A safety net mechanism and intelligent control method for high-power CO2 lasers
By employing a three-layer control architecture for intelligent control of high-power pulsed CO2 lasers, combined with deep learning and safety strategy monitoring, the risk of AI runaway is mitigated, achieving high-precision and high-efficiency laser control and ensuring system safety and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHANGCHUN INST OF OPTICS FINE MECHANICS & PHYSICS CHINESE ACAD OF SCI
- Filing Date
- 2025-05-30
- Publication Date
- 2026-07-31
AI Technical Summary
High-power CO2 pulsed lasers pose a risk of AI inference model runaway in high-risk application scenarios. Traditional control methods exhibit limitations under multimodal sensor and multivariable coupling conditions, and are difficult to meet the requirements of high precision and safety controllability.
It adopts a three-layer control architecture: AI inference control layer, security network layer and basic control layer. Through deep learning and security policy monitoring, it can detect and switch to security mode in real time to ensure stable system operation.
It achieves high-precision and high-efficiency control under complex working conditions, while quickly reverting to the basic control mode at critical moments to avoid the risk of major loss of control caused by AI disorder, thereby improving system safety and reliability.
Smart Images
Figure CN120595667B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the interdisciplinary field of industrial automation and laser processing, and in particular to a safety net mechanism and intelligent control method for high-power CO2 lasers. Background Technology
[0002] CO2 pulsed lasers can generate high-power, high-energy-density lasers in the mid-infrared band, and are widely used in fields such as photolithography, metal cutting, welding, surface treatment, material drilling, and medical surgery. Compared with other types of lasers (such as fiber lasers and solid-state lasers), they have certain advantages in output efficiency and applicable material range. However, due to their high output energy and high requirements for environment, heat dissipation, optical path sealing, and optical components, they face many technical challenges and safety hazards in practical use. On the one hand, improper output power adjustment or command malfunction during high-power operation or high-load mode may lead to excessive ablation of the workpiece, damage to peripheral equipment, or even safety accidents. On the other hand, under high-power conditions, pressure components such as resonant cavities, RF power supplies, and cooling loops are often under near-limit loads. If heat dissipation is poor or flow is unstable, thermal runaway, unstable intracavity oscillations, and even component failure and damage can easily occur. In addition, in some high-speed processing scenarios, if the control system cannot adjust the laser pulse mode in time, energy fluctuations will further cause overall instability. Meanwhile, applications such as metal processing and medical intervention require extremely high precision and long-term stable operation of laser frequency, pulse width, and power RMS. This places stringent requirements on control algorithms and hardware reliability, which must meet micron-level or smaller processing precision while maintaining stable operation and safe controllability of the equipment.
[0003] With the continuous evolution of industrial and manufacturing technologies, traditional classical control methods such as finite state machines (FSMs) or PID controllers are gradually showing limitations when facing multimodal sensors, multivariable coupling, and dynamic operating conditions. Therefore, more and more organizations are attempting to introduce artificial intelligence (AI) technology into laser systems, using deep learning, expert systems, or reinforcement learning to achieve real-time parameter tuning, predictive maintenance, and anomaly detection, thereby achieving advantages such as adaptive control, predictive maintenance health management, and multi-task optimization. However, AI inference models also face the risk of runaway in the high-risk field of high-power lasers: when sensor noise accumulates, algorithmic vulnerabilities arise, or training data is insufficient, AI inference models are prone to outputting biased decisions; if critical sensors fail or communication links are interrupted, model predictions are highly likely to become disordered; even under normal operating conditions, if AI inference models encounter extreme situations that exceed the training distribution, they may make excessive or erroneous decisions due to insufficient confidence. Furthermore, industries such as military and medical have stringent regulations on safety and controllability. The opacity and potential disorder of AI inference models make it even more difficult to pass compliance reviews directly. Therefore, while achieving the high precision and efficiency brought by AI, a reliable security protection and emergency response mechanism is still needed to compensate for its uncertainties.
