A defect identification method of an optical pumped laser

Abstract

The application discloses a kind of defect identification methods of optically pumped laser, it is related to laser detection technical field, comprising: in the self-check window of starting, idle frame window, load switching window or allow window of reducing the rating inside determination diagnosis window and carry out multichannel synchronous response acquisition;Based on acquisition result, constructs multi-working condition response tensor, extracts relational feature, and generates drift normalization residual in combination with health baseline model;According to residual, component-level attribution, in-service health baseline freezing and retest stabilization control are carried out;According to attribution result, output risk level, disposal instruction and execute closed-loop feedback.The method can separate early latent defects under the condition that normal working condition drift, load fluctuation and control compensation coexist, improve the anti-drift capability, attribution divisibility and determination stability, reduce false alarm and enhance the pertinence of maintenance decision.

Description

Technical Field

[0001] This invention relates to the field of laser detection technology, specifically to a method for identifying defects in optically pumped lasers. Background Technology

[0002] Optically pumped solid-state lasers typically use a semiconductor pump source to excite a solid-state gain medium for oscillation or amplification output. They offer high electro-optical efficiency, small size, and easily achieve high repetition rates and high average power, making them commonly used in high-end industrial processing, scientific research, measurement, and high-average-power pulsed systems. Engineering applications for these devices generally include pump drive, power supply, temperature control and cooling, monitoring photodetector, output coupling, and transmission components, along with a controller to monitor operating parameters such as current, voltage, power, and temperature. As repetition rate, average power, and continuous operating time increase, the balance between pump module brightness, efficiency, heat dissipation, and stability becomes increasingly important for monitoring the pump module's operational status.

[0003] US Patent document US20210044081A1 (February 11, 2021) discloses a "laser oscillation device for direct diode laser and a fault diagnosis method for the laser oscillation device." The laser oscillation device provides laser light to a laser processing head and consists of a laser module, a power supply circuit, and a power controller. The laser module is composed of multiple laser diodes connected in series or parallel. The power supply circuit drives the laser module in a constant current manner, and the power controller controls the operation of the power supply circuit according to the drive commands from the system controller. The document discloses that during actual control, current and voltage sensors are used to detect the current and voltage in the laser module. A pre-defined diagnostic map area is generated for the corresponding region to determine whether it is a normal or abnormal region. A correspondence is established between the normal and abnormal regions and the cause of the fault. Thus, when an abnormality is diagnosed, corresponding fault information can be provided to the device, triggering a power supply shutdown or operation stoppage. In other words, the basic working idea of ​​the prior art is to use the current-voltage operating point and operating region as the basis for fault diagnosis to determine the status of the laser oscillation device and assist in fault recovery.

[0004] The aforementioned existing technology can quickly determine faults in direct diode laser oscillators based on current and voltage operating points, but it still has certain shortcomings in high-power, high-repetition-frequency optically pumped solid-state lasers.

[0005] First, the output state of optically pumped solid-state lasers is generally affected by a variety of physical factors. In addition to the pump source power, it also involves multiple coupled effects such as gain medium thermal effects, optical element absorption or contamination, backlight disturbances, cooling conditions, and control loop compensation behavior. The same output attenuation or fluctuation can have different causes. Second, judging solely based on the current-voltage operating point or a small amount of electrical quantity can only reflect faults such as power supply abnormalities, open circuits, or short circuits. It cannot directly distinguish early latent defects that gradually form, especially anomalies caused by temperature rise, load, slight misalignment, or control compensation masking. Third, if the above latent anomalies cannot be detected early during long-term continuous operation, it may lead to problems such as unstable output, deterioration of processing quality, accidental shutdown, increased maintenance and troubleshooting scope, and increased operating costs.

[0006] Therefore, it is necessary to solve the following technical problem: how to better address the issue of early detection of latent defects in optically pumped solid-state lasers under the influence of drift and multi-source coupling during normal operation. Summary of the Invention

[0007] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a defect identification method for optically pumped lasers. Based on the acquired data, a multi-condition response tensor is constructed, relational features are extracted, and a drift-normalized residual is generated in conjunction with a health baseline model. Component-level attribution, in-service health baseline freezing, and retesting stabilization control are performed based on the residuals. Risk levels and handling instructions are output based on the attribution results, and closed-loop feedback is executed. This method can separate early latent defects under conditions of normal operating drift, load fluctuation, and control compensation, improving drift resistance, attribution separability, and judgment stability, reducing false alarms, and enhancing the targeting of maintenance decisions; thus solving the technical problems described in the background art.

[0008] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: A defect identification method for an optically pumped laser includes determining a diagnostic window identifier and determining admission conditions when the power-on self-test window, idle frame window, load switching window, or allowed derating window is in a non-high-risk prohibited state under safety interlock conditions. When the admission conditions are met, a controlled perturbation is issued, and the driving current, driving voltage, monitoring photodiode signal, output power, temperature signal and return light signal are sampled synchronously before, during and after the perturbation to obtain a multi-channel synchronous response dataset; based on the dataset, a multi-condition response tensor is constructed and relational features are generated, and the relational features are processed by a healthy baseline model to obtain drift normalized residuals; The main conclusion, secondary conclusion, risk level and confidence level are obtained by matching the drift normalized residual with the anomaly template. Based on the risk level, the action instructions are generated. The action instructions include at least one of freezing the in-service health baseline, triggering the devaluation retest, maintenance scheduling or shutdown, and are fed back to the diagnostic window scheduling and health baseline model.

[0009] Furthermore, the diagnostic window identifier corresponds to one of the following: power-on self-test window, idle frame window, load switching window, and allow derating window. After receiving the diagnostic window identifier, the diagnostic controller first reads the safety interlock status and then determines whether the current window enters the active diagnostic branch, passive sampling branch, or bypass protection branch.

[0010] Furthermore, the criteria for determining the admission conditions include output stability margin, thermal margin, backlight risk margin, and controller saturation margin. The diagnostic controller forms the admission results of the diagnostic window based on each margin. When the admission results do not meet the active perturbation conditions, controlled perturbations are prohibited from being issued, and only passive sampling under natural operating conditions is retained.

[0011] Furthermore, the controlled perturbation is sent from the diagnostic controller to one of the pump driver, temperature controller, and timing controller, and is limited to a small, short-term, and rollbackable setpoint change; synchronous sampling covers the baseline segment before the perturbation, the perturbation action segment, and the recovery segment, and each channel uses a unified timestamp to write to a unified buffer.

[0012] Furthermore, in addition to driving current, driving voltage, monitoring photodiode signal, output power, temperature signal, and reflected light signal, the multi-channel synchronous response dataset also includes the adjustment and error terms of the stable power controller; the diagnostic controller caches the multi-channel synchronous response dataset according to the operating condition dimension, channel dimension, and time dimension, and writes the sampling quality identifier.

[0013] Furthermore, the process of constructing a multi-condition response tensor and generating relational features includes time alignment, dimensional normalization, and sampling quality screening of the multi-channel synchronous response dataset, and extracting at least two types of relational features from the slope, hysteresis, recovery time, channel ratio, phase difference, and coupling relationship between the controller compensation amount and the main output.

[0014] Furthermore, the health baseline model includes at least one of the factory health baseline, in-service health baseline, and post-maintenance reconstruction baseline. The diagnostic controller selects the health baseline model according to the current model status and projects relational features into the health baseline space to filter out normal thermal drift, load drift, and slight alignment changes, and generate drift normalized residuals.

[0015] Furthermore, the abnormal templates include pump source abnormal templates, gain medium thermal abnormal templates, optical element abnormal templates, and return light path abnormal templates. The diagnostic controller performs matching scoring on each abnormal template based on the drift normalized residual, and outputs the main conclusion and secondary conclusion according to the scoring results. When the matching score is insufficient to distinguish between the main and secondary conclusions, an uncertain state is output.

[0016] Furthermore, when the risk level is low and the confidence level meets the preset conditions, a slow update of the in-service health baseline is allowed; when the risk level is medium, the in-service health baseline is frozen and a protected devaluation retest is initiated; when the risk level is high and the devaluation retest still fails, a maintenance schedule is output or the system is shut down, and consistency checks are performed on multiple test results.

[0017] Furthermore, the handling instructions include: when the risk level is low and the attribution is stable, feeding back the current response characteristics to the health baseline model at a limited rate; when the risk level is medium, feeding back the results of this round to the diagnostic window scheduling logic to determine the next retest window; when the risk level is high, prohibiting further active perturbation and locking the protection priority; and when maintenance is completed, returning to step one to re-establish an effective baseline.

[0018] (III) Beneficial Effects This invention provides a method for defect identification in optically pumped lasers, which relates to smart sensor technology and has the following beneficial effects: By performing diagnostics only during the power-on self-test window, idle frame window, load switching window, or derating window, controlled micro-perturbations are collected and multiple channels are processed simultaneously under the condition that the diagnostic window access is met. This is coordinated with the normal service boundary of the laser and avoids instantaneous disturbances under high-risk conditions from directly entering the diagnostic system, thereby improving the reliability and consistency of subsequent data entering the analysis stage.

[0019] By placing the drive current, drive voltage, monitoring photodiode signal, output power, temperature signal, and reflected light signal into a synchronous sampling system under the same diagnostic window identifier, and through sampling quality identifiers and controller status record constraints, dispersed electrical, optical, and thermal signals are simultaneously extracted to obtain evidence of the same source, so as to continue to provide continuous input for abnormal changes masked by control compensation.

[0020] By constructing a multi-condition response tensor to extract relational features, and building drift-normalized residuals with the health baseline model, the diagnostic object changes from a single-moment absolute anomaly to a response relationship between controlled conditions. This preserves abnormal structural information under normal thermal drift, load fluctuations, and slight alignment changes, further increasing the separability of different defect sources.

