, the effective inventory at the previous moment... Current effective inventory The current window contains the quantity of materials received into the warehouse. If the time interval is 5 minutes, then This indicates that the instantaneous consumption rate within the event window, after inbound correction, is 3 mass units per minute. If the current window is in a pure replenishment phase and there are no outbound path events, this calculation result will not be included in the consumption sequence as a valid outbound consumption rate. (Reference) Figure 7 , Figure 7 The horizontal axis represents time, showing the instantaneous consumption rate change according to the state time. The vertical axis represents t / h, the mass consumed per unit time, in tons per hour. Figure 7The solid line curve depicts the time-series change sequence of the main indicator in the historical sampling window, while the gray dashed or dotted lines represent auxiliary sequences, threshold lines, or closed reference lines. The star-shaped marker corresponds to the current sampling window. Figure 7 The calculation is based on the consumption rate calculated using the difference between adjacent inventory levels, the amount of material fed in, and the time interval, and is used in conjunction with the reliability to determine whether the current output is stable. The current output covers 240 sampling windows, and the overall trend remains stable, with only slight fluctuations in some local windows; the main indicator ranges from -384.2894 to 3469.7293, with a mean of 1108.2166.

[0054] Furthermore, the semantic state release process also performs controlled updates to the cumulative consumption. The cumulative consumption is only updated by the data segments that have passed stage identification, confidence allocation, and conservation verification, including the discharge segment, the consumption segment at the end of the cutting silo, and the data segments written into the raw material silo consumption sequence after cross-silo segmentation. The replenishment segment, the static anomaly candidate segment, the action non-response window, and the transitional unconfirmed segment that does not meet the segmentation conditions do not participate in the cumulative consumption update. If a discharge segment has a complete gate state, feeder state, and weight drop boundary, but the radar level curve shows discontinuous changes due to isolated echoes, the fusion measurement processing part uses weight as the basis for updates. Changes and equipment events serve as the basis for consumption estimation. At the same time, the reliability of radar material level is reduced and written into the semantic state stream. If a certain static segment experiences a weight jump but there is no corresponding equipment event boundary, the cumulative consumption remains unchanged and the segment is written into the anomaly candidate queue. If a certain silo transition segment is divided into a raw material silo consumption part and a target silo storage part after cross-silo conservation verification, only the raw material silo consumption part is included in the cumulative consumption. The advantage of this embodiment is that the update path of the cumulative consumption is constrained by stage labels and rule semantics, which can avoid replenishment, static mutations and transition segments awaiting confirmation from being mistakenly included in consumption.

[0055] In this embodiment, the overall working process can be summarized as follows: continuously receiving multi-source state data, uniformly generating a material state sequence, identifying stage labels based on event boundary chains, allocating sensor credibility according to stages, jointly estimating material level and consumption using conservation relationships, maintaining the material level-quality relationship using a mapping update process, handling abnormal candidates using a credibility backoff process, processing warehouse switching data using warehouse transition identification and cross-warehouse conservation verification, and outputting computable states using a semantic state publishing process. There is a data closed loop between these processing flows. The material state sequence provides input for stage identification, stage labels provide conditions for credibility allocation and mapping sample admission, and the fused measurement results are the conservation residual and instantaneous consumption. Speed ​​provides the computational foundation, while the anomaly candidate queue and transition confirmation queue provide credibility and attribution information for the semantic state stream. The semantic state stream then serves as the input state for the automatic control of material demand, material respite, and bin switching. The closed loop does not change the existing structure of the conveyor belt, bin, gate, feeder, unloading trolley, and measuring equipment in the blast furnace trough feeding system. Instead, it completes the measurement of material level and consumption through data processing, state recognition, fusion estimation, and rule constraints. The advantage of this embodiment is that the system unifies radar material level, weight weighing, and equipment events into the same measurement logic without adding unrelated hardware structures, so that the material level, inventory, and consumption data under complex feeding conditions have a consistent calculation basis.

