A method for dynamic generation and execution of forklift inspection tasks

Abstract

The application discloses a forklift inspection task dynamic generation and execution method, and belongs to the technical field of forklift inspection. The method comprises the following steps: collecting forklift image or video data and performing identification processing to obtain current inspection objects and component structure state information; constructing a forklift digital portrait according to the visual identification result and basic file information; performing knowledge reasoning based on a pre-constructed inspection knowledge graph to automatically generate an inspection task set; in the inspection execution process, the corresponding inspection task is adaptively switched according to the recognized change of the inspection object; guide information is superimposed and displayed through an augmented reality device; the consistency of the recognized object and the inspection item is checked in real time, and an exception is prompted; when the inspection result is abnormal, an additional task set is automatically generated to guide execution; various types of data are associated to form a structured inspection data set, an inspection report is automatically generated, and closed-loop management of the whole inspection process is realized. The application improves the intelligent level and accuracy of forklift inspection.

Description

Technical Field

[0001] This invention relates to the field of forklift inspection technology, and in particular to a method for dynamically generating and executing forklift inspection tasks. Background Technology

[0002] Forklifts, as an important component of specialized motor vehicles used in factories and workshops, are widely used in warehousing, logistics, and manufacturing industries. According to relevant safety technical specifications, forklifts require initial and periodic inspections to ensure safe operation. Current forklift inspections primarily rely on manual experience or auxiliary systems based on fixed inspection checklists, which has the following shortcomings: (1) Inspection items are usually carried out according to a preset list and cannot be dynamically adjusted according to differences in equipment configuration, structural characteristics and on-site conditions; (2) Inspectors need to determine the inspection items corresponding to the current component themselves, which may easily lead to missed inspections, incorrect inspections, or inapplicable items being performed. (3) Most existing AR-assisted systems only provide static work instructions and cannot adjust the inspection task in real time according to the on-site recognition results; (4) Existing inspection systems typically only record the results after detecting anomalies, lacking the ability to automatically extend subsequent inspection tasks; (5) The inspection process, abnormal evidence collection, risk assessment and document generation are independent of each other and it is difficult to form a complete closed loop; (6) Complex working conditions such as changes in ambient lighting, component obstruction or dirt can lead to deviations in the single recognition of the digital image of the forklift.

[0003] Therefore, there is a need for an intelligent inspection method that can combine visual recognition, knowledge reasoning, and dynamic task management mechanisms to achieve automatic generation, dynamic expansion, and closed-loop execution of inspection tasks. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for dynamically generating and executing forklift inspection tasks.

[0005] The objective of this invention is achieved through the following technical solution: a method for dynamically generating and executing forklift inspection tasks, comprising the following steps: Step S1: Collect forklift image or video data, and use target detection, image classification and OCR recognition algorithms to recognize and process the forklift image or video data to obtain the current inspection object and its component structure status information. At the same time, obtain the spatial position information, observation angle information and time information of the current inspection object, and calculate the recognition confidence score of each object. Step S2: Based on the visual recognition results and the basic information of the forklift, construct a digital profile of the forklift to represent the topological connection relationship and structural dependency relationship between equipment components; Step S3: Based on the pre-built forklift inspection knowledge graph, combined with the forklift digital profile, the current inspection object, regulatory constraints and historical inspection data, knowledge reasoning is performed to automatically generate a set of inspection tasks that match the current equipment structure status; Step S4: During the inspection process, when a change is detected in the current inspection object, the system automatically switches to the inspection task corresponding to the inspection object, thereby achieving an adaptive mapping between the inspection object and the inspection task. Step S5: Use augmented reality devices to overlay and display guidance information for the current inspection task in real time, guiding inspection personnel to perform standardized inspection operations; Step S6: During the inspection process, when the recognition confidence of any object is lower than the preset threshold, the ambiguity resolution mechanism is triggered to perform a review and recognition, and the consistency between the current object and the inspection item is verified in real time. An error is detected in real time. Step S7: When it is detected that the test result does not meet the preset rules, the abnormal handling process is automatically triggered. An additional task set is automatically generated according to the abnormality type. The traceable chain from the occurrence of the abnormality to the generation of the additional task set is recorded, and the test personnel are guided to execute the additional task set. Step S8: Unify the association of various data generated during the inspection process with the corresponding inspection items to form a structured inspection dataset. Automatically generate inspection reports, anomaly analysis reports, and risk assessment reports based on the structured inspection dataset to dynamically update or switch inspection tasks while retaining original records, thus achieving closed-loop management of the entire inspection process. In step S1, the inspection content includes at least one of the following: manufacturing data review, modification data review, structural type inspection, vehicle appearance inspection, inspection of major load-bearing structural components, nameplate and safety sign inspection, power system inspection, transmission system inspection, driving system inspection, steering system inspection, hydraulic system inspection, braking system inspection, electrical and control system inspection, working device inspection, safety protection and protective device inspection, safety monitoring device inspection, explosion-proof performance inspection, and braking performance test.