[0021] The drift normalized residual is matched with pump source anomaly templates, gain medium thermal anomaly templates, optical component anomaly templates, and return light path anomaly templates, and the main conclusion, secondary conclusion, and confidence level are output. This elevates defect identification results beyond the anomaly alarm level to the component-level attribution level, enhancing the correspondence between maintenance actions and anomaly sources. Risk levels, handling instructions, and closed-loop feedback are integrated into the same operating process. The main control and safety interlock modules can continue operation, reduce operating limits, retest, schedule maintenance, or shut down based on the attribution results. The results, in turn, affect the diagnostic window scheduling and health baseline model, achieving a closed-loop feedback from data collection, modeling, attribution to handling. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall structure of the laser diagnostic system of the present invention; Figure 2 This is a schematic diagram of the diagnostic window identification and controlled perturbation distribution process of the present invention; Figure 3 This is a schematic diagram of the multi-channel synchronous sampling and three-stage buffer timing of the present invention; Figure 4 This is a schematic diagram of the multi-condition response tensor construction and relational feature extraction process of the present invention; Figure 5 This is a schematic diagram of the healthy baseline projection and drift normalization residual generation of the present invention; Figure 6 This is a schematic diagram of the component-level attribution and residual reliability determination process of the present invention; Figure 7 This is a schematic diagram of the closed loop of baseline freezing, rate limiting update, and rate reduction retesting in this invention. Detailed Implementation

[0023] 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.

[0024] Please see Figures 1-7 This invention provides a method for defect identification in optically pumped lasers. Step 1: Without breaking the main operational boundaries of the laser, converge the dispersed, asynchronous, and non-unique driving signals, optical output signals, thermal state signals, and return optical signals into a multi-channel synchronous response dataset within the same diagnostic window. Perform window admission, controlled perturbation, and quality screening at the acquisition source to obtain a unified input that can be used for relational feature construction in subsequent steps.

[0025] The following steps are all executed by the diagnostic controller, which is coupled to the pump driver, temperature controller, timing controller, monitoring photodiode, power sampling branch, temperature sensor, return light detection branch and main control interface respectively; when the model is configured with wavelength detection branch, beam spot detection branch, wavefront detection branch, polarization detection branch or stable power controller interface, the diagnostic controller is simultaneously connected to such expansion channel.

[0026] During continuous operation, high-power, high-repetition-rate optically pumped lasers are simultaneously affected by load fluctuations, heat accumulation, control compensation, optical path micro-offsets, and external coupling disturbances. The pump drive current, drive voltage, monitoring photodiode signal, output power, temperature signal, and return light signal are all affected by load fluctuations, heat accumulation, control compensation, optical path micro-offsets, and external coupling disturbances.

[0027] If a single channel value is directly captured at any given time, only a mixed result can be seen, making it impossible to determine whether the result is normal operating condition drift or an early defect signal. Therefore, step one does not pursue superficial multi-sampling, but first constrains the operating conditions in which sampling occurs, then constructs a controlled disturbance within those conditions, and then incorporates the responses before, during, and after the disturbance into the same time reference, so that each sampling record has a clear operating condition starting point, channel affiliation, and quality level. For ease of understanding, an exemplary field action can be used as an example: After the laser completes a processing pulse output, it enters a reserved idle frame. The main control interface sends an idle frame entry flag to the diagnostic controller. The diagnostic controller first reads the safety interlock status, then reads the current output power, current temperature, current return light level, and the power stabilization controller adjustment amount. When all states are within the allowable range, the diagnostic controller sends a short-term pump setpoint fine-tuning command to the pump driver, and captures the monitoring photodiode signal, output power signal, temperature signal, and return light signal before, during, and after the execution of this command, respectively, ultimately forming a continuous response segment with a unified timestamp. The visible actions on-site are: first, determine the window, then send out controlled perturbations, then sample synchronously, and then cache and encapsulate; the on-site result is that the data of each channel in the same window are aligned to a unified reference.

[0028] Among them, the diagnostic window that can carry out diagnostic actions is selected from the power-on self-test window, idle frame window, load switching window and allow derating window. A controlled perturbation record matching the current model and current operating status is generated in the diagnostic window, so that subsequent sampling is not unconditional capture, but unfolds around the clear operating condition trigger point.

[0029] If the diagnostic window is not defined first, the sampling behavior will be mixed with the main business output, resulting in three direct consequences: First, there is no start and end boundary between the sampling time and the control action, making it impossible to distinguish before and after the disturbance. Second, when the system is under high heat load or strong backlight, actively applying micro-disturbance itself will increase the risk. Third, the power stabilization controller of some models will continuously correct the drive quantity in the background, causing the sampling object and the compensation action to overlap, ultimately making the single captured value lose its clear meaning.

[0030] Therefore, window type, thermal margin, output stability margin, backlight risk margin, and controller saturation margin are treated as continuous variables in the same decision chain. Only when the chain is continuous can the controlled perturbation meet the entry conditions.

[0031] The diagnostic controller first receives the window candidate identifier from the main control interface, and then establishes a diagnostic window identifier for the time period to which the window candidate identifier belongs. Subsequently, the diagnostic controller reads the output stability margin. Heat margin Risk margin of return light and controller saturation margin This information is then written into the same window decision frame. To ensure consistent admission decisions across different device models, the diagnostic controller normalizes the four types of margins and generates diagnostic window admission coefficients:

[0032] Among them, the diagnostic window admission coefficient : Characterizes the overall capability of the current window to carry out proactive diagnostic actions; its value is positive; buffer constant. : To prevent positive constants with a denominator of zero.

[0033] Output stability margin : Represents the perturbation margin of the current output power relative to the target power, and takes a positive value; output stability margin The current output power is obtained from the output power sampling branch and read by the diagnostic controller. Power relative to current mission target And calculate using the following formula:

[0034] Where: Output stability margin : Represents the perturbation margin after the current output power deviates from the target power, and its value range is . Current output power : Obtained from the output power sampling branch; target power : Given by the current task parameters of the main control interface; buffer constant : A positive constant set to prevent the denominator from being zero.

[0035] Heat margin : Represents the remaining space between the current temperature state and the upper temperature boundary, and takes a positive value; thermal margin The temperature is obtained from the temperature sampling branch, and the diagnostic controller reads the current module temperature. Maximum allowable temperature of the model Compared with reference temperature And calculate using the following formula:

[0036] Among them: heat margin : Represents the remaining space between the current temperature and the upper temperature limit boundary, with a value range of . Current module temperature Temperature data is collected by a temperature sensor; upper limit temperature. : As specified by the model's temperature control configuration; Reference temperature : Reference temperature for healthy operation of this model; buffer constant : A positive constant set to prevent the denominator from being zero.

[0037] Controller saturation margin : Represents the remaining space between the current adjustment value of the steady power controller and the saturation boundary, and takes a positive value; controller saturation margin Obtained from the stable power controller interface, the diagnostic controller reads the current adjustment value of the controller. With the upper limit of the controller range And calculate using the following formula:

[0038] Where: controller saturation margin : Represents the remaining space between the controller adjustment amount and the saturation boundary, with a value range of . Current adjustment amount : Read in real time from the stable power controller interface; upper limit of range : Defined by the controller specification; buffer constant : A positive constant set to prevent the denominator from being zero.

[0039] Risk margin of return light : Represents the degree to which the current returned light level is close to the risk threshold, and is a non-negative value; returned light risk margin The returned light amplitude is obtained from the returned light detection branch and read by the diagnostic controller. risk threshold of aircraft model And obtained by normalization:

[0040] Among them: risk margin for backlighting : Indicates how close the current reflected light level is to the risk threshold, and takes a non-negative value; current reflected light amplitude. : Obtained from the return light detection branch; risk threshold : Given by the model calibration value; buffer constant : A positive constant set to prevent the denominator from being zero.

[0041] The physical meaning of this expression is that whether the diagnostic window allows active perturbation does not depend on any single margin, but on the degree to which the three types of bearable capabilities—output, thermal, and control—combined cover the risk of backlighting.

[0042] The diagnostic controller does not treat the power-on self-test window, idle frame window, load switching window, and derating allow window as four equal entry points. Instead, it first determines the timing position based on the window category, and then uses the diagnostic window admission coefficient. Determine the execution level. If the diagnostic window admission coefficient... If the active diagnostic threshold for the corresponding model is exceeded, the diagnostic controller enters the active perturbation branch; if the diagnostic window admission coefficient... If the threshold is below the active diagnostic threshold but still above the passive diagnostic threshold, the diagnostic controller enters the passive sampling branch; if the diagnostic window admission coefficient... If the threshold is below the passive diagnostic threshold, the diagnostic controller directly bypasses the current round of diagnosis and maintains the protection link as the priority. The key to this decision chain is not the threshold itself, but rather binding the window category with the overall margin, allowing the window's diagnostic capability and carrying capacity to be determined uniformly at the same point. As a supplement: active and passive diagnostic thresholds are configured parameters: active diagnostic threshold... :when Entering the active perturbation branch; passive diagnostic threshold. :when When entering the passive sampling branch; This round of bypass diagnosis.

[0043] Furthermore, the diagnostic controller writes the diagnostic window identifier. Simultaneously, the window's start time, end time, the reason for the previous window's exit, and the current window's admission branch type are written to the window, thus forming a complete window context. The direct result of this setting is that any subsequent sampling record is not isolated data, but is accompanied by upstream semantics such as which window it came from, why it could enter, and whether it entered an active or passive branch.