[0056] In a preferred embodiment, the system can replay the historical semantic state stream in the knowledge hub of intelligent sensing. The replay process does not change the real-time control input, but is only used to verify the material level quality mapping relationship, the credibility backoff rule, and the silo transition segmentation result. Specifically, the knowledge hub retrieves data segments of the same silo, the same material category, and the same stage label according to the semantic skeleton of equipment, material, workstation, working condition, event, and rule. The segments that have passed the conservation verification are used as reference segments, and the segments in the anomaly candidate queue and the transition confirmation queue are used as segments to be verified. By comparing the event boundary chain integrity, conservation residual, and silo location attribution result, the historical segments are re-marked. If the segment to be verified fills the missing event boundary in the subsequent data and forms a conservation closure, it can enter the delayed confirmation state. If it still cannot form event support, it remains in the anomaly candidate state. The marking result after the replay process can be used to update the material level quality mapping sample pool and stage identification rule, but does not directly cover the original observation record. The advantage of this embodiment is that the system can use the subsequently formed complete event chain to correct the state of the early confirmation segments, so that the measurement data processing has time continuity and traceability.

[0057] In this embodiment, the playback processing can also be used to verify the measurement responses corresponding to material demand events and material release events. After a material demand event occurs, the state acquisition and processing part searches the semantic state stream for the target silo, belt running status, unloading trolley positioning status, radar level rise boundary, and weight increase boundary. If the above states form a complete chain within the same event window or adjacent event windows, the material replenishment segment corresponding to the material demand event can be used as a mapping update candidate sample. If a material demand event exists but the level and weight do not change accordingly, the event window enters the action no-response queue. After a material release event occurs, the state acquisition and processing part searches for the gate status, feeder status, weight drop slope change, and level drop trend change. If the material release event does not match the discharge change, the corresponding segment enters the abnormal candidate queue. The above playback processing only uses the already generated semantic state stream and material state sequence, without introducing new sensor objects or additional hardware structures. The advantage of this embodiment is that a posterior verification relationship can be formed between the control event and the measurement response, thereby providing a more stable data foundation for real-time stage identification and confidence allocation.

[0058]

[0059] in, Indicates as of the date The cumulative consumption of each confirmed event window. Indicates the first Within each event window, the confirmed outbound consumption is determined through stage identification, confidence allocation, and conservation verification. For example, if only the 1st, 3rd, and 5th windows out of the first 5 event windows are confirmed as outbound segments, their confirmed outbound consumption amounts are 12, 15, and 10 respectively. If the 2nd window is a replenishment segment and the 4th window is a transitional segment awaiting confirmation, then... Unconfirmed transition segments and abnormal candidate segments are not included in the cumulative consumption summation. If, after cross-warehouse conservation verification, the fourth window is confirmed to have four mass units belonging to the tail end consumption of the raw material warehouse, then this confirmed portion will be included in the recalculation. Write the corresponding consumption sequence.

[0060] Furthermore, when handling continuous material supply scenarios involving multiple silos, the system maintains an independent material state sequence, stage label, reliability vector, material level quality mapping relationship, anomaly candidate queue, and cumulative consumption sequence for each silo. Simultaneously, it establishes the association boundaries between silos through the unloading trolley positioning status, belt operation status, and silo switching events. If the same belt continuously serves multiple silos within a short period, the state acquisition and processing unit splits the belt operation status into event windows for different silos based on the unloading trolley positioning status and silo switching events. If the actions of the same feeder or gate are related to multiple silos... The linking and stage identification processing part determines the material discharge path affiliation based on the mapping relationship between the equipment object and the silo object. If there is a silo switching transition between multiple silos, the cross-silo conservation verification process generates consumption sequence and inventory sequence according to the raw material silo and the target silo respectively. In this embodiment, the equipment events shared between silos are not repeatedly counted in multiple silos, but are allocated through the event boundary chain and silo location affiliation result. The advantage of this embodiment is that the shared equipment status in the multi-silo scenario can be decomposed into measurement constraints with silo affiliation, avoiding the same material infeed event or material discharge event being repeatedly used for multiple inventory calculations.

[0061] In this embodiment, the system can also classify and output the abnormal candidate queue. The classification output includes radar echo abnormal candidates, weighing impact abnormal candidates, event missed candidates, warehouse location conflict candidates, and unconfirmed warehouse transition candidates. Radar echo abnormal candidates are triggered by isolated sudden changes in radar material level during the static stage and no corresponding change in weight. Weighing impact abnormal candidates are triggered by a short-term spike in the weight curve and no corresponding change in radar material level. Event missed candidates are triggered by the material level and weight change directions being consistent but lacking equipment event boundaries. Warehouse location conflict candidates are triggered by the inconsistency between the unloading trolley positioning status and the target warehouse of the inventory change. Unconfirmed warehouse transition candidates are triggered by the cross-warehouse conservation check value not being closed or the segmentation condition not being met. The classification output does not change the material level and consumption calculation results, but is written into the knowledge center simultaneously with the semantic state stream, so that the automatic control system can identify whether there are unconfirmed fragments in the state source when reading the material demand, material delay, and warehouse location switching input states. The advantage of this embodiment is that the measurement results, abnormal attribution, and control input states are in the same semantic record, which facilitates the subsequent control logic to read according to the state credibility.