[0006] Preferably, the recognition confidence score is obtained by weighted summation of the basic confidence score, multi-frame voting consistency, and geometric overlap between the recognition box and the component's three-dimensional template box, with the weight of each calculation item being 1.

[0007] Preferably, in step S2, the forklift digital profile includes: equipment model, manufacturer, rated load, lifting height, configuration parameters, historical inspection records, historical defect records, and historical rectification records.

[0008] Preferably, in step S3, the forklift inspection knowledge graph includes: equipment type node, component node, inspection item node, regulatory standard node, judgment rule node, risk level node, and defect type node; based on the topological relationship of equipment components and the applicable conditions of regulations, the inspection standards are instantiated to dynamically generate a set of inspection tasks, and are updated in real time according to changes in the structural state of the components.

[0009] Preferably, the following steps are included before performing step S3: Time-series analysis and pattern recognition are performed on historical inspection records, historical defect records, and historical rectification records of the same equipment or equipment of the same model. Machine learning models are used to predict the probability of defects in each component and key inspection areas within the current inspection cycle. When generating the inspection task set, the inspection items of the corresponding components are assigned higher priority or key inspection sub-tasks are added based on the prediction results.

[0010] Preferably, the feature engineering of the machine learning model includes at least one of the following: service life, operating intensity, maintenance interval, and environmental parameters.

[0011] Preferably, in step S4, the adaptive mapping between the inspection object and the inspection task includes: when at least one of the four components—frame, mast, fork carriage, and forks—is identified, the main load-bearing structural component inspection task is automatically loaded, and prompts are given to perform external macroscopic inspection of the welds of the frame, mast, fork carriage, and forks, as well as inspection of the load-bearing structural components and connectors; changes in the current inspection object include changes in forklift model, individual changes of different frames of the same model, and structural configuration changes caused by the addition / removal of attachments.

[0012] Preferably, before performing the inspection, an inspection path planning step is also included: based on the component topology connection relationship and spatial location information in the digital profile of the forklift, with the goal of minimizing the movement path of the inspection personnel or maximizing the coverage efficiency, the optimal execution sequence of each inspection task is generated; and during the inspection execution process, the execution order of the remaining tasks is dynamically adjusted according to the real-time spatial location of the inspection personnel and the completed items.

[0013] Preferably, in step S5, the guidance information includes: inspection items, inspection methods, judgment criteria, legal provisions, precautions, and evidence collection requirements; the guidance information is dynamically bound to the currently identified inspection object and its corresponding inspection task, and the displayed content is updated synchronously as the inspection object is switched.

[0014] Preferably, in step S6, the real-time verification includes: determining whether the current identified object matches the inspection item, whether the inspection item is missing, whether the inspection order is reasonable, whether the inspection result meets the rule requirements, and whether there is an inapplicable item execution situation; when the judgment result is abnormal, real-time prompt information is sent to the inspection personnel through the augmented reality device or terminal device to guide the inspection personnel to correct the inspection operation.

[0015] Preferably, the ambiguity resolution mechanism includes: selecting anchor point components with identification confidence scores higher than preset scores based on the topological adjacency relationships recorded in the forklift inspection knowledge graph; calculating the expected pose and candidate identity of the low-confidence identification object based on the confirmed pose of the selected anchor point components and the relative pose constraints recorded in the forklift inspection knowledge graph; and updating the digital profile of the forklift. And / or, acquire the currently available auxiliary perception modalities of the augmented reality device to dynamically re-identify the target area of ​​the low-confidence recognition object and update the digital profile of the forklift.

[0016] Preferably, the auxiliary perception modality is selected from at least one of deep point cloud and infrared thermal image; when the augmented reality device is equipped with only a color camera, multi-frame voting consistency is used as the only auxiliary signal to achieve modality adaptive degradation; after updating the digital profile of the forklift through the ambiguity resolution mechanism, the dynamic switching of the inspection task or the generation of additional tasks is triggered.