[0044] In use, firstly, the original coarse triggering that only relied on window type is changed to fine triggering that combines window type with margin judgment, so that active perturbation will not be mistakenly triggered at high-risk boundaries; secondly, a unified working condition reference is fixed for subsequent steps; thirdly, the complete preservation of the window context enables subsequent steps to associate the residual differences of different windows with the window's carrying capacity.

[0045] When the diagnostic window enters the active perturbation branch, the diagnostic controller generates a controlled perturbation record in the following order: first, define the perturbation target; then, define the perturbation amplitude; and finally, define the perturbation duration. The preferred object for the controlled perturbation record is the pump driver, followed by the temperature controller or timing controller. In the same round of active diagnostics, the diagnostic controller issues a primary perturbation to only one primary object, maintaining the current settings for the remaining objects to avoid weakening interpretability due to the superposition of multiple perturbations.

[0046] The controlled perturbation amplitude is not a fixed constant, but is determined by the combined carrying capacity of the current window and the sensitivity of the observed object. Therefore, the diagnostic controller generates the controlled perturbation amplitude:

[0047] Among them, the amplitude of controlled perturbation : Characterizes the actual change range of the setpoint issued in this round of active diagnosis, with a positive value range; basic perturbation quantity The nominal perturbation start point pre-configured for the current model is a positive value; the diagnostic window admission coefficient. : The previous definition is used; Sensitivity suppression coefficient : The immediate sensitivity of the observed object to perturbations, a non-negative value.

[0048] Furthermore, when the window has a high capacity, the amplitude of the controlled perturbation can be appropriately amplified, while when the object is in a highly sensitive state, the amplitude of the controlled perturbation can be automatically compressed to balance observability and business boundaries.

[0049] If the primary target of this round is a pump driver, the diagnostic controller first writes the controlled perturbation record into the preload register of the pump driver, and then the timing controller releases the record at the beginning boundary of the window. If the primary target of this round is a temperature controller, the diagnostic controller writes the temperature control setpoint change in a segmented manner and limits the change holding time. If the primary target of this round is a timing controller, the diagnostic controller only adjusts the pulse interval or duty cycle, without changing the task identifier of the main pulse sequence. As an interleaved embodiment, a generalized implementation can be given: when the side pump amplification module enters the allowable derating window, the diagnostic controller determines that the diagnostic window admission coefficient is in the active perturbation branch, and then writes a segment of controlled perturbation record to the pump driver. The pump driver maintains the original setting in the first part of the window, performs controlled perturbation in the middle part, and restores the original setting in the last part. The monitoring photodiode and the output power sampling branch continue to work continuously throughout the entire window. When in use, firstly, the controlled perturbation is constrained to be a reversible, single-master-object, and duration-based diagnostic action to avoid confusion caused by multiple objects acting simultaneously; secondly, the amplitude of the controlled perturbation is coupled with the admission coefficient of the diagnostic window to ensure that windows under different carrying capacities can obtain perturbations that are discernible without being overly aggressive; thirdly, the controlled perturbation record already has four fields—object, amplitude, duration, and release boundary—before it is issued, which facilitates the organization of synchronous sampling according to the record.

[0050] Furthermore, the output diagnostic window identifier and controlled perturbation record are transformed into a multi-channel synchronous response dataset under a unified time reference. At the acquisition end, time consistency screening, signal integrity screening, spike contamination identification, and encapsulation of the output object from step one are completed, so that what enters step two is not loose raw values, but computable objects with a unified index.

[0051] Even after window admission and controlled perturbation generation are completed, if the sampling phase still uses the traditional method of independent start-up, independent stopping, and independent buffering for each channel, there will still be starting deviations, recovery deviations, and buffer delay deviations among the drive current, monitoring photodiode signal, output power, temperature signal, and returned light signal. For the subsequent processing chain that needs to extract slope, hysteresis, and phase difference, these deviations will directly change the geometry of the relational features, causing a response segment that should belong to normal recovery to be misinterpreted as abnormal coupling. Therefore, multi-channel synchronization is regarded as the essential conclusion of step one.

[0052] The diagnostic controller first generates a sampling time frame based on the output controlled perturbation record; this sampling time frame covers at least three continuous time zones: the baseline segment before the perturbation, the perturbation action segment, and the recovery segment. Subsequently, the diagnostic controller sends a unified trigger signal to each sampling channel and writes the return values ​​of each channel into a unified buffer.

[0053] For data in the unified buffer, the diagnostic controller first performs time offset correction, then unit normalization, and subsequently completes segmentation and encapsulation using sampling quality identifiers and window identifiers as indexes. To identify the synchronization level between channels at the source, the diagnostic controller constructs synchronization effectiveness coefficients:

[0054] Among them, the synchronous effective coefficient : Characterizes the degree of fit of the current round of sampling results into subsequent relational feature construction, positive value; Complete segment length: The effective sampling length that simultaneously satisfies the three consecutive existences of the initiation, effect, and recovery segments in the current diagnostic window, positive value; Time offset : Remaining time deviation of multi-channel response after unified triggering, non-negative value; Peak contamination level : The degree of contamination in the current sample segment caused by sudden noise, burr reflection, or instantaneous saturation, a non-negative value.

[0055] The effectiveness of synchronous sampling is not determined by any single channel, but rather by the ability of the complete segment length to cover time offset and spike contamination. The diagnostic controller is identified around the same diagnostic window. Perform unified timestamp synchronous sampling. The unified timestamp does not simply mean attaching a local time to each channel; rather, it uses the moment the diagnostic controller issues a unified trigger signal as time zero, and constructs a three-segment buffer consisting of the baseline segment before time zero, the perturbation action segment near time zero, and the recovery segment after time zero. Each sampling channel, when returning data, includes its channel number, window number, sampling sequence number, and local hardware timestamp. Before writing to the unified buffer, the diagnostic controller first performs time offset correction using the local hardware timestamp, and then writes the data to a fixed location according to the three-segment buffer. This results in a multi-channel synchronous response dataset with a natural structural framework of operating condition dimension, channel dimension, and time dimension.

[0056] When in use, firstly, subsequent steps directly compare the response order of different channels at the same time, without being affected by local caching; secondly, the three-stage caching naturally separates the three states before, during, and after the disturbance, which facilitates the extraction of slope, hysteresis, and recovery time in subsequent steps; and thirdly, the unified caching area fixes the input format for subsequent steps, so that the operation process can be the same for different models.

[0057] After completing the unified timestamp synchronization sampling, the diagnostic controller does not immediately send all the data to step two, but first based on the synchronization validity coefficient. Sampling quality screening is performed based on channel integrity, range occupancy status, and peak contamination patterns. If the synchronization effectiveness coefficient... If the current model's usable threshold is reached, and all basic channels contain complete three-segment data, the diagnostic controller will mark this round of data as a high-confidence sample; if the synchronization effectiveness coefficient... If the data is below the high confidence threshold but still above the retest trigger threshold, the diagnostic controller will mark this round of data as a low confidence sample; if the synchronization effective coefficient... If the data falls below the retest trigger threshold, the diagnostic controller discards the current round of data as feature input, retaining only error records and feeding them back to the diagnostic window scheduling chain in step one. The key here is not the well-known action of deleting bad data, but rather using low-confidence samples as the trigger source for subsequent retests, thus making the sampling quality problem itself a computable event in the diagnostic process.

[0058] After quality screening is completed, the diagnostic controller encapsulates the output object from step one. This output object consistently includes a multi-channel synchronous response dataset and a diagnostic window identifier. The output object contains six types of fields: controlled perturbation record, sampling quality identifier, controller status record, and abnormal branch record. When the model is configured with wavelength, beam spot, wavefront, or polarization branch, the corresponding data is appended to the same output object as extended fields, but the naming and order of the aforementioned six basic fields remain unchanged. Through this encapsulation method, what is received in step two is not a collection of fragmented data packets, but an output object from step one with a unified index, unified fields, and unified quality semantics.

[0059] When using it, firstly, retain the low-confidence samples and clarify their purpose to avoid the sampling failure information being swallowed up in the process; secondly, fix the order and meaning of the output object fields so that step two can be seamlessly continued without having to infer the upstream state again; thirdly, the extended channel adopts a plug-in access without changing the basic field skeleton, thus taking into account the unified implementation of both the basic version and the extended version.

[0060] If a certain model is not configured with a wavefront detection branch and a polarization detection branch, step one will still be executed according to the basic six-channel method, without affecting the main chain structure of the output object; if a certain model is configured with dual pump drivers, the diagnostic controller will prioritize the pump driver currently undertaking the main pumping task as the main object to send controlled perturbations, while the other pump driver will remain frozen; if the temperature signal of a certain model comes from distributed temperature sampling points, the diagnostic controller will first perform position mapping on the distributed temperature sampling points, and then output a unified temperature signal field.

[0061] Step 2: Output diagnostic window identifier in Step 1 Under constraints, the multi-channel synchronous response dataset is transformed from readable raw fragments into a relational feature set. Drift normalized residuals are generated under the constraints of the healthy baseline model, enabling step three to perform component-level attribution based on structured residuals, unaffected by drift under normal operating conditions.

[0062] All the above steps are implemented through the diagnostic controller. After receiving the output object from step one, the diagnostic controller completes time alignment, dimensional normalization and sampling quality selection in a unified buffer area, forms a multi-condition response tensor and a set of relational features in the relational feature space, and finally projects and outputs the drift normalized residuals on the healthy baseline model.

[0063] The diagnostic controller performs controlled perturbation acquisition once in two consecutive idle frames of the same laser, and then identifies the multi-channel response segments of the two acquisitions according to the same diagnostic window. The window context is merged into a set of multi-condition response tensor inputs, forming a single drift-normalized residual output, thereby providing a consistent criterion for the retest triggering and baseline freezing in step three.

