Laser Failure Prediction Methods, Equipment and Media

Abstract

This application relates to the field of laser fault diagnosis and safety control, and provides a method, device, and medium for predicting laser fault melting. The method includes: configuring a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor to collect multi-dimensional fault precursor parameters to identify fault precursor trends; constructing a multi-parameter time-series prediction model based on the laser melting fault mechanism, learning the fault evolution law through offline training, processing the parameters after feature extraction and fusion using the trained multi-parameter time-series prediction model, and outputting the melting risk level and expected melting time; generating a corresponding early warning signal based on the melting risk level output by the prediction model, and generating push processing suggestions; collecting new operating data to incrementally train the multi-parameter time-series prediction model to adapt to changes in fault evolution laws caused by laser component aging, changes in the operating environment, etc.

Description

Technical Field

[0001] This application relates to the field of laser fault diagnosis and safety control technology, and in particular to a method, device and medium for predicting laser fault melting. Background Technology

[0002] In the field of laser fault diagnosis and safety control technology, existing laser fault protection technologies mainly rely on passive protection methods such as overcurrent protection, overtemperature protection, and overvoltage protection. For example, triggering power-off by detecting whether the drive current exceeds the rated threshold has significant lag. Protection is only triggered when a fault has occurred (such as current exceeding the threshold or temperature exceeding the upper limit), making it impossible to predict the risk of melting and making it very easy to cause irreversible damage to the core components inside the laser (such as the pump source and gain medium).

[0003] Meanwhile, existing technologies generally suffer from the limitation of monitoring only a single parameter. Fuse failures are typically caused by a combination of factors, including current overload, abnormal temperature, and component aging. However, existing technologies only monitor one or a few parameters, making it difficult to comprehensively capture early signs of failure, resulting in a high false alarm and false negative rate (approximately 15%-20%). Furthermore, existing technologies lack the ability to identify fault evolution patterns and cannot predict the timing of fuse failure based on parameter change trends. This fails to provide maintenance personnel with sufficient emergency response time, severely impacting the stability of continuous equipment operation.

[0004] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0005] This application provides a method for predicting laser fault melting, aiming to solve the problem that in the field of laser fault diagnosis and safety control technology, existing laser fault protection technologies mainly rely on passive protection methods such as overcurrent protection, overtemperature protection, and overvoltage protection, which have significant lag. They can only trigger protection when a fault has occurred (such as current exceeding the threshold or temperature exceeding the upper limit), and cannot predict melting risks in advance, which can easily lead to irreversible damage to the core components inside the laser (such as pump source and gain medium).

[0006] In a first aspect, embodiments of this application provide a method for predicting the failure and melting of a laser, the method comprising: The system is equipped with a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor. It collects multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude. It extracts time-series features from the real-time collected multi-dimensional parameters and eliminates noise interference from single parameters through feature fusion to identify fault precursor trends. A multi-parameter time-series prediction model is constructed by using a long short-term memory network algorithm and combining it with the fault mechanism of laser fuse. The model learns the fault evolution law through offline training, and processes the parameters after feature extraction and fusion using the trained multi-parameter time-series prediction model to output the fuse risk level and the expected fuse time. Based on the circuit breaker risk level output by the prediction model, a corresponding early warning signal is generated, and a push notification for handling suggestions is generated. After each preset running time, new running data is collected, and the multi-parameter time series prediction model is incrementally trained to adapt to changes in the fault evolution pattern caused by laser component aging, changes in the operating environment, etc.

[0007] In some embodiments, the configuration of the drive current sensor, laser cavity temperature sensor, pump source voltage detection module, component loss monitoring module, and vibration sensor to collect multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude, includes: configuring the drive current sensor to collect drive current within a preset current range in real time; configuring the laser cavity temperature sensor to collect cavity temperature within a preset temperature range in real time; configuring the pump source voltage detection module to collect pump source input voltage within a preset voltage range in real time; configuring the component loss monitoring module to monitor the transmittance of the gain medium and the luminous efficiency of the pump source in real time to obtain component loss parameters; and configuring the vibration sensor to collect operating vibration amplitude within a preset amplitude range in real time, wherein the sampling frequency of each sensor is set to a preset frequency range.

[0008] In some embodiments, the step of extracting time-series features from real-time acquired multi-dimensional parameters and eliminating single-parameter noise interference through feature fusion to identify fault precursor trends includes: processing real-time acquired multi-dimensional parameters using a sliding window technique, calculating time-series features including the mean, rate of change, and peak factor of the parameters within the sliding window; inputting the extracted time-series features of each dimension into a preset feature fusion module, and performing weighted combination or dimensionality reduction processing on the extracted time-series features through a preset algorithm to eliminate single-parameter noise interference, so as to accurately identify the changing trends of fault precursors.

[0009] In some embodiments, the use of a Long Short-Term Memory (LSTM) network algorithm, combined with the laser fracturing fault mechanism, to construct a multi-parameter time-series prediction model includes: analyzing the laser fracturing fault mechanism, such as current overload, temperature rise, accelerated component aging, and the fracturing process; constructing a multi-parameter time-series prediction model based on the LTM network algorithm, comprising an input layer, a hidden layer, and an output layer, wherein the input layer receives the time-series features of multi-dimensional parameters, the hidden layer learns the time-series evolution features of the parameters, and the output layer outputs the fracturing risk-related results; and determining the network structure parameters of the multi-parameter time-series prediction model based on the fracturing risk-related results, including the number of hidden layers and the number of neurons, so that the multi-parameter time-series prediction model can conform to the progressive development law of fracturing faults.

[0010] In some embodiments, the step of learning fault evolution patterns through offline training and processing the parameters after feature extraction and fusion using a trained multi-parameter time series prediction model to output the circuit breaker risk level and the expected circuit breaker time includes: collecting operational data covering all stages of normal operation, minor anomalies, severe anomalies, and circuit breaker precursors as training samples, and including data from various scenarios such as single-factor gradual change and multi-factor sudden change; performing data preprocessing on the training samples, including time series alignment and outlier removal; inputting the preprocessed training samples into the constructed multi-parameter time series prediction model for offline training, enabling the multi-parameter time series prediction model to learn fault evolution patterns, and training until the prediction accuracy of the multi-parameter time series prediction model is greater than or equal to a preset accuracy and the early warning time is greater than or equal to a preset threshold; inputting the real-time parameters after feature extraction and fusion into the trained multi-parameter time series prediction model, and the multi-parameter time series prediction model outputs the circuit breaker risk level and the expected circuit breaker time after processing.

[0011] In some embodiments, generating a corresponding early warning signal based on the circuit breaker risk level output by the prediction model includes: classifying the circuit breaker risk level into low risk, medium risk, high risk, and emergency risk; generating an early warning signal with only remote notification when the output circuit breaker risk level is low risk; generating an early warning signal with audible and visual warning and remote push of early warning information when the output circuit breaker risk level is medium risk; generating an early warning signal with audible and visual warning, remote notification, and triggering power reduction operation when the output circuit breaker risk level is high risk; and generating an early warning signal with audible and visual warning, remote emergency notification, and immediately triggering power outage protection when the output circuit breaker risk level is emergency risk.

[0012] In some embodiments, generating push processing suggestions includes: establishing a processing suggestion database corresponding to the circuit breaker risk level, wherein the processing suggestion database stores specific processing measures for different risk levels; retrieving the corresponding specific processing measures from the processing suggestion database based on the circuit breaker risk level output by the multi-parameter time series prediction model, and pushing the specific processing measures to the operation and maintenance terminal, wherein the specific processing measures include, but are not limited to, adjusting the drive current, checking the heat dissipation of the laser, and checking the aging degree of components, one or more of these.

[0013] In some embodiments, the step of collecting new operational data and incrementally training a multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment includes: setting a preset operating time for each system cycle as an incremental training cycle; automatically collecting new operational data within each incremental training cycle, including data from normal and abnormal operating conditions; performing the same data preprocessing operations as during offline training on the new operational data, inputting it into the multi-parameter time-series prediction model for incremental training, updating the model parameters corresponding to the multi-parameter time-series prediction model, and enabling the multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment.

[0014] Secondly, this application provides a computer device including a processor, a memory, and a computer program stored in the memory and executable by the processor. The memory stores a strategy model, and the computer program, when executed by the processor, implements the method provided in any embodiment of this application.

[0015] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the method provided in any embodiment of this application.

