Intelligent control system and method for laser
By employing a multi-dimensional calibration process and an adaptive learning cycle, the problem of sensor calibration deviation in laser systems was solved, enabling real-time monitoring and compensation of sensor performance, and maintaining system stability and control accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENGSHENG (HUZHOU) TECHNOLOGY CO LTD
- Filing Date
- 2026-02-02
- Publication Date
- 2026-05-15
AI Technical Summary
In existing laser systems, sensor calibration parameters rely on static compensation curves and uniform calibration benchmarks, failing to incorporate the unique characteristics of individual sensors formed during actual operation due to working hours, load cycles, and environmental exposure history. This results in deviations between calibration actions and the actual degradation path of the sensors, making it impossible to identify and compensate for performance drift in real time, thus affecting control accuracy and system stability.
By establishing an initialization parameter set associated with the laser control sensor, a multi-dimensional calibration process is performed, triggering an adaptive learning cycle for cyclic correction, reconstructing the internal parameter mapping table, and deploying an online verification program to generate the final control strategy, thereby achieving real-time monitoring and compensation for sensor performance drift.
It achieves individualization and adaptability of sensor calibration benchmarks, monitors and compensates for performance drift in real time, maintains stable control quality and output consistency of laser system during long-term operation, and overcomes the adaptation deviation and discrete intervention problems in traditional methods.
Smart Images
Figure CN122051775A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automatic control technology, specifically to an intelligent control system and method for lasers. Background Technology
[0002] In the field of precision control of laser systems, the measurement accuracy of laser control sensors is crucial to ensuring stable output. Existing technologies typically employ calibration methods based on fixed periods or preset thresholds, with calibration parameters largely derived from factory settings or general empirical models. Such approaches rely on static compensation curves and uniform calibration benchmarks, failing to incorporate the unique characteristics of individual sensors developed during actual operation due to varying operating times, load cycles, and environmental exposure histories. This leads to a discrepancy between the calibration process and the sensor's actual degradation path.
[0003] Existing technologies generally lack the ability to process incremental changes in sensor performance online. Sensor parameter drift is a continuous and slow process, while traditional methods can only perform intermittent corrections at specific intervals or after a failure. During continuous operation, subtle performance degradation caused by factors such as component aging and cumulative temperature effects cannot be identified and compensated for in real time, resulting in a gradual decrease in control accuracy over time and potential risks to system stability. Therefore, an intelligent control method is needed that can dynamically construct calibration benchmarks based on the sensor's own historical data and continuously perform closed-loop corrections for performance drift throughout its entire lifecycle, in order to achieve more accurate and robust laser control. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent control system and method for lasers to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides a method for intelligent control of a laser, the method comprising: Establish an initialization parameter set associated with the laser control sensor, the initialization parameter set being derived from the laser system's operation log library and real-time feedback from the environmental monitoring unit; Based on the initialization parameter set, the multi-dimensional calibration process of the laser control sensor is activated. The multi-dimensional calibration process defines the calibration benchmark by parsing the historical operating modes of the laser control sensor. After completing the multi-dimensional calibration process, the adaptive learning cycle of the laser control sensor is triggered. The adaptive learning cycle is based on the performance drift data of the laser control sensor during continuous working periods and is cyclically corrected. Using the results of the cyclic correction, the internal parameter mapping table of the laser control sensor is reconstructed. The internal parameter mapping table is used to correlate the dynamic response curve of the laser control sensor with the power output range of the laser system. Based on the internal parameter mapping table, an online verification program for the laser control sensor is deployed. The online verification program verifies the response protocol of the laser control sensor by simulating load mutation events of the laser system. By integrating the aforementioned response protocol with the inherent characteristic library of the laser control sensor, the final control strategy for the laser control sensor is generated.
[0006] Preferably, the step of activating the multi-dimensional calibration process of the laser control sensor based on the initialization parameter set includes: The static configuration parameters and dynamic operating parameters of the laser control sensor are separated from the initialization parameter set; The static configuration parameters are traced back to the factory calibration records and maintenance files of the laser control sensor. The dynamic operating parameters are filtered to extract trend parameters that characterize the long-term stability of the laser control sensor. By integrating the factory calibration record, the maintenance archives, and the trend parameters, a composite calibration profile of the laser control sensor is constructed. Based on the composite calibration profile, an independent calibration target value is set for each tunable element of the laser control sensor; The calibration operations of all tunable components are performed synchronously, and external disturbance signals of the laser system are introduced during the calibration process to record the anti-interference calibration data of the laser control sensor. Based on the anti-interference calibration data, the independent calibration target value is compensated to form a multi-dimensional calibration output of the laser control sensor.
[0007] Preferably, the step of performing parameter filtering on the dynamic operating parameters to extract trend term parameters characterizing the long-term stability of the laser control sensor includes: The dynamic operating parameters are arranged in a time sequence to form the original parameter time stream of the laser control sensor; The original parameter time stream is processed using a sliding window algorithm to generate a smooth parameter sequence for the laser control sensor; The local variance and local mean of the smoothing parameter sequence within each window are calculated to form a local statistical feature set of the laser control sensor; Identify statistical features that exhibit monotonous variation patterns within the local statistical feature set and mark them as candidate trend features for the laser control sensor; The candidate trend features are subjected to correlation verification, and the feature subset with the highest correlation to the known aging model of the laser control sensor is selected. The parameter components that dominate the direction and rate of change are extracted from the feature subset and defined as trend term parameters characterizing the long-term stability of the laser control sensor.
[0008] Preferably, the adaptive learning period of the trigger laser control sensor includes: After the laser control sensor enters a stable working state, periodic performance sampling is initiated to collect data on the response delay and accuracy attenuation of the laser control sensor at different operating points. The response delay and accuracy attenuation data are input into the learning and evaluation network built into the laser control sensor, and the learning and evaluation network outputs the deviation spectrum between the current performance and the expected performance of the laser control sensor. The distribution pattern of the deviation spectrum is analyzed to identify the main and secondary modes of performance degradation of the laser control sensor; For the primary and secondary modes, incremental learning and reinforcement learning tasks for the laser control sensor are designed respectively. The incremental learning task and the reinforcement learning task are executed alternately, and the internal state evaluation vector of the laser control sensor is updated after each task is executed. The convergence of the internal state evaluation vector is monitored. When the internal state evaluation vector enters the stable region, the adaptive learning cycle of the laser control sensor is determined to be completed, and the current learning parameters are fixed.
