Tunneling support operation intelligent identification system based on AI technology

Abstract

The invention discloses a tunneling support operation intelligent identification system based on an AI technology, and belongs to the technical field of coal mine intelligent safety monitoring. The system comprises an alarm generation module, an alarm processing module, an alarm reporting module and an alarm reminding module. When the heading machine stops, the AI model is automatically triggered for reasoning, and abnormal behaviors are recognized in real time according to the operation sequence of wall knocking and top asking, temporary supporting and permanent supporting; a time sequence reverse checking mechanism is innovatively adopted to automatically check the execution condition of a preorder procedure within 30 minutes, so that the integrity of the operation procedure is ensured; knocking action, forepoling bar state, anchor rod installation and other key indexes are identified through a computer vision technology, an alarm is generated immediately when abnormity is found, and underground broadcast reminding is linked; and after the management personnel disposes, the misinformation data is uploaded to the group training center optimization model. According to the invention, intelligent supervision of the whole process of the tunneling support operation is realized, post responsibility investigation is converted into in-process intervention, and the operation normalization and safety are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent safety monitoring technology in coal mines, and in particular to an intelligent identification system for tunneling and support operations based on AI technology. Background Technology

[0002] Tunnel excavation is one of the most important and dangerous aspects of daily coal mine production. Its speed and quality directly affect the mine's rational production layout, stable mining continuity, and the effectiveness of proactive disaster prevention measures. In coal mine production, tunneling and support operations include three key processes: roof and sidewall inspection, temporary support, and permanent support. These processes must be strictly followed in sequence according to operating procedures. Any omission or improper operation of any process can lead to roof collapse, spalling, and other safety accidents. The confined space, harsh environment, dense personnel, and multi-process collaborative nature of tunneling operations make safety management extremely difficult, especially under complex geological conditions such as high stress and soft rock. The standardization of support operations is directly related to the safety of workers.

[0003] Currently, safety management in coal mine tunneling faces mainly relies on manual supervision and periodic inspections, which has several shortcomings: First, manual supervision has limited coverage, failing to achieve 24 / 7 monitoring and easily leading to blind spots in supervision; second, traditional video monitoring systems only provide image recording functions, lacking intelligent analysis capabilities and unable to automatically identify violations and missing procedures; third, there is a lack of effective monitoring methods for the temporal logical relationship between the three procedures of roof and wall inspection, temporary support, and permanent support, making it difficult to determine whether the work procedures are complete; fourth, the handling process after problems are discovered is not timely enough, often resulting in post-event accountability rather than in-process intervention, failing to effectively prevent accidents. Compared to coal mining faces, tunneling faces have a lower level of equipment intelligence, lower tunneling efficiency, and require more workers with higher labor intensity, becoming a key bottleneck restricting the intelligent construction of coal mines and the realization of less-staffed and unmanned safe and efficient production.

[0004] To ensure production safety and achieve unmanned or minimally manned operations at tunneling faces, intelligent decision-making and automated collaboration across systems have become inevitable trends. With the rapid development of artificial intelligence, particularly computer vision, it has become possible to deeply integrate AI technology with coal mine tunneling processes to build an intelligent operation recognition and safety management system. By using deep learning models to identify and analyze the operation process in real time, combined with temporal logic judgment and rule-based reasoning, intelligent supervision of the entire tunneling and support operation process can be achieved. Simultaneously, establishing a closed-loop optimization mechanism of "edge reasoning - manual verification - model iteration" can continuously improve the system's recognition accuracy and practicality. Therefore, there is an urgent need to develop an AI-based intelligent recognition system for tunneling and support operations to improve the safety and reliability of these operations, further enhance the enterprise's scientific and technological strength and the level of innovation achievement transformation, and achieve the strategic goals of building a technologically strong enterprise and a brand-strong nation. Summary of the Invention

[0005] To address the technical problems of existing technologies, such as limited coverage of tunneling and support operation supervision, lack of intelligent recognition capabilities, inability to monitor the sequential logic of processes, and insufficient accident prevention capabilities, this invention provides an intelligent recognition system for tunneling and support operations based on AI technology. Through the deep integration of computer vision and multi-process sequential logic analysis, it achieves automated monitoring and safety management of the entire process of mine support operations.