[0062] In a preferred embodiment, the system can cache and restore the material state sequence at the edge. The edge-side cache stores data segments that have not yet completed stage identification or conservation verification according to the bin identifier, event window identifier, and state time sequence. If data transmission is interrupted or some systems are temporarily without data, the state acquisition and processing part will keep the received radar level gauge data, weight sensor data, and equipment event data in the original event window and will not forcibly output inventory changes for incomplete windows. After data recovery, the time scale tuning process continues to complete the event window according to the composite alignment anchor point. The stage identification processing part reconstructs the event boundary chain within the corresponding time range. The fusion measurement processing part performs inventory and consumption calculations based on the completed material state sequence. If the recovered data still lacks key boundaries, the corresponding segment enters the anomaly candidate queue or the transitional pending confirmation queue. In this embodiment, the breakpoint recovery processing does not rely on supplementing manual reports as the measurement basis, but continues to use a unified time scale, event boundary chain, and material conservation relationship to complete data merging. The advantage of this embodiment is that short-term data loss will not immediately destroy the continuity of the measurement sequence, and the recovered data can still enter the inventory and consumption calculation through the same fusion logic.

[0063] In this embodiment, the system can also set reading conditions for the input status output to the automatic control system. The reading conditions include clear stage labels, clear warehouse location results, cumulative consumption not depending on unconfirmed segments, current material level or inventory status with sensor reliability records, and the completion of calculation of the conservation residual. When the reading conditions are met, the semantic status release process marks the corresponding status as an executable input status. When the reading conditions are not met, the status is marked as a pending confirmation input status. The automatic control system can read the executable input status for control judgment of material demand, material delay, and warehouse location switching. The pending confirmation input status is still stored in the knowledge center for display, traceability, or subsequent verification. This embodiment isolates the measurement data processing results from the automatic control input through reading conditions, avoiding abnormal candidate segments, incomplete warehouse segmentation segments, and unsupported mutation segments from directly entering the control input. The advantage of this embodiment is that the status provided by the material level and consumption measurement system can be distinguished according to executability, and there is a clear correspondence between the control input and the measurement reliability.

[0064] Furthermore, the system can divide the outputs of material level, effective inventory, instantaneous consumption rate, and cumulative consumption into real-time observation output and confirmation measurement output. The real-time observation output is formed by radar level gauge data, weight sensor data, and equipment event data within the current event window after preliminary time-scale adjustment. It is used to display the current silo space occupancy status, material quality status, and event status. The confirmation measurement output is formed by data after completing stage identification, credibility allocation, conservation verification, and semantic status publication. It is used to enter the knowledge center and automatic control input status. The real-time observation output can be marked as pending confirmation. The confirmation measurement output must have stage tags, silo attribution, and credibility records. In this embodiment, the real-time observation output and the confirmation measurement output share the same material status sequence and do not generate independent data links. When the real-time observation output changes, the confirmation measurement output is updated after the event window is closed. If the real-time observation output has a sudden change but fails to pass the event boundary and conservation verification, the confirmation measurement output maintains the previous valid state and records the abnormal candidate. The advantage of this embodiment is that the system can take into account both real-time status presentation and constrained measurement output, avoiding the direct use of unconfirmed instantaneous readings as the basis for automatic control.

[0065] In this embodiment, each process of the integrated measurement and processing section can be implemented by a computer program. When the program is executed in the processor, it reads the material state sequence, calls the stage identification rules, confidence allocation rules, material level quality mapping relationship, conservation verification rules, and semantic state release rules. The program data structure includes an event window table, a state tuple table, a stage label table, a confidence vector table, a mapping sample table, an anomaly candidate table, a transition confirmation table, and a semantic state flow table. The event window table stores start and end boundaries and anchor point information. The state tuple table stores radar level gauge data, weight sensor data, and equipment event data. The stage label table stores replenishment, discharge, settling, silo cutting, and coupling labels. The confidence vector table stores the participation ratio of various types of data. The mapping sample table stores material level quality samples that meet the admission conditions. The anomaly candidate table stores data segments that do not participate in the current inventory update. The transition confirmation table stores data segments that cannot be directly assigned during silo cutting. The semantic state flow table stores the final output state record. The advantage of this embodiment is that the data structure implemented by the software corresponds to the measurement logic, which is convenient for deployment, maintenance, and traceability in the blast furnace trough feeding system.

