Inventory real-time checking method and device for asrs full-automatic material storage system

Abstract

The application provides a kind of ASRS full-automatic material storage system inventory real-time inventory method, comprising: reading material label identification code and signal strength by fixed RFID antenna array, outputting RFID sensing data set by de-duplication coarse positioning, collecting real-time weight value by weight sensor, detecting storage location state change by sliding filter, and outputting weight sensing data set;After time alignment and space matching of two data sets, determine the occupancy state of storage location according to multi-source fusion rule and give confidence, update inventory real-time model by three-dimensional state matrix increment, and maintain material track tracking graph, compare with warehouse management system account total, output difference record, execute root cause reasoning based on material track and operation log, output root cause category and disposal suggestion;Trigger stacker-mounted RFID reader to directional inventory coverage update inventory model in idle period for low confidence difference or coverage blind area, trigger disposal process to closed loop according to difference level, periodically generate health degree evaluation and difference mode analysis report.

Description

Technical Field

[0001] This application relates to the fields of automated warehouse management and Internet of Things sensing technology, specifically to a method and apparatus for real-time inventory counting in an ASRS fully automated material storage system. Background Technology

[0002] The accuracy of inventory data in an automated storage and retrieval system (ASRS) is directly related to the supply of production materials and the efficiency of warehousing operations. Current ASRS inventory counts mainly adopt periodic full counts or sampling counts, with the count cycle on a weekly or monthly basis. During this period, there is a significant time difference between the inventory ledger data and the actual material status. When the discrepancy between the ledger and the actual material status accumulates to the point that it is discovered, it often leads to production line shutdowns due to material shortages or excessive inventory backlogs.

[0003] Current inventory counting methods largely rely on stacker cranes to retrieve materials one by one for manual verification or barcode scanning. A single full inventory count takes hours, and the warehouse must halt normal storage and retrieval operations during the count, sacrificing production capacity for updated inventory data. In terms of data maintenance, inventory data heavily depends on a single data source from the warehouse management system, lacking an independent physical sensing channel from the warehouse floor to cross-validate the ledger data. Discrepancies between the recorded and actual inventory data caused by data entry errors, system malfunctions, or non-standard operations accumulate and are difficult to detect in a timely manner without external verification.

[0004] In the handling of discrepancies after they are discovered, existing methods rely on manual verification of each location to find the root cause of the discrepancy. The average time taken to locate a single discrepancy is too long, and there is a lack of systematic root cause reasoning ability and intelligent inventory strategy scheduling, resulting in a disproportion between the investment of inventory resources and the improvement of inventory reliability. Summary of the Invention

[0005] To address the issues of existing ASRS inventory counting relying on downtime operations and lacking real-time cross-validation and intelligent discrepancy analysis, this application provides a real-time inventory counting method and apparatus for an ASRS fully automated material storage system. Through real-time, downtime-free sensing via a combination of fixed RFID arrays and location weight sensors, multi-source fusion status inference and confidence assessment, intelligent root cause reasoning assisted by material trajectory tracking, and dynamic inventory resource scheduling driven by risk priority, it supports real-time updates of inventory data and continuous assurance of consistency between records and physical inventory.

[0006] To solve at least one of the above problems, this application provides the following technical solution: Firstly, this application provides a real-time inventory counting method for an ASRS fully automated material storage system, comprising: reading material tag identification codes and signal strengths at preset intervals using fixed RFID antenna arrays on each shelf layer; outputting an RFID sensing dataset after deduplication and coarse positioning; collecting real-time weight values ​​from location weight sensors; calculating weight changes through sliding filtering to detect changes in location status; and outputting a weight sensing dataset; performing time alignment and spatial nearest neighbor matching between the RFID sensing dataset and the weight sensing dataset; determining the location occupancy status and assigning fusion confidence based on the combination status of tag and weight detection according to multi-source fusion rules; incrementally updating the real-time inventory model with a three-dimensional state matrix and simultaneously maintaining a material trajectory tracking map; and performing a full comparison between the real-time inventory model and the warehouse management system ledger according to location mapping, material identification, and quantity, and outputting difference records, including: extracting the complete location change sequence of the difference material from its entry into the warehouse to the present from the material trajectory tracking map; extracting the relevant entry, exit, and transfer operation records of the material from the operation log; aligning the location change sequence with the operation records along the time axis; and identifying entries that have occurred but no corresponding location changes. The system infers that the entry was not registered; for discrepancies that cannot be explained by operation records, further analysis of historical sensor data is conducted. If RFID and weight data continuously conflict, it is inferred that the tag is faulty; if the weight of a single storage location fluctuates abnormally, it is inferred that the sensor is drifting; after ruling out the aforementioned causes, it is attributed to system data errors. A root cause analysis report containing discrepancy identifiers, recommended root causes, and handling suggestions is output. The real-time inventory model uses a three-dimensional array to represent the storage location status of shelf columns, layers, and depths. Array elements include material identifiers, occupancy status, the fusion confidence level, and update time fields. When the fused status data frame matches the model... When the current records are inconsistent, an incremental update is triggered, which only rewrites the elements of the changed storage location and simultaneously adds the location record to the material trajectory tracking map. For each discrepancy record, root cause reasoning is performed based on the material trajectory tracking map and operation log to output the root cause category and disposal recommendations. For low-confidence discrepancy records or RFID coverage blind spots, the stacker-mounted RFID reader is triggered to perform targeted inventory during idle periods, and the mobile inventory results are overwritten and updated to the real-time inventory model. The disposal process is triggered according to the discrepancy level and tracked to a closed loop. Sensor data is collected periodically to generate health assessment and discrepancy pattern analysis reports.

[0007] Furthermore, it also includes: installing the fixed RFID antenna at a preset interval on the crossbeams of each layer of the shelf and polling the location tags within the coverage area through the radio frequency switch at a preset time slot; after the reader receives the tag response, extracting the tag identification code and the signal strength; taking the maximum signal strength of the repeated records of the same tag in multiple antenna channels and calculating the approximate spatial coordinates of the tag according to the weighted centroid method; the location weight sensor collects the real-time weight value at a preset sampling period; the original weight sequence is filtered by the mean value through a sliding window to suppress vibration noise; the difference between the current weight and the weight of the previous period is calculated to obtain the weight change; when the positive value exceeds a preset threshold condition, it is judged as a weight increase event; when the negative value exceeds a preset empty threshold condition, it is judged as a weight decrease event; when the value is lower than a preset empty threshold condition, it is judged as an empty location.

[0008] Furthermore, it also includes: aligning the RFID sensing dataset and the weight sensing dataset to the same second-level time window based on the system master clock; performing nearest neighbor matching on the coarse spatial coordinates and physical coordinates of the storage location that fall within the same time window according to a preset matching distance as a threshold condition; determining that the storage location is occupied and taking a high value for the fusion confidence score when both sources detect materials; determining that confirmation is pending and taking a median value for the fusion confidence score when only a single source detects materials; the real-time inventory model represents the storage location status of shelf columns, layers, and depths using a three-dimensional array, with array elements containing material identifier, occupancy status, fusion confidence score, and update time fields; triggering an incremental update when the fusion status data frame is inconsistent with the current record in the model, only rewriting the elements of the changed storage location, and simultaneously adding position records to the material trajectory tracking graph.

[0009] Furthermore, it also includes: extracting the complete location change sequence of the discrepancy material from its entry into the warehouse to the present from the material trajectory tracking map; extracting the relevant entry, exit, and transfer operation records of the material from the operation log; aligning the location change sequence with the operation records along the time axis; identifying cases where there is an entry operation but no corresponding location change and inferring that the entry was not registered; further analyzing the historical sensor data for discrepancy records that cannot be explained by the operation records; inferring tag failure when RFID and weight data continuously conflict; inferring sensor drift when the weight of a single storage location continuously fluctuates abnormally; and classifying system data errors after ruling out the aforementioned causes. The root cause analysis report, containing discrepancy identifiers, recommended root causes, and handling suggestions, is then output.