[0006] To achieve the above objectives: An AI-based intelligent identification system for tunneling and support operations includes: The alarm generation module includes a model inference unit, an anomaly judgment unit, and a video capture unit. The model inference unit receives the tunneling machine status signal and triggers the AI ​​model to perform inference calculations when the tunneling machine cutting head changes from the start state to the stop state. The anomaly judgment unit performs anomaly judgment on the model inference results based on the preset operation sequence and rule base of the knocking and roof inspection process, temporary support process, and permanent support process. The video capture unit captures evidence video of the corresponding time period from the streaming media server based on the timestamp of the alarm information. The alarm processing module, connected to the alarm generation module, includes an alarm display unit and a handling record unit. The alarm display unit displays the alarm type, alarm time, description of the violation, and evidence video. The handling record unit records the management personnel's handling opinions and penalty decisions. The alarm reporting module is connected to the alarm processing module. It uploads alarm data marked as false alarms by the model and related videos to the group's artificial intelligence training center, and receives the optimized model issued by the training center.

[0007] Furthermore, the alarm generation module also includes a time-series lookup unit. When a temporary support operation is detected, the time-series lookup unit searches back 30 minutes of operation records to determine if there are any records of knocking on the roof or side walls. When a permanent support operation is detected, the time-series lookup unit searches back 30 minutes of operation records to determine if there are any records of temporary support.

[0008] Furthermore, the system also includes an alarm reminder module, which is connected to the alarm generation module. After an alarm is generated, the alarm reminder module automatically sends a control signal to the underground broadcasting system, which then plays a voice reminder containing the type of violation and rectification requirements.

[0009] Furthermore, the anomaly detection unit identifies the hammering action on the sidewalls, the observer's presence, and the coverage of the hammering range during the sidewall and roof hammering process; identifies the lifting status of the front beam, the contact status between the front beam and the roof slab, and the support installation position during the temporary support process; and identifies the drilling depth of the anchor bolt drilling rig, the amount of anchoring agent added, the installation operation of the anchor bolts and anchor cables, the mixing time, the tray position, and the tightening torque of the anchor bolts and anchor cables during the permanent support process.

[0010] Furthermore, the system also includes a process integrity scoring unit, which calculates the process integrity score according to the formula S = Σ(Wi × Ci × Ti) / Σ(Wi × Tmax), where Wi is the weight coefficient of the i-th process, with the weight W1=0.3 for the knocking and jacking process, the weight W2=0.3 for the temporary support process, and the weight W3=0.4 for the permanent support process; Ci is the completion coefficient of the i-th process, ranging from 0 to 1; Ti is the actual execution time of the i-th process; and Tmax is the standard execution time of each process. A critical alarm is generated when the score is lower than 0.7.

[0011] Furthermore, the system also includes an alarm accuracy calculation unit, which calculates the alarm accuracy according to the formula P = (TP + TN) / (TP + TN + FP + FN) × 100%, where TP is the number of correctly identified violations, TN is the number of correctly identified normal behaviors, FP is the number of false alarms, and FN is the number of missed alarms. When the accuracy is lower than 85%, the model optimization process is automatically triggered.

[0012] Furthermore, the model inference unit is deployed on an edge computing device, and the AI ​​model is a computer vision model based on deep learning, which achieves multimodal recognition by recognizing the action characteristics of operators, equipment status, and object detection.

[0013] Furthermore, the video capture unit's capture time is accurate to the second, and the alarm information includes alarm type, alarm time, alarm location, information of the violator, description of the violation, video start time, and video end time.