[0064] When the laser is in the closed-loop control state of the power controller, the output power and the monitoring photodiode signal are often pulled by the controller adjustment and maintained near the target. The output power value at a single moment is not directly equivalent to the healthy state. At the same time, the temperature signal and the driving voltage signal will drift slowly with heat accumulation. If the diagnosis is triggered only by the instantaneous threshold, normal drift is easily mistaken for abnormality.

[0065] Therefore, the multi-channel synchronous response segment from the pre-perturbation to the recovery stage in step one is used as the basic evidence unit. Multiple evidence units are divided into multi-condition response tensors according to the operating conditions. Normal drift is explained using the healthy baseline. The unexplainable part is output in the form of drift normalized residuals, which preserves the change of the response relationship caused by defects and reduces the common-mode disturbances caused by temperature rise, load change and slight alignment change. This serves as the direct input for template matching and risk classification in step three.

[0066] In this process, the multi-channel synchronous response dataset in the output object of step one is normalized into a comparable multi-condition response tensor input. Before entering the relational feature extraction, time alignment, dimension normalization, and sampling quality screening are performed to ensure that the subsequent relational features are not distorted due to channel delay, unit difference, or spike contamination.

[0067] Although step one aligns each channel to the same trigger zero time by using a unified timestamp, there may be two types of residual differences in actual engineering. First, the fixed delay of the hardware sampling link is different, and the response segment has a subsampling level offset at the inflection point of the perturbation action segment. Second, the channel range and noise floor are different, and the amplitude scale of the driving voltage and the return optical signal are different in the same time period. If the tensor is directly connected, the subsequent relational type will be masked by the large-scale channel.

[0068] The diagnostic controller is identified by the diagnostic window. Using the primary index, the system reads the three-segment buffer fragments of each channel within the same diagnostic window from the output object of Step 1, and simultaneously reads the perturbation release time, perturbation duration, and recovery boundary from the controlled perturbation record. The diagnostic controller first performs residual delay correction on each channel, then performs dimensional normalization on the corrected segments, and finally uses the synchronization effective coefficient... The sampling quality is screened using peak contamination patterns as a gating mechanism. The screened segments are then written into a multi-condition response tensor input buffer, which stores data in the order of condition dimension-channel dimension-time dimension.

[0069] The diagnostic controller does not treat uniform triggering as absolute alignment, but instead further utilizes perturbation recordings and channel response patterns to establish residual delay correction. For each channel, the diagnostic controller selects a stable interval from the baseline segment before perturbation as a baseline reference, and then locates the inflection point in the perturbation-affected segment where the channel exhibits a significant response to the perturbation. The inflection point location employs a fixed-form derivative threshold determination process: first, finite impulse response smoothing is performed on the segment; then, the first-order difference is calculated on the smoothed sequence; finally, the position where the difference sequence first exceeds the threshold is taken as the inflection point. The threshold is jointly determined by the absolute difference mean of the channel in the baseline segment and the range occupancy status, ensuring that channels with different noise levels still have consistent inflection point criteria. After locating the inflection point, the diagnostic controller performs time correction based on the offset of the inflection point relative to the perturbation release time, and resamples the corrected sequence to ensure that each channel is aligned on the same sampling grid. Resampling uses piecewise cubic spline interpolation as a preferred path; when the sampling frequency is low or the noise is strong, the diagnostic controller selects piecewise linear interpolation and superimposes a finite impulse response smoothing to avoid ringing introduced by higher-order interpolation.

[0070] The diagnostic controller performs three actions—positioning inflection points, shifting sequences, and resampling alignment—on each channel segment within the buffer. After completion, the rising edges of the perturbation action segments of each channel appear at the same sequence number position.

[0071] After residual delay correction, the diagnostic controller performs dimensional normalization and quality gating on each channel segment. Dimensional normalization does not use simple max-min scaling, as the maximum value is easily inflated by spike contamination. The diagnostic controller uses the steady-state mean of the baseline segment before the perturbation as the zero point and the difference between the steady-state plateau of the perturbation segment and the baseline as the scale, excluding sampling points identified as anomalous by spikes during the scale calculation. Spike identification uses a morphological criterion: if a point exhibits a single-point jump within a short window and recovers to its original level within two adjacent sampling periods, this point is marked as a spike candidate; if a spike candidate point appears simultaneously in the return light signal and the output power signal, the diagnostic controller marks it as a return spike contamination and passes this contamination label to subsequent gating. After dimensional normalization, the diagnostic controller uses the synchronization effectiveness coefficient... Channel integrity and spike contamination markers for segment classification: when the synchronization effectiveness coefficient When the usable threshold is reached and the spike contamination does not cross the critical interval of the perturbation action segment, the fragment enters the usable set; when the synchronization effectiveness coefficient When the value is below the available threshold but above the retest trigger threshold, the fragment enters the low confidence set and is written to the retest trigger queue; when the synchronization effective coefficient... When the three-segment structure is damaged due to the retest trigger threshold or channel loss, the segment enters the discard set and returns to the step one window scheduling chain.

[0072] When used, on the one hand, the channel responses after dimension normalization can be directly compared on the same scale, and the subsequent feature extraction will not be held hostage by the range difference; on the other hand, peak contamination is marked at the source and participates in gating, and mismatched data will not enter the healthy baseline projection through seemingly effective means, reducing the probability of misattribution triggering; low confidence sets will not be simply deleted, but will be explicitly used for retest triggering, and sampling quality issues will serve as inputs for subsequent stabilization control.

[0073] Furthermore, a multi-condition response tensor is established on the available set fragments of the sub-output and a relational feature set is extracted. The relational features are projected or fitted using a healthy baseline model to form drift normalized residuals and encapsulate them into the output object of step two, so that template matching and risk classification can be performed in the residual space in step three.

[0074] If only multi-channel segments are input into long sequences for subsequent attribution, two types of engineering problems arise: First, the sequence length changes with the window time, and subsequent template matching still needs to process sequences of different lengths, making it difficult to achieve consistency; second, the original sequence contains a large amount of low-frequency information unrelated to drift, obscuring the structural differences between defect signals. Relational feature extraction and healthy baseline projection are considered as an integrated chain: first, the sequence is compressed into a small number of relational features; then, common-mode drift is explained on the healthy baseline; finally, the unexplained portion is output as residual output. The key point of this chain is that relational features are not equivalent to general statistics; relational features are constructed from the condition transfer relationship and are tightly coupled with the controlled perturbation record from step one.

[0075] The diagnostic controller is identified by the diagnostic window. For indexing, different load case segments within the same window are organized into multi-load case response tensors according to the load case dimension, and the controlled perturbation amplitude is recorded for each load case segment. Diagnostic window admission coefficient Synchronous effective coefficient The diagnostic controller performs relational feature mapping on the tensor, outputting a set of relational features. Subsequently, the diagnostic controller selects a health baseline model corresponding to the machine model and performs projection to obtain the drift-normalized residual. The selection of the health baseline model follows a priority order: factory-made, in-service, and post-maintenance reconstruction. When the in-service health baseline is frozen, the diagnostic controller prioritizes using the in-service health baseline at the time of freezing. When the in-service health baseline is allowed to be updated, the diagnostic controller performs weighted fusion between the factory-made health baseline and the in-service health baseline. When the system completes maintenance and enters the post-maintenance reconstruction phase, the diagnostic controller uses the post-maintenance reconstruction baseline as the primary baseline and the factory-made health baseline as the constraint boundary to avoid excessive offset.

[0076] The diagnostic controller combines the available output segments into a multi-condition response tensor based on the operating condition dimension, generating a set of relational features around the controlled perturbation record. The diagnostic controller first defines the operating condition anchor points, i.e., the starting boundary of the perturbation action segment and the ending boundary of the recovery segment; then, for each channel, it calculates the response slope, response hysteresis, and recovery time within this interval. The slope describes the instantaneous response strength of the perturbation, the hysteresis describes the channel's response lag to the perturbation, and the recovery time represents the time it takes for the channel to return to the baseline from the perturbation state.

[0077] This section introduces a control coupling characteristic to describe the consistency of the coupling direction between the compensation amount and the main output of the stable power controller. This characteristic does not depend on absolute power, but on the relative synchronization relationship between the compensation amount and the output. To ensure that this coupling characteristic can still be compared under different window lengths, a coupling consistency coefficient is constructed within the perturbation action segment:

[0078] Among them, the coupling consistency coefficient The degree to which the compensation amount of the stable power controller is consistent with the direction of the output change, and the value is [value missing]. ; Compensation function The instantaneous adjustment of the drive setpoint by the stable power controller within the perturbation range; the range of which is determined by the controller's range; output rate of change. : The rate of change of output power over time, the range of which is determined by the rate of change of output power; the starting moment of integration. The starting boundary time of the perturbation action segment; the termination time of the integration. : Termination boundary time of the perturbation action segment; Buffer constant : To prevent positive constants with a denominator of zero.

[0079] When the compensation amount and the output change have the same direction, the coupling consistency coefficient tends to a positive value; when the compensation amount and the output change are opposite, the coupling consistency coefficient tends to a negative value; when the compensation amount and the output change are approximately unrelated, the coupling consistency coefficient tends to zero.

[0080] Diagnostic controller and The acquisition is given a clear path: the compensation function The regulation sampling sequence is directly from the steady power controller interface; the output rate of change. The output power sampling sequence is obtained by first-order differential and superimposed with finite impulse response for smoothing. The length of the smoothing window is determined by the number of sampling periods within the diagnostic window, so that the differential noise is not excessively amplified.