[0004] For high-risk, high-value applications of CO2 high-power pulsed lasers in industries, medical fields, and scientific research, a "two-layer" or "multi-layer" control architecture is commonly used to prevent potential risks caused by AI disorder: the AI control layer achieves efficient and intelligent scheduling under normal circumstances; the safety network mechanism layer monitors the output instructions of the AI inference model in real time, and once the model disorder is detected or the critical threshold is exceeded, the AI control signal is immediately cut off and the system switches to the underlying basic control (such as finite state machine FSM, PID basic control loop) to ensure that the system can still operate in a safe mode in emergency situations. Summary of the Invention
[0005] The present invention aims to solve the technical problems in the prior art by providing a safety net mechanism and intelligent control method for high-power CO2 lasers.
[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:
[0007] A safety net mechanism and intelligent control method for high-power CO2 lasers are disclosed. The applicable system includes, in sequence, a human-computer interaction unit, an AI inference control layer, a safety net layer, a basic control layer, an actuator, a high-power CO2 pulsed laser body, and a sensor unit; the AI inference control layer is also connected to the actuator and the sensor unit respectively.
[0008] in:
[0009] The human-computer interaction unit is used to provide a host computer management interface, monitor the status of each sensor in the system and the overall operation progress and status of the system in real time, and query the system operation log and alarm status.
[0010] The AI inference control layer is responsible for the preprocessing and fusion of system sensor data, and generates AI control commands in real time through deep learning models.
[0011] The security layer is used to monitor the security policies of AI output;
[0012] The basic control layer is used to perform safe takeover operations on the main body of the high-power CO2 pulsed laser. In the event of AI failure, it takes over the system and ensures that the system switches to a safe operation mode or a low-power maintenance mode.
[0013] Actuators are used to control the system's state;
[0014] The sensor unit is the system's data acquisition unit;
[0015] The safety net mechanism and intelligent control method for high-power CO2 lasers include the following steps:
[0016] Step 1: The AI inference control layer improves data quality by merging and preprocessing data and aligning multi-source data from sensor units using synchronized sampling timestamps;
[0017] Step 2: The security network layer determines in real time whether the command exceeds the security range;
[0018] Step 3: The basic control layer defines the basic working states in the state machine structure. When the safety net layer is triggered, the system switches to the corresponding state according to the received rollback command and runs in safety mode or gradually shuts down, reducing the laser output to a safe value or closing the shutter.
[0019] In the above technical solution, the AI inference control layer includes: a data preprocessing unit, a deep learning module, and AI control parameter confidence judgment;
[0020] The data preprocessing unit is used to perform preprocessing on the data collected by the sensor unit, including filtering, normalization and outlier removal. After that, data fusion and time alignment are performed, and the processed data is transmitted in batches to the deep learning module.
[0021] The deep learning module is used to receive data processed by the data preprocessing unit, perform real-time inference, and output AI control commands;
[0022] AI control parameter confidence assessment is used for diagnosis through model evaluation functions. If the confidence is too low or abnormal model inference is detected, a fault warning is issued.
[0023] In the above technical solution, step 1 specifically includes:
[0024] The AI inference control layer improves data quality by fusion and preprocessing data, using synchronous sampling timestamps to align multi-source data from various sensors, and employing filtering and / or outlier removal methods.
[0025] The inference engine, based on deep learning models, expert rules, and fuzzy logic, makes decisions on power settings, pulse frequency, cooling adjustment, and / or mechanical platform motion trajectory based on the current operating conditions.
[0026] When the system detects that a critical sensor is offline, data is abnormal, or the confidence of the model prediction drops sharply, the AI inference control layer actively marks the possible disorder and reports it to the safety net layer.
[0027] In the above technical solution, the security network layer includes: a security policy judgment module and a dynamic update configuration module; the security policy judgment module is equipped with a security policy library and a rule engine; the security policy library is used to store various thresholds and rule entries; the rule engine is used to receive AI instructions and current sensor data, compare them with the security policy library, and output the results.
[0028] The security policy judgment module is used to input the received instructions output by the AI inference control layer into a pre-set security policy library for judgment. The security of the AI control instructions is realized through fixed threshold judgment and dynamic fuzzy rule verification by the rule engine.
[0029] In the above technical solution, the specific steps of the security policy judgment module in determining the security of AI control commands through fixed threshold judgment and dynamic fuzzy rule verification are as follows:
[0030] If the rules engine determines that everything is normal, the AI instruction will be sent to the execution agency to continue execution;
[0031] If the rules engine determines that there is an anomaly, it will trigger an anomaly, and the security policy judgment module will block the AI command and write the anomaly type to the log.