[0010] Furthermore, it also includes: constructing a warehouse location inventory priority scoring model with a weighted sum of time decay factor, material value factor, material risk level factor and historical difference frequency factor as the scoring function; calculating scores for all warehouse locations every hour and sorting them in descending order to generate a priority ranking; automatically selecting the top warehouse locations to generate the mobile inventory task during periods when equipment utilization is below a preset load threshold; after the stacker crane runs to a preset positioning point in front of the target warehouse location, the load platform RFID reader performs close-range directional reading at a preset power and distance and triggers a vision camera to collect warehouse location images to extract appearance features for auxiliary verification; and fusing the RFID reading results of the mobile inventory, the current value of the weight sensor and the visual verification results to determine the actual status of the warehouse location based on a confidence level not lower than a preset threshold.

[0011] Furthermore, it also includes: classifying the discrepancy records into three preset levels according to severity. The first level is when the discrepancy between the book and the actual inventory affects the materials that can be shipped. The relevant storage location is immediately locked and an alarm is pushed to the warehouse management terminal, requiring on-site verification within a preset time limit. The second level is when attribute differences or low confidence levels need to be confirmed and are included in the daily inventory queue, requiring processing within a working day. After manual verification and confirmation of processing, the real-time inventory model and the warehouse management system ledger are updated to ensure consistency between the book and the actual inventory. The discrepancy records that are not processed within the time limit are automatically upgraded, and a daily processing report is generated to statistically analyze the total number of discrepancies, the distribution of each type, and the closure rate. On a weekly basis, discrepancy pattern clustering and sensor health trend analysis are performed based on the accumulated discrepancy records.

[0012] Secondly, this application provides an ASRS fully automated material storage system real-time inventory counting device, comprising: a multi-source sensing module, used to read material tag identification codes and signal strengths at preset intervals through fixed RFID antenna arrays on each shelf layer and output RFID sensing datasets; and a weight sensor at the storage location to collect real-time weight values ​​and detect storage location status change events to output weight sensing datasets; a data fusion and inventory modeling module, used to read the RFID sensing dataset and the weight sensing dataset, perform time alignment and spatial matching, determine the storage location status according to multi-source fusion rules and assign fusion confidence, incrementally update the real-time inventory model with a three-dimensional state matrix and simultaneously maintain the material trajectory tracking map; and a ledger-to-physical comparison and difference analysis module, used to perform a full comparison between the real-time inventory model and the warehouse management system ledger to detect difference records, including: extracting the complete location change sequence of the material from entry to the present from the material trajectory tracking map; extracting the relevant entry, exit, and transfer operation records of the material from the operation log; aligning the location change sequence with the operation records along the time axis; identifying when there is an entry operation but no corresponding location change and inferring that the entry was not registered; and for cases that cannot be explained by the operation records... The discrepancy record is further analyzed using historical sensor data. When RFID and weight data continuously conflict, it is inferred to be a tag malfunction; when the weight of a single storage location continuously fluctuates abnormally, it is inferred to be sensor drift; after ruling out the aforementioned causes, it is attributed to system data errors. A root cause analysis report containing discrepancy identifiers, recommended root causes, and handling suggestions is output. The real-time inventory model uses a three-dimensional array to represent the storage location status of shelf columns, layers, and depths. Array elements include material identifiers, occupancy status, the fusion confidence level, and update time fields. When the fused status data frame is inconsistent with the current record in the model, an incremental update is triggered, only rewriting the elements of the changed storage location. The system synchronously adds location records to the material trajectory tracking map; for each difference record, it performs root cause reasoning based on the material trajectory tracking map and operation log to output root cause category and handling suggestions; the mobile inventory execution module is used to trigger the stacker crane onboard mobile inventory task for low confidence differences or RFID coverage blind spots, perform targeted inventory during idle periods, and update the real-time inventory model with the three-source fusion results; the difference handling and optimization module is used to trigger corresponding handling processes according to the difference level and track the entire process status from detection to closure, and periodically generate sensor health assessment and difference pattern analysis reports.

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the real-time inventory counting method of the ASRS fully automated material storage system.

[0014] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the real-time inventory counting method of the ASRS fully automated material storage system.

[0015] Fifthly, this application provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the steps of the real-time inventory counting method of the ASRS fully automated material storage system.

[0016] As can be seen from the above technical solution, this application provides an ASRS fully automated material storage system's real-time inventory counting method and apparatus. It continuously senses the inventory status without interrupting normal storage and retrieval operations by real-time fusion of RFID and weight dual-source data. It reduces the false alarm rate of discrepancies by using multi-source cross-validation and confidence quantification. It automatically locates the root cause of discrepancies and generates disposal suggestions by using material trajectory tracking and operation log correlation reasoning. It drives the scheduling of mobile inventory counting resources during equipment idle periods by using a four-factor priority scoring based on time decay, material value, risk level, and historical discrepancies. It takes into account the real-time nature of inventory perception, the accuracy of discrepancy location, and the efficiency of inventory counting resource utilization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a flowchart illustrating the real-time inventory counting method of the ASRS fully automated material storage system in this application embodiment; Figure 2 This is a schematic diagram illustrating the process of RFID antenna array data acquisition and weight sensor data acquisition in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the process of multi-source data fusion and real-time inventory model update in the embodiments of this application; Figure 4 This is a schematic diagram of the differential root cause reasoning process in the embodiments of this application; Figure 5 This is a flowchart illustrating the mobile inventory priority scoring and targeted inventory process in an embodiment of this application. Figure 6 This is a schematic diagram of the differential classification and closed-loop tracking process in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0020] Considering the problems of existing ASRS inventory counting relying on downtime operations and lacking multi-source physical sensing cross-validation, this application provides a real-time inventory counting method for ASRS fully automated material storage systems. Through dual-source collaborative sensing of fixed RFID arrays and location weight sensors, multi-source fusion status inference, material trajectory tracking to assist root cause reasoning, and risk priority-driven mobile inventory counting resource scheduling, it achieves real-time inventory counting without downtime and continuous assurance of consistency between records and actual inventory.

[0021] This application provides an embodiment of a real-time inventory counting method for an ASRS fully automated material storage system, see [link to embodiment]. Figure 1 The method specifically includes the following: Step S101: The material tag identification code and signal strength are read by the fixed RFID antenna array on each layer of the shelf according to the preset cycle. After deduplication and coarse positioning, the RFID sensing dataset is output. The real-time weight value is collected by the weight sensor of the storage location and the weight change is calculated by sliding filter to detect the change of storage location status. The weight sensing dataset is then output.

[0022] like Figure 1 As shown, this step reads material tag data from fixed RFID antenna arrays deployed on each layer of the automated warehouse racking, and simultaneously collects real-time weight values ​​from weight sensors at the bottom of the storage locations, constructing a dual-source sensing data pipeline. The fixed RFID antennas are installed at preset intervals on the front of the beams of each racking layer and poll the reader module via an RF switch matrix at preset time slots. The reader operates in the UHF band, initiates multi-tag inventory commands at preset intervals, receives the tag backscattered signals, decodes them to obtain the tag identification code, and calculates the signal strength value by converting the receiving channel gain value and the received signal power.

[0023] Duplicate read records of the same tag appearing on multiple antenna channels are deduplicated. Grouping and aggregation are performed using the tag identification code as the grouping key. Within each group, records are sorted in descending order of signal strength value, and the record with the highest value and its corresponding antenna number is selected. Other duplicate records are discarded. Records with a maximum signal strength below a preset lower limit are marked as weak signals and are temporarily excluded from coarse localization.

[0024] The deduplicated tag records are then entered into the coarse positioning calculation unit. The weighted centroid method is used to spatially extrapolate the signal strength of multiple antennas. The normalized signal strength of each antenna is used as a weight, and a weighted average is taken of the known installation coordinates of each antenna in the warehouse's global coordinate system to obtain the tag's coarse spatial coordinates. The coarse positioning coordinates include three spatial components: column, layer, and depth.

[0025] When the coarse positioning coordinates exceed the warehouse boundary, an out-of-bounds anomaly marker is registered. Out-of-bounds records are not included in subsequent spatial matching but are transferred to the anomaly handling queue. The coarse positioning accuracy is determined by statistically analyzing the deviation distribution between the coarse positioning coordinates and the actual warehouse location center, using a batch of tag samples with known accurate warehouse locations as a reference. The RFID sensing dataset uses the tag identification code as the primary key and includes fields such as electronic product code, maximum signal strength, corresponding antenna number, coarse positioning coordinates, and inventory timestamp, for reading in step S102.