[0017] Preferably, in step S7, the additional task set includes at least one of the following: a fine measurement task for measuring the size, thickness, deformation, and wear of the abnormal area; a multi-angle evidence collection task including front view, side view, close-up view, and continuous video recording; a risk level assessment task that automatically matches regulatory clauses, defect levels, risk levels, and safety impact levels using a knowledge graph; a re-inspection and confirmation task that performs secondary testing, verification measurement, and manual confirmation on preset abnormal items; an inspection opinion generation task that automatically generates inspection opinions, defect descriptions, and rectification suggestions based on the abnormality type; and an opinion notification generation task that automatically generates special equipment inspection opinion notices according to regulatory requirements and automatically associates them with abnormal items.

[0018] Preferably, the fine measurement task, multi-angle evidence collection task, and re-inspection confirmation task are automatically arranged according to the preset execution order of additional tasks to form an anomaly handling workflow and guide the inspection personnel to execute them in sequence; the inspection opinions, defect descriptions, rectification suggestions, and special equipment inspection opinion notices are all automatically associated with the anomalies to form a traceable anomaly handling record chain.

[0019] Preferably, in step S8, the various types of data include: inspection tasks, abnormal items, image evidence, video evidence, measurement data, risk assessment results, and document information; the structured inspection dataset contains a unified association index between each inspection item and its corresponding data; the original records, inspection reports, abnormal analysis reports, and risk assessment reports are automatically generated by extracting corresponding data based on the unified association index, forming a traceable electronic archive for closed-loop management of the entire inspection process.

[0020] The beneficial effects of this invention are: 1) This invention achieves dynamic generation of inspection tasks through visual recognition and knowledge graph collaborative reasoning, rather than relying on a preset fixed list. It can automatically match and generate a set of inspection tasks based on the actual equipment structure status identified on site. Compared with existing solutions that rely on manual judgment or preset lists, it significantly improves the matching accuracy between inspection tasks and actual equipment status, and effectively avoids the problem of unsuitable projects being executed or applicable projects being omitted due to differences in equipment configuration.

[0021] 2) By establishing an adaptive mapping mechanism between the inspection object and the inspection task, the present invention automatically switches the corresponding inspection task once a change in the current observation object is detected during the inspection process. This realizes a dynamic execution mode of "inspecting what you see" without requiring inspectors to judge the inspection items corresponding to the current component. This greatly reduces the probability of missed and incorrect inspections and improves the efficiency of inspection execution.

[0022] 3) This invention uses augmented reality devices to overlay and display inspection guidance information dynamically bound to the currently identified object in real time, and verifies the consistency between the identified object and the inspection items in real time during the inspection process. When items are found to be missing, the order is unreasonable, or the results do not conform to the rules, it will actively prompt corrections, so that even if the inspectors are inexperienced, they can perform inspections according to the standardized process. This effectively reduces the reliance on human experience and improves the standardization and normalization of the inspection process.

[0023] 4) This invention uses an anomaly-driven automatic expansion mechanism for additional tasks. When the test results do not meet the preset rules, it automatically generates a set of additional tasks, including fine measurement, multi-angle evidence collection, risk level assessment, re-inspection confirmation, test opinion generation, and opinion notification generation. This makes anomaly handling no longer limited to simple recording, but forms a fully automated handling chain from anomaly discovery to supplementary testing, risk assessment, and document issuance, significantly improving the completeness and efficiency of anomaly handling.

[0024] 5) This invention unifies the association of inspection tasks, abnormal items, image evidence, measurement data, risk assessment results and document information to the corresponding inspection items to form a structured dataset, and automatically generates original records, inspection reports, abnormal analysis reports and risk assessment reports based on this dataset, thus opening up a closed-loop data process from task generation, execution guidance, abnormal handling to report issuance, solving the problem of independent links and data fragmentation in the existing technology, and realizing traceable closed-loop management of the entire inspection process.