[0032] When multiple anomalies or serious fault alarms occur in a short period of time, the system control is transferred from the AI inference control layer to the basic control layer, so that the system enters a safe operation mode or a low-power maintenance mode.
[0033] In the above technical solution, step 2 specifically involves the security network layer determining in real time whether an instruction exceeds the safe range through threshold comparison, timing pattern recognition, and / or hardware fault linkage.
[0034] In the above technical solution, step 3 specifically includes:
[0035] Determine the current operating state of the system and switch the state machine in the basic control layer to the corresponding state node;
[0036] If the checks and inferences within the state node can be satisfied after entering the state machine, the system will operate safely through the state machine and continue to send control commands to the actuators. If the relevant parameters cannot meet the threshold judgment after entering the state node, the system will enter a low-power operation state to wait for maintenance. After maintenance is completed and the system passes the self-test, it will return to the AI inference control layer to continue working.
[0037] The present invention has the following beneficial effects:
[0038] This invention presents a safety net mechanism and intelligent control method for high-power CO2 lasers. Through multi-sensor fusion and deep monitoring, it simultaneously reviews power commands, device temperature, gas flow rate, cooling loop pressure, RF power supply status, seed light oscillation mode, and laser amplifier status. A flexible and configurable safety policy library sets corresponding safety limits and thresholds for different application processes and environmental requirements. Upon detecting significant anomalies, it can switch to basic control mode in a very short time and automatically record fault trigger data, enabling rapid post-event tracing and algorithm improvement. In this way, while fully leveraging the real-time adaptive and high-precision control advantages of AI inference models, this invention successfully resolves the technical challenges of complex control nodes, susceptibility to runaway control, and extremely high failure costs in large-scale laser equipment. It provides a more robust and safer intelligent control solution for high-power pulsed lasers, significantly improving system safety and engineering practicality. Attached Figure Description
[0039] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0040] Figure 1 This is a schematic diagram of the overall system architecture.
[0041] Figure 2 This is a schematic diagram of the multi-sensor AI inference control layer process for a high-power CO2 pulsed laser.
[0042] Figure 3 This is a schematic diagram of the workflow of the safety net mechanism.
[0043] Figure 4 This is a schematic diagram of the basic control layer operation.
[0044] Figure 5 This is a timing diagram for a system exception.
[0045] Figure 6 This is a schematic diagram of the system operation monitoring and fault auditing interface.
[0046] The reference numerals in the figure are:
[0047] 101-Human-Computer Interaction Unit; 102-AI Inference Control Layer; 103-Safety Net Layer; 104-Basic Control Layer; 105-Actuator; 106-High-Power CO2 Pulsed Laser Main Body; 107-Sensor Unit;
[0048] 201 - Data Preprocessing Unit; 202 - Deep Learning Module; 203 - AI Control Parameter Confidence Judgment;
[0049] 301 - Dynamic configuration update module; 302 - Security policy judgment module; 303 - Security policy library; 304 - Rule engine; 305 - Abnormal trigger; 306 - Multiple abnormalities or serious fault alarms. Detailed Implementation
[0050] The inventive concept of this invention is as follows:
[0051] This invention aims to establish a three-layer architecture for high-power CO2 pulsed lasers, consisting of an AI inference control layer, a safety net layer for monitoring, and a basic control layer as a fallback. This architecture ensures high precision and efficiency under complex operating conditions while also allowing for rapid revert to the basic control mode at critical moments, thus avoiding significant risks of loss of control due to AI malfunction.
[0052] To achieve this goal, the system introduces advanced deep learning algorithms to improve its adaptive capabilities and control precision. When it detects disordered signs such as sensor failure, abnormal AI algorithm inference, or network interruption, it relies on a safety net mechanism for real-time monitoring and rapid intervention, making the entire laser operation more controllable and reliable.