[0026] The weight sensor for the cargo location uses a resistance strain gauge bridge sensor, installed below the support point of the load-bearing plate at the bottom of the cargo location. Signals from multiple sensors are summed and amplified to output a single analog voltage signal, which is then converted into a digital weight value by an analog-to-digital converter. The sensor continuously samples the original weight sequence at a preset sampling frequency.

[0027] When an ASRS stacker crane stores or retrieves materials, the mechanical impact caused by the insertion and withdrawal of the forks generates large-amplitude but short-duration weight spikes. This type of vibration noise is suppressed by using a sliding window mean filter. A pre-defined circular buffer is maintained; each new sampled value overwrites the oldest value, and the arithmetic mean of the current values ​​in the buffer is used as the filtered output.

[0028] The weight change is obtained by subtracting the filtered weight value from the filtered weight value recorded in the previous complete detection cycle. This change is compared to a preset threshold; a positive exceedance indicates a weight gain event, while a negative exceedance indicates a weight loss event. A filtered weight value below a preset empty load threshold indicates an empty storage location.

[0029] The weight-aware dataset uses the location number as the primary key and includes fields for event type enumeration, current filtered weight value, weight change, and judgment time, which are read by step S102 at the data fusion pipeline entry point. The event type enumeration values ​​include five types: weight increase, weight decrease, no change, empty location, and weight anomaly.

[0030] Step S102: Read the RFID sensing dataset and the weight sensing dataset, perform time alignment and spatial nearest neighbor matching, determine the location occupancy status according to the combination status of tag and weight detection and assign fusion confidence based on multi-source fusion rules, update the real-time inventory model incrementally with the three-dimensional state matrix and maintain the material trajectory tracking map simultaneously, compare the real-time inventory model with the warehouse management system ledger in three dimensions of location mapping, material identification and quantity, and output the difference records, and perform root cause reasoning on each difference record based on the material trajectory tracking map and operation log to output the root cause category and disposal suggestions.

[0031] like Figure 1 As shown, the RFID sensing dataset and weight sensing dataset output in step S101 are read into the data fusion pipeline inlet. The system master clock is a high-precision clock source deployed in the control cabinet, and each subsystem synchronizes with the master clock at a preset frequency. The data fusion engine pulls the latest data from the two types of data caches at fixed intervals and aligns them according to second-level time windows.

[0032] For time-aligned data pairs, perform spatial nearest neighbor matching. Calculate the distance between the coarse positioning coordinates of each RFID tag and the physical coordinates of all warehouse locations, where the physical coordinates are represented by the three-dimensional coordinates of the geometric center point of the location's bottom surface. Select the location with the smallest distance as the nearest neighbor location. If this minimum distance does not exceed a preset matching distance threshold, the match is successful, and the tag is associated with the location.

[0033] The storage location occupancy status is determined according to a preset multi-source fusion rule set. If both sources detect material, the storage location is determined to be occupied, and the fusion confidence level is set to high. If only one source detects material, the location is determined to be pending confirmation, and the fusion confidence level is set to medium. If the two sources detect contradictory information, the location is determined to be abnormal, and the confidence level is set to low. If multiple tags map to the same storage location within the same time window, it is marked as a crosstalk event, and the tag with the highest signal strength is selected as the preferred associated tag. The determination result where the fusion confidence level is lower than the write threshold is transferred to the pending confirmation queue for verification during the moving inventory check in step S103.

[0034] The real-time inventory status model uses a three-dimensional array as its core data structure. The three dimensions of the array correspond to the column number, shelf number, and depth number of the shelf, respectively. The array elements contain fields for material identification, occupancy status, fusion confidence score, and last update time. Upon startup, the model initializes by reading a snapshot saved before the last shutdown from persistent storage.

[0035] Model updates employ an incremental update strategy. When the fusion pipeline outputs a new fusion status data frame, the corresponding element in the 3D array is located based on the cargo location coordinates, and the new frame is compared with the element's current value. If any field changes, a write operation is triggered and the corresponding field is rewritten; otherwise, the process is skipped. Model snapshots are automatically saved to persistent storage at preset intervals.

[0036] The material tracking chart is updated synchronously with the real-time inventory model. The tracking chart uses the material identifier as a unique index, and each material maintains a tracking chain based on its location changes from receipt to the present. When the location coordinates corresponding to a material identifier in the model change, a location record containing the location coordinates and a timestamp is appended to the end of that material's tracking chain.

[0037] The inventory reconciliation is performed at a preset cycle. The reconciliation engine reads the occupancy status and material identification records of all storage locations from the real-time inventory model, queries the corresponding ledger records from the warehouse management system, and compares them one by one according to the storage location mapping dimension, material identification dimension, and quantity dimension. The results of the three reconciliations are output as discrepancy records and written to the discrepancy queue.

[0038] Each discrepancy record in the discrepancy queue is consumed sequentially by the root cause inference engine. The complete location change sequence of the discrepancy material is extracted from the material trajectory tracking map, and the relevant inbound, outbound, and transfer operation records for that material are extracted from the operation log. After being aligned along the timeline, conflict analysis is performed to infer the root cause category. An analysis report containing the discrepancy identifier, recommended root cause, and handling suggestions is output and pushed to the discrepancy handling pipeline for consumption by step S103.

[0039] Step S103: For low-confidence difference records or RFID coverage blind spots, trigger the stacker onboard RFID reader to perform targeted inventory during idle periods, update the real-time inventory model with the mobile inventory results, trigger the handling process according to the difference level and track it to the closed loop, and collect sensor data periodically to generate health assessment and difference pattern analysis reports.

[0040] like Figure 1 As shown, in step S102, difference records with a fusion confidence level lower than the writing threshold and storage locations in RFID fixed antenna coverage blind spots are included in the mobile inventory candidate queue. The mobile inventory is performed collaboratively by an RFID reader / writer mounted on the stacker crane's load platform and a vision camera.

[0041] A priority scoring model for inventory location counting is constructed, using a weighted sum of a time decay factor, a material value factor, a material risk level factor, and a historical difference frequency factor as the scoring function. The time decay factor increases with the time elapsed since the last inventory count, and the material value factor is read from the material unit price table and standardized according to the entire warehouse distribution. The material risk level factor is configured and graded by warehouse management personnel according to the degree of impact on production line stoppage. The historical difference frequency factor is taken as the proportion of historical difference days for that inventory location, and the sum of the four weights is normalized to ensure that the score is within a preset range.

[0042] The scoring calculation is performed once per hour across all warehouse locations. During the iteration, for each location, the time difference since the last inventory count, the value level and risk level of the stored materials, and the historical frequency of discrepancies are queried sequentially. These values ​​are then substituted into the scoring formula to calculate the comprehensive score, which is written into the inventory count scoring results table. The results are then sorted in descending order of score to generate a priority ranking.

[0043] The system checks the current stacker crane equipment utilization rate, which is defined as the ratio of the total estimated execution time of queued tasks in the current work queue to the scheduling cycle. When the equipment utilization rate is lower than the preset load threshold, the system selects the top-ranked storage locations from the sorted list in descending order of scores to generate a mobile inventory task. The proportion of the total stacker crane operation time occupied by the mobile inventory task on a single day does not exceed the preset resource constraint limit.

[0044] After receiving a mobile inventory task, the stacker crane moves from its current position to a preset positioning point directly in front of the target storage location. Upon reaching the positioning point, the RFID reader on the load platform initiates short-range directional reading, continuously reading multiple times in sub-cycles and taking the mode as the final RFID reading result. Simultaneously, a vision camera is triggered to capture an image of the storage location, which is then input into a feature extractor to output an appearance feature vector. This vector is then compared with a preset feature template corresponding to the material type to calculate the similarity score, which serves as the visual verification score.

[0045] The mobile RFID reading results, the current value of the weight sensor, and the visual verification score are fused using a weighted voting rule. If the fusion confidence level is not lower than a preset threshold, the actual state of the storage location is determined to be reliable, and the result is overwritten into the corresponding array element of the real-time inventory status model. If the fusion confidence level is lower than the threshold, the original model remains unchanged, and a mobile inventory count failure event is recorded.