[0016] This application utilizes multi-dimensional parameter monitoring and time-series prediction models to predict fuse failures in advance, upgrading passive protection to proactive early warning, avoiding irreversible damage to core components, and reducing maintenance costs. Through multi-dimensional parameter fusion and time-series feature analysis, the fault prediction accuracy is ≥98.5%, with a false alarm rate ≤3% and a false negative rate ≤1%, significantly outperforming existing single-parameter monitoring technologies. It can effectively identify various fuse failure precursors, such as gradual changes due to single factors and sudden changes due to multi-factor coupling, and its applicability covers different types of lasers, including industrial and laboratory-grade lasers. Tiered early warning provides maintenance personnel with sufficient processing time (≥30 seconds) and pushes processing suggestions in conjunction with the system, lowering the maintenance threshold. The model is automatically incrementally trained every preset running time to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment, ensuring long-term model adaptability and reducing maintenance costs. The monitoring module and algorithm unit can be added to existing laser control systems without replacing the entire equipment, resulting in low modification costs and wide adaptability.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart illustrating the steps of a laser fault fusing prediction method according to an embodiment of this application; Figure 2 This is an overall prediction and early warning block diagram corresponding to the laser fault melting prediction provided in one embodiment of this application; Figure 3 This is a flowchart of the LSTM prediction model provided in one embodiment of this application; Figure 4 This is a comparison diagram of the fault prediction effect provided in one embodiment of this application; Figure 5 This is a schematic block diagram of a laser fault fuse prediction system provided in one embodiment of this application; Figure 6 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0020] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0023] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0024] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] In the field of laser fault diagnosis and safety control technology, existing laser fault protection technologies mainly rely on passive protection methods such as overcurrent protection, overtemperature protection, and overvoltage protection. For example, triggering power-off by detecting whether the drive current exceeds the rated threshold has significant lag. Protection is only triggered when a fault has occurred (such as current exceeding the threshold or temperature exceeding the upper limit), making it impossible to predict the risk of melting and making it very easy to cause irreversible damage to the core components inside the laser (such as the pump source and gain medium).

[0027] Meanwhile, existing technologies generally suffer from the limitation of monitoring only a single parameter. Fuse failures are typically caused by a combination of factors, including current overload, abnormal temperature, and component aging. However, existing technologies only monitor one or a few parameters, making it difficult to comprehensively capture early signs of failure, resulting in a high false alarm and false negative rate (approximately 15%-20%). Furthermore, existing technologies lack the ability to identify fault evolution patterns and cannot predict the timing of fuse failure based on parameter change trends. This fails to provide maintenance personnel with sufficient emergency response time, severely impacting the stability of continuous equipment operation.

[0028] Therefore, a method is urgently needed to solve at least one of the above problems.

[0029] To solve the above problem, please refer to Figure 1 This application provides a method for predicting the failure and melting of a laser, applied to computer equipment. The computer equipment can be deployed on a single server or a server cluster. It can also be deployed on handheld terminals, laptops, wearable devices, or robots, etc. It should be noted that all information involved in the method provided in this application is extracted with the authorization of the relevant user and in accordance with relevant regulations, and will not infringe on user privacy.

[0030] The provided method for predicting laser failure includes steps S101 to S103, detailed below: Step S101. Configure the drive current sensor, laser cavity temperature sensor, pump source voltage detection module, component loss monitoring module and vibration sensor to collect multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters and operating vibration amplitude; extract time-series features from the real-time collected multi-dimensional parameters, and eliminate noise interference from single parameters through feature fusion to identify fault precursor trends.

[0031] Specifically, this step is the foundation for the perception and front-end processing of the entire circuit breaker prediction method. It addresses the limitations of existing technologies that rely on single-parameter monitoring. By building a multi-dimensional perception system that is strongly correlated with circuit breaker faults, it simultaneously completes feature extraction and noise filtering of time-series data, providing high-quality and highly relevant input data for subsequent prediction models. This is a prerequisite for achieving accurate prediction.

[0032] The multi-dimensional real-time acquisition of fault precursor parameters breaks through the limitations of traditional monitoring of three types of parameters: overcurrent, overtemperature, and overvoltage. It establishes a monitoring system covering four dimensions: electrical, thermal, optical, and mechanical. It collects precursor parameters that are strongly correlated with the entire evolution of laser fuse failure, rather than just collecting parameters that exceed limits after the failure occurs, thus achieving full-dimensional capture of fault precursors from the root cause.

[0033] Temporal feature extraction and multi-source feature fusion denoising extracts temporal features that reflect the gradual change trend of parameters from high-frequency acquired time-series data, rather than focusing only on the instantaneous value of parameters; by fusing multi-source features, it eliminates the misjudgment caused by random noise and environmental interference of a single parameter, accurately identifies the fault precursor trend hidden in normal fluctuations, and avoids the passivity of only discovering anomalies after a single parameter exceeds the limit.

[0034] The implementation of multi-dimensional parameter acquisition is achieved by simultaneously configuring five types of monitoring units to realize the synchronous acquisition of all precursor parameters: ① drive current sensor, which acquires laser drive current parameters in real time; ② laser cavity temperature sensor, which acquires the internal operating temperature parameters of the laser resonant cavity in real time; ③ pump source voltage detection module, which acquires the input voltage parameters of the pump source in real time; ④ component loss monitoring module, which acquires the loss parameters of two core components, namely gain medium transmittance and pump source luminous efficiency, in real time through optical detection unit; ⑤ vibration sensor, which acquires the operating vibration amplitude parameters of the entire laser and its core components in real time.

[0035] The sampling frequency of all monitoring units is uniformly set to 1kHz~5kHz to ensure the timing synchronization of parameters in all dimensions, and can capture abnormal fluctuations of parameters at the microsecond level, fully covering the rapid change process of the precursor to fuse failure.

[0036] The implementation of temporal feature extraction and fusion involves processing real-time multi-dimensional temporal parameters in frames using sliding window technology. Within each sliding window, core temporal features such as the sliding window mean, rate of change, peak factor, standard deviation, and kurtosis are calculated and extracted to quantify the changing trends and fluctuations of the parameters. The extracted temporal features are then input into the feature fusion module. Weighting coefficients are set based on the correlation between each parameter and the circuit breaker fault. Dimensionality reduction is achieved through weighted combination or principal component analysis (PCA) to eliminate random noise and environmental interference from single parameters, outputting fused features that accurately characterize the trend of fault precursors, thus completing the front-end data processing.

[0037] Step S102. Using a long short-term memory network algorithm and combining it with the laser fuse failure mechanism, a multi-parameter time series prediction model is constructed. The failure evolution law is learned through offline training. The trained multi-parameter time series prediction model is used to process the parameters after feature extraction and fusion, and output the fuse failure risk level and the expected fuse failure time.

[0038] Specifically, this step is the core decision-making center of the entire fault circuit prediction method. It addresses the core pain points of existing technologies, such as lagging passive protection, weak fault identification capabilities due to multi-factor coupling, and inability to predict fault evolution. By combining the LSTM time series model of the physical mechanism of circuit faults, it learns the evolution law of the entire fault cycle, achieving a core breakthrough from "responding after the fault occurs" to "predicting before the fault occurs".

[0039] The LSTM multi-parameter time-series prediction model based on fault mechanism construction abandons the black-box algorithm mode without physical constraints. It first clarifies the core evolution mechanism of laser fuse failure (a progressive chain reaction of current overload → temperature rise → accelerated component aging → performance degradation → final fuse failure). Then, it combines the strong learning ability of LSTM algorithm for long-term time-series dependencies to build a multi-parameter time-series prediction model that adapts to the progressive development law of fuse failure. This ensures that the learning direction of the model is consistent with the physical mechanism of the fault, and greatly improves the prediction reliability in multi-factor coupled scenarios.

[0040] The model is trained offline and inferred in real time. Offline training is completed using a sample set covering all fault stages and all application scenarios, allowing the model to fully learn the fault evolution rules under different working conditions. Strict accuracy and warning time thresholds are set to ensure the model's generalization and prediction capabilities. The fusion features output from step S101 are input into the trained model in real time to output accurate circuit breaker risk levels and expected circuit breaker times, enabling early prediction of faults.

[0041] The implementation of the prediction model is based on the Long Short-Term Memory (LSTM) network algorithm. A three-layer multi-parameter time-series prediction model is built, consisting of an input layer, a hidden layer, and an output layer. The input layer receives the multi-dimensional time-series fusion features output from step S101, with the number of neurons matching the dimensionality of the fusion features. The hidden layer adopts a multi-layer LSTM neuron structure to learn the temporal evolution features of multi-dimensional parameters, capturing the correlation between the long-term trend of parameter changes and circuit breaker failures, which conforms to the gradual development law of circuit breaker failures. The output layer outputs the results related to circuit breaker risk, with the number of neurons matching the number of risk levels.

[0042] The entire model construction process is constrained by the laser melting failure mechanism. The coupled evolution logic of "current-temperature-component loss-vibration" is embedded into the feature learning process of the model to avoid the model learning pseudo-correlation that is unrelated to the fault and improve the prediction stability under multi-factor coupled fault scenarios.

[0043] The implementation of model training and real-time inference includes: Offline training phase: First, collect operational data of the laser's entire lifecycle as training samples. The samples must cover all fault stages, including normal operation, minor anomalies, severe anomalies, and precursors to circuit breaker failure, with a cumulative sample size of ≥2000 sets. This also includes various circuit breaker failure scenarios, such as single-factor gradual changes and multi-factor coupled abrupt changes, ensuring full sample coverage. Preprocessing operations are performed on the training samples: time-series alignment (timestamp accuracy ±1ms), outlier removal using the 3σ criterion, and min-max normalization mapping to the 0~1 interval. The preprocessed samples are then input into the built model for offline training. Training hyperparameters such as the number of iterations, learning rate, and dropout coefficient are set. After training until the model's prediction accuracy is ≥98.5% and the early warning time is ≥30s, the model is solidified.