[0009] Preferably, inputting the response delay and accuracy attenuation data into the learning and evaluation network built into the laser control sensor includes: The response delay data is normalized and aligned to ensure that the delay data at different operating points have comparable time scales. The accuracy attenuation data is decomposed to identify the causes of attenuation, distinguishing between systematic attenuation caused by internal components of the laser control sensor and random attenuation caused by the external environment. Construct an input layer for the learning evaluation network, wherein the input layer simultaneously receives delayed data that has undergone normalized alignment processing and precision-attenuated data that has undergone causal decomposition; Configure the hidden layer structure of the learning evaluation network, wherein the hidden layer structure includes convolutional units for extracting temporal features of delayed data and recursive units for extracting features associated with precision decay data; Define the output layer function of the learning evaluation network. The output layer function fuses and calculates the features extracted by the hidden layer to generate a quantized performance deviation value and its corresponding confidence score. All performance deviation values and their confidence levels are arranged in order of operating point to form a deviation spectrum describing the difference between the current performance and the expected performance of the laser control sensor.
[0010] Preferably, the step of reconstructing the internal parameter mapping table of the laser control sensor using the result of the cyclic correction includes: Collect all cyclic correction records generated within a complete adaptive learning cycle. These cyclic correction records include the parameter identifiers of the laser control sensor that were adjusted, the adjustment magnitude, and the performance feedback after adjustment. Cluster analysis was performed on the cyclic correction records to identify the adjustment operation clusters that contributed the most to the performance improvement of the laser control sensor; A general parameter adjustment rule is extracted from the adjustment operation cluster. The general parameter adjustment rule defines the optimal adjustment direction and step size of a specific parameter of the laser control sensor under specific conditions. Retrieve the original internal parameter mapping table of the laser control sensor and identify mapping entries that conflict with or are redundant with the general parameter adjustment rules; The general parameter adjustment rules are used to cover or correct conflicting or redundant mapping entries and to supplement new working state mapping relationships not covered in the original internal parameter mapping table. The updated internal parameter mapping table is subjected to integrity and consistency checks to ensure that the laser control sensor can obtain definite parameter guidance by querying the internal parameter mapping table under any known operating state, thereby completing the iterative reconstruction of the internal parameter mapping table.
[0011] Preferably, the online verification procedure for deploying the laser control sensor includes: Based on historical fault data of the laser system and expert experience database, a set of simulated load mutation events is constructed, which cover the entire life cycle scenarios of the laser system from startup, steady-state operation to emergency shutdown. Configure an isolated verification environment for the laser control sensor, which can simulate the real input and output interfaces of the laser system without affecting the actual operating laser system; In the verification environment, the simulated load mutation events are injected sequentially, and the response data of the laser control sensor to each event is recorded throughout the process. Analyze the response process data to extract key response indicators of the laser control sensor, including event detection time, control command generation time, parameter adjustment completion time, and steady-state error after adjustment; The extracted key response indicators are compared with the performance thresholds preset in the laser control sensor to evaluate the laser control sensor's ability to cope with various load change events. Based on the assessment results of the aforementioned coping capabilities, a response protocol for the laser control sensor is generated. The response protocol specifies in detail the specific control logic and parameter switching sequence that the laser control sensor should adopt when facing sudden changes in load of different types and intensities.
[0012] Preferably, the analysis of the response process data and the extraction of key response indicators of the laser control sensor include: From the recorded response process data, the time period from the occurrence of the load mutation event signal to the output of the first control signal from the laser control sensor is extracted, and the time period data is analyzed at the microsecond level to determine the event detection time. The time interval from the generation of the control command to its complete transmission to the laser control sensor execution end is located from the response process data, and the duration of the time interval is calculated as the control command generation time. The time span from receiving the command to reaching and stabilizing at the target value for the relevant controlled parameters of the laser control sensor in the monitoring response process data is recorded as the parameter adjustment completion time. After the parameters are adjusted, a steady-state monitoring window is selected, and the average absolute deviation between the actual output value of the laser control sensor and the target command value is calculated. The average absolute deviation is used as the adjusted steady-state error. The process of extracting the event detection time, control command generation time, parameter adjustment completion time, and steady-state error after adjustment is repeated for all simulated load mutation events to form a dataset of key response indicators of the laser control sensor.
[0013] Preferably, the step of integrating the response protocol with the inherent characteristic library of the laser control sensor to generate the final control strategy for the laser control sensor includes: Access the inherent characteristic library of the laser control sensor, which stores the physical limiting parameters, material fatigue characteristic curves, and officially recommended operating range of the laser control sensor; The control logic and parameter switching sequence in the response protocol are compared with the physical limit parameters in the inherent characteristic library to ensure that no control action will cause the laser control sensor to work beyond its limits. Using the fatigue characteristic curve of the material, the life loss of high-frequency or high-intensity parameter switching actions in the response protocol is evaluated, and optimized alternatives are proposed for actions that may accelerate the aging of the laser control sensor. Within the officially recommended operating range, the response protocol is adaptively adjusted so that the adjusted protocol can meet the requirements of rapid response while ensuring that the laser control sensor can work in a highly efficient and stable region for a long time. The response protocol, after undergoing security verification, lifespan optimization, and working range adaptation adjustment, is compiled into a set of machine instructions that the laser control sensor can directly execute. The machine instruction set is then associated with the corresponding trigger conditions and encapsulated together as the final control strategy of the laser control sensor.
[0014] Preferably, the present invention also includes an intelligent control system for a laser, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the intelligent control method for a laser as described above.