[0081] When used, on the one hand, the relational feature set compresses the sequence into a set of features related to the working condition transmission relationship, eliminating the need for subsequent health baseline projection; on the other hand, the coupling consistency coefficient adds the control compensation behavior to the computable evidence, and the subsequent residuals will obtain a state with stable output but abnormal compensation and a state with changing output but normal compensation, thus improving interpretability.

[0082] After constructing the relational feature set, the diagnostic controller enters the health baseline projection and residual generation stage. The health baseline model is not considered an abstract concept, but is specifically parameterized as a set of computable mappings that can reproduce the projection and residual generation process on a given model.

[0083] The diagnostic controller organizes the set of relational features into relational feature vectors. and with a health baseline matrix With health bias vector The parameterized health baseline subspace transforms health baseline projection into a problem of solving parameter vectors. Simultaneously, to combat the perturbation of projection by a small number of residual spikes, the diagnostic controller selects weighted absolute deviation as the objective function, and provides the rules for determining the weights: the weights are determined by the synchronization effective coefficients. Diagnostic window admission coefficient Together with the channel quality identifier, higher-quality segments receive higher weights, and lower-confidence segments receive lower weights. The diagnostic controller then calculates the health parameter vector and constructs the projected values ​​and residuals. The calculation and residual generation are described as follows:

[0084] Among them, health parameter vector : The parameter solution that makes the relational feature vector closest to the health baseline, the range of which is determined by the column space of the health baseline matrix; parameter variable vector : Health parameter variables to be solved; weighting coefficients : No. The credibility weights of each relation feature component are positive; relation feature vector components : No. Relational feature values; health baseline matrix The baseline basis vector matrix constructed from healthy samples has the same dimensions as the relational feature vectors; the health bias vector... The bias term vector of the health baseline has the same dimension as the relation feature vector; health projection operator. : Projection of relational feature vectors onto the healthy baseline subspace; residual vector The difference between the relation feature vector and its health projection, with the same dimension as the relation feature vector; feature dimension. : The number of feature components contained in the relation feature vector, with a value that is a positive integer.

[0085] Health bias vector Construction: Collection Relationship feature vector under group health window The median value of each component is used to construct the health bias vector. :

[0086] Where: health bias vector components : indicates the first The steady-state center of each relation feature in a healthy state; the health feature vector : indicates the first Relationship feature vector obtained from secondary health window sampling; number of healthy samples : is a positive integer.

[0087] Health baseline matrix Construction: Decentralizing the health feature vector:

[0088] Then select from these decentralized vectors The health baseline matrix is ​​composed of 10 representative vectors as column vectors. .

[0089] Alternatively, a greedy maximum angle selection method can be used: first, select the decentralized vector with the largest norm as the first column; in each column, select the vector most similar to the selected column from the unselected vectors; stop only when the reconstruction error of the healthy verification sample is less than the threshold.

[0090] Solution method of projection operator: The optimization problem is solved by weighted minimum absolute deviation solver; optional engineering implementation paths include: linear programming solution, coordinate descent solution, and iterative weighted solution; coordinate descent solution is preferred because it is easy to embed into the software runtime environment of diagnostic controller.

[0091] The healthy baseline projection is not simply subtracting the mean. Instead, it uses the healthy baseline matrix and bias vector as structural constraints, and under weight gating, it finds the explanatory component that best reflects the healthy state. The remaining unexplained part is retained by the residual vector.

[0092] To ensure that a health baseline matrix is ​​available With health bias vector The diagnostic controller adopts three construction paths: First, during the factory delivery stage, several diagnostic window segments are collected from the calibration station to obtain the factory health baseline matrix. Second, during the in-service stage, segments are collected only under low-risk, high-confidence conditions to update the factory health baseline matrix. When updating, column vectors are used instead of full reconstruction, thus controlling the computational cost. Third, during the post-maintenance stage, the maintenance baseline matrix is ​​reconstructed using several diagnostic window segments after maintenance, and the column space of the maintenance baseline matrix is ​​used as the constraint boundary to avoid excessive baseline drift.

[0093] For example, the diagnostic controller sends the same controlled perturbation amplitude in two consecutive idle frames of the same end-pump oscillator. Two sets of multi-channel synchronous response segments are obtained; the diagnostic controller merges the two sets of segments into the same multi-condition response tensor according to the operating condition dimension, and integrates them into the same factory health baseline matrix. The projection is performed on the top, and the same residual vector is obtained in the end. The two estimates are performed. When the signs and magnitudes of the two residual vectors are consistent on the key feature components, the diagnostic controller encapsulates the residual vector into the output object of step two and passes it to step three.

[0094] In practice, firstly, the healthy baseline projection receives the common-mode changes caused by temperature rise and load variations, and the residual vector is more suitable for reflecting anomalies in the response relationship; secondly, the weighting coefficients... The introduction of this feature ensures that low-confidence segments do not dominate the projection, reducing the impact of peak residues on the residuals. Finally, the residual vector dimension is consistent with the relation feature vector, which facilitates direct component-level attribution in step three using template matching.

[0095] Specifically, the pump-side feature index set It must include at least the drive current-to-photodiode signal slope component, the drive voltage-to-photodiode signal slope component, and the drive signal-to-output power hysteresis component; and a set of intracavity hot-side feature indexes. It must include at least the temperature signal-output power recovery time component and the temperature signal-output power phase difference component; optical element side feature index set It must include at least the output power-wavefront relation component and the output power-beamspot relation component. If the model does not have a wavefront / beamspot channel, use the non-returning residual component related to the optical path instead; Returning side feature index set It includes at least the phase difference component between the return optical signal and the output power, and the coupling component between the return optical signal and the controller compensation.

[0096] The diagnostic controller defines the output object of step two as a fixed combination of fields, including at least the diagnostic window identifier. Controlled perturbation amplitude Diagnostic window admission coefficient Synchronous effective coefficient relational feature vectors Health projection operator output residual vector And sampling quality labeling.

[0097] Step 3: Identify the diagnostic window output in Step 2. Controlled perturbation amplitude Diagnostic window admission coefficient Synchronous effective coefficient relational feature vectors With residual vector Under the constraints, the residuals are converted into component-level attribution conclusions and stabilization control actions. Chronic defects are avoided from being written into the health baseline by freezing the rate limit update of the in-service health baseline. Under medium risk, protected degrading retesting is initiated and continuous result verification is performed.

[0098] The following steps are all executed by the diagnostic controller. The diagnostic controller outputs the main conclusion, secondary conclusion, uncertain state, confidence level, freeze flag and retest plan to the main control interface, and enters the freeze and protection branch under high heat load, strong backlight or controller saturation boundary.

[0099] Step two has projected normal thermal drift, load drift, and slight alignment variations onto the healthy projection operator as much as possible. And store the unexplained parts in the residual vector. .

[0100] residual vector The sources of residuals include at least two types: changes in response relationships caused by actual defects, or structural deviations caused by insufficient capacity of the diagnostic window or transient disturbances in the reflected light. Equating residuals with defects will lead to misattribution; if residuals are completely avoided, early signals will fail to materialize.

[0101] Therefore, the diagnostic window admission coefficient Synchronous effective coefficient As a gating mechanism, it ensures that the attribution conclusion and the retesting action are closed within the same chain. When the diagnostic window admission coefficient... Low range and synchronous effective coefficient When the load decreases, the diagnostic controller judges the state as insufficient load. Under the insufficient load state, it only allows the output of uncertain state and triggers retest, and does not allow the output of high-risk component conclusions, thereby avoiding amplification of misjudgment under boundary conditions.

[0102] residual vector The components are equivalent to the relation feature vector. Components, relational feature vector Based on relational features such as slope, hysteresis, recovery time, channel ratio, phase difference, and coupling consistency coefficient, if a conclusion is drawn based on a single component threshold, it will regress to anomaly diagnosis and easily misinterpret the overall rise in residual as a component anomaly.

[0103] Suppose that the mean value of a certain channel in the baseline segment before perturbation is The mean value during the perturbation period is The moment when the recovery phase reaches stability is ,but:

[0104] Where: response slope : Indicates a channel The rate of change of the effective segment relative to the baseline segment; :aisle The mean value during the perturbation period; :aisle The mean of the baseline segment before the perturbation; : Effective time difference between the active segment and the baseline segment; buffer constant : To prevent the denominator from being zero.

[0105] Recovery time is defined as the difference between the moment when the channel signal first returns to the baseline bandwidth and the moment when the perturbation ends. The baseline bandwidth consists of a fixed tolerance band of the mean baseline segment before the perturbation; the moment when the signal first enters and continuously maintains a certain number of sampling points within the tolerance band is considered the recovery time.

[0106]

[0107] Where: Channel ratio : Indicates a channel With channel The relative change amplitude under the same operating condition disturbance; , : The difference between the respective action segment and the baseline segment; buffer constant : To prevent the denominator from being zero.

[0108] Therefore, by constructing the residual strength, the residual confidence is obtained, and the residual vector is mapped according to the feature-component mapping table. Grouping the evidence into groups and obtaining primary and secondary conclusions based on the amount of evidence in each group, thus preserving uncertainty for scenarios with insufficient evidence.

[0109] The diagnostic controller reads the residual vector. Weighting coefficients Diagnostic window admission coefficient Synchronous effective coefficient Calculate the residual strength and construct the residual confidence level, then apply the residual vectors according to the feature-component mapping table. The component indexes are grouped and aggregated to obtain the pump-side evidence quantity, the cavity thermal-side evidence quantity, the optical element-side evidence quantity, and the return-side evidence quantity, and the main conclusion and secondary conclusion are obtained according to the magnitude of the evidence quantity.