[0044] In the real-time inference stage, the fused features output in real time in step S101 are input into the solidified multi-parameter time-series prediction model. After forward inference, the model outputs the circuit breaker risk level under the current operating conditions and the expected time of the circuit breaker failure in real time, thus completing the core prediction process.

[0045] Step S103. Based on the circuit breaker risk level output by the prediction model, generate a corresponding early warning signal and generate push processing suggestions; collect new running data for each preset running time, and incrementally train the multi-parameter time series prediction model to adapt to the changes in fault evolution patterns caused by laser component aging, changes in the operating environment, etc.

[0046] Specifically, this step is the execution closed loop and long-term adaptation guarantee of the entire fault circuit interruption prediction method. It solves the problems that existing technologies cannot reserve sufficient emergency time for operation and maintenance and the poor adaptability of the model for long-term operation. On the one hand, it achieves a balance between safety protection and continuous operation of equipment through graded early warning and emergency linkage. On the other hand, it ensures the prediction accuracy of the model throughout the entire life cycle of the laser through incremental self-optimization, thus forming a complete technical closed loop.

[0047] The tiered early warning and emergency response mechanism sets differentiated early warning and control strategies based on the circuit breaker risk level output by the model. In the low-risk stage, priority is given to ensuring the continuous operation of equipment, while in the high-risk stage, priority is given to ensuring the safety of core components. At the same time, it pushes operation and maintenance handling suggestions that match the risk level, which not only allows operation and maintenance personnel sufficient emergency handling time, but also avoids unnecessary downtime from affecting production continuity.

[0048] The model incremental self-optimization mechanism automatically collects new running data within a fixed incremental training period, performs incremental training on the model, dynamically updates the model parameters, and adapts to the fault evolution patterns throughout the entire life cycle, such as laser component aging, changes in the operating environment, and device replacement. This ensures high prediction accuracy for long-term model operation and reduces manual operation and maintenance costs.

[0049] The implementation of hierarchical early warning and processing suggestion push divides the fuse risk into four levels in advance: low risk, medium risk, high risk, and emergency risk. Corresponding early warning and linkage control strategies are set for different levels: for the low-risk level, only send a remote notification to the operation and maintenance management system, without triggering audible and visual warnings and control actions, and without affecting the normal operation of the equipment; for the medium-risk level, trigger audible and visual warnings, and at the same time remotely push early warning information to the operation and maintenance terminal to remind the operation and maintenance personnel to pay attention to the equipment status; for the high-risk level, continuously trigger audible and visual warnings and remote notifications, and at the same time automatically trigger the laser to run at a reduced power to slow down the deterioration speed of the fault and reserve emergency processing time for the operation and maintenance personnel; for the emergency risk level, immediately trigger audible and visual warnings and operation and maintenance emergency notifications, and at the same time trigger emergency power-off protection to cut off the laser drive power supply to avoid the occurrence of fuse faults and irreversible damage to the core components.

[0050] By synchronously establishing a processing suggestion database that matches the risk level and abnormal parameter type, according to the risk level and abnormal parameter dimension output by the model, the corresponding standardized processing measures are retrieved and pushed to the operation and maintenance terminal in real time, reducing the operation and maintenance threshold and improving the efficiency of fault handling.

[0051] The implementation of model incremental self-optimization presets that the system runs continuously for 72 hours as an incremental training cycle. After each cycle arrives, the system automatically collects the newly added operation data within this cycle, and the data needs to include two types of scenarios: normal working conditions and abnormal working conditions; after performing the same preprocessing and feature extraction operations as the offline training on the newly added data, input it into the固化 LSTM model for incremental training, and update the model parameters with a low learning rate to avoid destroying the core fault evolution rules learned by the model; after the incremental training is completed, verify that the model prediction accuracy still meets the requirements of ≥98.5% and the early warning time ≥30s. After the verification is passed, update the model parameters. If the verification fails, retain the original model parameters, and at the same time push a model exception notification to the operation and maintenance personnel to achieve the adaptive optimization of the entire life cycle of the model.

[0052] In some embodiments, the configuration of the drive current sensor, laser cavity temperature sensor, pump source voltage detection module, component loss monitoring module, and vibration sensor to collect multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude, includes: configuring the drive current sensor to collect drive current within a preset current range in real time; configuring the laser cavity temperature sensor to collect cavity temperature within a preset temperature range in real time; configuring the pump source voltage detection module to collect pump source input voltage within a preset voltage range in real time; configuring the component loss monitoring module to monitor the transmittance of the gain medium and the luminous efficiency of the pump source in real time to obtain component loss parameters; and configuring the vibration sensor to collect operating vibration amplitude within a preset amplitude range in real time, wherein the sampling frequency of each sensor is set to a preset frequency range.

[0053] This embodiment is a detailed implementation of the multi-dimensional fault precursor parameter acquisition step in step S101. It clarifies the functional positioning of each type of monitoring unit, the quantitative range of monitoring parameters, and the unified standard of sampling frequency. It transforms the technical solution of the acquisition step into a hardware configuration solution that can be directly implemented and quantitatively verified, further strengthening the technical boundaries of all-dimensional monitoring.

[0054] For the five core precursor parameters related to laser fuse failure, the configuration requirements of the corresponding sensing / detection modules, the preset range of the monitoring parameters, and the synchronous sampling frequency standard of the entire module are clearly defined. This ensures that the collected parameters are strongly correlated with the fuse failure, are time-synchronized, and meet the accuracy standards, fully covering the four dimensions of fault precursors: electrical, thermal, optical, and mechanical. This addresses the problem of missed or false detections caused by monitoring a single parameter at its root.

[0055] The drive current monitoring configuration uses a high-precision closed-loop Hall current sensor with a preset current monitoring range of 0~10A, which is compatible with the rated drive current range of industrial-grade fiber lasers. It is used to collect the drive current parameters of the laser pump source in real time and capture early signs of faults such as current overload and current fluctuation.

[0056] The cavity temperature monitoring configuration uses a platinum resistance temperature sensor with a preset temperature monitoring range of 15℃~80℃, covering the normal operating temperature and abnormal temperature rise range of the laser. It is installed inside the laser resonant cavity to collect cavity temperature parameters in real time and detect early signs of faults such as abnormal temperature rise.

[0057] The pump source voltage monitoring configuration uses a high-precision DC voltage detection module with a preset voltage monitoring range of 12V~48V, which is compatible with the rated input voltage range of industrial-grade laser pump sources. It is used to collect the input voltage parameters of the pump source in real time and capture early signs of faults such as voltage surges and voltage anomalies.

[0058] The component loss monitoring configuration integrates an optical transmittance detector and a luminous efficiency detector. The transmittance detector is used to monitor the laser transmittance of the gain medium in real time, and the luminous efficiency detector is used to monitor the electro-optical conversion efficiency of the pump source in real time. The two types of parameters together characterize the aging loss state of the laser core components and capture the precursors of fuse failure caused by component aging.

[0059] The vibration monitoring configuration uses piezoelectric vibration sensors with a preset vibration amplitude monitoring range of 0~5mm / s. These sensors are installed at the locations of core components such as the laser pump source and resonant cavity to collect the vibration amplitude parameters of the equipment in real time and detect early signs of mechanical loosening or abnormal vibration of components that could lead to fuse failure.

[0060] Sampling and synchronization configuration involves setting the sampling frequency of all monitoring units to a uniform 1kHz~5kHz and configuring a clock synchronization unit to ensure that the synchronization accuracy of the acquisition timestamp of all parameters is ≤1ms, thus ensuring the time series consistency of multi-dimensional parameters and providing a data foundation for subsequent time series feature analysis.

[0061] In some embodiments, the step of extracting time-series features from real-time acquired multi-dimensional parameters and eliminating single-parameter noise interference through feature fusion to identify fault precursor trends includes: processing real-time acquired multi-dimensional parameters using a sliding window technique, calculating time-series features including the mean, rate of change, and peak factor of the parameters within the sliding window; inputting the extracted time-series features of each dimension into a preset feature fusion module, and performing weighted combination or dimensionality reduction processing on the extracted time-series features through a preset algorithm to eliminate single-parameter noise interference, so as to accurately identify the changing trends of fault precursors.

[0062] This embodiment is a detailed implementation of the temporal feature extraction and feature fusion step in step S101. It clarifies the implementation method of feature extraction, the types of features extracted, and the specific algorithm for feature fusion. It transforms the feature processing step into a quantifiable and reproducible execution process, providing high-quality and highly discriminative feature inputs for subsequent prediction models and further reducing the false alarm rate.

[0063] By employing a sliding window technique to segment high-frequency time-series data into frames, multiple time-series features that can characterize parameter change trends are extracted, solving the problem that focusing only on instantaneous values ​​cannot capture progressive fault precursors. A weighted fusion algorithm is used to fuse multi-dimensional features, eliminating random noise and environmental interference from single parameters, improving the recognizability of fault precursor features, and avoiding misjudgments caused by fluctuations in single parameters.