[0015] Compared with the prior art, the beneficial effects of the present invention are: By analyzing the historical operating mode data stored in the laser control sensor, multi-dimensional features are extracted, including runtime distribution over time, temperature and humidity exposure records over the environmental dimension, and operating frequency in different power ranges over the load dimension. A calibration benchmark is then dynamically constructed based on these specific historical characteristics. This allows the calibration benchmark to be set in accordance with the sensor's actual, non-uniform performance degradation trajectory, with calibration parameters directly correlated to the sensor's individual usage history. This overcomes the inherent adaptation bias of traditional methods using uniform, static calibration models. The calibration process possesses traceability and adaptability tailored to the specific individual state of the sensor, thus establishing a more accurate measurement reference from the initial stage.
[0016] During continuous sensor operation, the system monitors and collects performance drift data in real time, including baseline offset of the output signal, response time trends, and gradual increases in noise levels. A closed-loop learning algorithm is used to recursively calculate and iteratively correct these continuous drift data, updating the sensor's internal compensation coefficients and calibration parameters online. This enables continuous tracking and immediate compensation for the sensor's slow performance degradation. It transforms performance maintenance from discrete, passive external intervention into a proactive, continuous self-adjusting process embedded within the workflow. This technology suppresses the time-varying growth of control errors caused by component aging and accumulated environmental stress, enabling the system to maintain stable control quality and output consistency during long-term operation. Attached Figure Description
[0017] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent control method for lasers described in this invention. Figure 2 Flowchart for activating the multi-dimensional calibration process; Figure 3 A flowchart for evaluating network data inputs for learning; Figure 4 A dual-index analysis diagram for a laser load abrupt change scenario; Figure 5 A radar chart comparing the control performance of lasers from multiple dimensions. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] Please see Figure 1 This invention provides a method for intelligent control of a laser. The method includes: establishing an initialization parameter set associated with a laser control sensor, the data source of which includes the laser system's operation log library and real-time feedback from an environmental monitoring unit. Based on this initialization parameter set, the method activates a multi-dimensional calibration process for the laser control sensor, which defines its calibration benchmark by parsing the sensor's historical operating modes. After the calibration process is completed, an adaptive learning cycle of the laser control sensor is triggered, which is cyclically corrected based on performance drift data generated by the sensor during continuous operation. Using the results of the cyclic correction, the method reconstructs the internal parameter mapping table of the laser control sensor, which is used to correlate the sensor's dynamic response curve with the power output range of the laser system. According to the reconstructed internal parameter mapping table, an online verification program for the laser control sensor is deployed, which verifies the response protocol followed by the sensor by simulating load mutation events in the laser system. The method integrates the response protocol obtained from the online verification with the inherent characteristic library of the laser control sensor to generate a final control strategy executable by the sensor, thereby completing the entire intelligent control closed loop.
[0020] In one embodiment of the present invention, see [reference] Figure 2The static configuration parameters and dynamic operating parameters of the laser control sensor are separated from the initialization parameter set. The static configuration parameters include the laser control sensor's serial number, factory-preset gain coefficient, and zero-point bias voltage value. The dynamic operating parameters include the sequence of operating temperature readings and output power feedback values collected per second during the laser control sensor's operation over the past 1000 hours. The static configuration parameters are traced back to their source by accessing the laser system's backend database to query the factory calibration record corresponding to the laser control sensor's serial number, obtaining the initial sensitivity data of the laser control sensor under standard temperature and power conditions, and simultaneously searching the laser system's maintenance logs to find the zero-point offset correction values recorded in the laser control sensor's three most recent calibration operation files. Parameter filtering is performed on dynamic operating parameters to extract trend parameters characterizing the long-term stability of the laser control sensor. The process involves arranging the operating temperature, a dynamic operating parameter, in timestamp order to form the raw parameter time-series stream of the laser control sensor. A sliding window algorithm is applied to process this raw parameter time-series stream, with the window size set to ten hours and sliding in one-hour increments. The moving average of the data within each window is calculated to generate a smoothed parameter sequence for the laser control sensor. The local variance and local mean of the smoothed parameter sequence within each ten-hour window are then calculated, forming a local statistical feature set of the laser control sensor composed of multiple window statistics. The statistical feature of a monotonically increasing local mean across twenty consecutive windows in a local statistical feature set is identified and marked as a candidate trend feature for the laser control sensor. The candidate trend features are then subjected to correlation verification, and the Pearson correlation coefficient between the trend feature and the known photoelectric conversion efficiency aging model of the laser control sensor is calculated. A subset of features with correlation coefficients higher than a preset threshold is selected. From this subset, parameter components with a dominant positive change direction and a change rate of 0.5 degrees Celsius per thousand hours are extracted and defined as trend term parameters characterizing the long-term stability of the laser control sensor. The extraction process of these trend term parameters conforms to the following mathematical relationship:
[0021] in: Indicates the first The trend parameter values calculated from each sampling point Is to assign the first [part] within the sliding window The weighting coefficients for each data point Indicates the smoothing parameter sequence at the th The change in the local mean of each window relative to the previous window. It is the first The time interval between windows, It is the size of the sliding window.
[0022] In some embodiments, a composite calibration profile of the laser control sensor is constructed by integrating initial sensitivity data obtained from the factory calibration record, zero-point offset correction values recorded in previous maintenance files, and extracted 0.5°C temperature rise trend parameters per thousand hours. This composite calibration profile is a three-dimensional data matrix, with the three dimensions representing the theoretical reference value, historical correction amount, and long-term drift prediction value, respectively. Based on the composite calibration profile, independent calibration target values are set for each tunable element of the laser control sensor. For example, a new bias voltage target value is set for the temperature compensation circuit of the laser control sensor, and a new amplification factor target value is set for the gain amplifier of the laser control sensor. The calibration operations of the internal temperature compensation circuit and gain amplifier of the laser control sensor are performed synchronously. During the calibration command issuance process, a 5Hz square wave signal is introduced as an external disturbance signal through the laser system's test interface to record the fluctuation data of the laser control sensor output under disturbance as anti-interference calibration data. It is understandable that, based on the maximum fluctuation amplitude and stabilization time recorded in the anti-interference calibration data, the independent calibration target value of the temperature compensation circuit is adjusted within a range of ±3%, and the independent calibration target value of the gain amplifier is adjusted within a range of ±5%, forming a multi-dimensional calibration output of the laser control sensor that includes the specific parameter values after compensation. Optionally, when constructing the composite calibration profile, in addition to integrating the basic data source, the historical task load rate of the laser control sensor is also introduced as a weighting factor to weight and correct the long-term drift prediction value, making the calibration profile more closely reflect the actual working intensity of the laser control sensor.