[0053] At the AI inference control level, this invention acquires key parameters such as device temperature, intracavity gas flow rate, optical power, beam quality, cooling fan bearing temperature, and cooling water temperature and flow rate through multi-sensor fusion. Optimal control commands (such as laser power and pulse frequency adjustment, and shutter opening mode) are then generated via deep learning and an expert system. With a self-diagnostic and feedback mechanism, the AI inference control layer monitors its own inference errors and input distribution changes. Upon detecting an anomaly, it proactively alerts the safety network. The safety network layer, through a safety policy library and rule engine, periodically (less than 20ms) compares the AI inference commands with real-time data. If a control command exceeds a safety threshold or the inference confidence level falls below a set threshold, the system immediately blocks the control command within milliseconds to seconds and triggers a rollback and recalculation mechanism, rewriting the AI inference control layer's output to ensure that the issued control output meets safety control standards. When the number of abnormal occurrences reaches a preset value or a serious alarm is triggered, control authority is transferred to the basic control layer, where a finite state machine then controls the system to ensure the laser enters a safe state or operates at low power pending maintenance.
[0054] In the basic control layer, this invention employs relatively simple and stable control logic such as finite state machines or PID loops, with built-in safety limiting and protection measures, such as emergency power-off and enhanced heat dissipation. Once the safety net triggers a backoff, this layer can maintain the laser power at an extremely low level or smoothly shut down, thereby preventing sudden energy jumps from damaging the cavity or surrounding equipment. When the cause of the fault is eliminated or safety is confirmed manually, the basic control layer allows control to be transferred back to the AI inference control layer, achieving a seamless switch between efficiency and safety.
[0055] Through this multi-layered architecture, the present invention has achieved significant improvements in security protection capabilities, system performance, scalability, fault auditing, and continuous optimization: dual monitoring inside and outside the AI inference model ensures that instantaneous loss of control scenarios are intercepted in time; in normal mode, it maximizes the advantages of AI in big data and self-learning; the basic control layer provides a stable backup solution to avoid downtime or serious damage caused by extreme operating conditions; the system's strategy and modular design also enable it to adapt to stricter regulations and diverse equipment requirements, and the fault audit records help maintenance personnel quickly trace the source and improve the model.
[0056] In summary, this invention provides powerful intelligent control and real-time safety backoff capabilities for high-power CO2 laser applications, achieving a significant improvement in both efficiency and stability during high-risk, high-precision operations. Its core concept is also applicable to other high-power or highly complex equipment (such as fiber lasers, medical laser surgery systems, and autonomous vehicles), and has broad applicability to industries that require a balance between large-scale production efficiency, stringent safety requirements, and advanced intelligent decision-making capabilities.
[0057] The present invention will now be described in detail with reference to the accompanying drawings.
[0058] The safety net mechanism and intelligent control method for high-power CO2 lasers of the present invention are designed specifically around the overall needs of the main body of CO2 pulsed laser and multi-layer control. The functions are divided into "AI inference control layer 102", "safety net layer 103" and "basic control layer 104", which are interconnected.
[0059] like Figure 1 As shown, the system applicable to the safety net mechanism and intelligent control method for high-power CO2 lasers of the present invention includes, in sequence: a human-computer interaction unit 101, an AI inference control layer 102, a safety net layer 103, a basic control layer 104, an actuator 105, a high-power CO2 pulsed laser body 106, and a sensor unit 107; the AI inference control layer 102 is also connected to the actuator 105 and the sensor unit 107 respectively; wherein:
[0060] The human-machine interface unit 101 serves as the primary means and interface for operators to monitor and control the system's status. It provides a host computer management interface, allowing real-time monitoring of the status of each sensor and the overall system operation progress and status. It also enables querying system operation logs and alarm information. Furthermore, manual confirmation is required before critical steps are performed. The human-machine interface unit 101 also allows for manual intervention in the system status.
[0061] The AI inference control layer 102 is responsible for the preprocessing and fusion of system sensor data, and generates AI control commands in real time through deep learning models.
[0062] Security layer 103 is used to monitor the security policies of AI output, including threshold management, fault detection algorithms, and rollback trigger logic.
[0063] The basic control layer 104, with a finite state machine as its core, is a more stable and secure control method than AI control commands. It performs a "safe takeover" operation on the main body 106 of the high-power CO2 pulsed laser. When the AI fails to control, it can take over the system in time and ensure that the system switches to a safe operation mode or a low-power maintenance mode.
[0064] The actuator 105 is the main means of controlling the state of the system, including: power supply, signal modulation frequency, power adjustment, shutter braking, and mechanical platform control. In this specific embodiment, the actuator 105 is a laser.