[0046] Discrepancy handling is categorized into three levels based on severity. Level 1 involves discrepancies between records and actual inventory that affect the availability of materials for shipment. The relevant storage locations are immediately locked, and an alarm is sent to the warehouse management terminal, requiring on-site verification within a preset timeframe. Level 2 involves attribute differences or low-confidence discrepancies awaiting confirmation. These are added to the daily inventory task queue, and automatically escalate if the handling timeout is exceeded. A daily handling report is generated, summarizing the total number of discrepancies, the distribution of each type, and the closure rate. Weekly, based on accumulated discrepancy records, discrepancy pattern clustering and sensor health trend analysis are performed, and adjustment suggestions are output.

[0047] In one embodiment of the real-time inventory counting method of the ASRS fully automated material storage system of this application, see [link to relevant documentation]. Figure 2 It can also specifically include the following: Step S201: Install the fixed RFID antenna at a preset interval on the crossbeams of each layer of the shelf and poll the location tags within the coverage area by using the radio frequency switch at a preset time slot. After the reader receives the tag response, it extracts the tag identification code and the signal strength. For the same tag, the maximum signal strength is taken from the repeated records in multiple antenna channels, and the coarse spatial coordinates of the tag are calculated according to the weighted centroid method.

[0048] This embodiment describes the deployment and data acquisition mechanism of the RFID antenna array, such as... Figure 2As shown, fixed RFID antennas are installed layer by layer along the shelf column on the front of the beams at preset intervals, with each antenna covering several adjacent storage locations. The antennas are connected to an RFID switch matrix via RF cables, and the switch matrix polls and switches each antenna to the reader module according to preset time slots.

[0049] The reader module sends a multi-tag inventory command to the corresponding antenna in each time slot. After receiving the tag backscattered signal, the antenna extracts the baseband signal and decodes it to obtain the tag identification code. The signal strength value is calculated by converting the gain value of the receiving channel with the received signal power, and each read record is written to the RFID raw data cache table.

[0050] Duplicate read records of the same tag appearing on multiple antenna channels are grouped by tag identification code. Within each group, records are sorted in descending order of signal strength value, and the record with the highest value and its corresponding antenna number is retained. Other duplicate records are discarded. The antenna corresponding to the highest signal strength value indicates the approximate spatial location range of the tag.

[0051] During the deduplication process, if the maximum signal strength is lower than the preset lower limit, it is marked as a weak signal record. The weak signal tag will not participate in the subsequent coarse positioning calculation. After the subsequent inventory cycle is retried, it will be transferred to the mobile inventory review queue for consumption by the mobile inventory in step S103 when the cumulative number of attempts reaches the preset number.

[0052] The coarse spatial coordinates of the tag are calculated using the weighted centroid method. The signal strength of each antenna is normalized and taken to a preset range. The known installation coordinates of the antennas in the global coordinate system of the warehouse are weighted and averaged using the normalized signal strength as the weight to obtain the coarse spatial coordinates of the tag. The coarse spatial coordinates include three spatial components: column, layer, and depth, with the unit of each component being meters.

[0053] When the coarse positioning coordinates are outside the warehouse boundary, an out-of-bounds anomaly is recorded. Out-of-bounds records are not included in spatial matching but are transferred to the anomaly handling queue for manual verification. The accuracy of the coarse positioning is determined by statistically analyzing the deviation distribution between the coarse positioning coordinates and the actual warehouse location center, using a sample of labels with known accurate warehouse location locations as a reference.

[0054] After the coarse positioning coordinates are written into the RFID sensing dataset, they are used for spatial nearest neighbor matching in step S102. The RFID sensing dataset uses the tag identification code as the primary key and includes fields such as electronic product code, maximum signal strength, corresponding antenna number, coarse positioning coordinates, and inventory timestamp.

[0055] Step S202: The cargo location weight sensor collects the real-time weight value according to a preset sampling period. The original weight sequence is filtered by taking the mean value through a sliding window to suppress vibration noise. The weight change is obtained by calculating the difference between the current weight and the weight of the previous period. When the weight exceeds the preset threshold condition in the positive direction, it is judged as a weight increase event; when it exceeds the threshold condition in the negative direction, it is judged as a weight decrease event; when it is lower than the preset empty cargo location threshold condition, it is judged as an empty cargo location.

[0056] This embodiment elaborates on the mechanism for data acquisition and state change detection of the weight sensor, such as... Figure 2 As shown, the weight sensor is a resistance strain gauge bridge sensor, installed below the support point of the load-bearing plate at the bottom of the loading bay. The signals from multiple sensors are summed and amplified to output a single analog voltage signal, which is then converted into a digital weight value by an analog-to-digital converter.

[0057] The sensor continuously samples the raw weight sequence at a preset sampling frequency. The mechanical impact caused by the insertion and retraction of the forks during material storage and retrieval by the ASRS stacker crane generates large-amplitude but short-duration weight spikes. A sliding window mean filter is used to suppress this type of vibration noise, maintaining a preset length of circular buffer.

[0058] Each time a new sample value is written to the buffer, it overwrites the oldest value. The arithmetic mean of the current values ​​in the buffer is then used as the filter output. During the initial filling phase of the filter, the average of the actual number of samples collected is used, without zero padding. The filter output value is written to the weight filter buffer, which is organized using the location number as the primary key.

[0059] The weight change is obtained by subtracting the filtered weight value from the filtered weight value recorded in the previous complete detection cycle. The change is compared with a preset change threshold. If the change exceeds the threshold positively, it is determined as a weight gain event; if it exceeds the threshold negatively, it is determined as a weight loss event; if it does not exceed the threshold, it is determined as no change.

[0060] When the filtered weight value is lower than the preset empty load threshold, the location is determined to be empty. When the deviation between the filtered weight value and the nominal weight of the material registered in the warehouse management system for that location exceeds a preset ratio, it is marked as a weight anomaly event. Potential causes include misplacement of materials, sensor zero-point drift, or elastomer fatigue.

[0061] When the weight sensor returns to a fixed zero value for multiple consecutive sampling cycles, a zero-point fault is determined and a fault marker is registered. This fault marker is consumed in the periodic health assessment in step S103 of the differential handling process and is used for sensor degradation early warning and maintenance plan formulation.

[0062] Each record in the weight-aware dataset contains an event type enumeration, the current filtered weight value, the weight change, and the judgment time field, which are written to the weight-aware data cache for reading in step S102. The event type enumeration includes five values: weight increase, weight decrease, no change, empty storage space, and weight anomaly.

[0063] In one embodiment of the real-time inventory counting method of the ASRS fully automated material storage system of this application, see [link to relevant documentation]. Figure 3 It can also specifically include the following: Step S301: Using the system master clock as a reference, align the RFID sensing dataset and the weight sensing dataset to the same second-level time window. Perform nearest neighbor matching on the coarse spatial coordinates and physical coordinates of the cargo location that fall into the same time window according to the preset matching distance as the threshold condition. When both sources detect materials, determine that the cargo location is occupied and take the high value of the fusion confidence. When only one source detects, determine that it needs to be confirmed and take the medium value of the fusion confidence.

[0064] This embodiment elaborates on the time alignment and spatial matching mechanisms, such as... Figure 3 As shown. The system master clock is a high-precision clock source deployed in the control cabinet. Each subsystem, including the RFID reader / writer, weight sensor acquisition unit, and data fusion engine, synchronizes with the master clock at a preset frequency, with the synchronization deviation controlled within milliseconds.

[0065] The data fusion engine retrieves the latest batch of data records from the RFID sensing data cache and the weight sensing data cache at fixed intervals. The acquisition time of each record is uniformly converted to the second-level granularity according to the system master clock. Two types of records falling into the same second-level time window index are regarded as data pairs within the same fusion cycle and enter the spatial matching process.

[0066] If there are only one type of data record within a certain time window and no corresponding data of the other type, then the single-source record is marked as a single-source record. Single-source records are processed using the single-source processing path, without entering the dual-source space matching, and are directly written into the pending confirmation status record using the median confidence score of the single-source fusion.

[0067] Spatial matching calculates the distance between the coarse positioning coordinates of each RFID tag and the physical coordinates of all warehouse locations. The physical coordinates of the locations are represented by the three-dimensional coordinates of the geometric center point of the location's bottom surface. These coordinates are obtained through total station calibration during warehouse construction and stored in a static configuration table. The distance is calculated using a three-dimensional Euclidean distance formula.