[0023] In one embodiment of the present invention, see [reference] Figure 3 In practice, after the laser control sensor completes the multi-dimensional calibration process and enters a stable working state, periodic performance sampling is initiated. The sampling period is set to once every 10 minutes, with each sampling lasting 30 seconds. Response delay data and accuracy attenuation data are collected at three different rated power operating points: 20%, 50%, and 80%. The response delay data is recorded as the number of milliseconds required for the laser control sensor to receive a power adjustment command and for its feedback signal to reach 95% of the target value. The accuracy attenuation data is recorded as the absolute value of the percentage error between the laser control sensor's reading and the high-precision reference instrument reading during steady-state output.
[0024] Before inputting the response delay data and accuracy attenuation data into the learning and evaluation network built into the laser control sensor, data preprocessing is performed. The response delay data from different operating points are normalized and aligned. This is done by using the maximum response delay data collected from the three operating points as a benchmark, dividing the delay data from each operating point by this benchmark value, so that all delay data are mapped to the 0-1 interval and have comparable time scales. The accuracy attenuation data is then decomposed based on the correlation analysis between the internal temperature sensor readings of the laser control sensor and external environmental vibration monitoring data. Accuracy attenuation highly synchronized with fluctuations in environmental vibration monitoring data is marked as random attenuation caused by the external environment. Accuracy attenuation that monotonically changes with the internal temperature of the laser control sensor and is unrelated to environmental vibration is marked as systematic attenuation caused by internal components of the laser control sensor.
[0025] The input layer of the learning evaluation network is constructed with five input nodes. Two nodes receive three operational point delay data points after normalization and alignment, and the other three nodes receive three operational point systematic and random precision decay data points after causal decomposition. The hidden layer structure of the learning evaluation network is configured, containing a one-dimensional convolutional unit with a kernel size of 3 for extracting temporal features from the delay data, and a long short-term memory recursive unit with 32 memory units for extracting associated features from the precision decay data. The output layer function of the learning evaluation network is defined. This function concatenates the feature vectors output by the convolutional unit and the recursive unit, and then calculates them through a fully connected layer to generate three quantized performance bias values and a corresponding confidence level for each bias value. The calculation of the performance bias values follows the following relationship:
[0026] in: Indicates the first Performance deviation value corresponding to each operating point This represents the total number of features extracted from the hidden layer. It is the first The working point corresponds to the first The weight coefficients of each feature It is the first one that the learning evaluation network actually extracts. 1 eigenvalue, This is the first time the laser control sensor has reached its expected performance state. The standard value of each feature, and the confidence level is based on the feature. The stability variance was calculated over five consecutive samplings. The performance deviation values and their confidence levels at all operating points were arranged in ascending order of operating point power, forming a deviation spectrum describing the difference between the current and expected performance of the laser control sensor.
[0027] In some embodiments, the distribution pattern of the analytical deviation spectrum is determined by analyzing the relative magnitudes and trends of three performance deviation values. The performance degradation mode corresponding to the operating point with the largest deviation value and a confidence level higher than 0.9 is identified as the primary mode, and the performance degradation mode with the second largest deviation value and continuous deterioration over three consecutive samples is identified as the secondary mode. For the primary and secondary modes, incremental learning and reinforcement learning tasks are designed for the laser control sensor, respectively. The incremental learning task is defined as exploring fine-tuning of the gain coefficient affecting the primary mode in steps of 0.01 based on the current parameters of the laser control sensor. The reinforcement learning task is defined as allowing the laser control sensor to explore a new set of compensation curve parameters different from the standard response under the load conditions corresponding to the secondary mode in a simulation environment.
[0028] It can be understood that incremental learning tasks and reinforcement learning tasks are executed alternately, with the execution order being one reinforcement learning task after completing one incremental learning task. After each task execution, a 5-dimensional internal state evaluation vector of the laser control sensor is updated based on the change in the performance deviation value of the laser control sensor before and after the task execution. Each dimension of the vector represents the health score of a key performance indicator. The convergence of the internal state evaluation vector of the laser control sensor is monitored. The monitoring standard is to calculate the rate of change of the Euclidean distance of the internal state evaluation vector after 10 consecutive updates. When the rate of change is less than 0.0001, the internal state evaluation vector of the laser control sensor is determined to have entered the stable region. When the internal state evaluation vector of the laser control sensor enters the stable region, the current adaptive learning cycle is considered to be completed, and the optimal gain coefficient adjustment determined by the incremental learning task and the new compensation curve parameters verified by the reinforcement learning task are solidified and written into the non-volatile configuration memory of the laser control sensor.
[0029] Optionally, when normalizing and aligning delayed data, the reference value used can be dynamically updated to the maximum delay value among the most recent 100 samples to adapt to long-term changes in the response characteristics of the laser control sensor. In some embodiments, when performing causal decomposition on the precision decay data, in addition to using correlation analysis, spectral analysis is also introduced to classify high-frequency components in the precision decay signal as random decay and low-frequency trend components as systematic decay. It can be understood that when the Long Short-Term Memory recursive unit processes the correlation features of precision decay data, its input sequence is the most recent 20 sets of systematic and random precision decay data arranged in chronological order.