[0065] The main body 106 of the high-power CO2 pulsed laser includes: a seed light system, a laser transmission and amplification link system, a laser amplifier, a cooling system, a gas mixing and distribution unit, and a radio frequency power supply.
[0066] The sensor unit 107 includes all the acquisition modules in the system, specifically including gas temperature, cooling water temperature, sensor temperature, gas pressure, cooling water pressure, beam transmission cavity pressure, gas flow rate, cooling water flow rate, component vibration, voltage, current, electrical power, laser power, laser spot image, and all other acquisition units.
[0067] The safety net mechanism and intelligent control method for high-power CO2 lasers of the present invention include the following steps:
[0068] Step 1: The AI inference control layer 102 aligns the multi-source data from the sensor unit 107 through data fusion and preprocessing using synchronous sampling timestamps; the inference engine makes decisions on power settings, pulse frequency, cooling adjustment and / or mechanical platform motion trajectory based on the current operating conditions; when the system detects that a key sensor is offline, data is abnormal or the confidence of the model prediction drops sharply, it actively marks it as "possibly out of order" and reports it to the safety net layer 103.
[0069] Specifically:
[0070] The AI inference control layer 102 aligns multi-source data from various sensors using synchronized sampling timestamps through data fusion and preprocessing, and improves data quality through filtering and outlier removal. The inference engine, based on deep learning models, expert rules, and fuzzy logic, makes decisions regarding power settings, pulse frequency, cooling adjustments, and the mechanical platform's motion trajectory, including the shutter. When the system detects a critical sensor going offline, data anomalies, or a sharp drop in model prediction confidence, the AI inference control layer 102 proactively marks the system as "potentially out of order" and reports it to the safety net layer 103 to prevent potential violations.
[0071] like Figure 2 As shown, the AI inference control layer 102 includes: a data preprocessing unit 201, a deep learning module 202, and an AI control parameter confidence judgment 203. The sensor unit 107 contains multiple sensing modules such as temperature, pressure, beam quality, and gas flow rate, specifically... Figure 1 Sensor unit 107 in the middle.
[0072] The data preprocessing unit 201 is used to preprocess the data collected by the sensor unit 107, including filtering, normalization and outlier removal, and then perform data fusion and time alignment, and transmit the processed data in batches to the deep learning module 202.
[0073] The deep learning module 202 is a pre-trained inference model that receives data processed by the data preprocessing unit, performs real-time inference, and outputs AI control commands such as modulation signal frequency, laser power, pulse frequency modulation parameters, and shutter control commands.
[0074] AI control parameter confidence judgment 203 is used for diagnosis through model evaluation function. If the confidence is too low or abnormal model inference is detected, a fault warning will be issued.
[0075] The security layer 103 is used to review the final instructions of the AI before they are actually executed. Only after the review is passed can they be executed.
[0076] Step 2: Security layer 103 performs real-time review of AI control output, including security interval verification, confidence assessment and historical behavior consistency check, to ensure that all issued instructions comply with the system security policy library;
[0077] Specifically:
[0078] Safety layer 103 uses threshold comparison (such as maximum power, maximum allowable temperature, etc.), timing pattern recognition (detecting sudden changes in pulse commands), and hardware fault linkage (sensor offline or pressure lower limit) to determine in real time whether the command exceeds the safe range.
[0079] Each main loop can calculate the model confidence score (maximum probability, entropy index) or the difference between the feature vector and the historical mean. If an abnormal increase is detected, it is labeled as "potentially out of order" and actively reported to the safety network so that the safety network can make a judgment based on multi-source information. Once an anomaly is determined, the safety network layer 103 immediately blocks or rewrites the current AI instruction and sends a "rollback command" to the basic control layer 104, while triggering alarms and fault log management. For instructions that pass normally, the safety network layer 103 will also mark them as "passed" and record them in the log, ensuring that the system has a traceable record of every decision, providing data support for subsequent algorithm iterations or fault troubleshooting.
[0080] The security layer 103 is an important defense against AI runaway or malfunction in this invention. Its core lies in a configurable security policy library 303 and a variety of anomaly detection algorithms.