[0064] The temporal feature extraction process employs a fixed-size sliding window to segment and process real-time acquired multi-dimensional temporal parameters. The sliding window size is preset to 5 seconds, and the sliding step size is preset to 1 second, enabling continuous monitoring of parameter trends. Within each sliding window, for each parameter, 12 core temporal features are calculated and extracted, specifically including: parameter mean, standard deviation, maximum value, minimum value, rate of change (first-order difference mean), peak factor, kurtosis, margin factor, skewness, impulse factor, waveform factor, and peak-to-peak value. This comprehensively quantifies the fluctuation characteristics, changing trends, and abnormal impact characteristics of the parameters, accurately capturing the gradual changes of fault precursors.

[0065] The multi-source feature fusion implementation process involves building a feature fusion module based on correlation weighting. First, offline fault mechanism analysis and Pearson correlation analysis are used to determine the correlation coefficient between each type of feature and the circuit breaker fault. Based on the correlation coefficient, the corresponding weights of the features are set (the higher the correlation with the fault, the greater the weight). All extracted time-series features are combined according to their corresponding weights to obtain a fused high-dimensional feature vector. Then, principal component analysis (PCA) is used to reduce the dimensionality of the high-dimensional feature vector, eliminating redundant features and noise interference. Finally, a fused feature with uniform dimension and high recognizability is output and input into the subsequent prediction model.

[0066] In some embodiments, the use of a Long Short-Term Memory (LSTM) network algorithm, combined with the laser fracturing fault mechanism, to construct a multi-parameter time-series prediction model includes: analyzing the laser fracturing fault mechanism, such as current overload, temperature rise, accelerated component aging, and the fracturing process; constructing a multi-parameter time-series prediction model based on the LTM network algorithm, comprising an input layer, a hidden layer, and an output layer, wherein the input layer receives the time-series features of multi-dimensional parameters, the hidden layer learns the time-series evolution features of the parameters, and the output layer outputs the fracturing risk-related results; and determining the network structure parameters of the multi-parameter time-series prediction model based on the fracturing risk-related results, including the number of hidden layers and the number of neurons, so that the multi-parameter time-series prediction model can conform to the progressive development law of fracturing faults.

[0067] This embodiment is a detailed implementation of the multi-parameter time series prediction model construction step S102. It clarifies the mechanism basis, network structure, functions of each layer and parameter settings of the model construction, and transforms the model construction scheme into a network structure that can be directly built and trained and verified. It enhances the adaptability of the model to the circuit breaker fault mechanism and solves the problems of poor generalization and weak recognition ability of black box models in multi-factor coupled scenarios.

[0068] First, by analyzing the failure mechanism of laser fuses, the progressive chain evolution law of the failure is clarified. Based on this, the input-output logic and network structure constraints of the model are determined. A multi-layer network structure is built based on the LSTM algorithm, and the number of neurons, core functions and hyperparameter settings of each layer are clarified, so that the feature learning direction of the model is fully consistent with the physical mechanism of the failure, and the correlation between the temporal evolution characteristics of multi-dimensional parameters and the fuse failure is accurately captured.

[0069] The mechanistic constraints for model construction are established by first analyzing the laser melting failure mechanism, clarifying the core evolution path as follows: current overload / voltage anomaly → abnormal rise in cavity temperature → accelerated aging of gain medium / pump source → component performance degradation → aggravated abnormal vibration → eventual melting failure. This path is a typical progressive temporal evolution process with multi-factor coupling effects. Based on this mechanism, the model input is determined to be the temporal fusion characteristics of multi-dimensional parameters, and the output is the melting risk level and the expected melting time, ensuring that the model's learning direction is consistent with the physical evolution law of the fault.

[0070] The specific construction of the LSTM network structure involves building a multi-parameter temporal prediction model containing an input layer, three hidden layers, and an output layer. The specific parameter settings are as follows: Input layer: 20 neurons, corresponding to the 20-dimensional fused feature vector output in Example 2, used to receive multi-dimensional temporal feature input; Hidden layer: 3 layers of LSTM network, each containing 64 LSTM neurons, followed by a dropout layer with a dropout coefficient of 0.2 to prevent overfitting; The 3-layer LSTM network extracts the temporal dependency features of multi-dimensional parameters layer by layer, learning the correlation between short-term parameter fluctuations and long-term fault evolution, adapting to the gradual development pattern of circuit breaker faults; Fully connected layer: Two fully connected layers are connected after the hidden layers, the first layer containing 32 neurons and the second layer containing 16 neurons, used to integrate and map the temporal features extracted by the LSTM layers; Output layer: 4 neurons, corresponding to the four circuit breaker risk levels: low risk, medium risk, high risk, and emergency risk, with an additional 1... Each regression neuron is used to output the estimated circuit breaker time, achieving synchronous output of risk level and fault occurrence time.

[0071] In some embodiments, the step of learning fault evolution patterns through offline training and processing the parameters after feature extraction and fusion using a trained multi-parameter time series prediction model to output the circuit breaker risk level and the expected circuit breaker time includes: collecting operational data covering all stages of normal operation, minor anomalies, severe anomalies, and circuit breaker precursors as training samples, and including data from various scenarios such as single-factor gradual change and multi-factor sudden change; performing data preprocessing on the training samples, including time series alignment and outlier removal; inputting the preprocessed training samples into the constructed multi-parameter time series prediction model for offline training, enabling the multi-parameter time series prediction model to learn fault evolution patterns, and training until the prediction accuracy of the multi-parameter time series prediction model is greater than or equal to a preset accuracy and the early warning time is greater than or equal to a preset threshold; inputting the real-time parameters after feature extraction and fusion into the trained multi-parameter time series prediction model, and the multi-parameter time series prediction model outputs the circuit breaker risk level and the expected circuit breaker time after processing.

[0072] This embodiment is a detailed implementation of the offline training and real-time inference process in step S102. It clarifies the requirements for training sample collection, data preprocessing, hyperparameter settings and standards for model training, as well as the execution process for real-time inference. It transforms the model training and inference process into a reproducible and verifiable standardized process, ensuring the model's prediction accuracy and early warning capabilities.

[0073] By collecting samples across all scenarios and stages, the model can learn the evolution patterns of all types of circuit breaker faults. Through standardized data preprocessing, the quality of sample data is improved, ensuring training effectiveness. By setting strict training hyperparameters and thresholds, the model's generalization ability and prediction accuracy are guaranteed. The execution process of real-time inference is clearly defined, enabling millisecond-level risk prediction output to meet the needs of real-time on-site early warning.

[0074] Training sample acquisition and preprocessing include: Sample acquisition: In a laboratory environment, various fuse failure scenarios such as laser current overload, abnormal temperature, component aging, vibration interference, and multi-factor coupling were simulated. 2500 sets of full-stage operational data were collected, including 1000 sets of normal operating conditions, 800 sets of slightly abnormal conditions, 500 sets of severely abnormal conditions, and 200 sets of fuse failure precursor conditions, fully covering the entire cycle of a fault from its inception to its occurrence. This also includes various scenarios such as gradual changes in a single factor and sudden changes in multiple factors, ensuring the comprehensiveness and balance of the samples. Data preprocessing: First, all sample data were time-aligned, with a unified timestamp accuracy of ±1ms to ensure the temporal synchronization of multi-dimensional parameters. Then, the 3σ criterion was used to remove outliers and dystopian values ​​from the data, eliminating interference from acquisition errors. Finally, the min-max normalization algorithm was used to map all parameter data to the 0~1 interval, eliminating the impact of differences in parameter dimensions on training, thus completing sample preprocessing.

[0075] The offline training process involves dividing preprocessed samples into training, validation, and test sets in an 8:1:1 ratio, and inputting them into the LSTM model built in Example 3 for training. The core hyperparameters are set as follows: 500 iterations, initial learning rate of 0.001, batch size of 32, Adam optimizer, and a combined loss function of cross-entropy loss (for risk level classification) and mean squared error loss (for predicted circuit breaker time regression). During training, the model accuracy is validated using the validation set after each iteration. If the validation set accuracy shows no improvement after 50 consecutive iterations, an early stopping mechanism is triggered to prevent overfitting. Training stops when the model's test set prediction accuracy is ≥98.5% and the average early warning time is ≥30s. The trained model parameters are then written to the processor's storage unit, completing model solidification. In this example, the final trained model achieved a prediction accuracy of 99.1% and an average early warning time of 65s, meeting the design requirements.

[0076] The real-time inference implementation process involves normalizing the fused feature vector output in step S101 according to the preprocessing standard during training during equipment operation, and then inputting it into the solidified LSTM model. The computational delay of the model in forward inference is ≤10ms, and the circuit breaker risk level corresponding to the current working condition and the expected time of circuit breaker occurrence are output in real time, thus completing the real-time prediction process.

[0077] In some embodiments, generating a corresponding early warning signal based on the circuit breaker risk level output by the prediction model includes: classifying the circuit breaker risk level into low risk, medium risk, high risk, and emergency risk; generating an early warning signal with only remote notification when the output circuit breaker risk level is low risk; generating an early warning signal with audible and visual warning and remote push of early warning information when the output circuit breaker risk level is medium risk; generating an early warning signal with audible and visual warning, remote notification, and triggering power reduction operation when the output circuit breaker risk level is high risk; and generating an early warning signal with audible and visual warning, remote emergency notification, and immediately triggering power outage protection when the output circuit breaker risk level is emergency risk.