[0030] In one embodiment of the invention, all cyclic correction records generated within a complete adaptive learning cycle are collected. A complete adaptive learning cycle lasts 72 hours, generating 360 cyclic correction records. Each cyclic correction record includes the parameter identifier of the laser control sensor that was adjusted, the adjustment range, and the performance feedback after adjustment. The adjustment range is expressed as a numerical increment or percentage. The performance feedback record after adjustment is the percentage reduction in steady-state error or the percentage improvement in response speed of the laser control sensor in the next sampling cycle after parameter adjustment. Cluster analysis is performed on the 360 cyclic correction records using the K-means algorithm based on Euclidean distance. The number of clusters is set to five. The adjustment operation cluster that contributes the most to the overall performance improvement of the laser control sensor is identified among the five clusters. The contribution is measured by the percentage reduction in average steady-state error brought about by all operations within the cluster. The adjustment operation cluster with the largest contribution contains 50 records, and its average steady-state error reduction percentage is 20%.
[0031] A general parameter adjustment rule was extracted from the adjustment operation cluster that contributed the most. The relationship between parameter identifiers, adjustment ranges, and corresponding operating conditions in fifty records within the cluster was analyzed. The general parameter adjustment rule is stated as follows: "When the laser system operating temperature is higher than 45 degrees Celsius and the steady-state error monitored by the laser control sensor is greater than 2%, the value of the internal temperature compensation gain parameter K_t of the laser control sensor is increased by 0.01 to 0.03." This rule defines the optimal adjustment direction and step size for specific parameters of the laser control sensor under specific conditions. The original internal parameter mapping table of the laser control sensor was retrieved. The original internal parameter mapping table is stored in the form of a two-dimensional database table. Mapping entries in the original internal parameter mapping table that conflict with or are redundant with the general parameter adjustment rule were identified.
[0032] In some embodiments, when performing cluster analysis on cyclic correction records, a sliding window analysis based on operation time series is introduced, with a window size of ten consecutive operations, to identify micro-clusters that frequently occur within a short time window and have similar adjustment patterns. The identification of micro-clusters follows the following relationship:
[0033] in: This represents a quantification of the clustering contribution of a micro-cluster within a sliding window. This indicates the number of cyclic correction records contained in the window. The first one in the window The performance improvement resulting from this adjustment. Indicates assigning the first The weighting coefficient of this operation, the weighting coefficient It is calculated based on the reciprocal of the time interval between the current operation and the previous operation. This can be understood as calculating the time interval of each sliding window. Values can help locate sequences of operations that occur intensively and have significant effects in a short period of time. These sequences are of high reference value for refining general parameter adjustment rules.
[0034] Optionally, when supplementing new working state mapping relationships, the parameter values in the newly added records can be set to the median value of the parameter adjustment magnitude calculated from the adjustment operation cluster that contributes the most, rather than directly using the result of a single operation, to enhance the robustness of the parameter settings. In some embodiments, consistency verification is implemented by constructing a state transition graph model, where the nodes of the graph represent different working states in the internal parameter mapping table, and the edges represent possible transitions between states. The verification process verifies that along any path in the graph, the parameter instruction sequence associated with the node will not cause unexpected jumps or oscillations in the output of the laser control sensor.
[0035] In one embodiment of the invention, a set of simulated load surge events is constructed based on historical fault data of the laser system and an expert experience database. The historical fault data comes from 120 abnormal event reports recorded during the laser system's operation over the past three years. The expert experience database consists of 50 typical fault mode descriptions provided by the equipment manufacturer. The simulated load surge events cover the entire lifecycle of the laser system, from startup and steady-state operation to emergency shutdown. Specifically, these include simulating a 20% drop in power supply voltage within five milliseconds, simulating a 50% sudden decrease in cooling system flow, and simulating a sudden instability of the optical resonator causing a 150% instantaneous surge in output power. An isolated verification environment is configured for the laser control sensor. This isolated verification environment uses a separate hardware-in-the-loop simulator. The hardware-in-the-loop simulator can simulate the actual input / output interfaces of the laser system, including simulating the 16-bit analog input channel and 24 digital control signal output channels of the laser control sensor. All operations of the hardware-in-the-loop simulator are implemented through a physically isolated network and data bus, and will not affect the actual operating laser system.
[0036] In an isolated verification environment, simulated power supply voltage drop events, simulated cooling flow reduction events, and simulated instantaneous power surge events were sequentially injected. The response process data of the laser control sensor for each event was recorded throughout the process, with a sampling frequency of 1 kHz. The recorded content included all input signal waveforms, internal state variable sequences, and output control command sequences of the laser control sensor. The recorded response process data was analyzed to extract key response indicators of the laser control sensor. The time interval from the occurrence of the load mutation event signal to the output of the first control signal from the laser control sensor was extracted from the recorded response process data. Microsecond-level analysis was performed on this time interval data. The event detection time was determined by detecting the precise moment when the laser control sensor input signal exceeded a preset threshold and the rising edge of the first output control command. The time interval from the generation of the control command to its complete transmission to the laser control sensor's execution end was located from the response process data. The starting point of control command generation was defined as the setting of the output flag of the laser control sensor's internal decision algorithm, and the ending point was defined as the stabilization of the corresponding actuator's drive signal level. The duration of the calculated time interval was used as the control command generation time. The monitoring response process data includes the time span from receiving the command to reaching and stabilizing the controlled parameter of the laser control sensor at the target value. For example, the set voltage value of the temperature compensation circuit is used as a control parameter. The stability criterion is that the fluctuation range of this parameter value within one hundred consecutive sampling points is less than 0.1% of its range. This is recorded as the parameter adjustment completion time. After parameter adjustment is completed, a ten-second steady-state monitoring window is selected, and the average absolute deviation between the actual output value of the laser control sensor and the target command value is calculated. This average absolute deviation is taken as the adjusted steady-state error. The steady-state error is calculated as follows:
[0037] in: This represents the adjusted steady-state error. This indicates the total number of sampling points within the steady-state monitoring window. Indicates the laser control sensor in the first The actual output value of each sampling point This represents the target command value. The process of extracting the event detection time, control command generation time, parameter adjustment completion time, and steady-state error after adjustment is repeated for all simulated load mutation events to form a dataset of key response indicators for the laser control sensor. See Table 1; some data is recorded in Table 1.