[0110] The generation of feature-component mapping tables typically follows a reproducible calibration process: during the factory calibration phase, the diagnostic controller sequentially applies controlled perturbations to the pump drive setpoint, temperature control setpoint, and timing duty cycle using the power-on self-test window and idle frame window as reference windows, and records the relational feature vector under each perturbation. residual vector The component index is then divided into four sets according to the component boundaries: pump side, cavity thermal side, optical element side, and return side. The pump side set includes at least the driving current-monitoring photodiode signal coupling component and the driving voltage-monitoring photodiode signal slope component. The cavity thermal side set includes at least the temperature signal-output power recovery time component and the coupling consistency coefficient component. The return side set includes at least the return light signal-output power phase difference component. When the model uses a wavefront detection branch or a beam spot detection branch, the optical element side set includes at least the wavefront or beam spot related residual component.

[0111] This mapping table is fixed using component indices and does not require specific numerical values, allowing for reuse under different loads and temperatures. When the residual confidence level is lower than the attribution value, the diagnostic controller outputs an uncertain state and an evidence vector to enable retesting. The diagnostic controller first constructs the residual strength, representing the overall deviation of the residuals, and then correlates it with the weighting coefficients from step two. Semantic consistency is maintained. The residual strength is defined as follows:

[0112] Among them, residual strength : Characterizes the overall deviation of the residuals, with a positive value range; weighting coefficient : Characterize the first Each residual component has a confidence weight, with a positive value range; residual components : Represents the residual vector The Each component has a value range determined by the dimensional normalization scale of the corresponding feature components; feature dimension. : Represents the number of components in the residual vector, and its value ranges from positive integers; Then, the diagnostic controller maps the residual vector according to the feature-part mapping table. Group aggregation is performed, and group aggregation uses residual strength. Consistent weighted absolute aggregation ensures a greater contribution from high-confidence components. The mapping table is generated during factory calibration and can be replaced during post-maintenance reconstruction. The mapping table indexes relational feature components across four components: the pump side, the intracavity thermal side, the optical element side, and the return side. Each piece of evidence is derived from the source of the relational features. After evidence aggregation is complete, the diagnostic controller creates residual confidence levels to suppress misattributions caused by insufficient window capacity or sampling contamination. The residual confidence level is defined as:

[0113] Among them, residual confidence The residual is used to assess the reliability of component attribution, and its value range is [value range missing]. Diagnostic window admission coefficient : The window's comprehensive ability to perform proactive diagnostic actions, with a positive value range; synchronization effectiveness coefficient. : The degree of fit of the sampled features into the relational feature construction, with a positive value range; residual strength : Characterizes the overall deviation of the residuals, and the value range is positive; Diagnostic controller based on residual confidence The residual confidence level is obtained by taking the difference between the primary and secondary evidence quantities as the result. When the residual confidence level is high and the primary and secondary differences are sufficient, the output primary and secondary conclusions have high confidence levels. When the residual confidence level is moderate or the primary and secondary differences are insufficient, the output primary and secondary conclusions have medium confidence levels, and this result is used as a prerequisite for freezing the in-service baseline. When the threshold for triggering a retest is lowered, an uncertain state is output and the retest scheduling branch is entered.

[0114] To ensure the confidence level is verifiable, the diagnostic controller uses the ratio of the difference between the primary and secondary evidence quantities to the maximum evidence quantity as the dissociation index, and correlates this dissociation index with the residual confidence level. They are jointly mapped to a three-segment confidence level; the mapping is implemented using a fixed threshold comparator, and the threshold is determined and fixed as a configuration item by healthy samples and post-maintenance samples of the same model during the factory calibration stage.

[0115] In addition to outputting the main conclusion and secondary conclusions, the diagnostic controller also outputs four fields: evidence vector, separation index, residual confidence level, and freeze recommendation flag, facilitating direct entry into the freeze or retest branch. In an interim embodiment, the diagnostic controller obtains the residual vector within the allowed reduction window and calculates the residual strength. After assessing the reliability of the residuals, the system outputs the main conclusion of pump-side anomaly and triggers a derating retest. The on-site result is that the system enters a protected derating state and waits for the next diagnostic window for retesting.

[0116] When used, residual reliability Diagnostic window admission coefficient Synchronous effective coefficient With residual strength The coupling is a single gated quantity, and the structured output of primary and secondary conclusions and uncertain states allows weak evidence stages to be absorbed by retesting without being over-asserted.

[0117] Furthermore, the confidence levels of the output main conclusions, secondary conclusions, uncertainties, and residuals are analyzed. Based on this, the in-service health baseline is frozen and the rate limit is updated. Under medium-risk conditions, protected depreciation retesting and consistency verification are initiated to prevent chronic defects in long-term operation from being written into the health baseline, while converting single accidental residuals into stable evidence.

[0118] If the in-service health baseline is not updated for a long period, seasonal temperature drift and post-maintenance regression will cause the health baseline to deviate from the true health status; if the in-service health baseline is not updated unconditionally, chronic defects will be absorbed into the health baseline. On the other hand, return light spikes and sampling link saturation will cause short-term residual anomalies, and direct upgrade shutdowns will affect services. Sub-step 302 binds the freeze / update separation with the two-stage retest: first, based on the residual confidence level... Residual strength Risk segmentation is performed based on the primary and secondary differences, and then baseline update permit status, reduction retest plan and consistency verification rules are executed within the corresponding segments.

[0119] The diagnostic controller receives the main conclusion, secondary conclusion, uncertainties, and residual confidence level. Residual strength Evidence quantity vector and diagnostic window identifier The diagnostic controller divides the status into low-risk, medium-risk, high-risk, and retest pending segments, and implements different baseline strategies in each segment: the low-risk segment allows rate-limited updates to the active health baseline, the medium-risk segment freezes the active health baseline, and the high-risk segment rejects the active health baseline and locks the protection priority; when in the retest pending or medium-risk segment, the diagnostic controller uses the diagnostic window identifier... The next available diagnostic window is used as the retest window, and the controlled perturbation amplitude is preferred for reuse. To maintain comparability.

[0120] After the retest is completed, the diagnostic controller performs a consistency check based on the consistency of the main conclusions, the consistency of the primary and secondary order, and the consistency of the symbols of the critical components of the residuals. Based on this, it either upgrades maintenance prompts or maintains an uncertain state. The retest scheduling adopts a window type priority mechanism: the diagnostic controller prioritizes selecting an idle frame window as the retest window, followed by a window that allows derating, and then a load switching window. When none of the above windows are available, the diagnostic controller suspends the retest plan and forces a retest to be performed in the next power-on self-test window. To avoid endless retesting loops, the diagnostic controller identifies the same diagnostic window. Associate a retest count record. When the retest count reaches the preset upper limit, the diagnostic controller will no longer schedule retests but will directly output a maintenance prompt or shutdown suggestion, and will maintain the in-service health baseline locked flag until the post-maintenance reconstruction phase ends.

[0121] The diagnostic controller restricts in-service health baseline updates to low-risk, rate-limited write behavior. At the implementation level, the diagnostic controller allocates a rolling window buffer for each machine, identified by the diagnostic window identifier. Save the projection results of the most recent preset number of high-confidence diagnostics as a key. and the corresponding synchronization efficiency coefficient When all consecutive entries in the cache meet the high confidence level and the synchronization efficiency coefficient Within the available range, the diagnostic controller can only trigger a rate-limited write operation once, and the write operation is driven by a steady state rather than an occasional window.

[0122] Rate-limited writing means that the projection result is only displayed when the diagnostic window outputs high confidence and no abnormalities for a preset number of consecutive cycles. The corresponding healthy samples are written into the active health baseline, and each time a basis vector of the health baseline matrix is ​​written or the health bias vector is updated. A portion of these factors contribute to the long-term drift of the health baseline.

[0123] Freezing refers to the diagnostic controller outputting a freeze flag when the system is in the medium-risk range or when the output is at medium confidence level. During the validity period of the freeze flag, the health baseline matrix at the time of the freeze is used. and health bias vector Write, refuse to write; refuse to update means that when in a high-risk segment or with strong backlighting, or at the controller saturation boundary, the diagnostic controller upgrade is frozen to a locked flag and remains so until the end of the post-maintenance rebuild phase. In an equivalent embodiment, the update of the health baseline matrix can be either column replacement or segmented update of the health bias vector, but it shares the same freeze and rate-limited update boundaries as rate-limited writes.

[0124] As a supplement: Define the degree of separation between primary and secondary elements:

[0125] Among them: primary and secondary separation degree : Indicates the degree of separation between the largest component's evidentiary value and the second largest component's evidentiary value, with a value range of . Maximum component evidence : Represents the maximum value among the four types of component evidence; the second largest component evidence value. : Represents the second largest value among the four types of component evidence; buffer constant : To prevent the denominator from being zero.

[0126] Furthermore, define the overall risk level:

[0127] Among them: comprehensive risk level : Represents the overall risk level when the current residual is both significant and reliable; residual strength : Indicates the overall degree of deviation; residual confidence level : Indicates the reliability of using the current residual for decision-making.

[0128] Then it is stipulated that: when the residual confidence level Below the confidence threshold or When the value is below the separation threshold, an uncertain state is output; when the overall risk level is... When the risk level is in the low-risk range, output "low risk"; when the overall risk level is... When the risk level is in the medium-risk range, freeze the active-duty health baseline and trigger a deduction retest; when the overall risk level is... If the risk level is high and the retest results are consistent, output a maintenance prompt or shutdown suggestion.