[0066] In a preferred embodiment, when generating a material status sequence, the system can also perform verification access on report data within the same event window. The report data is not used as a direct basis for calculating real-time inventory and instantaneous consumption rate, but rather as a periodic consistency verification object. The status acquisition and processing part converts the raw material consumption information, tank storage information, and upstream and downstream system records in the report into report boundary states, and compares them with the cumulative consumption, effective inventory, and warehouse location results in the semantic state stream. If the report data is consistent with the semantic state stream after conservation verification, the measurement sequence within the corresponding period remains in a confirmed state. If the report data is inconsistent with the semantic state stream, the system does not directly replace the real-time measurement results with the report, but instead writes the event window corresponding to the inconsistent time period into the abnormal candidate queue and performs replay processing. In this embodiment, the access of report data still follows the semantic skeleton of equipment, materials, workstations, working conditions, events, and rules, avoiding incorrect overwriting between report fields and real-time measurement points due to naming inconsistencies. The advantage of this embodiment is that cross-system data can be used as consistency maintenance input without disrupting the main measurement chain formed by sensor fusion measurement and material conservation calculation.

[0067] In this embodiment, when the system completes a full measurement cycle, the state acquisition and processing section continuously acquires input data and generates a material state sequence. The stage identification and processing section determines whether the current segment belongs to the replenishment stage, discharge stage, static stage, silo cutting stage, or coupling stage based on the event boundary chain. The fusion measurement and processing section assigns the credibility of radar level gauge data, weight sensor data, and equipment event data according to the stage label, updates the effective inventory according to the material conservation relationship, and calculates the instantaneous consumption rate. The material level quality mapping update process maintains the mapping relationship when candidate samples meet the admission conditions. The credibility rollback process removes data that does not meet the event support and conservation conditions. Data segments with constant relationships are transferred to the anomaly candidate queue. The silo switching transition identification process and the cross-silo conservation verification process handle the tail consumption, initial feeding, and transitional unconfirmed segments during silo switching. The semantic state publishing process outputs a structured semantic state stream and writes it into the knowledge center of intelligent perception. The complete measurement loop is repeatedly executed after each event window is closed, and the relevant windows are replayed during data recovery, silo switching confirmation, or delayed confirmation of anomaly segments. The advantage of this embodiment is that the measurement loop uses the event window as the basic processing unit, which can cover the continuous operation, intermittent unloading, slowing down, silo switching, and anomaly observation scenarios of the entire process of feeding materials on the blast furnace trough.

Claims

1. A material level and consumption measurement system based on sensor fusion, characterized in that, It includes a status acquisition and processing section for receiving radar level gauge data, weight sensor data, belt running status, gate status, feeder status, unloading trolley positioning status, material demand events, material delay events, and bin switching events throughout the entire process of feeding materials on the blast furnace bin; The status acquisition and processing section generates a material status sequence according to a unified time scale; The material state sequence input stage identification and processing part divides the feeding process into the feeding stage, the discharging stage, the static stage and the silo cutting stage according to the equipment event boundary; The fusion measurement and processing section calls different sensor confidence combinations according to the stage type, and performs joint estimation of material level, effective inventory, instantaneous consumption rate and cumulative consumption based on the conservation relationship between the amount of material entering the warehouse, the amount of material in the warehouse and the amount of material consumed when leaving the warehouse.

2. The material level and consumption measurement system based on sensor fusion according to claim 1, characterized in that, The status acquisition and processing section includes a time scale tuning process, which uses equipment action edges, weight change inflection points, radar material level change inflection points, and silo switching events as composite alignment anchor points to establish event windows for data from different sampling periods. Within the same event window, radar level gauge data is converted into silo space occupancy status, weight sensor data is converted into material quality status, and belt running status, gate status, feeder status, and unloading trolley positioning status are converted into material flow direction constraint status. The material flow direction constraint status, along with material demand events and material delay events, are written into the material status sequence.

3. The material level and consumption measurement system based on sensor fusion according to claim 2, characterized in that, The stage identification and processing section constructs an event boundary chain based on the material state sequence. The event boundary chain includes at least the unloading trolley arrival boundary, the belt allowable operation boundary, the gate or feeder action boundary, the weight change boundary, the radar material level change boundary, and the bin switching boundary. For data segments between adjacent event boundaries, stage labels are generated according to the feeding direction, discharging direction, action duration, and silo affiliation. Data segments that simultaneously exhibit both feeding and discharging states are marked as coupled stages, allowing the fusion measurement and processing section to use a confidence combination independent of the static stage.