[0038] Table 1: Record of Key Response Indicators for Simulated Load Sudden Events
[0039] Understandably, the extracted key response indicators are compared with the performance thresholds preset within the laser control sensor. These performance thresholds are pre-set as follows: event detection time no greater than 10 milliseconds, control command generation time no greater than 5 milliseconds, parameter adjustment completion time no greater than 50 milliseconds, and steady-state error no greater than 0.5%. This comparison evaluates the laser control sensor's ability to handle various load surge events. Based on the evaluation results, a response protocol for the laser control sensor is generated. This protocol is defined in a structured document. For example, it stipulates that when faced with a "power supply voltage drop of 20%" event, the laser control sensor should issue a sequence of commands to "switch to the backup power bus and increase the rectifier gain" within 5 milliseconds after detecting the event. When faced with a "power surge of 150% instantaneously," it should immediately trigger the control logic to "quickly shut down the pump source and insert an optical attenuator."
[0040] In some embodiments, when constructing simulated load surge events, in addition to single event injection, a composite scenario in which two events occur consecutively within a short time interval is also simulated, such as simulating a power fluctuation followed immediately by a cooling failure, to verify the response protocol of the laser control sensor to complex sudden conditions. Optionally, when analyzing response process data to determine the event detection time, the preset threshold used can be dynamically adjusted according to the average power level of the laser system currently in operation to improve the sensitivity and accuracy of detection. In some embodiments, the controlled parameters on which the parameter adjustment completion time is recorded can simultaneously include multiple physical quantities such as temperature, voltage, and current, and the time when the latest stable target value is reached is taken as the final parameter adjustment completion time. It is understood that the length of the steady-state monitoring window can be adaptively set according to the thermal stability time constant of the laser system under different loads to ensure that the calculated steady-state error can truly reflect the long-term performance after adjustment.
[0041] See Figure 4 This is a dual-indicator analysis chart for laser load surge scenarios. "Voltage drop + cooling failure" can be classified as a high-risk scenario, requiring strengthened corresponding control strategies. For the response shortcomings in combined scenarios, the detection logic or parameter adjustment strategies can be adjusted to narrow the performance gap compared to single scenarios. Based on response data from different scenarios, differentiated fault handling procedures can be developed to improve the laser system's anti-interference capability. The core value of this chart lies in the performance evaluation and strategy optimization of laser intelligent control. It intuitively demonstrates that the control performance is optimal in the "single power surge" scenario, indicating that the control logic for this event is mature. The performance shortcomings in combined scenarios can guide the iteration direction of subsequent control strategies.
[0042] In one embodiment of the present invention, the inherent characteristic library of the laser control sensor is accessed. The inherent characteristic library of the laser control sensor is stored in the encrypted storage area of the main controller of the laser system. The inherent characteristic library stores the physical limit parameters, material fatigue characteristic curves, and officially recommended operating range of the laser control sensor. The physical limit parameters include the maximum allowable operating temperature of the laser control sensor being 85 degrees Celsius, the maximum withstand current being 500 mA, and the highest sampling rate being 2 kHz. The material fatigue characteristic curves record the percentage of light response decay of the core photodiode of the laser control sensor under different stress levels of switching cycles of 100,000, 1 million, and 10 million times in the form of a data table. The officially recommended operating range specifies that the long-term operating temperature range of the laser control sensor is 10 to 40 degrees Celsius, and the recommended operating current is 100 to 300 mA. The control logic and parameter switching sequence in the response protocol generated by the online verification program are compared with the physical limit parameters in the inherent characteristic library of the laser control sensor for safety verification. The response protocol includes a control logic that "immediately increases the internal bias current of the laser control sensor to 450 mA and maintains it for 100 milliseconds when a power overshoot is detected." The verification process involves comparing the 450 mA current value in this logic with the maximum withstand current of 500 mA in the inherent characteristic library to confirm that it does not exceed the physical limit. The response protocol also includes a parameter switching sequence that "sets the temperature control target of the laser control sensor to 50 degrees Celsius in emergency heat dissipation mode." The verification process involves comparing 50 degrees Celsius with the maximum allowable operating temperature of 85 degrees Celsius to confirm safety.
[0043] Using material fatigue characteristic curves from the inherent characteristic library of laser control sensors, the lifespan loss of high-frequency or high-intensity parameter switching actions in the response protocol is assessed. The response protocol defines a high-frequency action of "switching the internal gain level of the laser control sensor every thirty seconds" to cope with periodic load fluctuations. The assessment process involves querying the fatigue data of the corresponding gain switching element in the material fatigue characteristic curve, calculating the equivalent stress cycle number corresponding to each switch, and assessing the total damage degree of this action after 10,000 hours of continuous operation based on the linear cumulative damage theory. An optimized alternative is proposed for actions that may accelerate the aging of the laser control sensor. The optimized alternative is to modify "switching the internal gain level of the laser control sensor every thirty seconds" to "adaptively adjusting the switching cycle according to the load fluctuation amplitude, with a minimum interval of not less than sixty seconds". Within the officially recommended operating range in the inherent characteristic library of the laser control sensor, the response protocol is adaptively adjusted. The officially recommended operating range specifies that the long-term operating temperature range of the laser control sensor is 10 to 40 degrees Celsius. Although the "50-degree Celsius target in emergency heat dissipation mode" in the response protocol does not exceed the physical limit, it exceeds the long-term recommended range. The adaptive adjustment is to lower the temperature target value in this mode from 50 degrees Celsius to 40 degrees Celsius and correspondingly extend the heat dissipation execution time. This ensures that the adjusted protocol can meet the requirements of rapid response while ensuring that the laser control sensor can operate in a highly efficient and robust range for a long time.
[0044] The response protocol, after undergoing safety verification, lifespan optimization, and operating range adaptation, is compiled into a machine instruction set directly executable by the laser control sensor. The compilation process uses a dedicated compiler provided by the laser control sensor manufacturer to convert the high-level control logic descriptions in the protocol into a series of binary opcodes for the laser control sensor's microcontroller. For example, the logic of "increasing the bias current to 450 mA" is compiled into a specific write instruction sequence for the corresponding digital potentiometer register. The compiled machine instruction set is then associated with the corresponding trigger conditions in the response protocol, such as "when the laser system main control unit sends an overshoot alarm code 0x5A." The association operation uses this alarm code as an index to establish a mapping table with the compiled binary instruction sequence in the laser control sensor's flash memory. This mapping table is then encapsulated into the final control strategy of the laser control sensor. The final control strategy is stored in an encrypted configuration file and loaded into the laser control sensor's runtime memory upon startup.