[0078] This embodiment is a detailed implementation of the graded early warning signal generation step S103. It clarifies the classification criteria for the four levels of circuit breaker risk, as well as the early warning signals and linkage control actions corresponding to each risk level. It transforms the graded early warning strategy into a standardized response process that can be directly executed, thereby achieving a balance between the continuity of equipment operation and the safety protection of core components.

[0079] Based on the evolution stage of the fuse failure, the risk of component damage, and the expected fuse failure time, the risk is divided into 4 levels. Different early warning and linkage control strategies are set for different levels: in the low and medium risk stages, the focus is on early warning and reminders without interfering with the normal operation of the equipment to ensure production continuity; in the high and emergency risk stages, the focus is on safety protection, triggering power reduction or power outage protection to avoid irreversible damage to core components, while reserving sufficient emergency handling time for maintenance personnel.

[0080] Risk levels and corresponding response strategies are categorized into four levels based on expected shutdown time and the degree of parameter anomaly. The specific execution process is as follows: Low-risk level: The judgment criterion is that the model output expected shutdown time > 120s, and the parameters have no obvious anomalies, only slight fluctuations. The corresponding response is: only send remote status notifications to the operation and maintenance management system and the operation and maintenance personnel's terminals, without triggering any light and sound alarms or control actions, and the laser maintains normal operation at rated power. Medium-risk level: The judgment criterion is that the model output expected shutdown time is 60s~120s, and a single type of parameter shows an abnormal trend, but does not exceed the rated threshold. The corresponding response is: trigger the audible and visual alarm (volume ≥ 85dB, luminous intensity ≥ 500cd), and simultaneously push warning information and abnormal parameter details remotely to the operation and maintenance personnel's terminals, without triggering power adjustment or power-off control, the laser maintains normal operation, and reminds the operation and maintenance personnel to pay attention to the equipment status and prepare for emergency handling. High-risk level: The judgment criteria are a model output predicted fuse failure time of 30s~60s, multiple parameters showing an abnormal upward trend, approaching the rated threshold; the corresponding response is: continuously triggering audible and visual warnings and remote notifications, while automatically sending a power reduction command to the laser control system to reduce the laser's operating power to 70% of the rated power, slowing down the rate of fault deterioration, allowing sufficient on-site handling time for maintenance personnel, and simultaneously pushing emergency handling suggestions. Emergency risk level: The judgment criteria are a model output predicted fuse failure time <30s, parameters about to exceed the rated threshold, and rapid fault deterioration; the corresponding response is: immediately triggering the highest level audible and visual warnings and emergency telephone and SMS notifications to maintenance personnel, while simultaneously sending a trigger command to the laser's emergency power-off switch to immediately cut off the laser pump source drive power, forcibly shutting down the system to avoid fuse failure and prevent irreversible damage to core components such as the pump source and gain medium. Linkage control execution guarantee: All linkage control commands are connected to the laser's original control system and emergency power-off circuit via hard wiring, and a manual priority control switch is set up, allowing maintenance personnel to manually take over control at any time to avoid malfunctions and ensure equipment operation safety.

[0081] In some embodiments, generating push processing suggestions includes: establishing a processing suggestion database corresponding to the circuit breaker risk level, wherein the processing suggestion database stores specific processing measures for different risk levels; retrieving the corresponding specific processing measures from the processing suggestion database based on the circuit breaker risk level output by the multi-parameter time series prediction model, and pushing the specific processing measures to the operation and maintenance terminal, wherein the specific processing measures include, but are not limited to, adjusting the drive current, checking the heat dissipation of the laser, and checking the aging degree of components, one or more of these.

[0082] This embodiment is a detailed implementation of the processing suggestion push step in step S103. It clarifies the logic for building the processing suggestion database, the matching rules for processing suggestions, and the push method. It transforms the processing suggestion push into a feasible solution that accurately matches the risk level and anomaly type, reducing the operation and maintenance threshold and improving the efficiency and accuracy of fault handling.

[0083] By building a standardized handling suggestion database covering all risk levels and all anomaly types in advance, and based on the circuit breaker risk level and real-time anomaly parameter dimensions output by the model, corresponding implementable handling measures are retrieved through matching rules and pushed to operation and maintenance personnel through multiple channels, realizing the synchronous push of "risk warning - handling guidance", solving the problems of operation and maintenance personnel having no standardized handling guidance and low fault handling efficiency.

[0084] The database of handling suggestions is built according to a two-dimensional architecture of "risk level + abnormal parameter type". A standardized database of handling suggestions is constructed, and all handling measures in the database have been verified on-site during laser fault maintenance, ensuring operability and safety. The core categories and contents are as follows: Classified by risk level: For four risk levels—low, medium, high, and urgent—corresponding handling measures are set. The higher the risk level, the stronger the urgency and intervention of the handling measures. Classified by abnormal parameter type: For five types of abnormal scenarios—current abnormality, temperature abnormality, voltage abnormality, component loss abnormality, and vibration abnormality—targeted investigation and handling measures are set to achieve precise matching between abnormality type and handling suggestions.

[0085] Example of core database content: Low-risk general handling suggestions: Monitor the trend of equipment parameter changes, record abnormal fluctuations, and conduct a comprehensive inspection of the laser heat dissipation system, drive circuit, and core optical components during the next routine maintenance; Medium-risk - Temperature anomaly handling suggestions: Immediately check whether the outlet water temperature and flow rate of the laser water chiller are normal, clean the dust in the heat dissipation channel, check whether the resonant cavity seal is intact, and verify whether the temperature sensor data acquisition is normal; High-risk - Current anomaly handling suggestions: Immediately manually reduce the laser operating power, check on-site whether there are any loose connections or short circuits in the drive circuit, verify whether the pump source drive current is balanced, and check whether there is any local damage to the pump source; General emergency risk handling recommendations include: immediately disconnect the main power supply of the equipment, do not forcibly restart the equipment, thoroughly disassemble and inspect the pump source, gain medium, drive circuit, and power supply circuit for damage, replace faulty components and complete no-load testing before gradually restarting the equipment.

[0086] The process for matching and pushing handling suggestions involves the system first extracting the parameter type and severity of the current anomaly after the model outputs the circuit breaker risk level. Then, based on the matching rule of "risk level + anomaly type", the system retrieves the corresponding standardized handling measures from the handling suggestion database. At the same time, personalized handling suggestions are generated by combining the equipment model and operating conditions. The suggestions are then pushed simultaneously through four channels: the operation and maintenance management platform, the operation and maintenance personnel's mobile APP, SMS, and on-site touch screens, ensuring that operation and maintenance personnel can obtain handling guidance in a timely manner and improve the efficiency of fault handling.

[0087] In some embodiments, the step of collecting new operational data and incrementally training a multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment includes: setting a preset operating time for each system cycle as an incremental training cycle; automatically collecting new operational data within each incremental training cycle, including data from normal and abnormal operating conditions; performing the same data preprocessing operations as during offline training on the new operational data, inputting it into the multi-parameter time-series prediction model for incremental training, updating the model parameters corresponding to the multi-parameter time-series prediction model, and enabling the multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment.

[0088] This embodiment is a detailed implementation of the incremental self-optimization step in step S103. It clarifies the incremental training cycle, data acquisition requirements, training process and verification standards, and transforms the model self-optimization into a standardized process that can be executed automatically and does not affect the normal operation of the equipment. This ensures the prediction accuracy of the model throughout the entire life cycle of the laser and adapts to the deviation of fault evolution patterns caused by component aging and environmental changes.

[0089] By setting a fixed incremental training period, new operational data is automatically collected within the period. After standardized preprocessing, the model is incrementally trained with a low learning rate. The model parameters are updated without destroying the core fault patterns already learned by the model, adapting to the state changes throughout the laser's entire life cycle. Strict verification standards are set, and parameters are only updated when the incrementally trained model meets the accuracy requirements, ensuring the stability and reliability of the model's operation.

[0090] The incremental training cycle and data collection are based on a preset system that runs continuously for 72 hours as an incremental training cycle. After each cycle, the system automatically starts the incremental training process during the equipment's idle period (such as equipment standby or non-production period) without affecting the normal production operation of the equipment. The system automatically collects new operating data within the cycle, with no less than 100 sets of data collected. These data must include both normal and abnormal operating condition data. If there is no abnormal operating condition data within the cycle, abnormal operating condition data from the historical validation set can be reused to supplement the data, ensuring the balance of the training samples and avoiding overfitting of the model.

[0091] The incremental training process involves performing preprocessing operations identical to those used in offline training on the newly collected running data, including temporal alignment, outlier removal, normalization, and temporal feature extraction. This ensures that the format and standards of the incremental training data are completely consistent with those of the offline training data. The preprocessed new data is then input into the already fixed LSTM model for incremental training. The core training parameters are set as follows: no more than 50 iterations, a learning rate of 0.0001 (far lower than the offline training learning rate), and the parameters of the first two LSTM layers are frozen. Only the parameters of the last LSTM layer and the fully connected layer are updated to avoid the incremental training disrupting the core fault evolution patterns already learned by the model, thus ensuring the model's stability.