[0025] 6) To address the technical problem of single-time recognition deviation caused by complex working conditions such as changes in on-site lighting, component obstruction, or contamination, a visual recognition confidence assessment and ambiguity resolution mechanism is introduced to dynamically re-identify low-confidence recognition objects, thereby ensuring that the visual information on which subsequent inspection tasks are generated and switched has higher reliability. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See Figure 1 This invention provides a technical solution: a method for dynamically generating and executing forklift inspection tasks, comprising the following steps: Step S1: Collect forklift image or video data, and use target detection, image classification and OCR recognition algorithms to recognize and process the forklift image or video data to obtain the current inspection object and its component structure status information. At the same time, obtain the spatial position information, observation angle information and time information of the current inspection object, and calculate the recognition confidence score of each object. Step S2: Based on the visual recognition results and the basic information of the forklift, construct a digital profile of the forklift to represent the topological connection relationship and structural dependency relationship between equipment components; Step S3: Based on the pre-built forklift inspection knowledge graph, combined with the forklift digital profile, the current inspection object, regulatory constraints and historical inspection data, knowledge reasoning is performed to automatically generate a set of inspection tasks that match the current equipment structure status; Step S4: During the inspection process, when a change is detected in the current inspection object, the system automatically switches to the inspection task corresponding to the inspection object, thereby achieving an adaptive mapping between the inspection object and the inspection task. Step S5: Use augmented reality devices to overlay and display guidance information for the current inspection task in real time, guiding inspection personnel to perform standardized inspection operations; Step S6: During the inspection process, when the recognition confidence of any object is lower than the preset threshold, the ambiguity resolution mechanism is triggered to perform a review and recognition, and the consistency between the current object and the inspection item is verified in real time. An error is detected in real time. Step S7: When it is detected that the test result does not meet the preset rules, the abnormal handling process is automatically triggered. An additional task set is automatically generated according to the abnormality type. The traceable chain from the occurrence of the abnormality to the generation of the additional task set is recorded, and the test personnel are guided to execute the additional task set. Step S8: Link all types of data generated during the inspection process to the corresponding inspection items to form a structured inspection dataset. Automatically generate inspection reports, anomaly analysis reports, and risk assessment reports based on the structured inspection dataset to dynamically update or switch inspection tasks, while retaining the original records to achieve closed-loop management of the entire inspection process.

[0029] In this embodiment, the present invention achieves intelligent generation, adaptive execution, automatic anomaly handling, and closed-loop management of inspection tasks through visual recognition, equipment digital profile construction, knowledge graph reasoning, and dynamic task expansion mechanisms, thereby improving the accuracy, standardization, and intelligence level of forklift inspection.

[0030] This invention constructs a dynamic inspection task generation mechanism driven by inspection objects based on visual recognition and knowledge graph collaborative reasoning. This mechanism enables inspection tasks to no longer rely on preset static processes, but to generate and switch corresponding inspection tasks in real time based on the equipment objects and their structural states identified on site. This achieves dynamic binding and adaptive matching between inspection tasks and the state of inspection objects.

[0031] In some embodiments, the inspection content in step S1 includes at least one of the following: review of manufacturing data, review of modification data, structural type inspection, vehicle appearance inspection, inspection of major load-bearing structural components, inspection of nameplates and safety signs, inspection of power system, inspection of transmission system, inspection of driving system, inspection of steering system, inspection of hydraulic system, inspection of braking system, inspection of electrical and control system, inspection of working device, inspection of safety protection and protective device, inspection of safety monitoring device, inspection of explosion-proof performance, and braking performance test.

[0032] In this embodiment, S1: Visual perception and inspection object identification: Forklift images or video data are acquired through AR glasses, industrial cameras, or mobile terminals. The current inspection object is identified using target detection, image classification, and OCR recognition algorithms. The inspection content includes: review of manufacturing documents, review of modification documents, structural type inspection, overall vehicle appearance inspection, inspection of major load-bearing structural components, inspection of nameplates and safety signs, inspection of the power system, inspection of the transmission system, inspection of the driving system, inspection of the steering system, inspection of the hydraulic system, inspection of the braking system, inspection of the electrical and control systems, inspection of the working devices, inspection of safety protection and safeguards, inspection of safety monitoring devices, explosion-proof performance inspection, and braking performance testing.

[0033] At the same time, it obtains the spatial position, viewing angle, and time information of the current object.

[0034] In some embodiments, the recognition confidence score is obtained by weighted summation of the base confidence score, multi-frame voting consistency, and geometric overlap between the recognition box and the component's 3D template box, with the weight of each calculation item being 1.