[0014] The aforementioned AI-based intelligent identification system for tunneling and support operations includes an alarm generation module, an alarm processing module, an alarm reporting module, and an alarm notification module. The alarm generation module includes a model inference unit, an anomaly judgment unit, a time-series lookup unit, and a video capture unit. When the tunneling machine's cutting head changes from the start state to the stop state, the model inference unit automatically triggers the AI ​​model to perform inference calculations, judging behavioral anomalies in the process in real time based on the operational sequence of the roof and side impact assessment, temporary support, and permanent support procedures. The time-series lookup unit checks back 30 minutes for roof and side impact assessment records when temporary support is detected, and checks back 30 minutes for temporary support records when permanent support is detected, ensuring the integrity of the procedures. The anomaly judgment unit identifies the impact action on the roof and side in the roof and side impact assessment, the observer's presence status, the lifting status of the forward beam and the roof contact status in the temporary support procedure, and key indicators such as the anchor bolt drilling depth, the amount of anchoring agent applied, and the anchor bolt and cable tightening torque in the permanent support procedure. The video capture unit accurately captures evidence video from the streaming media server based on the timestamp of the alarm information. The alarm processing module is used by management personnel to review, handle, and decide on penalties for alarm information. The alarm reporting module uploads alarms and related data that are classified as model false alarms to the group's artificial intelligence training center. The training center retrains and optimizes the model before issuing an update. The alarm reminder module automatically activates the downhole broadcast system to provide real-time voice reminders after an alarm is generated. The system also includes a process integrity scoring unit, which calculates process integrity according to the formula S = Σ(Wi×Ci×Ti) / Σ(Wi×Tmax), and an alarm accuracy calculation unit, which evaluates system performance according to the formula P = (TP + TN) / (TP + TN + FP + FN) × 100%. When the index falls below the threshold, the optimization process is automatically triggered.

[0015] The beneficial effects of this invention are: This system achieves deep integration of artificial intelligence with tunneling processes, technologies, and equipment. Through the deep binding of AI models with support procedures, it pioneered a time-series reverse-checking mechanism to verify procedure compliance, effectively ensuring the complete execution of three key procedures: roof and sidewall inspection, temporary support, and permanent support. It has constructed an intelligent recognition system integrating multimodal AI technologies, comprehensively utilizing motion recognition, equipment status detection, and object detection technologies to achieve precise monitoring of the entire operation process, with an accuracy rate exceeding 93%. A real-time alarm and linkage response mechanism has been established; the system automatically generates alarm information and links it with underground broadcast alerts. The approach shifted from post-incident accountability to in-process intervention, effectively preventing safety accidents. An innovative closed-loop optimization system, "edge reasoning - manual verification - model iteration," was constructed. Through continuous learning from false alarm data, the system gains self-evolution capabilities, constantly improving recognition accuracy and practicality. Intelligent supervision of the entire tunneling operation process was achieved, significantly reducing manual supervision costs and improving operational standardization. This provides crucial technical support for the intelligent construction of coal mines and the realization of the goals of reduced manpower and unmanned operation, ensuring personnel safety, standardizing operational behavior, guaranteeing support quality, and promoting the intelligent transformation of coal mine safety production management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a diagram illustrating the overall architecture of the present invention, showing the four-layer architecture of the system; Figure 2 The flowchart for the monitoring of the tunneling machine's rockfall and roof operation in this invention details the AI ​​inference triggered by the stopping of the tunneling machine. Figure 3 The flowchart for monitoring temporary support operations of this invention illustrates the time-series reverse lookup mechanism; Figure 4 This is a flowchart of the permanent support operation monitoring process of the present invention, which illustrates the multi-step monitoring process of permanent support. Figure 5 This is a closed-loop diagram of the alarm processing and model optimization of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0019] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0020] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0021] See Figures 1 to 5 As shown This invention provides an AI-based intelligent identification system for tunneling and support operations. This system achieves intelligent monitoring and management of the entire process of underground tunneling and support operations in coal mines through deep integration of deep learning models with tunneling and support procedures. The system mainly comprises four core components: an alarm generation module, an alarm processing module, an alarm reporting module, and an alarm notification module. Through real-time monitoring and intelligent identification of three key operational stages—roof and sidewall inspection, temporary support, and permanent support—it ensures the safety and standardization of tunneling and support operations.