[0092] Model validation and parameter updates are performed after incremental training. A fixed standard test set is used to validate the updated model, with validation metrics including: prediction accuracy ≥ 98.5%, early warning time ≥ 30 seconds, false positive rate ≤ 3%, and false negative rate ≤ 1%. If all validation results meet the criteria, the model parameters are automatically updated, replacing the original parameters and completing the incremental optimization. If the validation results fail to meet the criteria, the parameter update is automatically abandoned, the original model parameters are retained, and a model anomaly notification is pushed to operations personnel to remind them to investigate data anomalies or model problems, ensuring the reliability of model operation.

[0093] Through the incremental self-optimization mechanism in this embodiment, the model can be guaranteed to continuously adapt to changes in state caused by component aging, changes in operating environment, and device replacement throughout the 5-10 year life cycle of the laser, always maintaining high prediction accuracy and significantly reducing manual maintenance costs.

[0094] In some embodiments, in order to address the problems of passive response, weak multi-factor coupling identification capability, and lack of prediction and early warning in existing laser fuse protection, a prediction model is established by monitoring multi-dimensional fault precursor parameters and combining them with fault evolution laws. This enables accurate prediction and graded early warning of laser fuse faults, allowing sufficient time for operation and maintenance, reducing the risk of component damage, and ensuring stable equipment operation.

[0095] The technical solution provided in this embodiment includes: 1. Establish a multi-dimensional fault precursor monitoring unit: Configure a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module (monitoring the transmittance of the gain medium and the luminous efficiency of the pump source), and a vibration sensor to collect drive current (0~10A), cavity temperature (15℃~80℃), pump source input voltage (12V~48V), component loss parameters, and operating vibration amplitude (0~5mm / s) in real time, with the sampling frequency set to 1kHz~5kHz.

[0096] 2. Constructing a fuse failure prediction model: Using the Long Short-Term Memory (LSTM) network algorithm, combined with the mechanism of laser fuse failure (such as current overload - temperature rise - accelerated component aging - fuse failure), a multi-parameter time series prediction model is constructed. The model learns the failure evolution law through offline training. The training samples cover data of all stages, including normal operation, minor anomalies, severe anomalies, and fuse failure precursors, with a cumulative sample size of ≥2000 sets. The model includes various scenarios such as single-factor gradual change and multi-factor sudden change. The model prediction accuracy is ≥98.5%, and the early warning time is ≥30s.

[0097] 3. Feature Engineering and Trend Analysis: Time-series feature extraction (such as sliding window mean, rate of change, peak factor) is performed on multi-dimensional parameters collected in real time. Feature fusion is used to eliminate noise interference from single parameters and accurately identify the trend of fault precursors.

[0098] 4. Tiered early warning and emergency response: Based on the circuit breaker risk level (low risk, medium risk, high risk, emergency risk) output by the prediction model, corresponding early warning signals (audible and visual warning, remote notification) are generated. High risk and above levels will automatically trigger power reduction operation or early power outage protection, and the operation and maintenance management system will push handling suggestions.

[0099] 5. Model self-optimization and update: Every 72 hours of system operation, new operating data (including normal and abnormal operating conditions) is automatically collected to incrementally train the prediction model and adapt to changes in fault evolution patterns caused by laser component aging, changes in the operating environment, etc.

[0100] The key points of the invention corresponding to this embodiment include: 1. Multi-dimensional monitoring covers all early warning parameters of faults, overcoming the limitations of single-parameter monitoring; 2. The LSTM algorithm captures the time-series evolution characteristics of parameters, which aligns with the gradual development pattern of circuit breaker faults, enabling accurate prediction; 3. Model design based on fault mechanisms improves prediction reliability in scenarios with multiple coupled factors; 4. A tiered early warning and emergency response mechanism to balance equipment operation continuity with fault protection safety; 5. Self-optimization and updates ensure long-term model adaptability and reduce operation and maintenance costs.

[0101] The beneficial effects of this embodiment include: 1. Achieve proactive prediction and early warning: Predict fuse failure 30s to 120s in advance, upgrading from passive protection to proactive early warning, avoiding irreversible damage to core components, and reducing maintenance costs by more than 80%; 2. High prediction accuracy: Multi-dimensional parameter fusion + time series feature analysis, fault prediction accuracy ≥98.5%, false alarm rate ≤3%, false alarm rate ≤1%, significantly better than existing technologies; 3. Adaptable to multiple fault scenarios: It can effectively identify various precursors to fuse failure, such as gradual changes due to a single factor or sudden changes due to multiple factors coupled together, and its application scope covers different types of lasers, including industrial and laboratory lasers; 4. Highly user-friendly operation and maintenance: Tiered early warning provides operation and maintenance personnel with ample processing time (≥30s), and pushes processing suggestions in conjunction with the system, lowering the threshold for operation and maintenance; 5. Strong compatibility: Monitoring modules and algorithm units can be added to the existing laser control system without replacing the entire equipment, resulting in low modification costs and wide adaptability.

[0102] like Figure 2 As shown, the overall prediction and early warning framework includes a multi-dimensional monitoring unit, a feature extraction module, an LSTM prediction model module, a risk grading module, an early warning linkage module, a model self-optimization module, a laser, and an operation and maintenance management system; the signal flow is: monitoring unit → feature extraction module → prediction model module → risk grading module → early warning linkage module → laser / operation and maintenance system, and the running data → self-optimization module → prediction model module.

[0103] like Figure 3 As shown, the provided LSTM prediction model workflow includes: Training phase: collecting samples from all stages → data preprocessing (time alignment, outlier removal) → feature extraction → model training → validation (accuracy ≥ 98.5%, warning time ≥ 30s) → model solidification; Working phase: real-time parameter monitoring → feature extraction → inputting into the model → outputting risk level → triggering warning / linkage control.

[0104] like Figure 4 As shown in the figure, the fault prediction effect comparison chart provides a timeline of response between existing passive protection and the predictive early warning of this invention under the same fuse failure scenario, intuitively demonstrating the advantages of early warning and the protection effect of core components.

[0105] For example, the provided industrial-grade fiber laser fuse failure prediction and early warning system includes: 1. Hardware configuration includes: (1) Multi-dimensional monitoring unit: current sensor (LA58-P, accuracy ±0.02A), cavity temperature sensor (PT100, accuracy ±0.1℃), voltage detection module (AV200, accuracy ±0.1V), component loss monitoring module (transmittance detector, accuracy ±0.5%; luminous efficiency detector, accuracy ±1%), vibration sensor (YD-100, accuracy ±0.01mm / s), sampling frequency 3kHz; (2) Control and computing module: Embedded processor (Intel Core i5-12400H) + FPGA coprocessor, integrated LSTM computing acceleration module, computing latency ≤10ms; (3) Early warning and linkage unit: sound and light alarm (volume ≥85dB, light intensity ≥500cd), 4G remote communication module, laser power adjustment module (adjustment range 0~100% rated power), emergency power cut-off switch; (4) Laser: Industrial grade fiber laser, rated power 1000W, rated drive current 6A, normal operating temperature range of cavity 20℃~40℃, rated input voltage of pump source 24V.

[0106] 2. The model training steps include: (1) Sample collection: In a laboratory environment, different fuse failure scenarios (current overload, abnormal temperature, component aging, vibration interference and multi-factor coupling) were simulated, and 2,500 sets of full-stage operation data were collected, covering normal operation (1,000 sets), minor abnormality (800 sets), serious abnormality (500 sets), and fuse failure precursor (200 sets). (2) Data preprocessing: The time series data is aligned (timestamp accuracy ±1ms), outliers are removed using the 3σ criterion, and the parameters are normalized (mapped to the 0~1 interval). (3) Feature extraction: Features were extracted using a sliding window (window size 5s), including 12 types of time-series features such as parameter mean, standard deviation, rate of change, peak factor, and kurtosis; (4) Model training: Construct an LSTM model (20 neurons in the input layer, 3 layers of hidden layers × 64 neurons, and 4 neurons in the output layer corresponding to 4 levels of risk), set the number of iterations to 500, the learning rate to 0.001, and the dropout coefficient to 0.2, and stop training when the prediction accuracy reaches 99.1% and the average warning time reaches 65s; (5) Model solidification: Write the trained model parameters into the processor storage unit.

[0107] 3. The work steps include: (1) Real-time monitoring: Each sensor transmits multi-dimensional parameter data to the computing module every 333μs; (2) Feature extraction: The operation module extracts time-series features in real time using a sliding window to eliminate noise interference; (3) Risk prediction: Input the feature data into the LSTM model and output the real-time circuit breaker risk level (low / medium / high / emergency) and the expected circuit breaker time; (4) Tiered response: Low risk → remote notification only; Medium risk → audible and visual warning + remote push processing suggestions; High risk → 30% power reduction operation + tiered warning; Emergency risk → immediate power outage protection + emergency maintenance notification; (5) Model self-optimization: Every 72 hours of operation, 100 sets of new running data are automatically collected to incrementally train the model and update the parameters to adapt to changes in equipment status.

[0108] 4. Effectiveness verification includes: Simulated current overload + temperature anomaly coupled fuse failure scenario (drive current increases from 6A to 8A, cavity temperature increases from 30℃ to 65℃): Existing overcurrent protection: Power is cut off when the current reaches 8A (33% above the rated threshold), at which point the pump source has already suffered partial damage; The method of this invention triggers a high-risk warning when the current rises to 6.8A and the cavity temperature rises to 42°C (expected to blow after 38 seconds). Maintenance personnel promptly reduce the power to handle the situation, without causing any damage to any components. The prediction accuracy rate is 99.3%, and the warning response is timely and effective.