[0035] In this embodiment, during the visual recognition process in step S1, the recognition confidence score of each recognition object is calculated synchronously; during the real-time verification process in step S6, when the recognition confidence score is lower than a preset threshold, the ambiguity resolution mechanism is automatically triggered to perform multimodal re-recognition (such as fusing infrared images, depth images or laser point cloud data to re-recognize the target area), and the current inspection object and its structural state information are dynamically updated according to the re-recognition results, thereby ensuring that the visual information on which the subsequent inspection tasks are generated and switched has higher reliability.

[0036] In some embodiments, in step S2, the forklift digital profile includes: equipment model, manufacturer, rated load, lifting height, configuration parameters, historical inspection records, historical defect records, and historical rectification records.

[0037] In this embodiment, S2: Equipment Digital Profile Construction: A digital profile of the forklift is constructed based on the visual recognition results and basic file information. The digital profile includes: equipment model, manufacturer, rated load, lifting height, configuration parameters, historical inspection records, historical defect records, and historical rectification records. This forms a structured digital model for the current equipment.

[0038] In some embodiments, in step S3, the forklift inspection knowledge graph includes: equipment type node, component node, inspection item node, regulatory standard node, judgment rule node, risk level node, and defect type node; the inspection standard is instantiated according to the topological relationship of the equipment components and the applicable conditions of regulations to dynamically generate a set of inspection tasks, and is updated in real time according to the changes in the component structure state.

[0039] In this embodiment, S3: Dynamic inspection task generation based on knowledge graph: Constructing a forklift inspection knowledge graph. The knowledge graph includes: equipment type nodes, component nodes, inspection item nodes, regulatory standard nodes, judgment rule nodes, risk level nodes, and defect type nodes.

[0040] The system combines digital profiles, current identification objects, regulatory constraints, and historical inspection data to perform knowledge reasoning. It automatically generates a set of inspection tasks corresponding to the current device.

[0041] In some embodiments, the following steps are included before performing step S3: Time-series analysis and pattern recognition are performed on historical inspection records, historical defect records, and historical rectification records of the same equipment or equipment of the same model. Machine learning models are used to predict the probability of defects in each component and key inspection areas within the current inspection cycle. When generating the inspection task set, the inspection items of the corresponding components are assigned higher priority or key inspection sub-tasks are added based on the prediction results.

[0042] In some embodiments, the feature engineering of the machine learning model includes at least one of service life, operating conditions, maintenance intervals, and environmental parameters.

[0043] In some embodiments, the adaptive mapping of the inspection object and the inspection task in step S4 includes: when at least one of the four components of frame, mast, fork carriage and forks is identified, the main load-bearing structural component inspection task is automatically loaded, and prompts are given to perform external macroscopic inspection of the welds of the frame, mast, fork carriage and forks, and inspection of load-bearing structural components and connectors; changes in the current inspection object include changes in forklift model, individual changes of different frames of the same model, and structural configuration changes caused by the addition / removal of attachments.

[0044] In this embodiment, S4: Adaptive mapping between inspection object and inspection task: When the system detects a change in the currently observed object, it automatically switches to the corresponding inspection task. For example: Identifies one of the four components: frame, mast, fork carriage, or forks, and automatically loads an inspection of the main load-bearing structural components. It also prompts that the following should be performed: a macroscopic external inspection of the welds on the frame, mast, fork carriage, or forks, as well as an inspection of the load-bearing structural components and connections. This achieves the goal of "inspecting what you see."

[0045] In some embodiments, before performing the inspection, an inspection path planning step is also included: based on the component topology connection relationship and spatial location information in the forklift digital profile, with the goal of minimizing the movement path of the inspection personnel or maximizing the coverage efficiency, an optimal execution sequence for each inspection task is generated; and during the inspection execution process, the execution order of the remaining tasks is dynamically adjusted according to the real-time spatial location of the inspection personnel and the completed items.

[0046] In some embodiments, in step S5, the guidance information includes: inspection items, inspection methods, judgment criteria, legal provisions, precautions, and evidence collection requirements; the guidance information is dynamically bound to the currently identified inspection object and its corresponding inspection task, and the displayed content is updated synchronously as the inspection object is switched.

[0047] In this embodiment, S5: AR-guided inspection execution: The system displays in real-time, via AR glasses, the inspection items, inspection methods, judgment criteria, regulatory clauses, precautions, and key evidence collection requirements. This guides inspection personnel in completing standardized inspection operations.