[0081] like Figure 3 As shown, the process within the security layer 103 begins with receiving the AI instructions output by the AI inference control layer 102. The security layer 103 mainly includes a security policy judgment module 302 and a dynamic configuration update module 301. The security policy judgment module 302 is primarily used to input the received instructions from the AI inference control layer 102 into a pre-set security policy library 303 for judgment, and uses a rule engine 304 to implement fixed threshold judgment and dynamic fuzzy rule verification to ensure the security of the AI control instructions.
[0082] If the rule engine 304 determines that the operation is normal, the AI instruction will be sent to the execution mechanism 105 to continue execution; if the rule engine determines that the operation is abnormal, an abnormality trigger 305 will be triggered, the security policy judgment module 302 will block the AI instruction and write the abnormality type to the log. If there are multiple abnormalities in a short period of time or a serious fault alarm 306 occurs, the system control will be transferred from the AI inference control layer 102 to the basic control layer 104, so that the system enters a safe operation mode or a low-power maintenance mode, etc.
[0083] The security policy library 303 is used to store various thresholds (such as laser power limit, pulse jump rate, safe temperature range, etc.) and rule entries (such as "trigger an alarm if the temperature is continuously higher than X℃ for more than Y seconds").
[0084] The rule engine 304 receives AI instructions and current sensor data, compares them with the security policy library 303, and outputs a "pass" or "abnormal" result.
[0085] The exception trigger 305 is responsible for sending the fault type and severity to the outside world, as well as sending a rollback signal to the basic control layer 104.
[0086] The dynamic update configuration module 301 is used to modify the policies in the security policy library 303. It can dynamically update the rules during system runtime, making the judgment rules more flexible. At the same time, it allows for hot updates of security policies or adjustments of rule priorities based on different process requirements or equipment status.
[0087] Step 3: The basic control layer 104 defines multiple operating states through the built-in state machine structure. When it receives a "back-off command" from the safety network layer 103, the system can immediately switch from the current working state to the preset safety mode or shutdown state, and perform safety actions such as reducing the laser output power to the safety threshold and closing the optical shutter mechanism to prevent accidental laser firing or system malfunction.
[0088] Specifically:
[0089] The basic control layer 104 defines several basic states such as "standby", "warm-up", "operation", "safe mode" and "fault" in the state machine structure. When the safety net layer 103 is triggered, the system can quickly switch to the safe mode or gradually shut down, reduce the laser output to a safe value or close the optical shutter.
[0090] In this invention, the basic control layer 104 plays the final role of providing backup and safety protection. At this time, the basic control layer 104 no longer receives high-level scheduling instructions from the AI inference control layer 102 to avoid further risks caused by disorder. After the fault is resolved and manually confirmed, the basic control layer 104 can gradually return from the safe mode to the normal operating state and re-grant parameter tuning permissions to the AI inference control layer 102; during the recovery process, a power ramp-up mechanism is set to prevent sudden energy surges.
[0091] like Figure 4 As shown, when the safety net layer 103 is triggered, the system's control authority will be handed over to the basic control layer 104. First, it will determine the current operating state of the system and switch the state machine in the basic control layer 104 to the corresponding state node. If the checks and inferences within the state node are satisfied after entering the state machine, the system can automatically control safe operation and continue sending control commands to the actuator 105. If the relevant parameters fail to meet the threshold judgment after entering the state node, the system will enter a low-power operation state to await maintenance. After maintenance is completed and the system passes self-test, it can return to the AI inference control layer 102 to continue working.
[0092] The timing details of the system applicable to the safety net mechanism and intelligent control method for high-power CO2 lasers of the present invention under abnormal triggering are as follows:
[0093] like Figure 5 As shown, the entire system can be abstracted into an AI inference control layer 102, a security network layer 103, a basic control layer 104, and a laser actuator 105.
[0094] Under normal circumstances, at time T1, the system generates AI control instructions via the AI inference control layer 102 and sends them to the security network layer 103 for review. If no abnormal trigger occurs, the instructions are filtered (at time T2), and the AI control instructions are forwarded to the execution mechanism 105. The execution mechanism 105 will adjust the corresponding parameters and perform actions according to the corresponding control instructions (at time T3).