[0045] In some embodiments, the linear cumulative damage theory used in lifetime loss assessment follows the following relationship:
[0046] in: It indicates the cumulative degree of damage to a specific material or component. This indicates the number of stress levels assessed. The response protocol is expected to be in effect during the lifecycle of the laser control sensor, at the [number]th [year]. The number of actions that occur under a certain stress level. This indicates the value obtained from the material fatigue characteristic curve, corresponding to the first... The number of cycles required for the material or component to fail at a given stress level. This can be understood as the calculated cumulative damage level... When the value approaches or exceeds a threshold, the corresponding parameter switching action is determined to need optimization.
[0047] Optionally, in addition to comparing numerical limits, safety verification also includes composite safety verification of the combined effects of multiple parameters. For example, verifying whether the combined power consumption exceeds the maximum heat dissipation capacity of the laser control sensor package when simultaneously increasing the operating current and sampling rate of the laser control sensor. When adaptively adjusting the operating range of the response protocol, a penalty function can be introduced into the adjustment process. This function weighs the performance gains of the control action against the degree of deviation from the officially recommended operating range, and can automatically solve for an approximately optimal adjustment scheme that satisfies the constraints. It can be understood that the compiled machine instruction set needs to undergo a round of simulated execution verification before packaging. This simulated execution verification is performed in a virtual laser control sensor environment to confirm that the instruction set can trigger the expected actions without errors.
[0048] See Figure 5 This is a radar chart comparing the performance of laser control across multiple dimensions. The optimized performance comprehensively surpasses the original performance in all dimensions, indicating that the optimization achieved a full-dimensional performance improvement in laser control, without any weaknesses in any single dimension. This directly demonstrates the effectiveness of the optimized control strategy, avoiding the imbalance problem of "improvement in one dimension while decline in others," reflecting the holistic nature of the optimization solution. The original performance was relatively weak in dimensions such as "anti-interference capability and event detection speed," suggesting that the control logic before optimization was insufficient in handling sudden scenarios and suppressing interference. The results of the full-dimensional improvement support the large-scale application of subsequent laser control strategies and also clarify the direction for the next round of optimization: "further amplifying advantageous dimensions and refining accuracy."
[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for intelligent control of lasers, characterized in that, The method includes: Establish an initialization parameter set associated with the laser control sensor, the initialization parameter set being derived from the laser system's operation log library and real-time feedback from the environmental monitoring unit; Based on the initialization parameter set, the multi-dimensional calibration process of the laser control sensor is activated. The multi-dimensional calibration process defines the calibration benchmark by parsing the historical operating modes of the laser control sensor. After completing the multi-dimensional calibration process, the adaptive learning cycle of the laser control sensor is triggered. The adaptive learning cycle is based on the performance drift data of the laser control sensor during continuous working periods and is cyclically corrected. Using the results of the cyclic correction, the internal parameter mapping table of the laser control sensor is reconstructed. The internal parameter mapping table is used to correlate the dynamic response curve of the laser control sensor with the power output range of the laser system. Based on the internal parameter mapping table, an online verification program for the laser control sensor is deployed. The online verification program verifies the response protocol of the laser control sensor by simulating load mutation events of the laser system. By integrating the aforementioned response protocol with the inherent characteristic library of the laser control sensor, the final control strategy for the laser control sensor is generated.
2. The intelligent control method for lasers as described in claim 1, characterized in that, The step of activating the multi-dimensional calibration process of the laser control sensor based on the initialization parameter set includes: The static configuration parameters and dynamic operating parameters of the laser control sensor are separated from the initialization parameter set; The static configuration parameters are traced back to the factory calibration records and maintenance files of the laser control sensor. The dynamic operating parameters are filtered to extract trend parameters that characterize the long-term stability of the laser control sensor. By integrating the factory calibration record, the maintenance archives, and the trend parameters, a composite calibration profile of the laser control sensor is constructed. Based on the composite calibration profile, an independent calibration target value is set for each tunable element of the laser control sensor; The calibration operations of all tunable components are performed synchronously, and external disturbance signals of the laser system are introduced during the calibration process to record the anti-interference calibration data of the laser control sensor. Based on the anti-interference calibration data, the independent calibration target value is compensated to form a multi-dimensional calibration output of the laser control sensor.
3. The intelligent control method for lasers as described in claim 2, characterized in that, The step of filtering the dynamic operating parameters to extract trend parameters characterizing the long-term stability of the laser control sensor includes: The dynamic operating parameters are arranged in a time sequence to form the original parameter time stream of the laser control sensor; The original parameter time stream is processed using a sliding window algorithm to generate a smooth parameter sequence for the laser control sensor; The local variance and local mean of the smoothing parameter sequence within each window are calculated to form a local statistical feature set of the laser control sensor; Identify statistical features that exhibit monotonic variation patterns within the local statistical feature set and mark them as candidate trend features for the laser control sensor; The candidate trend features are subjected to correlation verification, and the feature subset with the highest correlation to the known aging model of the laser control sensor is selected. The parameter components that dominate the direction and rate of change are extracted from the feature subset and defined as trend term parameters characterizing the long-term stability of the laser control sensor.
4. The intelligent control method for lasers as described in claim 1, characterized in that, The adaptive learning period of the trigger laser control sensor includes: After the laser control sensor enters a stable working state, periodic performance sampling is initiated to collect data on the response delay and accuracy attenuation of the laser control sensor at different operating points. The response delay and accuracy attenuation data are input into the learning and evaluation network built into the laser control sensor, and the learning and evaluation network outputs the deviation spectrum between the current performance and the expected performance of the laser control sensor. The distribution pattern of the deviation spectrum is analyzed to identify the main and secondary modes of performance degradation of the laser control sensor; For the primary and secondary modes, incremental learning and reinforcement learning tasks for the laser control sensor are designed respectively. The incremental learning task and the reinforcement learning task are executed alternately, and the internal state evaluation vector of the laser control sensor is updated after each task is executed. The convergence of the internal state evaluation vector is monitored. When the internal state evaluation vector enters the stable region, the adaptive learning cycle of the laser control sensor is determined to be completed, and the current learning parameters are fixed.