[0109] Please see Figure 5 As shown, Figure 5 This is a schematic diagram of the structure of a laser fault-prediction system 200 provided in this application embodiment. The laser fault-prediction system 200 is used to execute the steps of the laser fault-prediction method shown in the above embodiments. The laser fault-prediction system 200 can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, laptop computer, wearable device, or robot.

[0110] like Figure 5 As shown, the laser fault fuse prediction system 200 includes: The data acquisition unit 201 is configured with a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor. It collects multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude. It performs time-series feature extraction on the real-time collected multi-dimensional parameters and eliminates noise interference from single parameters through feature fusion to identify fault precursor trends. The feature extraction unit 202 is used to construct a multi-parameter time series prediction model by using a long short-term memory network algorithm combined with the laser fuse failure mechanism. It learns the failure evolution law through offline training, and uses the trained multi-parameter time series prediction model to process the parameters after feature extraction and fusion, and outputs the fuse failure risk level and the expected fuse failure time. It is recommended that generation unit 203 be used to generate corresponding early warning signals based on the circuit breaker risk level output by the prediction model, and generate push processing suggestions; every preset running time, new running data is collected to incrementally train the multi-parameter time series prediction model to adapt to changes in fault evolution patterns caused by laser component aging, changes in the operating environment, etc.

[0111] In some embodiments, the configuration of the drive current sensor, laser cavity temperature sensor, pump source voltage detection module, component loss monitoring module, and vibration sensor to collect multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude, includes: configuring the drive current sensor to collect drive current within a preset current range in real time; configuring the laser cavity temperature sensor to collect cavity temperature within a preset temperature range in real time; configuring the pump source voltage detection module to collect pump source input voltage within a preset voltage range in real time; configuring the component loss monitoring module to monitor the transmittance of the gain medium and the luminous efficiency of the pump source in real time to obtain component loss parameters; and configuring the vibration sensor to collect operating vibration amplitude within a preset amplitude range in real time, wherein the sampling frequency of each sensor is set to a preset frequency range.

[0112] In some embodiments, the step of extracting time-series features from real-time acquired multi-dimensional parameters and eliminating single-parameter noise interference through feature fusion to identify fault precursor trends includes: processing real-time acquired multi-dimensional parameters using a sliding window technique, calculating time-series features including the mean, rate of change, and peak factor of the parameters within the sliding window; inputting the extracted time-series features of each dimension into a preset feature fusion module, and performing weighted combination or dimensionality reduction processing on the extracted time-series features through a preset algorithm to eliminate single-parameter noise interference, so as to accurately identify the changing trends of fault precursors.

[0113] In some embodiments, the use of a Long Short-Term Memory (LSTM) network algorithm, combined with the laser fracturing fault mechanism, to construct a multi-parameter time-series prediction model includes: analyzing the laser fracturing fault mechanism, such as current overload, temperature rise, accelerated component aging, and the fracturing process; constructing a multi-parameter time-series prediction model based on the LTM network algorithm, comprising an input layer, a hidden layer, and an output layer, wherein the input layer receives the time-series features of multi-dimensional parameters, the hidden layer learns the time-series evolution features of the parameters, and the output layer outputs the fracturing risk-related results; and determining the network structure parameters of the multi-parameter time-series prediction model based on the fracturing risk-related results, including the number of hidden layers and the number of neurons, so that the multi-parameter time-series prediction model can conform to the progressive development law of fracturing faults.

[0114] In some embodiments, the step of learning fault evolution patterns through offline training and processing the parameters after feature extraction and fusion using a trained multi-parameter time series prediction model to output the circuit breaker risk level and the expected circuit breaker time includes: collecting operational data covering all stages of normal operation, minor anomalies, severe anomalies, and circuit breaker precursors as training samples, and including data from various scenarios such as single-factor gradual change and multi-factor sudden change; performing data preprocessing on the training samples, including time series alignment and outlier removal; inputting the preprocessed training samples into the constructed multi-parameter time series prediction model for offline training, enabling the multi-parameter time series prediction model to learn fault evolution patterns, and training until the prediction accuracy of the multi-parameter time series prediction model is greater than or equal to a preset accuracy and the early warning time is greater than or equal to a preset threshold; inputting the real-time parameters after feature extraction and fusion into the trained multi-parameter time series prediction model, and the multi-parameter time series prediction model outputs the circuit breaker risk level and the expected circuit breaker time after processing.

[0115] In some embodiments, generating a corresponding early warning signal based on the circuit breaker risk level output by the prediction model includes: classifying the circuit breaker risk level into low risk, medium risk, high risk, and emergency risk; generating an early warning signal with only remote notification when the output circuit breaker risk level is low risk; generating an early warning signal with audible and visual warning and remote push of early warning information when the output circuit breaker risk level is medium risk; generating an early warning signal with audible and visual warning, remote notification, and triggering power reduction operation when the output circuit breaker risk level is high risk; and generating an early warning signal with audible and visual warning, remote emergency notification, and immediately triggering power outage protection when the output circuit breaker risk level is emergency risk.

[0116] In some embodiments, generating push processing suggestions includes: establishing a processing suggestion database corresponding to the circuit breaker risk level, wherein the processing suggestion database stores specific processing measures for different risk levels; retrieving the corresponding specific processing measures from the processing suggestion database based on the circuit breaker risk level output by the multi-parameter time series prediction model, and pushing the specific processing measures to the operation and maintenance terminal, wherein the specific processing measures include, but are not limited to, adjusting the drive current, checking the heat dissipation of the laser, and checking the aging degree of components, one or more of these.

[0117] In some embodiments, the step of collecting new operational data and incrementally training a multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment includes: setting a preset operating time for each system cycle as an incremental training cycle; automatically collecting new operational data within each incremental training cycle, including data from normal and abnormal operating conditions; performing the same data preprocessing operations as during offline training on the new operational data, inputting it into the multi-parameter time-series prediction model for incremental training, updating the model parameters corresponding to the multi-parameter time-series prediction model, and enabling the multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment.

[0118] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the laser fault fusing prediction system and its modules described above can be found in the corresponding contents of the various embodiments of the laser fault fusing prediction method, and will not be repeated here.

[0119] The aforementioned method for predicting laser failure can be implemented as a computer program, which can be used in various ways, such as... Figure 5 It runs on the device shown.

[0120] Please see Figure 6 , Figure 6 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0121] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any laser fault-prevention method.

[0122] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0123] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to perform any laser fault prediction method.

[0124] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0125] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0126] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: The system is equipped with a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor. It collects multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude. It extracts time-series features from the real-time collected multi-dimensional parameters and eliminates noise interference from single parameters through feature fusion to identify fault precursor trends. A multi-parameter time-series prediction model is constructed by using a long short-term memory network algorithm and combining it with the fault mechanism of laser fuse. The model learns the fault evolution law through offline training, and processes the parameters after feature extraction and fusion using the trained multi-parameter time-series prediction model to output the fuse risk level and the expected fuse time. Based on the circuit breaker risk level output by the prediction model, a corresponding early warning signal is generated, and a push notification for handling suggestions is generated. After each preset running time, new running data is collected, and the multi-parameter time series prediction model is incrementally trained to adapt to changes in the fault evolution pattern caused by laser component aging, changes in the operating environment, etc.

[0127] In some embodiments, the configuration of the drive current sensor, laser cavity temperature sensor, pump source voltage detection module, component loss monitoring module, and vibration sensor to collect multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude, includes: configuring the drive current sensor to collect drive current within a preset current range in real time; configuring the laser cavity temperature sensor to collect cavity temperature within a preset temperature range in real time; configuring the pump source voltage detection module to collect pump source input voltage within a preset voltage range in real time; configuring the component loss monitoring module to monitor the transmittance of the gain medium and the luminous efficiency of the pump source in real time to obtain component loss parameters; and configuring the vibration sensor to collect operating vibration amplitude within a preset amplitude range in real time, wherein the sampling frequency of each sensor is set to a preset frequency range.

[0128] In some embodiments, the step of extracting time-series features from real-time acquired multi-dimensional parameters and eliminating single-parameter noise interference through feature fusion to identify fault precursor trends includes: processing real-time acquired multi-dimensional parameters using a sliding window technique, calculating time-series features including the mean, rate of change, and peak factor of the parameters within the sliding window; inputting the extracted time-series features of each dimension into a preset feature fusion module, and performing weighted combination or dimensionality reduction processing on the extracted time-series features through a preset algorithm to eliminate single-parameter noise interference, so as to accurately identify the changing trends of fault precursors.

[0129] In some embodiments, the use of a Long Short-Term Memory (LSTM) network algorithm, combined with the laser fracturing fault mechanism, to construct a multi-parameter time-series prediction model includes: analyzing the laser fracturing fault mechanism, such as current overload, temperature rise, accelerated component aging, and the fracturing process; constructing a multi-parameter time-series prediction model based on the LTM network algorithm, comprising an input layer, a hidden layer, and an output layer, wherein the input layer receives the time-series features of multi-dimensional parameters, the hidden layer learns the time-series evolution features of the parameters, and the output layer outputs the fracturing risk-related results; and determining the network structure parameters of the multi-parameter time-series prediction model based on the fracturing risk-related results, including the number of hidden layers and the number of neurons, so that the multi-parameter time-series prediction model can conform to the progressive development law of fracturing faults.