[0022] In the specific implementation of the system, high-definition cameras and sensor equipment need to be deployed at the tunneling face to collect on-site video streams and tunneling machine operating status data. When the system platform detects that the tunneling machine's cutting head changes from the start state to the stop state, this status change signal will automatically trigger the AI ​​model to start inference calculations, at which point the entire process enters the intelligent monitoring stage. The process monitoring module will automatically generate a new process operation record and enable real-time video monitoring, allowing managers to view real-time video on-site through the system interface. During the side and roof tapping operation, the AI ​​model uses computer vision technology to identify the action characteristics of the workers. When it detects that a worker has started side and roof tapping, the model will send a message containing the identification results in real time. After receiving the message, the platform will perform anomaly judgment according to the preset rule base, specifically including judging whether there are workers tapping the side, whether there is an observer on site, whether the tapping action is standardized, and whether the tapping range covers the entire working face, among other dimensions. If any anomaly is found, the system will immediately generate corresponding alarm information.

[0023] When the AI ​​model identifies that the site has entered the temporary support operation phase, the system automatically activates a time-series reverse lookup mechanism, checking back T minutes (default setting is 30 minutes) of historical operation records to see if there are any valid records of roof and side tapping actions within this time window. If no roof and side tapping actions are detected or the operation does not meet the specifications, the system will immediately generate an alarm message "Roof and side tapping operation not performed according to procedures". During the temporary support operation process, the AI ​​model will continuously identify multiple key actions and equipment statuses, including whether the front beam is correctly raised, whether the front beam effectively contacts the top, whether the temporary support is installed in the correct position, and whether the support density meets the standards. The platform will make rule judgments on each identification result based on the real-time message data pushed by the model. Once a situation that does not comply with the operation procedures is found, the system will generate the corresponding type of alarm. For the identification of permanent support operations, when the model detects that workers have started permanent support operations, it will also trigger a time-series reverse lookup mechanism, checking back T minutes for valid records of temporary support to ensure the integrity and continuity of the operation process. During the permanent support process, the system will comprehensively monitor and identify multiple key technical indicators, such as anchor bolt installation, anchor cable installation, drilling depth of anchor bolt drilling rig, amount of anchoring agent, operation standardization of anchor bolt and anchor cable installation, whether the mixing time is sufficient, whether the upper tray position is correct, and whether the torque of tightening anchor bolt and anchor cable meets the standard. The platform will judge in real time whether each operation step meets the requirements of the operation procedure based on the message data pushed by the model. Any operation that does not meet the standard will trigger an immediate alarm.

[0024] After an alarm is generated, the system will automatically execute several linked operations: First, the alarm information will be displayed in real time on the management platform interface, including detailed information such as alarm type, alarm time, alarm location, information of the violating personnel, and description of the violation; second, based on the timestamp information of the evidence video contained in the alarm information, the system will automatically and accurately extract the corresponding time period of evidence video clips from the marked video on the streaming media server, with the start and end times of the video accurate to the second, ensuring the integrity and accuracy of the evidence; third, the alarm reminder module will automatically link with the underground broadcast system to provide real-time voice alarm reminders, including the type of violation and rectification requirements, ensuring that on-site personnel can receive safety reminders in a timely manner and rectify the situation immediately. After receiving the alarm information, the management personnel above or below ground need to handle the alarm promptly. The handling process includes steps such as viewing alarm details, watching evidence video, judging the accuracy of the alarm, filling in handling opinions, and issuing penalty decisions. If management personnel determine that an alarm is a false alarm after verification, they can mark the alarm as a false alarm and upload the false alarm case along with relevant video data and annotation information to the group's artificial intelligence training center through the alarm reporting module. The training center will use this false alarm data to retrain and optimize the AI ​​model. After the training is completed, the optimized and iterated new model will be redeployed to the edge computing devices in each mine for deployment and updates, thus forming a closed-loop optimization mechanism of "edge inference - manual review - model iteration".