[0068] For the closest storage location, check if the distance does not exceed a preset matching distance threshold. This threshold is determined based on the statistical distribution of coarse positioning accuracy, using a preset percentile of the deviation distribution as the threshold value. If the threshold is not exceeded, the match is successful and the RFID tag is associated with the storage location; if the threshold is exceeded, the tag is marked as spatial mismatch and moved to the confirmation queue.

[0069] When multiple tags' coarse location coordinates are mapped to the same storage location within the same time window, it is marked as a crosstalk event. The tag with the highest signal strength value among the multiple tags is selected as the preferred associated tag for that storage location, and the remaining tags are marked as crosstalk candidates. Crosstalk candidate tags are not currently involved in determining the occupancy status of that storage location.

[0070] The multi-source fusion state inference rule set is defined based on a combination of two detection quantities: tag detection and weight detection. When both sources detect material, the location is determined to be occupied, and the fusion confidence score is set to a high value. When only one source detects material, the location is determined to be pending confirmation, and the confidence score is set to a medium value. The fusion state data frame contains the fields of location coordinates, determination state, fusion confidence score, and fusion time, which serve as the input for model update in step S302.

[0071] Step S302: The real-time inventory model uses a three-dimensional array to represent the location status of shelf columns, layers and depths. The array elements contain material identifier, occupancy status, fusion confidence and update time fields. When the fusion status data frame is inconsistent with the current record in the model, an incremental update is triggered, which only rewrites the elements of the changed location and simultaneously adds a location record to the material trajectory tracking graph.

[0072] This embodiment elaborates on the structure and update mechanism of the real-time inventory status model, such as... Figure 3 As shown, the real-time inventory status model uses a three-dimensional array as its core data structure. The three dimensions of the array correspond to the shelf column number, layer number, and depth number. The array elements contain fields such as material identifier, occupancy status, fusion confidence score, and last update time.

[0073] During startup loading, the model initializes its 3D array by reading a snapshot saved before the last shutdown from persistent storage. Snapshots are automatically saved at preset intervals; each time a snapshot is saved, the 3D array is serialized into a structured format, and a snapshot timestamp is appended before being written to persistent storage. Snapshot saving uses an asynchronous write method to reduce the time overhead on the main fusion pipeline process.

[0074] When the fusion pipeline outputs a new fusion status data frame, the model update unit locates the corresponding element in the 3D array based on the location coordinates in the data frame. It compares each field of the new frame with the current value of the element. If any field—material identifier, occupancy status, or fusion confidence—changes, a write operation is triggered, and the corresponding field is rewritten.

[0075] Elements that have not changed are skipped in this round of update operations. This incremental update strategy uses the storage location as the smallest update granularity, only rewriting elements in changed storage locations while keeping the remaining storage location records unchanged, avoiding the computational overhead of a full refresh. The update time field is written to record the update moment for each update operation.

[0076] The material tracking map is updated synchronously with the real-time inventory model. The tracking map uses the material identifier as a unique index, and each material maintains a tracking chain based on its location changes from receipt to the present. When the location coordinates corresponding to a material identifier in the model change, a location node is added to the end of that material's tracking chain.

[0077] Each location node contains the location coordinates, timestamp, and a fusion confidence field for the corresponding time. The trajectory chain is sorted in ascending order by timestamp, supporting fast retrieval by material identifier key and range queries by time range. The trajectory tracking graph is persisted to a graph database to support trajectory queries across sessions.

[0078] The inventory reconciliation is performed by the comparison engine at preset intervals. The comparison engine reads the occupancy status and material identification records of all storage locations from the real-time inventory model, queries the corresponding ledger records from the warehouse management system ledger, and compares them one by one according to the storage location mapping dimension, material identification dimension, and quantity dimension. If any dimension is inconsistent, a discrepancy record is output and written to the discrepancy queue for the root cause reasoning engine to consume in sequence.

[0079] In one embodiment of the real-time inventory counting method of the ASRS fully automated material storage system of this application, see [link to relevant documentation]. Figure 4 It can also specifically include the following: Step S401: Extract the complete location change sequence of the differential material from its entry into the warehouse to the present from the material trajectory tracking map, extract the relevant entry, exit and transfer operation records of the material from the operation log, align the location change sequence with the operation records on the time axis, and infer that the entry into the warehouse is not registered when there is an entry operation but no corresponding location change.

[0080] This embodiment elaborates on root cause reasoning based on the correlation between trajectory and logs, such as... Figure 4 As shown. After each difference record in the difference queue is consumed by the root cause reasoning engine, the complete location change sequence of the material from its first entry into the warehouse to the present is extracted from the material trajectory tracking map based on the material identifier in the difference record.

[0081] The location change sequence is a list of nodes arranged in ascending order of timestamps. Each node contains the location coordinates, timestamp, and a fusion confidence field for the corresponding time. Retrieve all inbound, outbound, and transfer operation records related to this material identifier from the warehouse management system's operation logs.

[0082] The operation record includes fields for operation type, operation time, source location, target location, and operator. The operation type enumeration includes three possible values: inbound, outbound, and transfer. After the operation records are sorted in ascending order by operation time, they are aligned and compared with the location change sequence to form two parallel timelines.

[0083] The comparison algorithm uses the timestamp of the operation record as the center and searches for a corresponding node in the location change sequence within a preset time window. The width of the time window is determined based on the statistical upper limit of the sum of the system master clock time synchronization deviation, data acquisition delay, and operation execution time.

[0084] If there are inbound operation records but no new nodes appear in the location change sequence within the corresponding time window, it is inferred that the inbound operation was not registered, meaning the material was actually shelved and put into storage, but the system did not record its location information. The confidence level of this inference depends on the completeness of the trajectory coverage within the time window.

[0085] If a node in the location change sequence appears at a time when there are no operation records within the corresponding time window, and this node is the end node of the trajectory chain, it is inferred that the outbound operation was not registered, meaning the material has been taken out but the system has not recorded the outbound operation. If the location coordinates of the latest node in the location change sequence are inconsistent with the current location coordinates recorded in the ledger, and there is no corresponding transfer record in the operation log, it is inferred that the material was misplaced.

[0086] The three types of root causes derived from the analysis of discrepancies between operation records and trajectories are collectively classified as operation-related root causes. The confidence level of operation-related root causes is positively correlated with the node completeness of the trajectory tracking graph and the recording granularity of the operation log. Discrepancies that cannot be explained by operation-related root causes are transferred to the sensor historical data analysis in step S402.

[0087] Step S402: Further analyze the historical sensor data for the discrepancy records that cannot be explained by the operation log. If the RFID and weight data continue to conflict, it is inferred that the tag is faulty. If the weight of a single storage location continues to fluctuate abnormally, it is inferred that the sensor is drifting. After ruling out the above reasons, it is attributed to system data error. Output the root cause analysis report containing the discrepancy identifier, recommended root cause, and handling suggestions.

[0088] This embodiment elaborates on root cause reasoning based on historical sensor data analysis, such as... Figure 4 As shown. For discrepancies that cannot be explained after the operation record and trajectory contradiction analysis in step S401, further analysis is conducted on the sensor historical data of the discrepancy location and material within the past preset time period.

[0089] If the RFID read records and weight sensor records of the discrepancy material's location show a persistent discrepancy for at least a predetermined inventory period, and the RFID read success rate of the material shows a monotonically decreasing trend in recent times, then it is inferred to be a tag malfunction. The confidence level for tag malfunction is the normalized value of the product of the slope of the decreasing RFID read success rate and the number of periods of discrepancy.

[0090] If the RFID read success rate suddenly drops to an extremely low level instead of decreasing gradually, and the weight sensor continues to detect the presence of material normally, it is inferred that the tag has physically detached from the material surface. This inference differs from tag failure in that the read success rate curve changes in a step-like manner, while the latter shows a gradual decline.

[0091] If the weight sensor data for a single cargo location shows sustained abnormal fluctuations exceeding the normal noise range in a recent period, while the data from other sensors remain normal, it is inferred that the weight sensor is drifting. The drift detection method involves taking the weight filter value sequence for that cargo location over a past period, fitting a trend line using linear regression, and determining whether drift is valid based on a preset significance level.