[0129] The diagnostic controller designs the retest as a protected derating retest. "Protected" means that before the retest window arrives, the diagnostic controller sends a derating command to the main control interface, causing the pump drive setpoint to decrease to the allowable range and remain stable. Subsequently, within the stable range, the controlled perturbation amplitude is reused. Perform step one (sampling) and step two (residual generation). After the retest, the diagnostic controller performs a consistency check on the two results. The consistency check sequentially checks whether the main conclusion is consistent, whether the order of the main and secondary conclusions is consistent, and the residual vector. The key component symbols are consistent; if consistency is passed, it is upgraded to a maintenance prompt or shutdown suggestion; if consistency fails, it remains frozen and outputs an uncertain state, while the retest is scheduled to the next more appropriate diagnostic window.

[0130] The order of the two-stage retest is as follows: In the first stage, the controlled perturbation amplitude is reused under the protected derating state. Perform retesting and consistency verification; if the first phase of consistency fails but the residual confidence level is high... When the signal increases, the diagnostic controller enters the second stage of retesting. In the second stage, after maintaining a longer stable period within the next idle frame window, the controlled perturbation amplitude is reused again. Sample and generate residual vector If the second phase still fails the consistency check, the diagnostic controller outputs a maintenance prompt or shutdown suggestion and locks the protection priority. In an interim embodiment, the diagnostic controller freezes the in-service health baseline after the initial diagnosis and triggers a derating retest. When the retest consistency is passed, a maintenance prompt is output and the system remains frozen. The visible result on-site is that the system continues to run, but the maintenance queue is clearly defined.

[0131] When used, derating and retesting transforms single random residuals into repeatable evidence, and consistency checks make maintenance prompts independent of single samples, thereby reducing erroneous shutdowns and improving the stability of conclusions.

[0132] Step 4: Based on the main conclusions, secondary conclusions, uncertainties, risk levels, freeze indicators, retest plans, and residual confidence levels established in Step 3. Under the constraints, the diagnostic results are transformed into handling instructions and maintenance scheduling information that directly constrain the operation of the laser. The handling results are then fed back to the in-service health baseline at a limited rate. At the same time, the window scheduling parameters are fed back to step one, and the baseline selection status is fed back to step two, thus forming a sustainable closed-loop link. This closed-loop link does not change the existing safety interlock priority of the laser, nor does it change the main business control weight of the pump driver, temperature controller, and timing controller. Instead, it provides clear exit conditions, retest limits, and recovery paths for the diagnostic process within the allowable range of the safety interlock.

[0133] In step three, the main conclusion, secondary conclusion, and uncertainty state indicate the attribution direction of the anomaly source; the risk level indicates the constraint of the anomaly on continued operation; the freeze flag indicates whether the active health baseline can continue to absorb new samples; the retest plan indicates the type of diagnostic window that will be occupied by the next diagnosis and the derating conditions; and the residual confidence level... This indicates the degree to which decisions can be made based on the residuals in this round.

[0134] If this information is not converted into instructions that can be recognized by the pump driver, temperature controller, timing controller, and safety interlock module in this step, the diagnostic conclusion will remain at the information level and will not be able to constrain system behavior. If the handling results are not fed back into steps one and two, the next round of diagnosis will still enter the window with the original parameters, resulting in repeated triggering and baseline drift.

[0135] Specifically, the attribution results, risk level, freeze flag, and retest plan from step three are presented as a set of executable treatment instructions, and the diagnostic window in the treatment instruction set is marked. Controlled perturbation amplitude Diagnostic window admission coefficient Synchronous effective coefficient residual vector With residual reliability The output of step four is encapsulated so that the main control interface and the maintenance scheduling interface obtain the decision basis and execution conditions in the same message.

[0136] The field operation of optically pumped lasers often faces three constraints simultaneously: the first constraint comes from the safety interlock module, whose response must precede any diagnostic actions; the second constraint comes from the main operational control loop, whose objective is usually to maintain output power or pulse energy within a specified range; and the third constraint comes from maintenance availability, which requires minimizing unplanned downtime without sacrificing safety. Step three generates a risk level and retest plan, but simplifying it to text alarms will fail to meet the control system's requirements for command fields, triggering conditions, and durations; equating it to shutdown will fail to meet maintenance availability requirements. Therefore, diagnostic conclusions are described as executable actions in the form of command type-command boundary-command holding-command release, while simultaneously identifying the actions and diagnostic window. The timing boundaries are consistent.

[0137] After receiving the output from step three, the diagnostic controller first determines whether the safety interlock status is a high-risk prohibited state. When the safety interlock status is a high-risk prohibited state, the diagnostic controller does not generate derating retest or continue-run instructions, but instead writes the protection priority lock flag into the output object and limits the handling instructions to shutdown or isolation instructions. When the safety interlock status is not a high-risk prohibited state, the diagnostic controller determines the appropriate handling instructions based on the risk level and residual reliability. A set of handling instructions is generated. This set includes at least one or more of five categories: continue operation instructions, derated operation instructions, retest trigger instructions, maintenance scheduling instructions, and shutdown suggestion instructions. Each category is assigned execution conditions and duration boundaries. Subsequently, the diagnostic controller encapsulates this instruction set into an output object (step four) and sends this output object to both the main control interface and the maintenance scheduling interface.

[0138] The diagnostic controller translates the risk level and freeze flag into a set of handling instructions with boundary conditions. The continue-run instruction is used to characterize situations where the risk level is in the low-risk range and the residual confidence level is low. The state reaching the high confidence interval is constrained by the following boundary constraints: under the given task objectives, the pump drive setpoint, the temperature control setpoint, and the timing controller main pulse task remain unchanged. The derating operation command characterizes a state where the risk level is in the medium-risk or undetermined retest phase. Its boundary constraints involve switching the pump drive setpoint to the derating level and maintaining a stable range so that the controlled perturbation amplitude can be reused in subsequent retests. The retest trigger instruction is used to indicate that the next diagnostic window is occupied when the retest plan is valid. Its boundary constraint is that the retest window type must be consistent with the retest plan output in step three, and the in-service health baseline is frozen during the retest. The maintenance scheduling instruction is used to indicate a high-risk state where the consistency check has passed or the retest has reached its limit. Its boundary constraint is that the frozen or locked flag is maintained until the post-maintenance reconstruction phase ends. The shutdown recommendation instruction is used to indicate a state where the safety interlock is prohibited or the high-risk state persists. Its boundary constraint is that the shutdown conditions are bound to the existing shutdown interface of the safety interlock module without bypassing the safety interlock module.

[0139] In the equivalent embodiment, the derating operation command can be either a pump driver gear switching or a timing controller duty cycle switching. Both are constrained by reducing the average pump load and maintaining a stable segment, without changing the semantics of the risk level. The handling command is no longer an abstract alarm, but an executable action with execution conditions, persistence boundaries, and release conditions. This allows the main control interface to schedule within the same control cycle. The risk level and freeze flag serve as command boundaries, preventing the absorption of new samples into the in-service health baseline under medium-risk and high-risk conditions.

[0140] After completing the instruction type mapping, the diagnostic controller encapsulates the set of treatment instructions and key evidence fields into the output object of step four. The output object of step four is fixed as the diagnostic window identifier. Main conclusion, secondary conclusion, uncertainty status indicator, risk level, freeze indicator, retest plan, controlled perturbation amplitude. Diagnostic window admission coefficient Synchronous effective coefficient Residual reliability With residual vector Abstract index; residual vector The summary index refers to the symbol combination and grouping of key residual components, excluding the original waveform. The maintenance scheduling interface can understand the source of the anomaly, provided bandwidth allows. When the diagnostic controller issues this output object, it uses a dual-channel approach: the main control interface triggers continued operation, derating operation, and retesting; the maintenance scheduling interface triggers maintenance scheduling and shutdown recommendations.

[0141] To ensure consistent parsing for the two types of interfaces across different communication links, the output object in step four preferably uses a fixed field order to form a command frame. This command frame must at least include a message number and a diagnostic window identifier. The diagnostic controller sends an instruction, which includes a type code, execution condition field, persistence boundary field, and acknowledgment field. After sending the instruction, the controller waits for the acknowledgment field to return. If the acknowledgment field does not return within a preset waiting period, the controller resends the instruction only once and binds the resend record to the message number to avoid control-side oscillations caused by duplicate instructions. The diagnostic controller uses the same message number for both channels to ensure that the main control interface and the maintenance scheduling interface identify the same diagnostic window. The conclusion is consistent; when any channel returns a failed confirmation, the diagnostic controller records the failure and identifies it in the diagnostic window. Bind and enter the subsequent exit boundary judgment to avoid repeated triggering due to inconsistent control links.

[0142] For example, after the laser is processed, it enters an idle frame window. The diagnostic controller receives the return-side abnormal conclusion from step three as the main conclusion, the risk as medium risk segment, and freezes the flag and retest plan. The diagnostic controller generates a derating operation command and a retest trigger command, encapsulates the output object and sends it to the main control interface. The main control interface switches the pump driver to the derating level to maintain the stable segment. The field output power is maintained in the stable segment corresponding to the derating level and enters the next idle frame window to reserve the retest window.

[0143] When used, the diagnostic conclusions are simultaneously delivered to both the control and maintenance entities and remain consistent, reducing information fragmentation; the output object retains sufficient evidence fields to support subsequent interpretation, but does not introduce redundant waveform transmissions that would affect real-time performance.

[0144] Furthermore, the generated and issued handling instructions are treated as closed-loop actions. Based on the handling results, the diagnostic window scheduling parameters in step one are updated, the health baseline selection status in step two is updated, and the freeze flag and retest count in step three are updated. When the exit conditions are met, the current diagnostic cycle ends or the process enters maintenance / shutdown state, thus enabling the process to have convergent boundaries rather than an infinite loop.