[0025] To quantify the accuracy of the evaluation system's identification and the effectiveness of its alarms, this invention designs two key calculation formulas. The first is a process integrity scoring formula, used to assess the completeness of the support operation processes: Process integrity score S = Σ(Wi × Ci × Ti) / Σ(Wi × Tmax) Among them, Wi represents the weight coefficient of the i-th process. The weight of the knocking and sounding process W1 = 0.3, the weight of the temporary support process W2 = 0.3, and the weight of the permanent support process W3 = 0.4. The weight coefficient is determined according to the importance of each process to the overall safety; Ci represents the completion coefficient of the i-th process, and its value range is from 0 to 1. When the process is fully executed according to the regulations, Ci = 1; when it is partially executed, 0 < Ci < 1; when it is not executed, Ci = 0; Ti represents the actual execution time of the i-th process (unit: minute); Tmax represents the standard execution time of each process. The standard time for knocking and sounding T1max = 15 minutes, the standard time for temporary support T2max = 20 minutes, and the standard time for permanent support T3max = 25 minutes. The calculation process is as follows: First, calculate the weighted score of each process Wi×Ci×Ti. For example, if the knocking and sounding process is fully executed and takes 12 minutes, the score of this process is 0.3×1×12 = 3.6; the temporary support process is partially executed (completion degree 0.8) and takes 18 minutes, and the score is 0.3×0.8×18 = 4.32; the permanent support process is fully executed and takes 23 minutes, and the score is 0.4×1×23 = 9.2; then calculate the total score 3.6 + 4.32 + 9.2 = 17.12; finally, divide by the theoretical maximum value 0.3×15 + 0.3×20 + 0.4×25 = 20.5, and the process integrity score S = 17.12 / 20.5 = 0.835, that is, 83.5%. When the score is lower than 0.7, the system will issue a serious alarm.

[0026] The second is the calculation formula for the alarm accuracy rate, which is used to evaluate the recognition accuracy of the AI model: Alarm accuracy rate P = (TP + TN) / (TP + TN + FP + FN) × 100% Among them, TP represents the number of correctly recognized violation behaviors, that is, the number of cases where the model determines a violation and it is actually a violation; TN represents the number of correctly recognized normal behaviors, that is, the number of cases where the model determines normal and it is actually normal; FP represents the number of false alarms, that is, the number of cases where the model determines a violation but is actually normal; FN represents the number of missed alarms, that is, the number of cases where the model determines normal but is actually a violation. The specific calculation process is as follows: Assume that within a one-month operation cycle, the system has monitored 1000 support operations in total, where TP = 680 times (correctly recognize violations), TN = 250 times (correctly recognize normal), FP = 40 times (false alarms), and FN = 30 times (missed alarms). Then the alarm accuracy rate P = (680 + 250) / (680 + 250 + 40 + 30) × 100% = 930 / 1000 × 100% = 93%. When the accuracy rate is lower than 85%, the system will automatically trigger the model optimization process, requiring the group training center to retrain the model.

[0027] Example 1 Taking the actual application at a coal mine tunneling face as an example, the mine deployed the system of this invention at the 3205 tunneling face. During the morning shift on March 15, 2024, after the tunneling machine completed a cycle of cutting operations and stopped, the system automatically triggered the AI ​​model to begin reasoning. At 8:35 AM, the model detected two workers entering the working face to begin support work. However, the system, through time-series analysis, found no record of roof and side inspection operations within the previous 30 minutes. It immediately generated a Level 1 alarm for "Failure to perform roof and side inspection operations according to procedures," and simultaneously activated the underground broadcast system to issue a voice reminder: "Attention at the 3205 working face, please perform roof and side inspection operations before proceeding with support." Upon receiving the reminder, the workers immediately stopped their current operation and performed 15 minutes of roof and side inspection operations according to procedures. During this time, the system detected workers using long-handled tools to tap and inspect the roof and sides, and an observer was monitoring from a safe position. All actions complied with the requirements of the procedures. Temporary support work began at 8:50 AM. The system detected that the forward support beam was correctly raised and effectively contacted the roof, and the temporary support installation was up to standard. At 9:10 AM, the permanent support phase began. The system monitored in real time that the drilling depth of the anchor bolt drill reached the design requirement of 2.4 meters, the application of three anchoring agents met standards, and the anchor bolt installation torque reached the specified value of 250 N·m. The entire operation was standardized and orderly. Through the application of this system, the violation rate of support operations at this working face decreased from an average of 15% per month before implementation to 3%, and the process integrity score improved from 0.72 to 0.91, effectively ensuring the safety and standardization of tunneling support operations and achieving a major breakthrough in intelligent safety management of coal mine tunneling faces.