[0095] In abnormal situations, such as during the operation from time T4 to T5, if the AI inference control layer 102 generates a new control command, but an anomaly occurs during the review process by the security network layer 103, the current AI command will be reverted. After time T6, if the AI control command is repeatedly intercepted and reverted, or if a serious alarm occurs, the security network layer 103 will be triggered to transfer control authority to the basic control layer 104. After the basic control layer 104 takes over the system, if it can still operate normally, it will enter safe mode to continue operating; if it still cannot operate normally, it will enter low-power maintenance mode at time T7. At the same time, corresponding commands are issued to the actuator 105. If the fault is serious, a complete shutdown or increased cooling circulation can be performed. At time T8, the security network layer 103 records the fault log and alarms the human-machine interaction unit 101. At time T9, after maintenance personnel or automatic detection confirm that the system has recovered, it will switch back to AI control mode.
[0096] The present invention relates to a system operation monitoring and fault auditing interface for a safety net mechanism and intelligent control method applicable to high-power CO2 lasers, such as... Figure 6 As shown: Key operating parameters (power, temperature, AI model confidence level, current process mode) and safety net status (whether it has been triggered, cause of failure, recording time) can be visually monitored on the host computer or operation panel; after a failure occurs, the data trajectory can be replayed and the specific scenario of the safety net being triggered can be viewed, which is convenient for operators to diagnose and correct the model.
[0097] To illustrate the actual operation of the system more intuitively, this invention provides typical scenarios of high-speed cutting and AI disorder failure rollback. In the high-speed cutting scenario, the AI inference control layer 102 adaptively adjusts parameters based on the plate thickness and path information, and, in conjunction with regular monitoring by the safety net, ensures that the system maintains a high-power, high-precision cutting state without abnormal triggering; efficiency is improved by approximately 15% compared to traditional PID. In the AI disorder and failure rollback scenario, when a sensor provides erroneous feedback causing the AI to issue an over-limit power command, the safety net layer determines that it has crossed the line and quickly interrupts the output. Simultaneously, it causes the basic control layer 104 to enter a safe mode. Finally, the fault log is used to pinpoint the problem, helping maintenance personnel quickly replace the sensor and gradually restore normal operation.
[0098] Based on a multi-layered control design, this invention achieves a good balance between system efficiency and safety: the AI inference control layer 102 is responsible for highly flexible adaptive control, while the safety net layer 103 intercepts risks at an early stage, greatly reducing the harm of loss of control and downtime losses. For different operating conditions or different process requirements, rapid migration and deployment can be achieved simply by adapting the safety policy library 303 and threshold configuration. Simultaneously, both the AI inference control layer 102 and the safety policies support online updates, allowing AI engineers to iteratively improve algorithms and strategies based on historical fault records, significantly enhancing the overall system's self-learning and self-evolution capabilities.
[0099] In summary, this invention provides a highly efficient, precise, and secure intelligent control system for CO2 pulsed lasers through a multi-layered collaborative approach: an AI inference control layer 102, a security network layer 103 for monitoring, and a basic control layer 104 as a safety net. Unlike traditional single finite state machines or redundant hardware solutions, this invention significantly expands the adaptability of CO2 pulsed lasers under dynamic operating conditions while maintaining minimum safe and feasible control, and utilizes real-time anomaly detection and log auditing to form a closed-loop optimization. Its application scope can be extended to other types of lasers (fiber optics, solid-state, etc.) and high-risk industrial equipment such as lithography machines and plasma processing equipment, providing a feasible path that balances safety and efficiency for a wider range of intelligent manufacturing fields.