[0092] Discrepancies that remain unexplained after excluding all the aforementioned categories are classified as system data errors. System data errors include misalignment caused by timestamp time synchronization deviations, data loss due to communication packet loss, and data inconsistencies caused by database write conflicts.

[0093] For each of the aforementioned root causes, the posterior probability of each root cause is calculated based on the completeness of the chain of evidence. The root cause category with the highest posterior probability is selected as the recommended root cause. The root cause analysis report is output in a structured format, including a difference indicator, recommended root cause, evidence summary, and recommended action fields. If the recommendation confidence level is lower than the preset reporting threshold, a manual review suggestion mark is added to the report.

[0094] The root cause analysis report is pushed to the discrepancy handling pipeline. Each root cause category has a corresponding pre-set handling action template for the handling terminal to choose from. The handling action template includes action options such as locking the storage location, pushing an alarm, adding to the inventory queue, and marking the log. The discrepancy classification handling module executes the actions according to the discrepancy level.

[0095] In one embodiment of the real-time inventory counting method of the ASRS fully automated material storage system of this application, see [link to relevant documentation]. Figure 5 It can also specifically include the following: Step S501: Construct a warehouse location priority scoring model. The scoring function is the weighted sum of time decay factor, material value factor, material risk level factor and historical difference frequency factor. Calculate the score of all warehouse locations every hour and sort them in descending order to generate a priority ranking. When the equipment utilization rate is lower than the preset load threshold, automatically select the top warehouse locations to generate the mobile inventory task.

[0096] This embodiment elaborates on the inventory priority scoring model and task scheduling mechanism, such as... Figure 5 As shown, the scoring model calculates the score using a weighted sum of the time decay factor, material value factor, risk level factor, and historical difference frequency factor. The four factors are normalized to a uniform interval and then summed according to preset weights.

[0097] The time decay factor increases monotonically with the time elapsed since the last inventory count, using an exponential decay complement form. The longer the time since the last inventory count, the closer the factor approaches its maximum value. The material value factor is read from the material unit price table in the warehouse management system and standardized to a preset range according to the quantile of each material's unit cost in the overall warehouse distribution.

[0098] Material risk level factors are configured and graded by warehouse staff according to the degree of impact of the material on production line downtime. Low, medium, and high risks are respectively mapped to three quantiles of preset interval values, configured as a static parameter table, and updated and maintained by warehouse staff when new material types are added or production lines are changed.

[0099] The historical discrepancy frequency factor is calculated as the percentage of days with discrepancies recorded for that storage location within a preset number of days. The upper limit is truncated to the maximum value to control the excessive influence of a single factor on the score. When the percentage exceeds the upper limit, the maximum value is used to prevent historically high-frequency discrepancy storage locations from locking up inventory resources for extended periods.

[0100] The scoring calculation performs a full warehouse location traversal once every preset period. During the traversal, for each location, the time difference since the last inventory count, the value level and risk level of the stored materials, and historical difference frequency statistics are sequentially queried, and these are substituted into the scoring formula to calculate the comprehensive score. The calculation results are written into the inventory count scoring results table, and sorted in descending order of score to generate a priority ranking.

[0101] After updating the scoring results table, the system checks the current stacker crane equipment utilization rate. Equipment utilization rate is defined as the ratio of the total estimated execution time of queued tasks in the current work queue to the scheduling cycle. When the equipment utilization rate is lower than the preset load threshold, the system selects storage locations from the sorted list in descending order of scores to generate a moving inventory task.

[0102] The proportion of daily mobile inventory counting tasks occupying the stacker crane's total operating time does not exceed the preset resource constraint limit. The system continuously monitors the cumulative execution time of daily mobile inventory counting tasks against the stacker crane's total runtime. When the cumulative proportion approaches the limit, the generation of new mobile inventory counting tasks stops. After task generation, tasks are written to the mobile inventory counting task queue for automatic retrieval and execution by the stacker crane when idle.

[0103] Step S502: After the stacker crane moves to the preset positioning point in front of the target storage location, the RFID reader on the load platform performs close-range directional reading at a preset power and distance, and triggers the vision camera to collect images of the storage location to extract appearance features for auxiliary verification. The RFID reading results of the mobile inventory, the current value of the weight sensor and the vision verification results are fused together, and the actual status of the storage location is determined according to a confidence level not lower than a preset threshold condition.

[0104] This embodiment elaborates on the sensing operations when the stacker crane performs a moving inventory task, such as... Figure 5 As shown. After receiving a moving inventory task, the stacker crane moves from its current position to a preset positioning point directly in front of the target storage location. The coordinates of the positioning point are pre-calibrated in the warehouse coordinate system, and the horizontal distance between the positioning point and the storage location is controlled within a preset error range.

[0105] Upon reaching the positioning point, the RFID reader on the load platform initiates near-field directional reading. The main lobe beamwidth of the near-field antenna is selected and designed to ensure that the reading area covers the front area of ​​a single cargo location. Multiple readings are performed continuously in sub-cycles, and the mode result is taken as the final RFID reading result.

[0106] If different tag identification codes appear in multiple reads, crosstalk is considered to exist, and the RFID individual confidence level is reduced. The crosstalk confidence level discount factor is determined based on the number and distribution variance of different tags in the read results; the more different tags there are, the greater the confidence level discount.

[0107] The vision camera is synchronously triggered to acquire a color image of the target storage location. The camera uses a global shutter mode to ensure that the image is free of motion blur even under slight vibrations from the stacker crane. The acquired image is fed into a deployed feature extractor to output a feature vector of the storage location's appearance. The cosine similarity is then calculated between this vector and a preset feature template for the corresponding material type in the material information database to serve as the visual verification score.

[0108] The visual camera automatically adjusts exposure parameters and retakes the image when it is too dark or overexposed. If the number of retakes exceeds a preset limit, the last captured image is used for fusion and an illumination anomaly marker is added. The illumination anomaly marker reduces the weighting coefficient of the visual verification score during fusion, and the discount ratio is determined according to the severity of the illumination anomaly.

[0109] The mobile RFID reading result, the current value of the weight sensor, and the visual verification score are fused using a weighted voting rule. The weight of each sensing source is allocated according to its reliability statistics in historical data, and reliability is measured by the consistency ratio between the judgment result of that sensing source and the result of manual review.

[0110] When the fusion confidence level is not lower than the preset fusion threshold, the actual status of the storage location is determined to be a reliable state, and the result is overwritten into the corresponding array element of the real-time inventory status model. When the fusion confidence level is lower than the fusion threshold but not lower than the preset temporary storage threshold, the model is temporarily written and a "to be reviewed" mark is added. When the fusion confidence level is lower than the temporary storage threshold, the original model record remains unchanged, and a moving inventory count failure event is registered.

[0111] In one embodiment of the real-time inventory counting method of the ASRS fully automated material storage system of this application, see [link to relevant documentation]. Figure 6 It can also specifically include the following: Step S601: Difference records are divided into three preset levels according to severity. Discrepancies between accounts and actual inventory that affect deliverable materials are classified as Level 1, and the relevant storage locations are immediately locked and an alarm is pushed to the warehouse management terminal, requiring on-site verification within a preset time limit. Attribute differences or low confidence levels are classified as Level 2 and included in the daily inventory queue, requiring processing within a working day.

[0112] This embodiment elaborates on the differential classification and handling, such as Figure 6 As shown. After root cause reasoning is completed, discrepancy records are automatically assigned a handling level based on severity. When the discrepancy type is location discrepancy or quantity discrepancy and the materials in the discrepancy location are marked as affecting shippable materials in the warehouse management system, it is determined to be Level 1.

[0113] When the difference type is a difference in material identification or attribute, or a low-confidence difference with a fusion confidence level within a preset medium range, it is classified as Level 2. When the difference is a warning difference indicating a decrease in sensor read rate but not yet affecting accuracy, or a transient difference indicating a data delay timeout but eventually arriving, it is classified as Level 3.

[0114] The first level of discrepancy triggers an immediate handling process. The handling actions are, in sequence, locking, alarming, and timing. The locking action involves sending a location lock command to the warehouse management system to prevent subsequent storage and retrieval operations from being scheduled to that location. The alarm action involves sending an alarm notification containing the discrepancy identifier and location coordinates to the warehouse management terminal via a message push channel.