[0028] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0029] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent identification system for tunneling and support operations based on AI technology, characterized in that, include: The alarm generation module includes a model inference unit, an anomaly judgment unit, and a video capture unit. The model inference unit receives the tunneling machine status signal and triggers the AI ​​model to perform inference calculations when the tunneling machine cutting head changes from the start state to the stop state. The anomaly judgment unit performs anomaly judgment on the model inference results based on the preset operation sequence and rule base of the backing and roof inspection process, temporary support process, and permanent support process. The video capture unit captures evidence video of the corresponding time period from the streaming media server based on the timestamp of the alarm information. The alarm processing module, connected to the alarm generation module, includes an alarm display unit and a handling record unit. The alarm display unit displays the alarm type, alarm time, description of the violation, and evidence video. The handling record unit records the management personnel's handling opinions and penalty decisions. The alarm reporting module is connected to the alarm processing module. It uploads alarm data marked as model false alarms and related videos to the group's artificial intelligence training center, and receives the optimized model issued by the training center.

2. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, The alarm generation module also includes a time-series lookup unit. When a temporary support operation is detected, the time-series lookup unit searches back 30 minutes of operation records to determine if there are any records of knocking on the roof or side walls. When a permanent support operation is detected, the time-series lookup unit searches back 30 minutes of operation records to determine if there are any records of temporary support.

3. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, It also includes an alarm reminder module, which is connected to the alarm generation module. After an alarm is generated, it automatically sends a control signal to the underground broadcasting system, which then plays a voice reminder containing the type of violation and rectification requirements.

4. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, The anomaly detection unit identifies the hammering action on the sidewall, the observer's presence, and the coverage of the hammering range during the sidewall and roof hammering process; identifies the lifting status of the front beam, the contact status between the front beam and the roof slab, and the support installation position during the temporary support process; and identifies the drilling depth of the anchor bolt drilling rig, the amount of anchoring agent added, the installation operation of the anchor bolts and anchor cables, the mixing time, the tray position, and the tightening torque of the anchor bolts and anchor cables during the permanent support process.

5. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, It also includes a process integrity scoring unit, which calculates the process integrity score according to the formula S = Σ(Wi × Ci × Ti) / Σ(Wi × Tmax), where Wi is the weight coefficient of the i-th process, with the weight W1=0.3 for the knocking and jacking process, the weight W2=0.3 for the temporary support process, and the weight W3=0.4 for the permanent support process; Ci is the completion coefficient of the i-th process, ranging from 0 to 1; Ti is the actual execution time of the i-th process; and Tmax is the standard execution time of each process. A critical alarm is generated when the score is lower than 0.

7.

6. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, It also includes an alarm accuracy calculation unit, which calculates the alarm accuracy according to the formula P = (TP + TN) / (TP + TN + FP + FN) × 100%, where TP is the number of correctly identified violations, TN is the number of correctly identified normal behaviors, FP is the number of false alarms, and FN is the number of missed alarms. When the accuracy is lower than 85%, the model optimization process is automatically triggered.

7. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, The model inference unit is deployed on an edge computing device, and the AI ​​model is a computer vision model based on deep learning, which achieves multimodal recognition by recognizing the action characteristics of operators, equipment status, and object detection.

8. The intelligent identification system for tunneling and support operations based on AI technology according to claim 1, characterized in that, The video capture unit captures data with a time accuracy down to the second. The alarm information includes alarm type, alarm time, alarm location, information of the violator, description of the violation, video start time, and video end time.

Images