[0048] In some embodiments, the ambiguity resolution mechanism includes: selecting anchor point components with identification confidence scores higher than preset scores based on the topological adjacency relationships recorded in the forklift inspection knowledge graph; calculating the expected pose and candidate identity of the low-confidence identification object based on the confirmed pose of the selected anchor point components and the relative pose constraints recorded in the forklift inspection knowledge graph; and updating the digital profile of the forklift. And / or, acquire the currently available auxiliary perception modalities of the augmented reality device to dynamically re-identify the target area of ​​the low-confidence recognition object and update the digital profile of the forklift.

[0049] In this embodiment, the ambiguity resolution based on topological adjacency addresses the uncertainty of the identity or pose of objects that have been detected but whose confidence level is below a threshold. The calculated expected pose and candidate identity are used to update the digital profile of the forklift and trigger the dynamic switching of inspection tasks or the generation of additional inspection tasks.

[0050] In some embodiments, the auxiliary perception modality is selected from at least one of depth point cloud and infrared thermal image; when the augmented reality device is equipped with only a color camera, multi-frame voting consistency is used as the only auxiliary signal to achieve modality adaptive degradation; after updating the digital profile of the forklift through the ambiguity resolution mechanism, dynamic switching of inspection tasks or generation of additional tasks is triggered.

[0051] In some embodiments, step S6 includes real-time verification: determining whether the current identified object matches the inspection item, whether the inspection item is missing, whether the inspection order is reasonable, whether the inspection result meets the rule requirements, and whether there is an inapplicable item execution situation; when the judgment result is abnormal, real-time prompt information is sent to the inspection personnel through augmented reality device or terminal device to guide the inspection personnel to correct the inspection operation.

[0052] In this embodiment, S6: Real-time verification of the inspection process. The system determines in real time whether the current object matches the inspection item, whether any inspection item is missing, whether the inspection order is reasonable, whether the inspection results meet the rule requirements, and whether there are any inapplicable items being executed. Real-time prompts are given when an anomaly is detected.

[0053] In some embodiments, in step S7, the additional task set includes at least one of the following: a fine measurement task for measuring the size, thickness, deformation, and wear of the abnormal area; a multi-angle evidence collection task including front view, side view, close-up view, and continuous video recording; a risk level assessment task that automatically matches regulatory clauses, defect levels, risk levels, and safety impact levels using a knowledge graph; a re-inspection and confirmation task that performs secondary inspection, verification measurement, and manual confirmation on preset abnormal items; an inspection opinion generation task that automatically generates inspection opinions, defect descriptions, and rectification suggestions based on the abnormality type; and an opinion notification generation task that automatically generates special equipment inspection opinion notices according to regulatory requirements and automatically associates them with abnormal items.

[0054] In this embodiment, a dynamic expansion mechanism for inspection tasks based on anomaly identification is implemented: when the system detects that the inspection result does not meet preset rules, an anomaly handling process is automatically triggered. An additional task set is automatically generated based on the anomaly type.

[0055] (1) Fine measurement task for abnormal areas: Automatically add the following to abnormal areas: size measurement, thickness measurement, deformation measurement, and wear measurement. And automatically record the measurement results.

[0056] (2) Multi-angle evidence collection task for abnormal areas: Automatically generate supplementary evidence collection requirements. These include: front view; side view; close-up details; and continuous video recordings. Ensure the integrity of evidence in abnormal areas.

[0057] (3) Anomaly risk level assessment task: Knowledge graph automatic matching: corresponding legal provisions, corresponding defect level, risk level, and degree of safety impact. Generate risk assessment results.

[0058] (4) Re-inspection and Confirmation Task: For key anomalies, a re-inspection and confirmation task is automatically generated. Requirements: secondary inspection, verification measurement, and manual confirmation. This reduces the risk of misjudgment.

[0059] (5) Inspection opinion generation task, which automatically generates the following based on the type of abnormality: inspection opinion, defect description, and rectification suggestions.

[0060] (6) Opinion Notification Generation Task: Automatically generate special equipment inspection opinion notifications according to regulatory requirements and automatically associate them with abnormal items.

[0061] In some embodiments, the fine measurement task, multi-angle evidence collection task, and re-inspection confirmation task are automatically arranged according to the preset execution order of additional tasks to form an anomaly handling workflow and guide the inspection personnel to execute them in sequence; the inspection opinions, defect descriptions, rectification suggestions, and special equipment inspection opinion notices are all automatically associated with the anomalies to form a traceable anomaly handling record chain.