[0100] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A safety net mechanism and intelligent control method for high-power CO2 lasers, characterized in that, The applicable system includes, in sequence: a human-computer interaction unit (101), an AI inference control layer (102), a security network layer (103), a basic control layer (104), an actuator (105), a high-power CO2 pulsed laser body (106), and a sensor unit (107); the AI inference control layer (102) is also connected to the actuator (105) and the sensor unit (107) respectively; in: The human-computer interaction unit (101) is used to provide a host computer management interface to monitor the status of each sensor in the system and the overall operation progress and status of the system in real time, and to query the system operation log and alarm status. The AI inference control layer (102) is responsible for the preprocessing and fusion of system sensor data, and generates AI control commands in real time through deep learning models; The security layer (103) is used to monitor the security policies of the AI output; The basic control layer (104) is used to perform a safe takeover operation on the main body (106) of the high-power CO2 pulsed laser. In the event of AI failure, it takes over the system and ensures that the system switches to a safe operation mode or a low-power maintenance mode. The actuator (105) is used to control the system state; The sensor unit (107) is the system's data acquisition unit; The safety net mechanism and intelligent control method for high-power CO2 lasers include the following steps: Step 1: The AI inference control layer (102) improves data quality by aligning multi-source data from sensor unit (107) through data fusion and preprocessing and using synchronous sampling timestamps; Step 2: The security network layer (103) determines in real time whether the command exceeds the security range; Step 3: The basic control layer (104) defines the basic working state in the state machine structure. When the safety net layer (103) is triggered, the system switches to the corresponding state according to the received rollback command and runs in safety mode or gradually stops, reducing the laser output to a safe value or closing the shutter. The AI inference control layer (102) includes: a data preprocessing unit (201), a deep learning module (202), and an AI control parameter confidence judgment (203). The data preprocessing unit (201) is used to perform preprocessing on the data collected by the sensor unit (107), including filtering, normalization and outlier removal, and then perform data fusion and time alignment, and transmit the processed data in batches to the deep learning module (202). The deep learning module (202) is used to receive the data processed by the data preprocessing unit (201), perform real-time inference, and output AI control commands. AI control parameter confidence judgment (203) is used to diagnose through the model evaluation function. If the confidence is too low or abnormal model inference is detected, a fault warning is issued.
2. The safety net mechanism and intelligent control method for high-power CO2 lasers according to claim 1, characterized in that, Step 1 is as follows: The AI inference control layer (102) improves data quality by using data fusion and preprocessing, aligning multi-source data from various sensors with synchronous sampling timestamps, and using filtering and / or outlier removal methods. The inference engine, based on deep learning models, expert rules, and fuzzy logic, makes decisions on power settings, pulse frequency, cooling adjustment, and / or mechanical platform motion trajectory based on the current operating conditions. When the system detects that a critical sensor is offline, data is abnormal, or the confidence of the model prediction drops sharply, the AI inference control layer (102) actively marks the possible disorder and reports it to the safety net layer (103).
3. The safety net mechanism and intelligent control method for high-power CO2 lasers according to claim 1, characterized in that, The security network layer (103) includes: a security policy judgment module (302) and a dynamic update configuration module (301); the security policy judgment module (302) is equipped with a security policy library (303) and a rule engine (304); the security policy library (303) is used to store various thresholds and rule entries; the rule engine (304) is used to receive AI instructions and current sensor data, compare them with the security policy library (303), and output the results; The security policy judgment module (302) is used to input the received instructions output by the AI inference control layer (102) into the pre-set security policy library (303) for judgment, and realize the security of AI control instructions through the rule engine (304) with fixed threshold judgment and dynamic fuzzy rule verification.
4. The safety net mechanism and intelligent control method for high-power CO2 lasers according to claim 3, characterized in that, The specific steps of the security policy judgment module (302) in performing fixed threshold judgment and dynamic fuzzy rule verification of the security of AI control commands are as follows: If the rules engine (304) determines that the operation is normal, the AI instruction will be sent to the execution mechanism (105) to continue execution; If the rule engine (304) determines that there is an exception, then the exception is triggered (305), the security policy judgment module (302) blocks the AI instruction and writes the exception type to the log; When multiple abnormalities or serious fault alarms (306) occur in a short period of time, the system control is transferred from the AI inference control layer (102) to the basic control layer (104) to take over, so that the system enters a safe operation mode or a low-power maintenance mode.
5. The safety net mechanism and intelligent control method for high-power CO2 lasers according to claim 1, characterized in that, Step 2 specifically involves the security network layer (103) determining in real time whether an instruction exceeds the safe range through threshold comparison, timing pattern recognition, and / or hardware fault linkage.
6. The safety net mechanism and intelligent control method for high-power CO2 lasers according to claim 1, characterized in that, Step 3 specifically involves: Determine the current working state of the system and switch the state machine in the basic control layer (104) to the corresponding state node; If the checks and inferences within the state node can be satisfied after entering the state machine, the system will operate safely through the state machine and continue to send control commands to the actuator (105); if the relevant parameters cannot meet the threshold judgment after entering the state node, the system will enter a low-power operation state to wait for maintenance. After maintenance is completed and the system passes the self-test, the system will return to the AI inference control layer (102) to continue working.