[0130] In some embodiments, the step of learning fault evolution patterns through offline training and processing the parameters after feature extraction and fusion using a trained multi-parameter time series prediction model to output the circuit breaker risk level and the expected circuit breaker time includes: collecting operational data covering all stages of normal operation, minor anomalies, severe anomalies, and circuit breaker precursors as training samples, and including data from various scenarios such as single-factor gradual change and multi-factor sudden change; performing data preprocessing on the training samples, including time series alignment and outlier removal; inputting the preprocessed training samples into the constructed multi-parameter time series prediction model for offline training, enabling the multi-parameter time series prediction model to learn fault evolution patterns, and training until the prediction accuracy of the multi-parameter time series prediction model is greater than or equal to a preset accuracy and the early warning time is greater than or equal to a preset threshold; inputting the real-time parameters after feature extraction and fusion into the trained multi-parameter time series prediction model, and the multi-parameter time series prediction model outputs the circuit breaker risk level and the expected circuit breaker time after processing.

[0131] In some embodiments, generating a corresponding early warning signal based on the circuit breaker risk level output by the prediction model includes: classifying the circuit breaker risk level into low risk, medium risk, high risk, and emergency risk; generating an early warning signal with only remote notification when the output circuit breaker risk level is low risk; generating an early warning signal with audible and visual warning and remote push of early warning information when the output circuit breaker risk level is medium risk; generating an early warning signal with audible and visual warning, remote notification, and triggering power reduction operation when the output circuit breaker risk level is high risk; and generating an early warning signal with audible and visual warning, remote emergency notification, and immediately triggering power outage protection when the output circuit breaker risk level is emergency risk.

[0132] In some embodiments, generating push processing suggestions includes: establishing a processing suggestion database corresponding to the circuit breaker risk level, wherein the processing suggestion database stores specific processing measures for different risk levels; retrieving the corresponding specific processing measures from the processing suggestion database based on the circuit breaker risk level output by the multi-parameter time series prediction model, and pushing the specific processing measures to the operation and maintenance terminal, wherein the specific processing measures include, but are not limited to, adjusting the drive current, checking the heat dissipation of the laser, and checking the aging degree of components, one or more of these.

[0133] In some embodiments, the step of collecting new operational data and incrementally training a multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment includes: setting a preset operating time for each system cycle as an incremental training cycle; automatically collecting new operational data within each incremental training cycle, including data from normal and abnormal operating conditions; performing the same data preprocessing operations as during offline training on the new operational data, inputting it into the multi-parameter time-series prediction model for incremental training, updating the model parameters corresponding to the multi-parameter time-series prediction model, and enabling the multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment.

[0134] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the steps implemented by the processor described above and their corresponding specific working processes can be referred to the corresponding contents in the various embodiments of the laser fault fuse prediction method, and will not be repeated here.

[0135] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the laser fault-prediction method provided in any embodiment of this application.

[0136] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0137] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the failure and fuse burning of a laser, characterized in that, include: The system is equipped with a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor. It collects multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude. It extracts time-series features from the real-time collected multi-dimensional parameters and eliminates noise interference from single parameters through feature fusion to identify fault precursor trends. A multi-parameter time-series prediction model is constructed by using a long short-term memory network algorithm and combining it with the fault mechanism of laser fuse. The model learns the fault evolution law through offline training, and processes the parameters after feature extraction and fusion using the trained multi-parameter time-series prediction model to output the fuse risk level and the expected fuse time. Based on the circuit breaker risk level output by the prediction model, a corresponding early warning signal is generated, and a push notification for handling suggestions is also generated. For each preset running time, new running data is collected, and the multi-parameter time series prediction model is incrementally trained to adapt to changes in fault evolution patterns caused by laser component aging, changes in the operating environment, etc.

2. The method according to claim 1, characterized in that, The configuration includes a drive current sensor, a laser cavity temperature sensor, a pump source voltage detection module, a component loss monitoring module, and a vibration sensor. It collects multi-dimensional fault precursor parameters in real time, including drive current, cavity temperature, pump source input voltage, component loss parameters, and operating vibration amplitude. A drive current sensor is configured to collect drive current within a preset current range in real time; a laser cavity temperature sensor is configured to collect cavity temperature within a preset temperature range in real time; a pump source voltage detection module is configured to collect pump source input voltage within a preset voltage range in real time; a component loss monitoring module is configured to monitor gain medium transmittance and pump source luminous efficiency in real time to obtain component loss parameters; and a vibration sensor is configured to collect operating vibration amplitude within a preset amplitude range in real time. The sampling frequency of each sensor is set to a preset frequency range.

3. The method according to claim 1, characterized in that, The process of extracting time-series features from real-time acquired multi-dimensional parameters, eliminating noise interference from single parameters through feature fusion, and identifying pre-fault trends includes: The sliding window technique is used to process the multi-dimensional parameters collected in real time, and the time-series characteristics including the mean, rate of change and peak factor of the parameters are calculated within the sliding window. The extracted time-series features of each dimension are input into a preset feature fusion module. The extracted time-series features are weighted and combined or dimensionality reduced by a preset algorithm to eliminate noise interference from single parameters, so as to accurately identify the changing trend of fault precursors.

4. The method according to claim 1, characterized in that, The method employs a long short-term memory network algorithm, combined with the laser fuse failure mechanism, to construct a multi-parameter time-series prediction model, including: By analyzing the failure mechanism of laser fuses, such as current overload, temperature rise, accelerated component aging and the fuse process; Based on the Long Short-Term Memory (LSTM) network algorithm, a multi-parameter time series prediction model is constructed, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the time series features of multi-dimensional parameters, the hidden layer is used to learn the time series evolution features of the parameters, and the output layer is used to output the results related to circuit breaker risk. Based on the results related to circuit breaker risk, the network structure parameters of the multi-parameter time series prediction model are determined. The network structure parameters include the number of hidden layers and the number of neurons, so that the multi-parameter time series prediction model can conform to the gradual development law of circuit breaker failure.

5. The method according to claim 4, characterized in that, The process involves learning fault evolution patterns through offline training, processing the extracted and fused parameters using a trained multi-parameter time-series prediction model, and outputting the circuit breaker risk level and predicted circuit breaker time, including: The training samples are collected from operational data covering all stages, including normal operation, minor anomalies, severe anomalies, and circuit breaker precursors, and include data from various scenarios, including single-factor gradual changes and multi-factor sudden changes. The training samples are preprocessed, including time alignment and outlier removal. The preprocessed training samples are input into the constructed multi-parameter time series prediction model for offline training, so that the multi-parameter time series prediction model learns the fault evolution law and is trained until the prediction accuracy of the multi-parameter time series prediction model is greater than or equal to the preset accuracy and the early warning time is greater than or equal to the preset threshold. The real-time parameters after feature extraction and fusion are input into the trained multi-parameter time series prediction model. After processing, the multi-parameter time series prediction model outputs the circuit breaker risk level and the expected circuit breaker time.

6. The method according to claim 1, characterized in that, The step of generating a corresponding early warning signal based on the circuit breaker risk level output by the prediction model includes: The risk levels of circuit breakers are divided into low risk, medium risk, high risk, and emergency risk. When the output fuse failure risk level is low risk, a warning signal is generated that only provides remote notification; when the output fuse failure risk level is medium risk, a warning signal is generated that provides audible and visual warnings and remotely pushes the warning information; when the output fuse failure risk level is high risk, a warning signal is generated that provides audible and visual warnings, remote notifications, and triggers reduced power operation; when the output fuse failure risk level is emergency risk, a warning signal is generated that provides audible and visual warnings, remote emergency notifications, and immediately triggers power outage protection.

7. The method according to claim 6, characterized in that, The generated push notification processing suggestions include: Establish a database of handling suggestions corresponding to the risk level of circuit breaker, which stores specific handling measures for different risk levels; Based on the circuit breaker risk level output by the multi-parameter time series prediction model, the corresponding specific handling measures are retrieved from the handling suggestion database and pushed to the operation and maintenance terminal. The specific handling measures include, but are not limited to, adjusting the drive current, checking the heat dissipation of the laser, and checking the aging degree of components, one or more of these measures.

8. The method according to claim 1, characterized in that, The process of collecting new operational data and incrementally training a multi-parameter time-series prediction model to adapt to changes in fault evolution patterns caused by laser component aging and changes in the operating environment includes: The system is set to run for a preset duration as one incremental training cycle. In each incremental training cycle, newly added operational data is automatically collected, including data from normal and abnormal operating conditions. After performing the same data preprocessing operations as during offline training, the newly added operational data is input into the multi-parameter time series prediction model for incremental training, updating the model parameters corresponding to the multi-parameter time series prediction model, so that the multi-parameter time series prediction model can adapt to the changes in fault evolution patterns caused by laser component aging and changes in the operating environment.

9. A computer device, characterized in that, The computer device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, the memory storing a strategy model, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the method as described in any one of claims 1-8.

Images