[0062] In some embodiments, in step S8, the various types of data include: inspection tasks, abnormal items, image evidence, video evidence, measurement data, risk assessment results, and document information; the structured inspection dataset contains a unified association index between each inspection item and its corresponding data; the original records, inspection reports, abnormality analysis reports, and risk assessment reports are automatically generated by extracting corresponding data based on the unified association index, forming a traceable electronic archive for closed-loop management of the entire inspection process.

[0063] In this embodiment, the test results are recorded in a structured manner, which uniformly associates the test tasks, abnormal items, image evidence, video evidence, measurement data, risk assessment results and document information with the corresponding test items to form a structured test dataset.

[0064] Inspection reports are automatically generated based on structured data, including: original records; inspection reports; anomaly analysis reports; and risk assessment reports. This achieves closed-loop management of the entire inspection process.

[0065] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for dynamically generating and executing forklift inspection tasks, characterized in that: Includes the following steps: Step S1: Collect forklift image or video data, and use target detection, image classification and OCR recognition algorithms to recognize and process the forklift image or video data to obtain the current inspection object and its component structure status information. At the same time, obtain the spatial position information, observation angle information and time information of the current inspection object, and calculate the recognition confidence score of each object. Step S2: Based on the visual recognition results and the basic information of the forklift, construct a digital profile of the forklift to represent the topological connection relationship and structural dependency relationship between equipment components; Step S3: Based on the pre-built forklift inspection knowledge graph, combined with the forklift digital profile, the current inspection object, regulatory constraints and historical inspection data, knowledge reasoning is performed to automatically generate a set of inspection tasks that match the current equipment structure status; Step S4: During the inspection process, when a change is detected in the current inspection object, the system automatically switches to the inspection task corresponding to the inspection object, thereby achieving an adaptive mapping between the inspection object and the inspection task. Step S5: Use augmented reality devices to overlay and display guidance information for the current inspection task in real time, guiding inspection personnel to perform standardized inspection operations; Step S6: During the inspection process, when the recognition confidence of any object is lower than the preset threshold, the ambiguity resolution mechanism is triggered to perform a review and recognition, and the consistency between the current object and the inspection item is verified in real time. An error is detected in real time. Step S7: When it is detected that the test result does not meet the preset rules, the abnormal handling process is automatically triggered. An additional task set is automatically generated according to the abnormality type. The traceable chain from the occurrence of the abnormality to the generation of the additional task set is recorded, and the test personnel are guided to execute the additional task set. Step S8: Link all types of data generated during the inspection process to the corresponding inspection items to form a structured inspection dataset. Automatically generate inspection reports, anomaly analysis reports and risk assessment reports based on the structured inspection dataset to dynamically update or switch inspection tasks, while retaining the original records to achieve closed-loop management of the entire inspection process. The ambiguity resolution mechanism includes: based on the topological adjacency relationship recorded in the forklift inspection knowledge graph, selecting anchor point components with identification confidence scores higher than preset scores, calculating the expected pose and candidate identity of low-confidence identification objects based on the confirmed pose of the selected anchor point components and the relative pose constraints recorded in the forklift inspection knowledge graph, and updating the digital profile of the forklift. And / or, acquire the currently available auxiliary perception modalities of the augmented reality device to dynamically re-identify the target area of ​​the low-confidence recognition object and update the digital profile of the forklift; The auxiliary perception modality is selected from at least one of deep point cloud and infrared thermal image; when the augmented reality device is equipped with only a color camera, multi-frame voting consistency is used as the only auxiliary signal to achieve modality adaptive degradation; after updating the digital profile of the forklift through the ambiguity resolution mechanism, the dynamic switching of the inspection task or the generation of additional tasks is triggered.

2. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S1, the inspection contents include at least one of the following: review of manufacturing data, review of modification data, structural type inspection, vehicle appearance inspection, inspection of major load-bearing structural components, inspection of nameplates and safety signs, inspection of power system, inspection of transmission system, inspection of driving system, inspection of steering system, inspection of hydraulic system, inspection of braking system, inspection of electrical and control system, inspection of working device, inspection of safety protection and protective device, inspection of safety monitoring device, inspection of explosion-proof performance, and braking performance test.

3. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: The recognition confidence score is obtained by weighted summation of the basic confidence score, multi-frame voting consistency, and geometric overlap between the recognition box and the component's 3D template box, with the weight of each calculation item being 1.

4. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S2, the digital profile of the forklift includes: equipment model, manufacturer, rated load, lifting height, configuration parameters, historical inspection records, historical defect records, and historical rectification records.

5. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S3, the forklift inspection knowledge graph includes: equipment type node, component node, inspection item node, regulatory standard node, judgment rule node, risk level node, and defect type node; based on the topological relationship of equipment components and the applicable conditions of regulations, the inspection standards are instantiated to dynamically generate a set of inspection tasks, and are updated in real time according to changes in the structural state of the components.

6. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: The following steps are included before performing step S3: Time-series analysis and pattern recognition are performed on historical inspection records, historical defect records, and historical rectification records of the same equipment or equipment of the same model. Machine learning models are used to predict the probability of defects in each component and key inspection areas within the current inspection cycle. When generating the inspection task set, the inspection items of the corresponding components are assigned higher priority or key inspection sub-tasks are added based on the prediction results.

7. The method for dynamically generating and executing forklift inspection tasks according to claim 6, characterized in that: The feature engineering of the machine learning model includes at least one of the following: service life, operating intensity, maintenance interval, and environmental parameters.

8. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S4, the adaptive mapping between the inspection object and the inspection task includes: when at least one of the four components, namely the frame, mast, fork carriage, and forks, is identified, the main load-bearing structural component inspection task is automatically loaded, and prompts are given to perform external macroscopic inspection of the welds of the frame, mast, fork carriage, and forks, as well as inspection of the load-bearing structural components and connectors; changes in the current inspection object include changes in forklift model, individual changes of different frames of the same model, and structural configuration changes caused by the addition / removal of attachments.

9. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: Before the inspection is carried out, the inspection path planning step is also included: based on the component topology connection relationship and spatial location information in the digital profile of the forklift, the optimal execution sequence of each inspection task is generated with the goal of minimizing the movement path of the inspection personnel or maximizing the coverage efficiency; and during the inspection execution process, the execution order of the remaining tasks is dynamically adjusted according to the real-time spatial location of the inspection personnel and the completed items.

10. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S5, the guidance information includes: inspection items, inspection methods, judgment basis, legal provisions, precautions and evidence collection requirements; the guidance information is dynamically bound to the currently identified inspection object and its corresponding inspection task, and the displayed content is updated synchronously as the inspection object is switched.

11. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S6, real-time verification includes: determining whether the current identified object matches the inspection item, whether the inspection item is missing, whether the inspection order is reasonable, whether the inspection result meets the rule requirements, and whether there is an inapplicable item execution situation; when the judgment result is abnormal, real-time prompt information is sent to the inspection personnel through augmented reality device or terminal device to guide the inspection personnel to correct the inspection operation.

12. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S7, the additional task set includes at least one of the following: a fine measurement task for measuring the size, thickness, deformation, and wear of the abnormal area; a multi-angle evidence collection task including front view, side view, close-up view, and continuous video recording; a risk level assessment task that automatically matches regulatory clauses, defect levels, risk levels, and safety impact levels using a knowledge graph; a re-inspection and confirmation task that performs secondary detection, verification measurement, and manual confirmation of preset abnormal items; and an inspection opinion generation task that automatically generates inspection opinions, defect descriptions, and rectification suggestions based on the abnormality type. And the task of automatically generating special equipment inspection opinion notices and automatically associating them with anomalies, in accordance with regulatory requirements.

13. The method for dynamically generating and executing forklift inspection tasks according to claim 12, characterized in that: The detailed measurement task, multi-angle evidence collection task, and re-inspection confirmation task are automatically arranged according to the preset execution order of additional tasks to form an anomaly handling workflow and guide the inspection personnel to execute them in sequence; the inspection opinions, defect descriptions, rectification suggestions, and special equipment inspection opinion notices are all automatically associated with the anomalies to form a traceable anomaly handling record chain.

14. The method for dynamically generating and executing forklift inspection tasks according to claim 1, characterized in that: In step S8, the various types of data include: inspection tasks, abnormal items, image evidence, video evidence, measurement data, risk assessment results, and document information; the structured inspection dataset contains a unified association index between each inspection item and its corresponding data; the original records, inspection reports, abnormal analysis reports, and risk assessment reports are automatically generated by extracting the corresponding data based on the unified association index, forming a traceable electronic archive for closed-loop management of the entire inspection process.

Images