4. The material level and consumption measurement system based on sensor fusion according to claim 1, characterized in that, The fusion measurement and processing part includes a dynamic reliability allocation process. In the material replenishment stage, the target silo is defined by the positioning status of the unloading trolley and the belt running status. In the material discharge stage, the discharge path is defined by the gate status and the feeder status. In the static stage, the lack of effective equipment action events is used as the state maintenance condition. In the silo cutting stage, the silo ownership relationship before and after the switch is used as the constraint condition. At different stages, credibility levels are assigned to radar level gauge data, weight sensor data, and equipment event data, and it is prohibited to directly use mutation data without event support as the basis for inventory updates.

5. The material level and consumption measurement system based on sensor fusion according to claim 1, characterized in that, The fusion measurement and processing part also includes a material level quality mapping update process. The material level quality mapping update process extracts radar material level change, weight change, material category, silo identification and working condition from the replenishment segment, discharge segment and static segment confirmed by the stage tag, and generates material level quality mapping relationship corresponding to different silos and different materials. Before updating the mapping relationship, data segments that do not satisfy the material conservation relationship, storage location relationship, or equipment event boundary relationship are removed, and the retained data segments are used as update samples for subsequent material location and inventory conversion.

6. The material level and consumption measurement system based on sensor fusion according to claim 4, characterized in that, The dynamic credibility allocation process also includes a credibility rollback process, which compares the radar level change direction, weight change direction, and material flow direction constraint status within the same event window. When the direction of radar level change is inconsistent with the direction of weight change and there is no corresponding equipment event boundary, the corresponding data segment will be transferred to the anomaly candidate queue. When the weight change exhibits impact characteristics and the radar level change does not meet the response conditions corresponding to the stage label, the equipment event data is retained as a basis for subsequent attribution.

7. The material level and consumption measurement system based on sensor fusion according to claim 5, characterized in that, The material level quality mapping update process also includes a segmented mapping maintenance process. The segmented mapping maintenance process divides the material level quality mapping relationship into low-level segment, middle-level segment and high-level segment according to the silo space occupancy status, and maintains candidate mapping samples corresponding to the replenishment stage, discharge stage and static stage in each segment. Before entering the formal mapping relationship, the candidate mapping sample must simultaneously meet the following conditions: complete event boundary chain, unique silo ownership, closed loop between incoming material quantity and outgoing material consumption quantity, and no isolated echo abrupt change in radar level curve.

8. The material level and consumption measurement system based on sensor fusion according to claim 1, characterized in that, The stage identification and processing part also includes a silo switching transition identification process, which establishes a raw material silo tail-end status window and a target silo start-up status window after detecting a silo switching event. Within the tail section status window of the raw material silo, the tail material consumption is defined by the feeder status, gate status, and weight drop boundary. Within the starting status window of the target silo, the feeding is defined by the unloading trolley positioning status, belt running status, and radar level rise boundary. The overlapping data segments between the two status windows are marked as transitional segments awaiting confirmation.

9. The material level and consumption measurement system based on sensor fusion according to claim 8, characterized in that, The transitional unconfirmed segment is processed through a cross-warehouse conservation verification process, which combines the weight decrease of the raw material warehouse tail state window, the weight increase of the target material warehouse initial state window, the belt conveyor status, and the unloading trolley positioning status into a cross-warehouse constraint group. When there is a conflict in warehouse location ownership within the cross-warehouse constraint group, the transitional unconfirmed segment is segmented according to the order of the event boundary chain and written into the raw material warehouse consumption sequence and the target material warehouse inventory sequence respectively. Data segments that do not meet the segmentation conditions do not participate in the cumulative consumption update.

10. The material level and consumption measurement system based on sensor fusion according to claim 1, characterized in that, The fusion measurement and processing part also includes a semantic state release process. The semantic state release process takes the material level, effective inventory, instantaneous consumption rate, cumulative consumption, stage label, sensor credibility, abnormal candidate queue and warehouse location result after the cross-warehouse conservation verification process, and generates a semantic state stream of the entire process of feeding material on the blast furnace trough according to the semantic skeleton of equipment, material, work station, working condition, event and rule. The semantic state stream is written into the knowledge center of intelligent perception as the input state for automatic control of material demand, material delay and warehouse location switching.

Images