[0145] If the diagnostic process lacks clear feedback and exit conditions, two types of adverse effects may occur: First, consecutive windows may repeatedly trigger the same type of diagnosis and repeatedly occupy idle frames, causing the main service time slots to be squeezed; second, post-maintenance status changes may not be absorbed by the health baseline, resulting in the system still being treated as abnormal after recovery. By ensuring clear feedback targets, limiting feedback rates, and verifying exit conditions, the diagnostic process can not only run continuously but also exit when the risk stabilizes and re-enter a healthy state during the post-maintenance reconstruction phase.

[0146] The diagnostic controller receives execution feedback from the main control interface regarding the handling instructions. This feedback includes at least whether derated operation has taken effect, whether the retest window has been successfully reserved, whether the shutdown suggestion has been taken over by the safety interlock module, and whether the maintenance schedule has been registered. The diagnostic controller updates the update permission status of the in-service health baseline based on the execution feedback and the freeze flag, and updates the next diagnostic window scheduling parameters according to the retest plan. When the handling instruction is to continue operation and the risk level remains in the low-risk segment with high confidence, the diagnostic controller allows the in-service health baseline to perform a rate-limited update and resets the retest count to zero. When the handling instruction is derated operation or a retest is triggered, the diagnostic controller maintains the freeze flag and increments the retest count. When the handling instruction is a maintenance schedule or a shutdown suggestion, the diagnostic controller locks the protection priority and stops active perturbation. Subsequently, the diagnostic controller determines whether to end the current process or enter the next diagnostic cycle based on the exit criteria.

[0147] The diagnostic controller categorizes feedback into three types of objects and locks them respectively: the first type of object is the diagnostic window scheduling parameters in step one, including window priority, window admission threshold and retest window type; the second type of object is the health baseline selection status in step two, including the selection identifier and freeze identifier of factory health baseline, in-service health baseline and post-maintenance reconstructed baseline; the third type of object is the retest count and consistency verification status in step three.

[0148] The diagnostic controller employs rate-limited backfeedback for three types of objects: when the risk level is in the low-risk range and the residual confidence level is high. When maintaining a high confidence interval, the diagnostic controller only allows writing an update permission flag to the in-service health baseline once after a preset number of consecutively met conditions has ended, and then passes this update permission flag to step two. When the risk level is in the medium-risk segment or the retest pending segment, the diagnostic controller passes a freeze flag to step two and prohibits any update permission, while passing the retest window type and derating requirements to step one, making the next diagnostic window selection more inclined towards idle frame windows rather than load switching windows. When the risk level is in the high-risk segment or the safety interlock prohibition state, the diagnostic controller passes a lock protection priority flag to step one and prohibits active perturbation. In an equivalent embodiment, limited rate backfeedback can be achieved by allowing one update after the window count reaches a threshold, or by allowing one update after the time interval reaches a threshold. Both use the freeze flag as a hard gate, disallowing updates during the effective period of the freeze flag.

[0149] When in use, the feedback objects are locked into three categories: window scheduling, baseline selection, and retest status, to avoid irrelevant parameters being involved and causing confusion in terminology; the limited rate backflow ensures that the update of the in-service health baseline is driven by a stable low-risk state, rather than by a single window, thereby reducing the probability of chronic defect writes.

[0150] After feedback feedback is completed, the diagnostic controller performs an exit condition determination. The exit condition determination is based on the risk level, freeze flag, retest count, and master control execution feedback: when the risk level is continuously in the low-risk range and the master conclusion is consistently empty or consistently low-risk, and the residual reliability can still be maintained after the freeze flag is lifted. When the risk level is in the high confidence range, the diagnostic controller determines that the current diagnostic cycle has ended and returns to normal monitoring. When the risk level is in the medium risk range and the retest count has not reached the upper limit, the diagnostic controller maintains the freeze flag and enters the next diagnostic cycle. The next cycle must occupy the specified window type according to the retest plan. When the risk level is in the high risk range and the retest count has reached the upper limit, or when the main controller executes the feedback indication to stop the machine and the safety interlock module has taken over, the diagnostic controller determines that the diagnostic cycle has terminated and enters the maintenance or shutdown state. At the same time, it locks the protection priority and stops active perturbation. When the maintenance schedule feedback indicates that the maintenance has been completed and the baseline reconstruction after maintenance has been completed, the diagnostic controller releases the lock flag and allows the process to return to step one to re-establish a valid diagnostic window sequence.

[0151] In the interleaved embodiment, after two derating retests, the diagnostic controller still obtains the same master conclusion and receives maintenance scheduling registration feedback. The diagnostic controller switches the process exit condition to maintenance waiting state and stops active perturbation. The visible result on site is that the laser continues to run at the derating level and waits for maintenance personnel to arrive to replace parts or clean optical components. After maintenance is completed, the diagnostic controller re-enters the power-on self-test window to establish a maintenance post-reconstruction baseline and remove the freeze flag.

[0152] When in use, the exit condition provides a convergence boundary for the diagnostic link, preventing infinite retesting from occupying the window; the baseline reconstruction after maintenance is included in the recovery path of the exit condition, so that the status changes after maintenance can be identified by the process and return to the healthy operating range.

[0153] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0154] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0155] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0156] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included 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 defect identification in an optically pumped laser, characterized in that: include, When the power-on self-test window, idle frame window, load switching window, or allowed derating window is in a state where the safety interlock is not in a high-risk prohibited state, determine the diagnostic window identifier and determine the admission conditions. When the admission conditions are met, a controlled perturbation is issued, and the driving current, driving voltage, monitoring photodiode signal, output power, temperature signal and return light signal are sampled synchronously before, during and after the perturbation to obtain a multi-channel synchronous response dataset. Based on the dataset, a multi-condition response tensor is constructed and relational features are generated. The relational features are then processed by a health baseline model to obtain drift-normalized residuals. The main conclusion, secondary conclusion, risk level and confidence level are obtained by matching the drift normalized residual with the anomaly template. Based on the risk level, the action instructions are generated. The action instructions include at least one of freezing the in-service health baseline, triggering the devaluation retest, maintenance scheduling or shutdown, and are fed back to the diagnostic window scheduling and health baseline model.

2. The defect identification method for an optically pumped laser according to claim 1, characterized in that: The diagnostic window identifier corresponds to one of the following: power-on self-test window, idle frame window, load switching window, and allow derating window. After receiving the diagnostic window identifier, the diagnostic controller first reads the safety interlock status and then determines whether the current window enters the active diagnostic branch, passive sampling branch, or bypass protection branch.

3. The defect identification method for an optically pumped laser according to claim 2, characterized in that: The criteria for determining the admission conditions include output stability margin, thermal margin, backlight risk margin, and controller saturation margin. The diagnostic controller forms the admission results of the diagnostic window based on each margin. When the admission results do not meet the active perturbation conditions, controlled perturbations are prohibited from being issued, and only passive sampling under natural operating conditions is retained.

4. The defect identification method for an optically pumped laser according to claim 3, characterized in that: The controlled perturbation is sent from the diagnostic controller to one of the pump driver, temperature controller and timing controller, and is limited to a small, short-term, rollback-able setpoint change; synchronous sampling covers the baseline segment before the perturbation, the perturbation action segment and the recovery segment, and each channel uses a unified timestamp to write to a unified buffer.

5. The defect identification method for an optically pumped laser according to claim 4, characterized in that: In addition to driving current, driving voltage, monitoring photodiode signal, output power, temperature signal, and reflected light signal, the multi-channel synchronous response dataset also includes the adjustment and error terms of the power stabilization controller. The diagnostic controller caches the multi-channel synchronous response dataset according to the operating condition dimension, channel dimension, and time dimension, and writes the sampling quality identifier.

6. The defect identification method for an optically pumped laser according to claim 5, characterized in that: The process of constructing a multi-condition response tensor and generating relational features includes time alignment, dimensional normalization, and sampling quality screening of the multi-channel synchronous response dataset, and extracting at least two types of relational features from the slope, hysteresis, recovery time, channel ratio, phase difference, and coupling relationship between the controller compensation amount and the main output.

7. The defect identification method for an optically pumped laser according to claim 6, characterized in that: The health baseline model includes at least one of the factory health baseline, in-service health baseline, and post-maintenance reconstruction baseline. The diagnostic controller selects the health baseline model according to the current model status and projects relational features into the health baseline space to filter out normal thermal drift, load drift, and slight alignment changes, and generate drift normalized residuals.

8. The defect identification method for an optically pumped laser according to claim 7, characterized in that: The abnormal templates include pump source abnormal templates, gain medium thermal abnormal templates, optical element abnormal templates, and return light path abnormal templates. The diagnostic controller performs matching scoring on each abnormal template based on the drift normalized residual, and outputs the main conclusion and secondary conclusion according to the scoring results. When the matching score is insufficient to distinguish between the main and secondary conclusions, an uncertain state is output.

9. The defect identification method for an optically pumped laser according to claim 8, characterized in that: When the risk level is low and the confidence level meets the preset conditions, slow updates to the in-service health baseline are allowed; when the risk level is medium, the in-service health baseline is frozen and a protected devaluation retest is initiated; when the risk level is high and the devaluation retest still fails, a maintenance schedule is output or the system is shut down, and consistency checks are performed on multiple test results.

10. The defect identification method for an optically pumped laser according to claim 9, characterized in that: The handling instructions further include: when the risk level is low and the attribution is stable, feeding back the current response characteristics to the health baseline model at a limited rate; when the risk level is medium, feeding back the results of this round to the diagnostic window scheduling logic to determine the next retest window; when the risk level is high, prohibiting further active perturbation and locking the protection priority; and when maintenance is completed, returning to step one to re-establish an effective baseline.

Images