[0115] The first level of discrepancy triggers a preset timeout for verification, requiring warehouse personnel to arrive on-site within the time limit to perform manual verification and confirm or correct the judgment result. The on-site verification method involves warehouse personnel using a handheld terminal to scan the location label and verify the material information against the system judgment result. The verification receipt is then submitted to the discrepancy handling pipeline via the handheld terminal.

[0116] Level 2 discrepancies trigger the same-day handling process, appending the discrepancy record to the end of the daily inventory task queue for targeted review by subsequent mobile inventory tasks during the day's working hours. If a Level 2 discrepancy is not resolved before the end of the day's working hours, it is automatically escalated to Level 1 and handled according to the Level 1 locking and alarm procedures.

[0117] Level 3 difference triggers silent handling, only writing the difference record to the difference log table and adding a mark, without affecting normal access job scheduling. When the cumulative number of Level 3 difference records reaches the preset alarm threshold, a summary notification is triggered to the maintenance terminal, which includes the cumulative number of differences and the distribution of the main difference types.

[0118] The execution and completion of each action are tracked through a difference record status tracking table. The status tracking table uses the difference identifier as the primary key and includes fields for current status, action level, action handler, action time, and action result audit. The status field can take four values: pending action, in progress, closed loop, and timeout escalation.

[0119] Step S602: After manual verification and confirmation that the disposal is completed, update the real-time inventory model and the warehouse management system ledger to ensure consistency between the ledger and the actual inventory. The discrepancy records for disposal that have not been completed within the time limit are automatically upgraded and a disposal report is generated daily to count the total number of discrepancies, the distribution of each type, and the closure rate. On a weekly basis, discrepancy pattern clustering and sensor health trend analysis are performed based on the accumulated discrepancy records.

[0120] This embodiment elaborates on the statistical and periodic analysis of the handling of differences after closed-loop processing, such as... Figure 6 As shown. After manual verification and confirmation of the processing, the corresponding array elements in the real-time inventory model and the corresponding records in the warehouse management system ledger are updated synchronously to ensure consistency between the records and the actual inventory. The update operation first writes to the real-time inventory model and then to the ledger.

[0121] Atomicity is ensured by a transactional guarantee mechanism for both writes. If either write fails, all executed write operations are rolled back, the synchronization failure event is registered, and the failed difference record is added back to the processing queue to await the next round of synchronization. Successfully written difference records are marked as closed-loop and a closed-loop timestamp is written to them.

[0122] A daily variance handling report for the previous calendar day is generated at a preset time each day. The report includes statistics on the total number of variances, distribution by handling level, distribution by root cause type, distribution by storage location, average handling time, and the closure rate. The closure rate is defined as the ratio of the number of closed variances to the total number of variances to be handled that day.

[0123] If a Level 1 discrepancy is triggered and no confirmation receipt is received within a preset time limit, the system will automatically escalate the alarm to the supervisor's terminal and increase the alarm push frequency, repeating the push at shorter intervals until a confirmation receipt is received. For Level 2 discrepancies that are not resolved within the preset timeout threshold, the resolution level will be automatically escalated.

[0124] A weekly sensor health assessment report is generated every Sunday. Using each fixed RFID antenna and each weight sensor as the assessment unit, the effective read rate and its weekly trend slope are calculated over the past week. The trend slope is fitted with linear regression; a sensor degradation warning is generated when the slope is negative and its absolute value exceeds a preset degradation threshold.

[0125] The weight sensor health assessment includes an additional check for zero-point drift. The deviation between the average weight reading during consecutive empty storage locations and the factory zero point is calculated. If the absolute value of the deviation exceeds a preset multiple of the sensor's accuracy class, a zero-point drift alarm is generated and sent to the maintenance terminal.

[0126] The discrepancy pattern analysis retrieves all closed-loop discrepancy records from the past week. These records are then cross-grouped by storage location, material type, root cause category, and occurrence time, and clustering algorithms are used to identify high-frequency discrepancy clustering patterns. The number of clusters is determined based on the internal structure of the data using the elbow rule, and a list of high-frequency discrepancy areas and a list of high-frequency discrepancy materials are output.

[0127] To perform all or part of the method described above, this application provides an embodiment of an apparatus for implementing the real-time inventory counting method of the ASRS fully automated material storage system, the apparatus specifically comprising the following components: The multi-source sensing module reads material tag identification codes and signal strengths from fixed RFID antenna arrays on each shelf layer at preset intervals and outputs RFID sensing datasets. It also collects real-time weight values ​​from location weight sensors and detects location status changes, outputting weight sensing datasets. Within this module, an antenna switching control subunit uses an RF switch matrix to poll and drive each antenna in preset time slots. A deduplication subunit removes duplicate records from multiple antennas by sorting them in descending order of signal strength. A coarse positioning subunit calculates the tag's spatial coordinates using a weighted centroid method. A weight acquisition subunit samples at a preset frequency and outputs filtered weight values ​​after sliding window mean filtering. A status change detection subunit compares the weight change with preset threshold conditions and outputs a judgment event indicating weight increase, weight decrease, or no change.

[0128] The data fusion and inventory modeling module is used to read the RFID sensing dataset and the weight sensing dataset, perform time alignment and spatial matching, determine the location status according to multi-source fusion rules, and assign fusion confidence. It incrementally updates the real-time inventory model with a three-dimensional state matrix and simultaneously maintains the material trajectory tracking map. Within this module, the time alignment subunit aligns the two types of data using a second-level time window based on system master clock synchronization; the spatial matching subunit matches and associates tags and locations using nearest neighbor matching; the fusion inference subunit performs status determination and confidence assignment according to multi-source fusion rules; the model update subunit rewrites the three-dimensional array elements of changing locations using an incremental strategy; and the trajectory tracking subunit adds location nodes to the trajectory chain using material identifiers as indexes.

[0129] The inventory comparison and discrepancy analysis module is used to perform a full comparison between the real-time inventory model and the warehouse management system ledger to detect discrepancies. For each discrepancy record, root cause inference is performed based on the material trajectory tracking map and operation logs to output the root cause category and handling suggestions. The comparison subunit within this module compares and outputs discrepancy records one by one according to the dimensions of location mapping, material identification, and quantity. The root cause inference subunit uses trajectory and log alignment contradiction analysis to cover operation-related root causes and sensor historical data analysis to cover equipment-related root causes. The posterior probability calculation subunit evaluates the probability of each root cause category and selects the highest probability as the recommended root cause.

[0130] The mobile inventory execution module is used to trigger mobile inventory tasks on the stacker crane during idle periods for low confidence differences or RFID coverage blind spots, and to update the real-time inventory model with the three-source fusion results. Within this module, a priority scoring subunit periodically calculates the inventory location priority ranking using a four-factor weighted scoring function. A task scheduling subunit selects inventory locations in descending priority order during periods of low equipment load, generates inventory tasks, and executes scheduling under daily time and coverage constraints. A mobile sensing subunit drives the RFID reader and vision camera on the payload platform to perform close-range reading and visual verification. The three-source fusion subunit determines the actual status of the inventory location using a weighted voting rule and writes the results into the inventory model.

[0131] The Difference Handling and Optimization module triggers corresponding handling procedures based on difference levels and tracks the entire process from detection to closure, periodically generating sensor health assessment and difference pattern analysis reports. Within this module, the graded handling subunit executes lockout alarms, queued review, and silent recording according to three severity levels. The timeout escalation subunit automatically escalates the handling level of unhandled differences based on preset timeout conditions. The status tracking subunit maintains the status fields of difference records, audit information for the handler and handling time. The statistical reporting subunit generates daily and weekly reports including difference distribution, handling efficiency, and sensor health indicators. The pattern analysis subunit uses clustering algorithms to identify high-frequency difference areas and material types and outputs adjustment suggestions.

[0132] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0133] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0134] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method.