5. The intelligent control method for lasers as described in claim 4, characterized in that, The step of inputting the response delay and accuracy attenuation data into the learning and evaluation network built into the laser control sensor includes: The response delay data is normalized and aligned to ensure that the delay data at different operating points have comparable time scales. The accuracy attenuation data is decomposed to identify the causes of attenuation, distinguishing between systematic attenuation caused by internal components of the laser control sensor and random attenuation caused by the external environment. Construct an input layer for the learning evaluation network, wherein the input layer simultaneously receives delayed data that has undergone normalized alignment processing and precision-attenuated data that has undergone causal decomposition; Configure the hidden layer structure of the learning evaluation network, wherein the hidden layer structure includes convolutional units for extracting temporal features of delayed data and recursive units for extracting features associated with precision decay data; Define the output layer function of the learning evaluation network. The output layer function fuses and calculates the features extracted by the hidden layer to generate a quantized performance deviation value and its corresponding confidence score. All performance deviation values and their confidence levels are arranged in order of operating point to form a deviation spectrum describing the difference between the current performance and the expected performance of the laser control sensor.
6. The intelligent control method for lasers as described in claim 1, characterized in that, The process of reconstructing the internal parameter mapping table of the laser control sensor using the results of the cyclic correction includes: Collect all cyclic correction records generated within a complete adaptive learning cycle. These cyclic correction records include the parameter identifiers of the laser control sensor that were adjusted, the adjustment magnitude, and the performance feedback after adjustment. Cluster analysis was performed on the cyclic correction records to identify the adjustment operation clusters that contributed the most to the performance improvement of the laser control sensor; A general parameter adjustment rule is extracted from the adjustment operation cluster. The general parameter adjustment rule defines the optimal adjustment direction and step size of a specific parameter of the laser control sensor under specific conditions. Retrieve the original internal parameter mapping table of the laser control sensor and identify mapping entries that conflict with or are redundant with the general parameter adjustment rules; The general parameter adjustment rules are used to cover or correct conflicting or redundant mapping entries and to supplement new working state mapping relationships not covered in the original internal parameter mapping table. The updated internal parameter mapping table is subjected to integrity and consistency checks to ensure that the laser control sensor can obtain definite parameter guidance by querying the internal parameter mapping table under any known operating state, thereby completing the iterative reconstruction of the internal parameter mapping table.
7. The intelligent control method for lasers as described in claim 1, characterized in that, The online verification procedure for deploying the laser control sensor includes: Based on historical fault data of the laser system and expert experience database, a set of simulated load mutation events is constructed, which cover the entire life cycle scenarios of the laser system from startup, steady-state operation to emergency shutdown. Configure an isolated verification environment for the laser control sensor, which can simulate the real input and output interfaces of the laser system without affecting the actual operating laser system; In the verification environment, the simulated load mutation events are injected sequentially, and the response data of the laser control sensor to each event is recorded throughout the process. Analyze the response process data to extract key response indicators of the laser control sensor, including event detection time, control command generation time, parameter adjustment completion time, and steady-state error after adjustment; The extracted key response indicators are compared with the performance thresholds preset in the laser control sensor to evaluate the laser control sensor's ability to cope with various load change events. Based on the assessment results of the aforementioned coping capabilities, a response protocol for the laser control sensor is generated. The response protocol specifies in detail the specific control logic and parameter switching sequence that the laser control sensor should adopt when facing sudden changes in load of different types and intensities.
8. The intelligent control method for a laser as described in claim 7, characterized in that, The analysis of the response process data extracts key response indicators of the laser control sensor, including: From the recorded response process data, the time period from the occurrence of the load mutation event signal to the output of the first control signal from the laser control sensor is extracted, and the time period data is analyzed at the microsecond level to determine the event detection time. The time interval from the generation of the control command to its complete transmission to the laser control sensor execution end is located from the response process data, and the duration of the time interval is calculated as the control command generation time. The time span from receiving the command to reaching and stabilizing at the target value for the relevant controlled parameters of the laser control sensor in the monitoring response process data is recorded as the parameter adjustment completion time. After the parameters are adjusted, a steady-state monitoring window is selected, and the average absolute deviation between the actual output value of the laser control sensor and the target command value is calculated. The average absolute deviation is used as the adjusted steady-state error. The process of extracting the event detection time, control command generation time, parameter adjustment completion time, and steady-state error after adjustment is repeated for all simulated load mutation events to form a dataset of key response indicators of the laser control sensor.
9. The intelligent control method for lasers as described in claim 1, characterized in that, The process of integrating the response protocol with the inherent characteristic library of the laser control sensor to generate the final control strategy for the laser control sensor includes: Access the inherent characteristic library of the laser control sensor, which stores the physical limiting parameters, material fatigue characteristic curves, and officially recommended operating range of the laser control sensor; The control logic and parameter switching sequence in the response protocol are compared with the physical limit parameters in the inherent characteristic library to ensure that no control action will cause the laser control sensor to work beyond its limits. Using the fatigue characteristic curve of the material, the life loss of high-frequency or high-intensity parameter switching actions in the response protocol is evaluated, and optimized alternatives are proposed for actions that may accelerate the aging of the laser control sensor. Within the officially recommended operating range, the response protocol is adaptively adjusted so that the adjusted protocol can meet the requirements of rapid response while ensuring that the laser control sensor can work in a highly efficient and stable region for a long time. The response protocol, after undergoing security verification, lifespan optimization, and working range adaptation adjustment, is compiled into a set of machine instructions that the laser control sensor can directly execute. The machine instruction set is then associated with the corresponding trigger conditions and encapsulated together as the final control strategy of the laser control sensor.
10. A smart control system for a laser, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent control method for a laser as described in any one of claims 1 to 9.