[0135] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for real-time inventory counting in an ASRS fully automated material storage system, characterized in that, The method includes: The fixed RFID antenna array on each shelf layer reads the material tag identification code and signal strength at a preset cycle. After deduplication and coarse positioning, the RFID sensing dataset is output. The real-time weight value is collected by the warehouse weight sensor and the weight change is calculated by sliding filter to detect the change of warehouse status, and the weight sensing dataset is output. The RFID sensing dataset and the weight sensing dataset are read and time-aligned and spatially matched for nearest neighbor matching. Based on the combined state of tag and weight detection, the location occupancy status is determined according to multi-source fusion rules, and a fusion confidence level is assigned. The real-time inventory model is incrementally updated with a three-dimensional state matrix, and the material trajectory tracking map is maintained simultaneously. The real-time inventory model is fully compared with the warehouse management system ledger in terms of location mapping, material identification, and quantity, and the difference records are output. This includes: extracting the complete location change sequence of the difference material from its entry into the warehouse to the present from the material trajectory tracking map; extracting the relevant entry, exit, and transfer operation records of the material from the operation log; aligning the location change sequence with the operation records on the time axis; and inferring that there is an entry operation but no corresponding location change when there is an entry operation. Unregistered goods entering the warehouse; for discrepancies that cannot be explained by operation logs, further analysis of historical sensor data is conducted. When RFID and weight data continuously conflict, it is inferred to be a tag malfunction; when the weight of a single storage location continuously fluctuates abnormally, it is inferred to be sensor drift; after ruling out the aforementioned causes, it is attributed to system data errors. The real-time inventory model uses a three-dimensional array to represent the storage location status of shelf columns, layers, and depths. The array elements contain material identifiers, occupancy status, the fusion confidence level, and update time fields. When the fusion status data frame is inconsistent with the current record in the model, an incremental update is triggered, rewriting only the elements of the changed storage location and simultaneously adding a location record to the material trajectory tracking graph. For each discrepancy record, root cause reasoning is performed based on the material trajectory tracking graph and operation logs to output the root cause category and handling suggestions. For low-confidence difference records or RFID coverage blind spots, the stacker-mounted RFID reader is triggered to perform targeted inventory during idle periods. The mobile inventory results are then overwritten and updated in the real-time inventory model. The handling process is triggered according to the difference level and tracked to a closed loop. Sensor data is collected periodically to generate health assessment and difference pattern analysis reports.

2. The real-time inventory counting method for the ASRS fully automated material storage system according to claim 1, characterized in that, The process of reading material tag identification codes and signal strengths at preset intervals using fixed RFID antenna arrays on each shelf layer includes: The fixed RFID antenna is installed on the crossbeams of each layer of the shelf at a preset interval and the location tags within the coverage area are polled by the radio frequency switch at a preset time slot. After the reader receives the tag response, it extracts the tag identification code and the signal strength. For the same tag, the maximum signal strength is taken from the repeated records in multiple antenna channels, and the coarse spatial coordinates of the tag are calculated according to the weighted centroid method. The cargo location weight sensor collects the real-time weight value according to a preset sampling period. The original weight sequence is filtered by average value taking through a sliding window to suppress vibration noise. The weight change is obtained by calculating the difference between the current weight and the weight of the previous period. When the weight exceeds the preset threshold condition in a positive direction, it is judged as a weight increase event; when it exceeds the threshold condition in a negative direction, it is judged as a weight decrease event; and when it is lower than the preset empty cargo location threshold condition, it is judged as an empty cargo location.

3. The real-time inventory counting method for the ASRS fully automated material storage system according to claim 2, characterized in that, The step of reading the RFID sensing dataset and performing time alignment and spatial nearest neighbor matching with the weight sensing dataset includes: Using the system master clock as a reference, the RFID sensing dataset and the weight sensing dataset are aligned to the same second-level time window. For the coarse spatial coordinates and physical coordinates of the cargo location that fall into the same time window, nearest neighbor matching is performed according to the preset matching distance as the threshold condition. When both sources detect materials, the cargo location is determined to be occupied and the fusion confidence score is set to a high value. When only a single source detects, it is determined to be pending confirmation and the fusion confidence score is set to a medium value.

4. The real-time inventory counting method for the ASRS fully automated material storage system according to claim 1, characterized in that, The method of triggering the stacker-mounted RFID reader to perform targeted inventory checks during idle periods for low-confidence difference records or RFID coverage blind spots includes: A priority scoring model for inventory locations is constructed, using a weighted sum of time decay factor, material value factor, material risk level factor, and historical difference frequency factor as the scoring function. The scoring of all warehouse locations is calculated every hour and sorted in descending order to generate a priority ranking. When the equipment utilization rate is lower than the preset load threshold, the top warehouse locations are automatically selected to generate a mobile inventory task. After the stacker crane moves to the preset positioning point in front of the target storage location, the RFID reader on the load platform performs close-range directional reading at a preset power and distance, and triggers the vision camera to collect images of the storage location to extract appearance features for auxiliary verification. The RFID reading results of the mobile inventory, the current value of the weight sensor and the vision verification results are fused together, and the actual status of the storage location is determined according to the confidence level of not less than the preset threshold condition.

5. The real-time inventory counting method for the ASRS fully automated material storage system according to claim 1, characterized in that, The process of triggering handling based on the difference level and tracking it to a closed loop includes: The discrepancies are classified into three preset levels according to their severity. The first level is when the discrepancy between the records and the actual inventory affects the materials that can be shipped. The relevant storage locations are immediately locked and an alarm is pushed to the warehouse management terminal. On-site verification is required within a preset time limit. The second level is when the discrepancy in attributes or low confidence is to be confirmed. The discrepancies are included in the daily inventory queue and are required to be handled within a working day. After manual verification and confirmation of the disposal, the real-time inventory model and the warehouse management system ledger are updated to ensure consistency between the ledger and the actual inventory. The discrepancy records for disposal that have not been completed within the time limit are automatically upgraded and a disposal report is generated daily to count the total number of discrepancies, the distribution of each type, and the closure rate. On a weekly basis, discrepancy pattern clustering and sensor health trend analysis are performed based on the accumulated discrepancy records.

6. A real-time inventory counting device for an ASRS fully automated material storage system, characterized in that, The device includes: The multi-source sensing module is used to read the material tag identification code and signal strength through the fixed RFID antenna array of each shelf layer at a preset cycle and output the RFID sensing dataset. The real-time weight value is collected by the warehouse weight sensor and the warehouse status change event is detected and the weight sensing dataset is output. The data fusion and inventory modeling module is used to read the RFID sensing dataset and the weight sensing dataset, perform time alignment and spatial matching, determine the location status according to the multi-source fusion rules and assign fusion confidence, incrementally update the real-time inventory model with the three-dimensional state matrix and synchronously maintain the material trajectory tracking map. The inventory comparison and discrepancy analysis module is used to perform a full comparison between the real-time inventory model and the warehouse management system ledger to detect discrepancies. This includes: extracting the complete location change sequence of the discrepancy material from its entry into the warehouse to the present from the material trajectory tracking map; extracting the relevant entry, exit, and transfer operation records of the material from the operation log; aligning the location change sequence with the operation records along the timeline; and identifying cases where there is an entry operation but no corresponding location change, inferring that the entry was not registered. For discrepancies that cannot be explained by the operation records, further analysis of historical sensor data is performed; if RFID and weight data continuously conflict, it is inferred that the tag is faulty. When a fault occurs or the weight of a single storage location fluctuates abnormally, it is inferred to be sensor drift; after ruling out the aforementioned causes, it is attributed to system data error. The real-time inventory model uses a three-dimensional array to represent the storage location status of shelf columns, layers, and depths. Array elements include material identifier, occupancy status, fusion confidence level, and update time fields. When the fusion status data frame is inconsistent with the current record in the model, an incremental update is triggered, rewriting only the elements of the changed storage location and simultaneously adding a location record to the material trajectory tracking graph. For each discrepancy record, root cause reasoning is performed based on the material trajectory tracking graph and operation log to output the root cause category and handling suggestions. The mobile inventory execution module is used to perform targeted inventory counts during idle periods for stacker crane mobile inventory tasks triggered by low confidence differences or RFID coverage blind spots, and to update the real-time inventory model with the three-source fusion results. The difference handling and optimization module is used to trigger corresponding handling processes according to the difference level and track the status of the entire process from detection to closure, and periodically generate sensor health assessment and difference pattern analysis reports.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the real-time inventory counting method of the ASRS fully automated material storage system according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the real-time inventory counting method of the ASRS fully automated material storage system as described in any one of claims 1 to 5.

Images